初始化项目,由ModelHub XC社区提供模型

Model: W-61/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-05-29 01:10:42 +08:00
commit 181eecd0b9
501 changed files with 166909 additions and 0 deletions

36
.gitattributes vendored Normal file
View File

@@ -0,0 +1,36 @@
*.7z filter=lfs diff=lfs merge=lfs -text
*.arrow filter=lfs diff=lfs merge=lfs -text
*.bin filter=lfs diff=lfs merge=lfs -text
*.bz2 filter=lfs diff=lfs merge=lfs -text
*.ckpt filter=lfs diff=lfs merge=lfs -text
*.ftz filter=lfs diff=lfs merge=lfs -text
*.gz filter=lfs diff=lfs merge=lfs -text
*.h5 filter=lfs diff=lfs merge=lfs -text
*.joblib filter=lfs diff=lfs merge=lfs -text
*.lfs.* filter=lfs diff=lfs merge=lfs -text
*.mlmodel filter=lfs diff=lfs merge=lfs -text
*.model filter=lfs diff=lfs merge=lfs -text
*.msgpack filter=lfs diff=lfs merge=lfs -text
*.npy filter=lfs diff=lfs merge=lfs -text
*.npz filter=lfs diff=lfs merge=lfs -text
*.onnx filter=lfs diff=lfs merge=lfs -text
*.ot filter=lfs diff=lfs merge=lfs -text
*.parquet filter=lfs diff=lfs merge=lfs -text
*.pb filter=lfs diff=lfs merge=lfs -text
*.pickle filter=lfs diff=lfs merge=lfs -text
*.pkl filter=lfs diff=lfs merge=lfs -text
*.pt filter=lfs diff=lfs merge=lfs -text
*.pth filter=lfs diff=lfs merge=lfs -text
*.rar filter=lfs diff=lfs merge=lfs -text
*.safetensors filter=lfs diff=lfs merge=lfs -text
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
*.tar.* filter=lfs diff=lfs merge=lfs -text
*.tar filter=lfs diff=lfs merge=lfs -text
*.tflite filter=lfs diff=lfs merge=lfs -text
*.tgz filter=lfs diff=lfs merge=lfs -text
*.wasm filter=lfs diff=lfs merge=lfs -text
*.xz filter=lfs diff=lfs merge=lfs -text
*.zip filter=lfs diff=lfs merge=lfs -text
*.zst filter=lfs diff=lfs merge=lfs -text
*tfevents* filter=lfs diff=lfs merge=lfs -text
tokenizer.json filter=lfs diff=lfs merge=lfs -text

81
README.md Normal file
View File

@@ -0,0 +1,81 @@
---
library_name: transformers
base_model: jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128
tags:
- alignment-handbook
- new-dpo
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517
This model is a fine-tuned version of [jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128](https://huggingface.co/jackf857/qwen3-8b-base-sft-ultrachat-4xh200-batch-128) on the HuggingFaceH4/ultrafeedback_binarized dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6076
- Fcm Dpo/beta: 0.0040
- Margin Dpo/margin Mean: 54.4214
- Margin Dpo/margin Std: 86.7093
- Logps/chosen: -331.5330
- Logps/rejected: -370.1833
- Logps/ref Chosen: -280.4167
- Logps/ref Rejected: -264.6455
- Kl/chosen Kl Mean: -51.1163
- Kl/rejected Kl Mean: -105.5378
- Kl/mean: -78.3270
- Kl/std: 74.4785
- Logits/chosen: 1.4424
- Logits/rejected: 1.5060
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-07
- train_batch_size: 4
- eval_batch_size: 2
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 8
- total_train_batch_size: 128
- total_eval_batch_size: 8
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1
### Training results
| Training Loss | Epoch | Step | Validation Loss | Fcm Dpo/beta | Margin Dpo/margin Mean | Margin Dpo/margin Std | Logps/chosen | Logps/rejected | Logps/ref Chosen | Logps/ref Rejected | Kl/chosen Kl Mean | Kl/rejected Kl Mean | Kl/mean | Kl/std | Logits/chosen | Logits/rejected |
|:-------------:|:------:|:----:|:---------------:|:------------:|:----------------------:|:---------------------:|:------------:|:--------------:|:----------------:|:------------------:|:-----------------:|:-------------------:|:--------:|:-------:|:-------------:|:---------------:|
| 4.9448 | 0.4188 | 200 | 0.6027 | 0.0090 | 26.4387 | 42.3830 | -286.1130 | -296.7805 | -280.4167 | -264.6455 | -5.6962 | -32.1350 | -18.9156 | 36.5734 | 1.6826 | 1.7371 |
| 4.7852 | 0.8377 | 400 | 0.6076 | 0.0040 | 54.4214 | 86.7093 | -331.5330 | -370.1833 | -280.4167 | -264.6455 | -51.1163 | -105.5378 | -78.3270 | 74.4785 | 1.4424 | 1.5060 |
### Framework versions
- Transformers 4.51.0
- Pytorch 2.3.1+cu121
- Datasets 2.21.0
- Tokenizers 0.21.4

28
added_tokens.json Normal file
View File

@@ -0,0 +1,28 @@
{
"</think>": 151668,
"</tool_call>": 151658,
"</tool_response>": 151666,
"<think>": 151667,
"<tool_call>": 151657,
"<tool_response>": 151665,
"<|box_end|>": 151649,
"<|box_start|>": 151648,
"<|endoftext|>": 151643,
"<|file_sep|>": 151664,
"<|fim_middle|>": 151660,
"<|fim_pad|>": 151662,
"<|fim_prefix|>": 151659,
"<|fim_suffix|>": 151661,
"<|im_end|>": 151645,
"<|im_start|>": 151644,
"<|image_pad|>": 151655,
"<|object_ref_end|>": 151647,
"<|object_ref_start|>": 151646,
"<|quad_end|>": 151651,
"<|quad_start|>": 151650,
"<|repo_name|>": 151663,
"<|video_pad|>": 151656,
"<|vision_end|>": 151653,
"<|vision_pad|>": 151654,
"<|vision_start|>": 151652
}

27
all_results.json Normal file
View File

@@ -0,0 +1,27 @@
{
"epoch": 0.9989528795811519,
"eval_KL/chosen_KL_mean": -48.39694595336914,
"eval_KL/mean": -76.14442443847656,
"eval_KL/rejected_KL_mean": -103.89189147949219,
"eval_KL/std": 75.29983520507812,
"eval_fcm_dpo/beta": 0.003546761581674218,
"eval_logits/chosen": 1.4365565776824951,
"eval_logits/rejected": 1.5009914636611938,
"eval_logps/chosen": -328.81365966796875,
"eval_logps/ref_chosen": -280.41668701171875,
"eval_logps/ref_rejected": -264.6455383300781,
"eval_logps/rejected": -368.53741455078125,
"eval_loss": 0.6136210560798645,
"eval_margin_dpo/margin_mean": 55.49495315551758,
"eval_margin_dpo/margin_std": 87.28593444824219,
"eval_runtime": 88.1538,
"eval_samples": 2000,
"eval_samples_per_second": 22.688,
"eval_steps_per_second": 2.836,
"total_flos": 0.0,
"train_loss": 4.986182768639779,
"train_runtime": 7647.9571,
"train_samples": 61135,
"train_samples_per_second": 7.994,
"train_steps_per_second": 0.062
}

30
config.json Normal file
View File

@@ -0,0 +1,30 @@
{
"architectures": [
"Qwen3ForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"bos_token_id": 151643,
"eos_token_id": 151643,
"head_dim": 128,
"hidden_act": "silu",
"hidden_size": 4096,
"initializer_range": 0.02,
"intermediate_size": 12288,
"max_position_embeddings": 32768,
"max_window_layers": 36,
"model_type": "qwen3",
"num_attention_heads": 32,
"num_hidden_layers": 36,
"num_key_value_heads": 8,
"rms_norm_eps": 1e-06,
"rope_scaling": null,
"rope_theta": 1000000,
"sliding_window": null,
"tie_word_embeddings": false,
"torch_dtype": "float32",
"transformers_version": "4.51.0",
"use_cache": true,
"use_sliding_window": false,
"vocab_size": 151936
}

21
eval_results.json Normal file
View File

@@ -0,0 +1,21 @@
{
"epoch": 0.9989528795811519,
"eval_KL/chosen_KL_mean": -48.39694595336914,
"eval_KL/mean": -76.14442443847656,
"eval_KL/rejected_KL_mean": -103.89189147949219,
"eval_KL/std": 75.29983520507812,
"eval_fcm_dpo/beta": 0.003546761581674218,
"eval_logits/chosen": 1.4365565776824951,
"eval_logits/rejected": 1.5009914636611938,
"eval_logps/chosen": -328.81365966796875,
"eval_logps/ref_chosen": -280.41668701171875,
"eval_logps/ref_rejected": -264.6455383300781,
"eval_logps/rejected": -368.53741455078125,
"eval_loss": 0.6136210560798645,
"eval_margin_dpo/margin_mean": 55.49495315551758,
"eval_margin_dpo/margin_std": 87.28593444824219,
"eval_runtime": 88.1538,
"eval_samples": 2000,
"eval_samples_per_second": 22.688,
"eval_steps_per_second": 2.836
}

6
generation_config.json Normal file
View File

@@ -0,0 +1,6 @@
{
"bos_token_id": 151643,
"eos_token_id": 151643,
"max_new_tokens": 2048,
"transformers_version": "4.51.0"
}

477
margin_logs/margins.jsonl Normal file
View File

@@ -0,0 +1,477 @@
{"epoch": 0.0, "step": 1, "batch_size": 128, "mean": 0.01704716682434082, "std": 0.7752009630203247, "min": -2.2227783203125, "p10": -0.8393455505371094, "median": 0.0678863525390625, "p90": 0.8599296569824219, "max": 2.196014404296875, "pos_frac": 0.5546875, "sample": [0.6739997863769531, -0.3828582763671875, -1.298828125, 0.071014404296875, 0.8088531494140625, 0.6873779296875, -0.081512451171875, 1.77093505859375, -1.4920806884765625, 1.769927978515625, 0.0503692626953125, -0.1540679931640625, 0.15589141845703125, -0.27809906005859375, -0.009723663330078125, 0.3800621032714844, 0.111541748046875, 0.2325439453125, 1.35693359375, -0.482269287109375, 0.2686614990234375, 1.01312255859375, -0.5781745910644531, -0.1348876953125, 0.15979766845703125, -1.054656982421875, 0.01500701904296875, 0.2495269775390625, 0.5402507781982422, 1.80511474609375, -0.8189926147460938, -0.17108154296875, 0.2899322509765625, -0.8205108642578125, 0.650146484375, -0.6688079833984375, 0.080841064453125, -1.4810791015625, 0.07396697998046875, -1.242919921875, 0.086151123046875, 1.62188720703125, -0.38132476806640625, 0.317474365234375, 2.196014404296875, 0.5562095642089844, 0.1424732208251953, -0.7183074951171875, -0.6070556640625, -0.6271820068359375, -0.5325164794921875, 0.67987060546875, -1.229156494140625, -1.91693115234375, -0.143280029296875, 0.2532958984375, -0.349273681640625, 0.42700958251953125, -0.262420654296875, 1.2039794921875, 0.364410400390625, 0.8435211181640625, 0.00079345703125, -0.19837188720703125, -0.5950164794921875, 0.362579345703125, 0.7775421142578125, -0.015625, -0.0785369873046875, -0.6810150146484375, 0.31806182861328125, -0.27123260498046875, 0.00982666015625, 0.098541259765625, -1.4693603515625, 0.20909881591796875, 0.87872314453125, 0.0, 0.4891510009765625, 0.100677490234375, -0.3092498779296875, 0.839263916015625, -0.5568389892578125, -0.5927276611328125, 1.82940673828125, -1.8787841796875, 0.4121856689453125, 0.1029205322265625, -0.05619049072265625, 0.41302490234375, -0.20123291015625, -1.11163330078125, 0.11328125, -0.6293182373046875, -0.4356689453125, 1.026641845703125, -0.4438629150390625, 0.06475830078125, 0.15139389038085938, 0.4244384765625, -0.3410911560058594, -0.8832931518554688, 0.38405609130859375, -0.6134033203125, 0.14745521545410156, -2.2227783203125, -0.3340606689453125, -0.4396820068359375, 0.543609619140625, 0.310455322265625, -0.6277847290039062, 0.28015899658203125, 0.1476898193359375, 0.3895263671875, 0.0498046875, -0.1731414794921875, 0.24071502685546875, -0.351348876953125, -0.0234375, 0.6951370239257812, 0.0606689453125, 0.8756256103515625, 0.680145263671875, 0.14440345764160156, 0.8532028198242188, -1.6363525390625, 1.4149551391601562, -0.476959228515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000001.npy"}
{"epoch": 0.0020942408376963353, "step": 2, "batch_size": 128, "mean": 0.08979731798171997, "std": 1.0063953399658203, "min": -2.3612060546875, "p10": -1.1013275146484374, "median": 0.015979766845703125, "p90": 1.2868972778320311, "max": 4.49609375, "pos_frac": 0.53125, "sample": [0.006134033203125, 0.6754608154296875, 0.06409072875976562, 0.3348197937011719, 0.104522705078125, 0.848785400390625, -1.3365631103515625, 0.40827178955078125, 1.093963623046875, -0.233673095703125, 0.2210693359375, 0.34949493408203125, 1.295623779296875, -0.01953887939453125, -0.0232696533203125, -0.88238525390625, 1.92608642578125, -0.0975494384765625, -0.0151214599609375, -0.137054443359375, 0.18506240844726562, -0.7067947387695312, -1.04547119140625, 0.01041412353515625, 0.3020477294921875, -0.03971099853515625, -0.283477783203125, 0.6910400390625, 0.15057373046875, -0.1564178466796875, 0.561614990234375, 0.05670166015625, -0.48236846923828125, 1.7666015625, -0.0575408935546875, -1.0196533203125, -2.3612060546875, 2.6371917724609375, -0.6959686279296875, 0.836669921875, -0.16046142578125, -0.033050537109375, 0.0088958740234375, 1.2831573486328125, 1.604461669921875, -0.9659347534179688, 0.1966876983642578, -1.5263671875, -0.4609222412109375, -0.87158203125, 4.49609375, -0.16178131103515625, -1.54376220703125, 0.0208740234375, 0.0, 0.33994483947753906, 0.185943603515625, -0.30196380615234375, 3.5068359375, 1.463623046875, 0.9294891357421875, -0.02225494384765625, -1.140228271484375, 0.1556549072265625, -0.2645263671875, 0.90374755859375, 0.5412216186523438, -1.59906005859375, -0.3666229248046875, -0.4569854736328125, -0.6967010498046875, 0.36395263671875, 0.13336181640625, -0.068084716796875, -0.563751220703125, 0.5245513916015625, 0.082733154296875, -0.73480224609375, 0.570953369140625, 0.2091522216796875, -0.021331787109375, -0.547821044921875, 1.6211090087890625, 0.8912696838378906, 0.40868377685546875, 2.679595947265625, 0.1652679443359375, -0.8451919555664062, -0.22149658203125, -0.6046524047851562, -0.001617431640625, 0.72979736328125, -1.3320388793945312, -0.18963623046875, 1.436309814453125, 0.23449325561523438, 0.4606475830078125, 0.095184326171875, 0.49684906005859375, -0.4890289306640625, 0.9012298583984375, 1.9844970703125, 1.4336700439453125, -0.7718658447265625, 0.1373443603515625, -1.352508544921875, -1.7950439453125, 0.286163330078125, -0.2695465087890625, 0.72283935546875, 0.463165283203125, -1.27398681640625, 0.45489501953125, 0.01837158203125, 0.01849365234375, 0.36487579345703125, -1.4290771484375, 0.8643035888671875, -0.195465087890625, 0.240478515625, -0.7327880859375, -1.255859375, -0.87091064453125, -1.08465576171875, 1.1551513671875, 0.01358795166015625, -0.0644378662109375, -1.950225830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000002.npy"}
{"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.0667317658662796, "std": 0.8198645710945129, "min": -2.37591552734375, "p10": -1.011199951171875, "median": 0.07687187194824219, "p90": 1.0807968139648436, "max": 1.989715576171875, "pos_frac": 0.53125, "sample": [0.494049072265625, -0.019641876220703125, -0.1837158203125, -0.036182403564453125, 0.2940826416015625, -0.10797119140625, 1.058013916015625, 0.317413330078125, 0.0, -0.525604248046875, 0.058837890625, -1.10650634765625, -0.715667724609375, 0.515716552734375, -0.42236328125, 0.208099365234375, -0.6510009765625, 1.83404541015625, 1.633148193359375, 0.35430908203125, -0.6342010498046875, 1.37811279296875, 0.70526123046875, 0.5212478637695312, -0.095550537109375, -0.12957763671875, 1.12872314453125, -1.5146484375, 0.79443359375, -0.516143798828125, 0.0703582763671875, 0.08338546752929688, -0.3818359375, 1.21051025390625, -0.39892578125, -0.33422088623046875, -0.7536468505859375, 0.03814697265625, -0.449554443359375, 1.989715576171875, -0.37432098388671875, 0.513671875, 1.252197265625, 1.3010101318359375, 0.5347900390625, -2.37591552734375, -0.6508636474609375, 0.431976318359375, 1.033905029296875, -1.110015869140625, 0.91229248046875, 0.22867965698242188, 0.3297119140625, 0.29500389099121094, 0.981536865234375, -0.116851806640625, 0.42346954345703125, 1.3902587890625, 0.30239105224609375, 0.3618659973144531, 0.0, -0.757965087890625, 1.389678955078125, -2.263885498046875, -0.6443252563476562, 0.5632781982421875, -1.090972900390625, -0.211456298828125, 0.09748458862304688, -0.050197601318359375, -0.8822021484375, 0.159149169921875, 0.914093017578125, -1.847747802734375, -0.08292388916015625, 0.25390625, 0.5157012939453125, -1.15411376953125, 0.2237548828125, -0.03350830078125, -0.0377655029296875, 1.0077972412109375, 0.9146995544433594, -0.13261032104492188, -0.088043212890625, 1.396484375, 0.37322998046875, 0.4630126953125, -0.16668701171875, 0.3434295654296875, 0.1820831298828125, -0.480743408203125, -0.722503662109375, -1.58819580078125, -0.009571075439453125, 0.2666473388671875, 0.12740325927734375, 0.37413787841796875, 0.6192054748535156, 0.09513473510742188, -0.6026611328125, 1.6822509765625, -0.178009033203125, 0.897857666015625, -0.007110595703125, 0.4471435546875, -0.6574249267578125, 0.77264404296875, -1.02081298828125, -0.3084754943847656, -1.41534423828125, -0.340576171875, 1.0602569580078125, 0.2105712890625, 0.61700439453125, 0.3296661376953125, 0.3885040283203125, -1.812347412109375, -0.389678955078125, -0.111053466796875, 0.25726318359375, -0.19788742065429688, 0.0230560302734375, -0.2529335021972656, -1.4273605346679688, 1.735687255859375, 0.43817901611328125, -1.007080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000003.npy"}
{"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": 0.042136386036872864, "std": 0.9621411561965942, "min": -2.52789306640625, "p10": -1.1272552490234373, "median": 0.045574188232421875, "p90": 1.2091110229492188, "max": 2.509307861328125, "pos_frac": 0.5390625, "sample": [-0.2943572998046875, 0.6842422485351562, 0.12587738037109375, 2.01702880859375, 0.232421875, -0.49796295166015625, 0.7163887023925781, 0.8746337890625, -1.076141357421875, 1.0120849609375, -1.557159423828125, -0.60003662109375, 0.005035400390625, -0.22509002685546875, 0.2987785339355469, 1.411285400390625, 1.6473846435546875, -2.09466552734375, -0.3594970703125, -0.0919189453125, -1.0723876953125, 0.5079345703125, 0.753387451171875, 2.1998291015625, -0.804046630859375, 1.2999725341796875, 2.509307861328125, 0.12498855590820312, -1.07470703125, 0.28737640380859375, -0.891510009765625, -0.17436599731445312, 0.604583740234375, -0.3134765625, 0.384735107421875, 1.1009521484375, -0.1355133056640625, -0.706329345703125, -0.4128265380859375, 0.334808349609375, -0.1261119842529297, 0.03277587890625, -0.6450576782226562, 0.416229248046875, -1.24652099609375, 0.4497528076171875, -1.4095916748046875, -0.5638504028320312, 0.882904052734375, -1.359710693359375, -0.191650390625, -0.2960357666015625, -0.63006591796875, -2.469512939453125, 0.037200927734375, -0.5251007080078125, -1.4183197021484375, 0.05316162109375, 0.15692138671875, -0.6307373046875, -1.432281494140625, -1.3563232421875, -0.22599029541015625, -0.957183837890625, 0.8846435546875, 0.016204833984375, 0.5868682861328125, 0.2009124755859375, 0.31815338134765625, -1.02410888671875, 1.2112579345703125, 0.169586181640625, 0.47393798828125, -1.0228271484375, 1.6881256103515625, -0.23486328125, -2.52789306640625, -0.43682861328125, 0.458740234375, 0.08670806884765625, 0.7168731689453125, 0.27816009521484375, -0.579681396484375, -1.038330078125, -0.09539794921875, 1.8663330078125, 1.0874786376953125, 0.375457763671875, 1.810943603515625, -0.27203369140625, -0.3956794738769531, 0.2891845703125, -0.8986892700195312, 0.5788421630859375, 0.865753173828125, -0.8009033203125, 0.8359375, -0.0541839599609375, -1.0188140869140625, -0.37567138671875, -1.61138916015625, -0.0212860107421875, -0.06554031372070312, 0.8197021484375, 0.116851806640625, -2.205078125, 0.04232025146484375, 0.116912841796875, -0.1793212890625, 2.14434814453125, 1.2686080932617188, 0.048828125, 0.33582305908203125, 1.20819091796875, 0.8128204345703125, 0.11566162109375, 1.08795166015625, 0.994171142578125, 0.782928466796875, -1.3458175659179688, -0.4736328125, 0.40960693359375, 1.792572021484375, 0.8863525390625, -0.3502197265625, 0.967193603515625, -0.07421875, 0.448974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000004.npy"}
{"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": 0.00983293354511261, "std": 0.9495989680290222, "min": -2.468414306640625, "p10": -1.0723976135253905, "median": 0.029529571533203125, "p90": 1.0070510864257811, "max": 4.458526611328125, "pos_frac": 0.5078125, "sample": [-0.5475616455078125, -0.5797119140625, 0.09175872802734375, -0.75665283203125, 0.4240875244140625, 1.0507049560546875, -0.814666748046875, 0.2365570068359375, 0.8741302490234375, -0.146636962890625, -0.7794189453125, 0.519073486328125, -0.96246337890625, -1.0806808471679688, -0.29740142822265625, 1.51287841796875, 0.42258453369140625, 0.545440673828125, 1.298004150390625, -2.468414306640625, -0.8271484375, 1.10137939453125, -0.022918701171875, 1.326416015625, -0.3245849609375, -0.917724609375, -0.1139068603515625, 0.4390907287597656, -0.18253326416015625, -0.17956924438476562, -0.68048095703125, -0.8125228881835938, 0.832916259765625, 0.0696868896484375, 0.6168975830078125, 0.2363433837890625, 0.052001953125, 0.0868988037109375, -2.102874755859375, 3.010009765625, 0.484893798828125, 0.902374267578125, -1.26470947265625, 0.534027099609375, 0.4862060546875, -0.051471710205078125, -1.015655517578125, -0.163299560546875, -0.2119007110595703, -1.06884765625, -0.703216552734375, -0.736053466796875, -1.30694580078125, -0.3852367401123047, 0.26950836181640625, -0.274993896484375, 0.8890380859375, -0.505096435546875, 0.3074188232421875, 1.2583770751953125, -1.85528564453125, 0.150299072265625, -0.4226531982421875, -0.500030517578125, -0.6968917846679688, -0.21728134155273438, 0.348114013671875, -0.02239990234375, -0.15185546875, 1.61138916015625, 0.845001220703125, 0.2445220947265625, 0.2384490966796875, -0.059295654296875, 0.10887908935546875, -0.109161376953125, -1.477783203125, 0.51519775390625, -0.286346435546875, 1.3054580688476562, -0.03497314453125, 0.96160888671875, 0.519561767578125, -0.070953369140625, 0.474517822265625, -0.51202392578125, -0.17327880859375, -1.950714111328125, 0.89947509765625, 0.5941162109375, -2.06390380859375, 0.09694099426269531, -1.74835205078125, 2.1485595703125, 0.818328857421875, -1.53271484375, -1.43280029296875, -0.06111907958984375, 0.0902862548828125, -0.18585205078125, -0.77008056640625, -1.164825439453125, 0.32938385009765625, 0.5472412109375, 0.1233367919921875, -0.649658203125, -0.3106689453125, -0.3264732360839844, 0.9523696899414062, -0.22259521484375, 0.3861083984375, -0.41405487060546875, 0.01988983154296875, 0.862945556640625, 1.116668701171875, 0.07600784301757812, 0.0959014892578125, 0.98834228515625, 0.0391693115234375, 1.226226806640625, 0.497039794921875, -0.521697998046875, 0.080413818359375, 0.07312774658203125, -0.81402587890625, 0.3363037109375, 0.2452545166015625, 4.458526611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000005.npy"}
{"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": -0.018405884504318237, "std": 0.9675299525260925, "min": -2.556396484375, "p10": -1.2205535888671875, "median": 0.0014295578002929688, "p90": 0.9692474365234374, "max": 2.95928955078125, "pos_frac": 0.5078125, "sample": [-1.1060333251953125, -1.601226806640625, -0.565277099609375, 1.806640625, 0.850616455078125, 0.7653961181640625, 1.024261474609375, 0.2397003173828125, 0.25060272216796875, -0.581451416015625, -0.0456085205078125, 1.00579833984375, 0.0749053955078125, 0.53082275390625, -1.5027008056640625, -0.30316162109375, -1.828857421875, 0.0012416839599609375, -0.3340320587158203, 0.511993408203125, -0.991485595703125, 0.9490089416503906, 0.474639892578125, 0.953582763671875, -0.14227294921875, -0.5408782958984375, 0.008056640625, 0.6954498291015625, 0.603668212890625, 0.18910789489746094, -0.5828933715820312, -0.4908447265625, 0.7093505859375, -0.694183349609375, -0.1186065673828125, 0.01629638671875, 0.1624755859375, 0.10076904296875, -0.2991790771484375, 2.484130859375, -0.5513763427734375, -0.868682861328125, -1.205291748046875, 0.9322509765625, 0.026483535766601562, -2.1195068359375, 0.3182964324951172, 1.41748046875, 0.5207443237304688, -1.00140380859375, -0.048187255859375, -0.8268280029296875, 1.083160400390625, -0.66290283203125, -0.592529296875, 0.22510147094726562, -0.62286376953125, -1.153717041015625, -0.25482177734375, 0.197021484375, 0.4862327575683594, 0.202484130859375, 0.8061065673828125, 0.388458251953125, -2.003875732421875, -0.220123291015625, 0.28261375427246094, 1.838134765625, -1.7696533203125, 0.6485595703125, 0.5284881591796875, 0.60528564453125, 2.95928955078125, -0.1645355224609375, 0.5564651489257812, -0.2323455810546875, 0.39434814453125, -0.13006591796875, 0.0, 0.001617431640625, 1.0561599731445312, -0.937774658203125, -0.4799957275390625, -0.0426483154296875, -0.1396636962890625, -0.402313232421875, -0.2017822265625, -0.0656585693359375, 0.25690460205078125, 0.1124420166015625, 0.481414794921875, 0.619232177734375, 0.6452674865722656, -1.488739013671875, 0.85736083984375, -0.73455810546875, -0.6961822509765625, 0.5253143310546875, 0.351165771484375, -1.66644287109375, 1.6624755859375, -1.40899658203125, -1.7152862548828125, -1.171722412109375, -0.27826690673828125, 0.4194793701171875, -0.716796875, -2.556396484375, 0.3870391845703125, -0.4770927429199219, -1.1553955078125, 0.08781623840332031, 2.686431884765625, 0.395172119140625, 0.677093505859375, 2.7259521484375, -0.453857421875, 0.777313232421875, -1.171661376953125, 0.695281982421875, -0.6720905303955078, -1.4932861328125, -0.01275634765625, -0.29022216796875, 1.100921630859375, -1.25616455078125, 0.168914794921875, -0.02909088134765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000006.npy"}
{"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": 0.015746206045150757, "std": 1.0094897747039795, "min": -2.59912109375, "p10": -1.2700958251953125, "median": -0.022674560546875, "p90": 1.3307464599609373, "max": 3.13507080078125, "pos_frac": 0.46875, "sample": [1.871856689453125, 2.245880126953125, -1.266387939453125, 0.362945556640625, -0.2275238037109375, 0.6087493896484375, 1.26654052734375, -0.7498779296875, -1.415924072265625, -0.5316390991210938, 0.45416259765625, 0.26727294921875, -1.900115966796875, 0.01201629638671875, -0.3265380859375, -0.387451171875, 0.7432403564453125, 0.7966384887695312, 0.0, 0.17893218994140625, 0.9646759033203125, -0.279754638671875, -0.45162200927734375, -0.3195648193359375, -0.9632568359375, -0.7396240234375, -1.097015380859375, -0.79656982421875, 0.712646484375, 0.591461181640625, -0.4272003173828125, 0.04559326171875, -0.122528076171875, -0.0165863037109375, 0.373748779296875, 3.13507080078125, 0.01032257080078125, 0.4718017578125, -1.9669189453125, -0.18958282470703125, -1.474609375, 0.27072906494140625, -0.56011962890625, 1.35003662109375, -1.786224365234375, 1.322479248046875, 2.10968017578125, 0.12020492553710938, 0.0153961181640625, 1.809814453125, 1.293975830078125, -0.40936279296875, 1.20208740234375, -0.4979400634765625, -0.50396728515625, 0.304534912109375, -1.2907257080078125, -0.020538330078125, -0.8208389282226562, 1.1915740966796875, -0.8015708923339844, 1.896270751953125, 0.43328094482421875, -1.27874755859375, -1.087371826171875, 1.516204833984375, -1.317962646484375, 0.3740806579589844, 1.2669677734375, -0.8121337890625, -0.046905517578125, 1.5745849609375, -0.23829078674316406, 0.2076263427734375, 0.1872100830078125, -0.2606201171875, 0.8158111572265625, -1.638031005859375, -2.59912109375, -1.093505859375, 0.06451416015625, 0.002223968505859375, 0.6514739990234375, 1.2131423950195312, -0.3211669921875, -0.5219345092773438, -1.911865234375, 0.078826904296875, -0.196380615234375, -0.373687744140625, -0.4434814453125, -0.1589508056640625, -0.020751953125, -0.4654655456542969, -0.7470550537109375, -0.53662109375, -0.1253814697265625, -0.35610008239746094, 0.34497833251953125, -0.4214630126953125, -0.620574951171875, -0.5539398193359375, -0.906219482421875, 0.2283935546875, 0.133880615234375, 0.136688232421875, 2.41925048828125, -0.02459716796875, -1.44256591796875, -0.79736328125, 0.6436920166015625, -0.5819168090820312, 0.21923828125, 1.2929229736328125, -0.148681640625, 0.208404541015625, -1.099395751953125, -0.2518768310546875, 1.0655517578125, 2.43231201171875, 0.9404144287109375, 0.1248779296875, 0.12816619873046875, -0.38416481018066406, -0.1602611541748047, -1.62371826171875, 1.673675537109375, 1.546630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000007.npy"}
{"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": -0.10379450023174286, "std": 1.0655173063278198, "min": -4.12451171875, "p10": -1.4086135864257812, "median": -0.020977020263671875, "p90": 1.1864440917968748, "max": 2.79180908203125, "pos_frac": 0.4921875, "sample": [-0.9260101318359375, 0.0054473876953125, 0.163818359375, -1.037506103515625, -0.272705078125, -0.86419677734375, 1.39739990234375, 0.055133819580078125, 0.03216361999511719, -2.134521484375, -0.059268951416015625, -1.584259033203125, 1.5955810546875, 0.699432373046875, -0.1905517578125, -0.40369415283203125, -0.8835296630859375, 0.3214111328125, 0.11736297607421875, 1.764190673828125, 0.92193603515625, 0.6802978515625, -2.5140838623046875, 2.0098876953125, -0.12542724609375, 0.35491943359375, -0.47918701171875, -1.482513427734375, -3.62969970703125, -2.00433349609375, -0.667083740234375, 2.79180908203125, 1.3480224609375, -0.7623291015625, -0.1045684814453125, -1.43548583984375, 0.8072967529296875, -0.2616424560546875, -0.66949462890625, 0.5747528076171875, -0.81805419921875, -1.41131591796875, -0.4215545654296875, -1.214447021484375, 0.186065673828125, -0.6103668212890625, 0.1595134735107422, -0.36411285400390625, -0.2138671875, 0.5369873046875, -1.276519775390625, -0.607757568359375, 0.04608154296875, -0.39635467529296875, -1.018524169921875, -1.4074554443359375, 0.830657958984375, 1.0057373046875, -1.47369384765625, -4.12451171875, 0.25732421875, -1.5750732421875, -1.5479736328125, 0.6497802734375, 0.43320465087890625, 2.08917236328125, -0.4517822265625, 0.3464202880859375, -0.12139892578125, 2.56097412109375, 0.178680419921875, -0.68670654296875, 0.694793701171875, 0.04803466796875, 0.96221923828125, 1.169708251953125, 1.99761962890625, 0.2847557067871094, 0.14743995666503906, -0.8816986083984375, -0.0105133056640625, 0.7772216796875, -0.928955078125, 0.05543327331542969, 0.683319091796875, 0.06990242004394531, -0.9126434326171875, 1.40380859375, 0.2207489013671875, 0.9365234375, 0.57379150390625, -0.90869140625, 0.596282958984375, 0.0603179931640625, -0.37223052978515625, -0.2290191650390625, 1.404296875, 1.50909423828125, 0.4187812805175781, -1.12481689453125, 0.3229827880859375, -0.322998046875, -0.100250244140625, 0.230255126953125, -0.055721282958984375, -0.07819366455078125, 0.5101165771484375, -0.03144073486328125, -0.9747314453125, -1.202117919921875, -0.3175926208496094, 0.269134521484375, 0.5880126953125, 0.3748779296875, -1.82940673828125, 0.213836669921875, 0.087921142578125, -1.011383056640625, 0.1749114990234375, 0.1959686279296875, -1.0567169189453125, -1.1568603515625, -0.2509765625, -0.49025726318359375, -0.28511810302734375, 0.4953460693359375, 1.225494384765625, -0.14421463012695312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000008.npy"}
{"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": 0.017461583018302917, "std": 0.8735716342926025, "min": -2.933380126953125, "p10": -1.051306915283203, "median": 0.044895172119140625, "p90": 1.1739105224609374, "max": 2.148284912109375, "pos_frac": 0.5234375, "sample": [0.06331634521484375, 0.5225830078125, 0.39032745361328125, 0.4341697692871094, -0.738861083984375, -0.448699951171875, -0.1020050048828125, 0.21437835693359375, 0.34100341796875, 0.00684356689453125, -0.478607177734375, -2.933380126953125, 0.26708984375, -0.565155029296875, -0.492431640625, -0.54083251953125, 1.186920166015625, -1.1725616455078125, 1.3762664794921875, -0.38165283203125, 0.007442474365234375, 2.148284912109375, -0.4937896728515625, 0.832763671875, -0.5029754638671875, 0.14669036865234375, -0.51214599609375, 0.634552001953125, 0.21051025390625, -1.7236328125, 0.2424163818359375, 2.1041259765625, -0.05103302001953125, -0.261383056640625, 0.201141357421875, -0.37033843994140625, 1.43218994140625, -0.358428955078125, 0.24526214599609375, -0.143829345703125, -0.889404296875, 0.302703857421875, 0.20672607421875, -0.1488494873046875, 1.1683349609375, 1.215728759765625, -0.90142822265625, -1.582611083984375, -2.401611328125, -0.59033203125, 1.7493896484375, -0.8846435546875, -0.37579345703125, 1.3033294677734375, 0.035125732421875, 0.4594764709472656, 1.0511474609375, -0.10528564453125, -0.27716064453125, -0.75946044921875, -0.1225128173828125, -0.40130615234375, -1.348388671875, -0.3310508728027344, -0.49676513671875, 1.030029296875, 0.05466461181640625, 1.865692138671875, -1.293212890625, 0.4514007568359375, -0.21026611328125, -0.0069828033447265625, -1.00927734375, 0.38892364501953125, 0.36370849609375, 0.8527374267578125, -1.32855224609375, 0.347137451171875, 0.11486434936523438, -0.08453369140625, 1.7540283203125, 0.14422607421875, 0.108551025390625, 0.2329559326171875, 0.9575042724609375, -0.987945556640625, 0.0, 0.439483642578125, 0.251129150390625, -1.03948974609375, -0.622802734375, 0.2751808166503906, 0.88128662109375, 0.23421478271484375, 1.589691162109375, 0.11572265625, -1.2679443359375, 0.057037353515625, -1.1081771850585938, 0.5916748046875, 0.13369369506835938, 0.0, 1.210540771484375, 1.497314453125, -0.56719970703125, 0.3048515319824219, 0.0, 0.6864013671875, -0.4470558166503906, 0.4307518005371094, 0.661651611328125, 0.5919647216796875, -0.754150390625, -0.080963134765625, -1.993499755859375, -0.0295257568359375, -0.4497528076171875, -0.5570068359375, 0.3197021484375, -0.344879150390625, -0.2831573486328125, 0.831298828125, 1.0715255737304688, 0.40679931640625, -1.0788803100585938, -1.60040283203125, 0.804412841796875, 0.716094970703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000009.npy"}
{"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": 0.019779890775680542, "std": 0.7800830602645874, "min": -2.363037109375, "p10": -1.02122802734375, "median": 0.05898284912109375, "p90": 0.9499160766601561, "max": 1.9639892578125, "pos_frac": 0.53125, "sample": [1.53497314453125, -0.0980072021484375, 0.17877197265625, -0.08722686767578125, 0.5590133666992188, -1.42462158203125, -0.17426300048828125, 0.5809173583984375, -0.08975982666015625, 0.08013916015625, -0.009765625, -0.3114738464355469, -0.45441436767578125, 0.146209716796875, 0.9991455078125, 0.1809234619140625, 0.3534889221191406, -1.3463134765625, -0.89093017578125, 0.36962890625, 0.294921875, -1.0090179443359375, -0.655364990234375, -0.404205322265625, 0.9288177490234375, -0.2652473449707031, 0.596160888671875, -0.45609283447265625, -1.583465576171875, 0.23004150390625, -0.18294525146484375, 0.65924072265625, -0.4278411865234375, 0.0301055908203125, 0.6162567138671875, -0.24725341796875, 0.644500732421875, 0.414825439453125, 1.614959716796875, 0.4569549560546875, -1.046478271484375, 0.5710678100585938, -0.27523231506347656, -1.010406494140625, 0.440399169921875, 0.822723388671875, -0.12414932250976562, 0.37329864501953125, 0.322021484375, -0.4402008056640625, -0.11945724487304688, -0.3142852783203125, 0.6336669921875, -1.1215972900390625, -0.839691162109375, 0.5241546630859375, -0.60235595703125, 1.5717010498046875, -0.22207260131835938, 0.430633544921875, -1.3087615966796875, -1.2349319458007812, 0.278228759765625, 0.11332321166992188, -0.25251007080078125, 0.67010498046875, 0.5800018310546875, -1.720306396484375, 0.04864501953125, 0.395751953125, -0.36272430419921875, 0.694915771484375, 1.9639892578125, 0.23602294921875, -0.7899551391601562, -0.47617340087890625, -0.1390228271484375, 0.880523681640625, -0.5416259765625, 1.50390625, -0.47475433349609375, 0.633575439453125, -0.44610595703125, 0.633148193359375, 0.43923377990722656, 1.333648681640625, -0.173583984375, -0.2518463134765625, 0.1756591796875, -0.118743896484375, 0.538482666015625, 0.05512237548828125, 0.2570343017578125, -2.363037109375, 1.25946044921875, 0.336181640625, 1.0463790893554688, 0.2442626953125, -0.4064483642578125, 0.4178924560546875, 0.316375732421875, -0.5927200317382812, 0.101654052734375, 0.92388916015625, 0.42779541015625, -0.8653945922851562, -0.5813217163085938, -1.778106689453125, 0.29933929443359375, -0.773956298828125, 0.015289306640625, 0.07442855834960938, 1.0010986328125, 0.0, -0.8301544189453125, -1.18316650390625, 0.06284332275390625, 0.086822509765625, -0.3389625549316406, -0.031890869140625, 1.48291015625, -0.04001617431640625, -0.864990234375, 0.33243560791015625, 1.5162353515625, -1.056915283203125, -1.303253173828125, 1.53106689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000010.npy"}
{"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": 0.01842629909515381, "std": 0.9578834772109985, "min": -2.661041259765625, "p10": -1.3388069152832032, "median": 0.09173583984375, "p90": 1.2718856811523438, "max": 2.21917724609375, "pos_frac": 0.546875, "sample": [-1.16607666015625, -1.01666259765625, 0.0946044921875, 0.974090576171875, 1.267669677734375, 0.13851547241210938, -0.613189697265625, -0.737152099609375, -0.77294921875, 0.278900146484375, 0.7137222290039062, 0.47737884521484375, 0.561767578125, -1.33416748046875, -0.342742919921875, 0.0888671875, -0.26446533203125, 0.11454010009765625, -0.378936767578125, -0.38751983642578125, 0.0061187744140625, -0.9575653076171875, 1.69036865234375, 2.0147705078125, 0.7389678955078125, -0.1060791015625, 1.6411895751953125, -0.2198944091796875, 1.092559814453125, -0.367523193359375, -0.60589599609375, 0.518890380859375, 0.05802154541015625, 0.77203369140625, 0.4940643310546875, 0.960174560546875, -2.304443359375, 1.3123321533203125, 1.252410888671875, 0.10044097900390625, 0.3556671142578125, -1.463531494140625, -1.743682861328125, -1.929443359375, 0.9336090087890625, 1.170440673828125, -0.31845855712890625, 1.38970947265625, 1.401123046875, -0.8917236328125, -0.4557952880859375, 0.45220947265625, -0.4858856201171875, -0.692535400390625, -0.611572265625, 0.6253662109375, -0.5796051025390625, -0.356536865234375, 0.00091552734375, 1.2817230224609375, -0.468505859375, 0.1297149658203125, -0.20635986328125, 0.87286376953125, -1.688995361328125, 0.40242767333984375, 1.37078857421875, -0.1648998260498047, -0.21735382080078125, 0.6206855773925781, 1.027923583984375, 0.39249420166015625, -0.092742919921875, -1.2257537841796875, 0.4556102752685547, -0.546722412109375, 0.1577911376953125, -0.02550506591796875, 0.589202880859375, 1.54071044921875, 0.26236724853515625, 0.7279052734375, -0.3389091491699219, 0.7620849609375, 0.555755615234375, 0.7928695678710938, -0.8679428100585938, -1.19525146484375, -2.661041259765625, -0.40399169921875, 0.1883544921875, -0.16703033447265625, -1.933929443359375, 0.019784927368164062, -1.4001922607421875, -1.3941650390625, 0.351470947265625, -1.43157958984375, 0.2106170654296875, -0.20647239685058594, -0.885986328125, 0.3624267578125, 2.21917724609375, -1.441650390625, 0.11175537109375, 0.723846435546875, 0.114715576171875, 0.210479736328125, 1.0820465087890625, -0.29217529296875, 1.09197998046875, -0.161285400390625, -1.286346435546875, -1.3496322631835938, 0.1356048583984375, -1.19354248046875, -1.9344482421875, 0.103546142578125, 0.8595123291015625, 1.60107421875, 1.44488525390625, -0.10205078125, 1.6177215576171875, 0.0127410888671875, 0.7499465942382812, -0.2359619140625, -0.8653106689453125, 1.002288818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000011.npy"}
{"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": -0.027375996112823486, "std": 0.9722285270690918, "min": -2.55889892578125, "p10": -1.2401878356933593, "median": -0.0316314697265625, "p90": 1.1706237792968746, "max": 2.5331268310546875, "pos_frac": 0.484375, "sample": [-1.328125, 0.8409423828125, 1.837982177734375, 0.982330322265625, -0.2516193389892578, -1.2489089965820312, -1.0703582763671875, -0.74285888671875, -0.46478271484375, -0.7173538208007812, -0.8622283935546875, 0.5180816650390625, 1.353973388671875, 0.862518310546875, 1.032562255859375, 0.4423980712890625, 0.267059326171875, 0.2619171142578125, -0.02777099609375, 0.1783447265625, -0.829742431640625, 1.787445068359375, 1.671661376953125, 0.5728225708007812, 2.22760009765625, -0.7249755859375, -1.123392105102539, -0.19561004638671875, 0.7125244140625, 0.78045654296875, -2.55889892578125, 1.34954833984375, 0.6726303100585938, -2.093902587890625, -0.6363525390625, 0.194793701171875, -0.14349746704101562, -0.4422760009765625, 0.320556640625, -0.2512054443359375, -0.45735931396484375, -1.72998046875, -0.707763671875, 0.854217529296875, 0.29880523681640625, -0.24052047729492188, 0.072235107421875, -0.0590057373046875, -0.67510986328125, 0.6569366455078125, 2.5331268310546875, 0.0933685302734375, 0.7166595458984375, -0.0931396484375, -0.1116180419921875, 0.03870391845703125, 0.06781005859375, -0.166656494140625, -1.1029586791992188, 0.0452423095703125, -1.058502197265625, -0.96136474609375, -1.45147705078125, 0.018218994140625, -0.2919273376464844, 1.0348663330078125, -0.33251190185546875, 0.68499755859375, -0.22772216796875, 0.788543701171875, -0.0336761474609375, -0.4410858154296875, 0.9020309448242188, -0.59228515625, -0.5343780517578125, -1.2364501953125, 0.32944488525390625, -0.5274124145507812, -0.9953460693359375, -0.52117919921875, 1.2823486328125, 0.4498748779296875, -0.2044677734375, 0.61761474609375, 0.2455902099609375, 1.12274169921875, -0.15660476684570312, 0.12542724609375, -1.888671875, 0.34918212890625, 0.757476806640625, 0.17052459716796875, -1.1629714965820312, 0.1468505859375, -0.685699462890625, 0.15167236328125, -1.224853515625, 0.4538116455078125, -0.9958953857421875, 1.36468505859375, -0.03143310546875, -0.031829833984375, 0.59185791015625, 0.901947021484375, -1.34295654296875, -0.16302490234375, 0.5582427978515625, 1.351776123046875, 0.0844573974609375, 1.98773193359375, 0.545684814453125, -0.8489990234375, -0.150238037109375, -0.1944580078125, 1.742645263671875, 0.690521240234375, -0.14090347290039062, -2.429931640625, -2.235443115234375, -0.645751953125, 0.452239990234375, -1.5460205078125, 1.824127197265625, -1.323333740234375, -0.3779296875, -1.61517333984375, 0.9175872802734375, -0.734222412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000012.npy"}
{"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": 0.01376672089099884, "std": 0.9082515835762024, "min": -3.26092529296875, "p10": -0.911505126953125, "median": 0.01381683349609375, "p90": 0.9086242675781248, "max": 3.16009521484375, "pos_frac": 0.5078125, "sample": [0.14984893798828125, -0.04187774658203125, -0.19533538818359375, 1.1326904296875, 0.6441650390625, 1.486083984375, 2.0709228515625, -0.0771484375, -0.162384033203125, -0.08483123779296875, 0.8531341552734375, -3.06768798828125, 0.0235443115234375, -0.3062286376953125, -0.20788002014160156, -0.06398200988769531, -0.18853759765625, 0.1112060546875, 0.2908477783203125, 0.36916351318359375, 0.18767166137695312, -1.67620849609375, 0.65509033203125, -0.02941131591796875, 0.668365478515625, -0.2377796173095703, -0.060260772705078125, -0.848175048828125, -0.557342529296875, 0.096771240234375, 0.967926025390625, 0.40836334228515625, -1.612274169921875, -0.300567626953125, 0.4863433837890625, 0.26303863525390625, -0.04195404052734375, 0.1983642578125, -0.47058868408203125, -0.3984222412109375, 0.44646644592285156, 0.43609619140625, 0.461578369140625, -0.06689453125, -0.0334930419921875, 0.1207427978515625, -1.3288116455078125, -1.268585205078125, 0.16668701171875, 0.7564697265625, 0.81219482421875, -1.1273956298828125, 0.18724441528320312, 1.1202545166015625, -1.0469970703125, 0.0, -0.9013671875, -0.489044189453125, 0.2078704833984375, -0.435943603515625, 0.3751373291015625, -0.91180419921875, -0.35671234130859375, 0.1102447509765625, 1.359832763671875, -0.78973388671875, 0.052032470703125, 0.645477294921875, 0.8449859619140625, 0.06252861022949219, -0.87255859375, -0.7578277587890625, -0.4212303161621094, 0.852935791015625, 0.02593994140625, 0.9712677001953125, -0.263214111328125, 3.16009521484375, 0.11003875732421875, 0.6361541748046875, 0.009857177734375, 0.696502685546875, 0.7645263671875, 0.090789794921875, 0.3770599365234375, 0.49700927734375, -0.35993194580078125, -0.16054916381835938, -0.163665771484375, 0.17980194091796875, 2.182464599609375, 0.35894012451171875, 0.14362335205078125, -0.15494537353515625, -0.1839447021484375, -0.6837692260742188, -0.16380691528320312, 0.53057861328125, -1.9810791015625, 1.55950927734375, -2.605255126953125, -0.42670440673828125, -0.101776123046875, -0.3296051025390625, -0.607269287109375, 1.52496337890625, -0.136810302734375, 1.30340576171875, -0.2918548583984375, 0.418243408203125, 0.23906326293945312, -0.911376953125, 0.6291351318359375, -0.3216552734375, 2.0716552734375, 0.70361328125, -0.24196624755859375, -1.681884765625, 0.883209228515625, 0.708740234375, -3.26092529296875, -1.084503173828125, -0.4014739990234375, 0.0177764892578125, -0.61700439453125, 0.21831130981445312, -0.7483673095703125, -0.0118408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000013.npy"}
{"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": 0.12003912031650543, "std": 0.8484699726104736, "min": -2.555419921875, "p10": -1.0591888427734375, "median": 0.02770233154296875, "p90": 1.3522994995117186, "max": 2.508270263671875, "pos_frac": 0.5, "sample": [1.34039306640625, 0.5387725830078125, -0.6843795776367188, 1.5721435546875, -0.41481781005859375, -0.2884979248046875, -1.2050018310546875, -0.49017333984375, 0.27040672302246094, 0.7398529052734375, -0.3459014892578125, -0.14235687255859375, -1.391357421875, -0.0484619140625, -0.491058349609375, 1.0979461669921875, -0.2042236328125, -0.214874267578125, -0.021572113037109375, 0.96881103515625, -0.1933135986328125, -0.15807723999023438, 0.50665283203125, 0.342529296875, -0.00701904296875, -0.0577392578125, -0.265869140625, 0.401275634765625, 0.676727294921875, 0.311859130859375, 1.14324951171875, 0.17352294921875, -1.27044677734375, 0.05963134765625, -0.269439697265625, -0.44769287109375, -0.052032470703125, -0.07214736938476562, 1.72998046875, -1.124237060546875, 0.3085460662841797, 0.524566650390625, -0.465179443359375, 1.593994140625, 0.0, 0.24286651611328125, -0.0669708251953125, 1.954254150390625, 0.0645599365234375, -0.1405792236328125, 0.191986083984375, -0.061767578125, -0.30263519287109375, 1.95733642578125, -0.419891357421875, -1.28448486328125, 0.95379638671875, 1.524322509765625, -1.187774658203125, 0.436279296875, 0.55511474609375, -1.382415771484375, 0.96636962890625, -0.08953857421875, 1.3800811767578125, -0.0755615234375, -0.4444580078125, -0.38525390625, 2.508270263671875, 0.825469970703125, 0.11357879638671875, -0.393341064453125, -0.19335556030273438, 1.564666748046875, 0.542449951171875, 0.36457061767578125, -0.4668121337890625, 0.783843994140625, 0.525360107421875, -0.1535186767578125, -0.39810943603515625, -1.407989501953125, 1.65289306640625, 0.2669677734375, 1.1304168701171875, -0.3813438415527344, 0.22501182556152344, 0.0554046630859375, -0.9864578247070312, -0.5801200866699219, 0.3191680908203125, 0.32527923583984375, 1.3997802734375, -0.2460174560546875, -0.759765625, -0.087158203125, -0.130126953125, -0.412933349609375, 1.62054443359375, 0.457611083984375, -1.502899169921875, -1.3389739990234375, 0.06365203857421875, -2.555419921875, -0.2777671813964844, 0.4168853759765625, 0.5083999633789062, -0.9694061279296875, 0.4528350830078125, -0.0136871337890625, 0.17852783203125, 0.259613037109375, -1.19140625, -0.34343719482421875, 0.288238525390625, -1.03131103515625, 0.3573150634765625, 0.786865234375, 0.3549652099609375, -0.022197723388671875, 1.7998046875, 0.9199371337890625, 0.495880126953125, -1.523101806640625, -0.02365875244140625, 0.8541259765625, 1.17218017578125, 0.8041839599609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000014.npy"}
{"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": -0.11122331023216248, "std": 1.0399363040924072, "min": -3.022552490234375, "p10": -1.4779083251953125, "median": -0.15240764617919922, "p90": 1.0354339599609375, "max": 4.16656494140625, "pos_frac": 0.421875, "sample": [1.033538818359375, -1.677978515625, -0.239654541015625, -0.17291259765625, -0.453765869140625, 0.6949462890625, 1.2939453125, -0.552093505859375, 0.074951171875, -0.27373504638671875, 0.4208850860595703, 0.63568115234375, -0.58795166015625, -0.2535667419433594, 0.7164459228515625, -1.516571044921875, -0.6682968139648438, -0.3168468475341797, -0.7850875854492188, 0.2583160400390625, -0.87762451171875, -0.42548370361328125, -0.639495849609375, -3.022552490234375, -0.5219039916992188, 0.0755157470703125, -0.1242828369140625, -1.58184814453125, -0.97674560546875, -0.04744720458984375, 1.84320068359375, 0.27197265625, -1.504974365234375, 0.24871063232421875, 0.065582275390625, -0.5573234558105469, 1.1680908203125, 0.56378173828125, -0.0604248046875, -0.6153564453125, -0.8116836547851562, -1.745758056640625, 1.4578857421875, -0.8940353393554688, -0.3270721435546875, -1.041778564453125, -0.1170196533203125, 1.4245452880859375, 0.0860595703125, 0.598541259765625, -0.16275596618652344, 0.27745819091796875, -0.953399658203125, 0.139923095703125, -0.39892578125, 0.016815185546875, -0.997344970703125, -0.073150634765625, 0.016733169555664062, 0.83197021484375, -0.17612457275390625, 0.4872932434082031, -1.46630859375, 0.0073490142822265625, -0.80657958984375, -0.636962890625, 0.2274932861328125, 0.209716796875, 0.12030029296875, -1.011444091796875, -0.8152732849121094, -0.20053482055664062, -0.3869171142578125, -0.264739990234375, -0.779327392578125, 1.236724853515625, 1.344482421875, 1.03985595703125, -0.6260223388671875, 1.452301025390625, 2.15484619140625, 0.0026397705078125, -0.5677490234375, -0.80731201171875, -1.839813232421875, -0.207977294921875, -0.699371337890625, -1.9088134765625, 0.279296875, 4.11737060546875, -0.16676712036132812, -0.0257720947265625, 0.10085296630859375, -1.840545654296875, 0.309051513671875, 4.16656494140625, -0.41256141662597656, -1.6824951171875, -0.2440185546875, 0.362060546875, -1.823394775390625, 2.741546630859375, -0.530059814453125, 0.5764007568359375, -0.07769775390625, 0.300262451171875, -0.732666015625, 0.16346359252929688, 0.3101959228515625, -0.03176116943359375, 0.16619110107421875, -0.2859649658203125, -0.5970802307128906, 0.79754638671875, -0.3627471923828125, -0.44902801513671875, -2.16302490234375, 0.6683578491210938, 0.8109970092773438, -0.43206787109375, 0.2183837890625, 0.11980438232421875, -0.45194244384765625, -2.220062255859375, 0.683685302734375, 0.268524169921875, -0.045803070068359375, -0.142059326171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000015.npy"}
{"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": 0.05807873606681824, "std": 0.9371537566184998, "min": -2.709716796875, "p10": -1.275537109375, "median": 0.04006195068359375, "p90": 1.2729553222656247, "max": 2.868377685546875, "pos_frac": 0.53125, "sample": [0.03850555419921875, 0.06746673583984375, 0.6178512573242188, 1.343109130859375, 0.405914306640625, 0.484161376953125, -0.75823974609375, 1.942718505859375, 1.068603515625, 0.72930908203125, 0.24622726440429688, 0.01790618896484375, 0.04161834716796875, -0.344482421875, 2.868377685546875, -0.6016464233398438, 0.8330078125, 1.242889404296875, -1.2091064453125, -0.18963623046875, 0.89312744140625, -0.00803375244140625, 0.21246337890625, 0.135955810546875, -0.185791015625, 0.16558837890625, 1.4285888671875, 2.140655517578125, -0.7848358154296875, -1.014923095703125, 0.635162353515625, 0.627838134765625, -2.709716796875, -0.917236328125, -1.625457763671875, -1.269622802734375, 1.6260986328125, -0.09558868408203125, -0.8872566223144531, -0.07893753051757812, 0.054534912109375, -1.03106689453125, 0.1251678466796875, 0.603271484375, -0.2945709228515625, -0.03255653381347656, -0.021148681640625, 0.4479522705078125, -0.0794677734375, 0.21735000610351562, -0.11293411254882812, 0.9935302734375, 0.709716796875, -1.66705322265625, -1.59295654296875, -0.14935302734375, 0.6116790771484375, -0.3428192138671875, 0.6247100830078125, -0.28851318359375, -0.45318603515625, -0.056041717529296875, 0.3193817138671875, 0.25091552734375, -1.289337158203125, -2.1787109375, 0.13778305053710938, 0.5913543701171875, 1.4326171875, -1.350830078125, 0.3028411865234375, -0.6297607421875, -1.7072982788085938, 1.41632080078125, 0.996917724609375, -0.4325714111328125, 0.089508056640625, 0.29986572265625, 0.03743743896484375, 1.107513427734375, -0.28418731689453125, -0.56134033203125, -1.3048858642578125, 0.534423828125, 0.6217041015625, 1.39141845703125, 0.0, 0.1832275390625, 0.790435791015625, -1.131195068359375, -0.7253952026367188, 1.042999267578125, -0.4872589111328125, -1.61065673828125, 0.44757080078125, -0.17449951171875, -0.018817901611328125, -0.276336669921875, -0.1361236572265625, 0.17313194274902344, 1.835784912109375, 0.693328857421875, -1.0692138671875, -0.119140625, -0.21514892578125, 0.09383392333984375, -0.032440185546875, -0.167877197265625, -0.2857666015625, -1.331878662109375, 0.02220916748046875, -0.18181610107421875, -1.69970703125, -0.619415283203125, -1.0040283203125, 0.768890380859375, 1.379241943359375, 0.458221435546875, -1.534576416015625, 0.669097900390625, -0.16376495361328125, 0.13323974609375, 0.37896728515625, 1.219207763671875, 2.01708984375, 0.482696533203125, 1.48846435546875, 0.9515380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000016.npy"}
{"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": 0.04243408143520355, "std": 0.841513991355896, "min": -2.158966064453125, "p10": -0.9576232910156249, "median": 0.02530670166015625, "p90": 1.1925384521484375, "max": 2.62152099609375, "pos_frac": 0.5, "sample": [-0.17607879638671875, -0.2578277587890625, -1.33319091796875, -0.4301300048828125, -0.93182373046875, 0.0827789306640625, 0.701324462890625, -2.158966064453125, 0.8307952880859375, 0.41512298583984375, 0.22100830078125, -0.562469482421875, -1.072540283203125, 2.23468017578125, 1.471160888671875, 0.30768585205078125, 1.1878509521484375, 1.039794921875, -1.017822265625, 1.37347412109375, -0.6246337890625, 1.306182861328125, -1.3677825927734375, -0.3161773681640625, -0.4374961853027344, 0.301849365234375, -1.695220947265625, -0.26922607421875, 0.20880126953125, -1.96197509765625, 0.1592254638671875, -0.4013671875, 0.45877838134765625, 0.77587890625, -0.66583251953125, 0.49078369140625, 0.21820068359375, -1.03900146484375, 0.11051177978515625, 0.33838653564453125, -0.0025634765625, 0.482025146484375, -0.5489654541015625, -0.045200347900390625, -0.43866729736328125, 0.2728729248046875, 1.2034759521484375, -0.3683452606201172, 0.7832794189453125, 0.6943511962890625, -0.740570068359375, 0.5587844848632812, -1.3656768798828125, 1.331207275390625, 0.33420562744140625, -0.35113525390625, -0.220947265625, -0.204498291015625, -0.7438888549804688, 0.32940673828125, -0.833038330078125, -0.49051666259765625, -0.3671722412109375, -0.4664764404296875, 0.065582275390625, -0.3277740478515625, -1.18927001953125, -0.22924041748046875, -0.230194091796875, 0.31729888916015625, 0.693939208984375, -0.114715576171875, -0.3004150390625, -0.909515380859375, 1.482696533203125, 0.739013671875, -1.5125579833984375, 0.7682571411132812, 1.144561767578125, 0.3074798583984375, 0.195526123046875, 1.17718505859375, -0.7029876708984375, -0.359466552734375, 0.44445037841796875, 0.775360107421875, 1.4007568359375, 0.3352203369140625, -0.55963134765625, -0.149261474609375, 0.13057708740234375, -0.40313720703125, 0.709686279296875, 1.00201416015625, -0.57568359375, -0.5072784423828125, -0.48679351806640625, -1.06732177734375, 0.4312744140625, 1.34747314453125, -0.056667327880859375, -0.64947509765625, 0.444122314453125, 1.50360107421875, -0.163238525390625, -0.572296142578125, 0.5713005065917969, 0.2527008056640625, 1.16180419921875, -0.80908203125, 0.1168670654296875, -0.5500564575195312, -0.1016845703125, 0.0531768798828125, -0.254608154296875, 0.1490478515625, 0.72479248046875, 2.62152099609375, 0.2452545166015625, -1.93280029296875, 0.2675285339355469, -0.04161834716796875, 0.60675048828125, -0.2429656982421875, -0.56207275390625, 1.4185791015625, -0.36798858642578125, 1.439300537109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000017.npy"}
{"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": -0.01105034351348877, "std": 0.9139746427536011, "min": -3.2373046875, "p10": -0.9375091552734375, "median": 0.003875732421875, "p90": 0.9544036865234374, "max": 2.644287109375, "pos_frac": 0.5, "sample": [-1.9792327880859375, -1.10076904296875, -0.12897491455078125, -0.267822265625, 0.3889732360839844, -1.755706787109375, 0.17544174194335938, -0.7143096923828125, -0.06548309326171875, 0.35544586181640625, -0.47039794921875, 0.5760040283203125, -0.1594085693359375, 0.31890869140625, 1.451904296875, -0.03985595703125, 0.06121826171875, 0.046478271484375, 0.9453582763671875, -0.7807540893554688, 0.49603271484375, 0.8165664672851562, 0.708770751953125, -0.1195068359375, 1.77239990234375, -0.3606128692626953, -0.87371826171875, 1.081268310546875, -0.9830322265625, 0.11934280395507812, -0.401611328125, -0.76556396484375, 0.20511627197265625, -0.3484039306640625, 0.1695404052734375, -1.0233230590820312, 0.25726318359375, 0.28533172607421875, -0.5714340209960938, 0.07861328125, 0.17736053466796875, -0.314208984375, -1.75946044921875, -0.811431884765625, -0.465179443359375, 0.2216949462890625, 0.210968017578125, 0.0155029296875, 0.9755096435546875, -0.295867919921875, 0.300262451171875, -0.578338623046875, 0.053619384765625, -0.12924957275390625, -0.02508544921875, -0.15594482421875, 2.27880859375, 0.6180419921875, 0.128936767578125, 0.45306396484375, 0.56243896484375, 0.5884323120117188, -0.1905517578125, -0.6519775390625, -3.2373046875, -0.579437255859375, -2.54541015625, 2.3393707275390625, -1.17266845703125, -1.25396728515625, -0.09663963317871094, 0.20719528198242188, -0.655731201171875, 1.27532958984375, 0.2984161376953125, -0.270477294921875, -2.143890380859375, -0.197723388671875, 0.6628646850585938, 0.0608673095703125, -0.36724853515625, -0.2464447021484375, -0.259521484375, 1.431549072265625, 0.3073577880859375, 0.6670684814453125, 0.1079254150390625, 1.49835205078125, 0.88189697265625, -0.623748779296875, -0.60833740234375, 0.20249557495117188, 0.4730987548828125, -0.6023941040039062, -0.4459228515625, 0.49407958984375, -0.15175247192382812, -0.30690765380859375, -0.345458984375, 0.34661102294921875, -0.78887939453125, 2.0787353515625, -0.61749267578125, -0.02557373046875, 0.22170257568359375, 0.79144287109375, 0.16009521484375, -0.404937744140625, -0.894744873046875, 0.500823974609375, -1.253570556640625, 0.802825927734375, -0.01214599609375, 2.644287109375, 1.623504638671875, 0.479400634765625, 0.0751190185546875, 1.458953857421875, -0.917999267578125, 0.1758880615234375, 0.83953857421875, -0.1609649658203125, 0.057159423828125, -0.599334716796875, 0.94232177734375, -0.00775146484375, -0.01617431640625, -2.31756591796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000018.npy"}
{"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": -0.05093524605035782, "std": 0.8296994566917419, "min": -2.30572509765625, "p10": -0.9961380004882813, "median": -0.01287841796875, "p90": 0.8488250732421875, "max": 3.2132568359375, "pos_frac": 0.4921875, "sample": [0.7979240417480469, -0.32708740234375, 0.797607421875, -0.10015869140625, 0.4107666015625, 0.4969482421875, 0.21649169921875, -0.17303466796875, -0.3273162841796875, -1.02532958984375, 0.07100486755371094, -0.06407737731933594, 0.803314208984375, -2.30572509765625, 0.7403335571289062, -0.398406982421875, -1.1147918701171875, -1.68115234375, 0.26067352294921875, 0.9605941772460938, 0.4521026611328125, 0.155242919921875, -0.34134864807128906, 0.08251953125, 0.8388214111328125, -0.5109100341796875, 0.1558971405029297, -0.5268783569335938, -1.67156982421875, 0.0308837890625, -1.808837890625, 1.55560302734375, -1.50262451171875, -0.2427978515625, -0.22088623046875, -0.5854072570800781, 0.149810791015625, -0.37860107421875, -0.021087646484375, 0.4270973205566406, 0.82476806640625, -0.2834339141845703, -0.660064697265625, 0.0297698974609375, -0.06341552734375, 1.23529052734375, -0.14672470092773438, 0.3852081298828125, -1.588348388671875, -0.8941574096679688, -0.43477630615234375, -1.819366455078125, 0.103118896484375, -1.0409393310546875, -0.8029937744140625, -0.579559326171875, 0.126251220703125, 0.038818359375, 0.03166961669921875, -0.08362579345703125, -0.6559982299804688, -0.02866363525390625, -1.31982421875, -0.19240951538085938, 0.779510498046875, -0.298248291015625, 0.85504150390625, 1.1641693115234375, -2.1337051391601562, -0.4506072998046875, 0.05108642578125, -0.7156982421875, -0.902557373046875, -0.68585205078125, 0.5534515380859375, -0.8852691650390625, 0.63189697265625, 0.4174346923828125, 0.491241455078125, 0.8150634765625, -0.809051513671875, -0.775390625, 0.29396629333496094, 0.03485107421875, 0.038482666015625, -0.6504364013671875, -0.446685791015625, -0.55133056640625, 0.00666046142578125, 0.68951416015625, 1.41058349609375, 0.47955322265625, 0.5994873046875, 0.4149169921875, 1.238433837890625, 1.050323486328125, 1.1528854370117188, -0.4972953796386719, 0.06490325927734375, 0.4952421188354492, 0.42987823486328125, -0.551605224609375, 0.079986572265625, -0.004669189453125, -0.45343017578125, 0.13128662109375, 0.117431640625, -0.26531982421875, 0.31078338623046875, 0.290130615234375, -0.2129364013671875, -0.8109054565429688, -0.206787109375, -0.9902191162109375, -1.00994873046875, 0.846160888671875, -0.53619384765625, -0.42388916015625, 3.2132568359375, 1.46490478515625, 1.6169586181640625, 0.194427490234375, 1.77484130859375, -0.153656005859375, 0.37041473388671875, -0.74017333984375, -0.5062408447265625, -0.676971435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000019.npy"}
{"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": 0.044177815318107605, "std": 0.9697855114936829, "min": -2.6187744140625, "p10": -1.032763671875, "median": 0.03990936279296875, "p90": 1.2211883544921873, "max": 2.856353759765625, "pos_frac": 0.53125, "sample": [-0.3072052001953125, 0.432647705078125, 0.757537841796875, 2.39410400390625, -0.481964111328125, 1.3724365234375, -0.615966796875, -2.04144287109375, 0.573486328125, -0.9880905151367188, -0.363677978515625, -1.050323486328125, 0.7469482421875, 1.209991455078125, -0.28253173828125, 0.4706878662109375, 0.99932861328125, -0.248504638671875, 0.2949676513671875, 0.2629280090332031, 0.2004852294921875, 0.0777587890625, 2.7796630859375, 0.011325836181640625, -0.22918128967285156, 0.341156005859375, -0.3227996826171875, -0.3879852294921875, -0.32726287841796875, -0.45916748046875, -0.014495849609375, -0.168121337890625, 0.01909637451171875, -0.51947021484375, -0.5766983032226562, -1.9415283203125, 0.272705078125, 0.201690673828125, -0.578216552734375, -1.35797119140625, -2.158905029296875, -0.29888916015625, -1.2645416259765625, 0.7848663330078125, -0.01324462890625, 0.131683349609375, -0.28570556640625, 0.06072235107421875, 0.204071044921875, -0.21605682373046875, 0.592864990234375, 0.15833282470703125, 0.0, 0.11480712890625, 1.524383544921875, 0.2088623046875, -0.16497230529785156, 0.36834716796875, 1.6024017333984375, 0.4976959228515625, 0.12700653076171875, -0.70208740234375, -0.81304931640625, 1.644775390625, 0.62335205078125, 1.007415771484375, 0.3092803955078125, 0.21247100830078125, 0.0887603759765625, 0.66278076171875, 0.26641845703125, -0.07318115234375, -0.371734619140625, 0.06333160400390625, 0.920196533203125, -0.534423828125, 0.06964874267578125, -1.932861328125, 0.9129104614257812, 1.34356689453125, 2.718353271484375, -0.3877105712890625, 0.6611251831054688, 0.004283905029296875, -0.1895751953125, -0.3677978515625, -1.025238037109375, -2.042236328125, 0.26319122314453125, 0.374267578125, -0.2574272155761719, 2.2691650390625, 2.33465576171875, -2.119476318359375, -0.2151947021484375, 1.6175537109375, 0.70867919921875, -0.7551956176757812, -0.6995468139648438, 0.00811004638671875, -2.6187744140625, -1.4342041015625, -0.177093505859375, -0.772705078125, -0.762542724609375, -0.4812469482421875, -0.5605316162109375, 0.64971923828125, -0.2432708740234375, 0.19183349609375, 1.247314453125, -0.3251800537109375, 0.6347732543945312, 0.550689697265625, 0.212921142578125, 0.4662132263183594, 2.856353759765625, -0.6506500244140625, -1.21026611328125, 0.11786460876464844, -0.235595703125, 0.350341796875, 0.7685546875, 0.0919036865234375, -0.629180908203125, -0.2452239990234375, 0.3271942138671875, -1.19207763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000020.npy"}
{"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": -0.018529891967773438, "std": 0.8829474449157715, "min": -3.021209716796875, "p10": -1.3691619873046874, "median": 0.0164794921875, "p90": 0.9536895751953123, "max": 2.05511474609375, "pos_frac": 0.5234375, "sample": [-0.5689697265625, 0.3575286865234375, -0.4330902099609375, 0.5216827392578125, -0.067901611328125, -2.32928466796875, -0.73663330078125, -0.374847412109375, 0.0944976806640625, -0.53472900390625, 0.68109130859375, -0.3707275390625, 0.303375244140625, -1.7586669921875, 0.703338623046875, -0.4693756103515625, 0.889404296875, 0.5013580322265625, -1.364776611328125, -0.09533309936523438, 0.68975830078125, 0.20141983032226562, -1.435791015625, 0.463836669921875, -0.2711334228515625, 0.8285064697265625, 1.87762451171875, -0.27764892578125, -0.27228546142578125, -1.4532318115234375, -0.167449951171875, 0.38067626953125, 1.62628173828125, -0.51324462890625, 0.58734130859375, 0.0467376708984375, -0.4632720947265625, 0.70294189453125, -3.021209716796875, 0.01202392578125, -0.7396621704101562, 0.31365966796875, -0.1756591796875, 0.14249420166015625, 0.16725921630859375, 1.12847900390625, 0.06297874450683594, -0.220245361328125, -0.30279541015625, 0.928558349609375, -0.3992767333984375, 0.1258544921875, 0.0, -0.38567352294921875, -1.096954345703125, 0.11669540405273438, -1.37939453125, -1.47137451171875, -0.78619384765625, -0.21382904052734375, 0.00299072265625, -0.43621826171875, 0.13235092163085938, 1.72967529296875, 0.240478515625, 0.3008270263671875, -0.2652740478515625, -0.56512451171875, 0.0178375244140625, 1.03900146484375, -0.337921142578125, -0.403961181640625, 0.038787841796875, -1.732086181640625, -1.0625686645507812, 1.668701171875, -0.355865478515625, 0.916900634765625, 1.019683837890625, 0.548126220703125, 0.5528030395507812, -0.3383445739746094, -0.191436767578125, -0.107696533203125, 0.1441650390625, 0.0881195068359375, 0.46453094482421875, 0.1856689453125, 0.2147216796875, 0.20316696166992188, -1.997467041015625, 0.18523406982421875, 1.70904541015625, 0.31198692321777344, 1.193084716796875, -0.007938385009765625, -0.016143798828125, 1.0123291015625, 1.944854736328125, 0.4779815673828125, -1.81695556640625, -0.7826080322265625, 2.05511474609375, 0.531890869140625, -0.30126953125, -1.815277099609375, 0.16228485107421875, 0.4655914306640625, 0.0151214599609375, 0.8109893798828125, -0.523895263671875, -0.1451568603515625, 0.627288818359375, -0.173309326171875, 0.7994613647460938, -0.249603271484375, 0.7502288818359375, 1.91033935546875, -0.415985107421875, -1.38372802734375, -1.73046875, -0.865753173828125, 0.13100051879882812, -0.087738037109375, 0.2781219482421875, 0.51678466796875, 0.11381912231445312, -0.0778656005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000021.npy"}
{"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": 0.12431301176548004, "std": 0.8225438594818115, "min": -2.11474609375, "p10": -0.7468765258789063, "median": 0.05136871337890625, "p90": 1.183169555664062, "max": 2.49359130859375, "pos_frac": 0.53125, "sample": [1.9115753173828125, 0.4588890075683594, 0.5153121948242188, 2.0125732421875, -0.024322509765625, -0.11241912841796875, 0.568328857421875, 0.121337890625, 0.09503173828125, 0.46343994140625, 1.10980224609375, -0.2348175048828125, -1.7829132080078125, 0.0597686767578125, -1.36578369140625, -0.582855224609375, -0.1622314453125, 0.619354248046875, -0.751800537109375, -0.1461944580078125, -0.0718231201171875, -0.11328887939453125, 0.335296630859375, -0.4161224365234375, 1.125518798828125, -0.3333282470703125, 0.39703369140625, 0.47406005859375, -0.28116607666015625, -0.06955337524414062, -1.02288818359375, -0.5265960693359375, 0.0, 0.4246826171875, 0.27459716796875, -0.799896240234375, 0.3561248779296875, -0.3928680419921875, 0.848114013671875, -1.650299072265625, -0.31427001953125, 0.8852691650390625, -0.694793701171875, 2.0050048828125, -0.8488006591796875, 0.0839996337890625, -2.11474609375, -1.9345703125, 0.8385009765625, 0.814727783203125, 0.17560195922851562, -0.09005355834960938, -0.699188232421875, -0.35028076171875, -0.0868682861328125, 0.341766357421875, 0.019588470458984375, 1.547210693359375, 1.87969970703125, 1.357177734375, 0.142303466796875, -0.49920654296875, 0.091949462890625, -0.5446548461914062, -0.31188392639160156, -0.2139739990234375, 1.5655517578125, -0.26654052734375, -0.7202987670898438, 0.101593017578125, 0.0017547607421875, -0.538970947265625, 0.5098495483398438, 0.19744873046875, -0.254547119140625, -0.26641845703125, 0.1143341064453125, -0.2025604248046875, -0.29034423828125, -0.628021240234375, -0.7447662353515625, -0.0563812255859375, 0.63372802734375, 0.11069869995117188, -0.0006256103515625, 1.6961669921875, 1.6908340454101562, -0.79119873046875, -0.74383544921875, -1.284576416015625, 2.4071044921875, -0.40789794921875, -0.16904067993164062, -0.0385284423828125, 0.25714111328125, 0.29959869384765625, 0.201416015625, -0.920684814453125, 1.31768798828125, 0.010467529296875, 0.183349609375, 0.15991592407226562, 1.40771484375, 0.8917236328125, 0.58685302734375, 0.3953094482421875, -0.0428924560546875, 1.09490966796875, 0.40008544921875, 0.12994384765625, 0.24921798706054688, 0.64306640625, 0.6198806762695312, -0.169525146484375, -0.6437168121337891, -0.0975341796875, 0.63287353515625, 2.49359130859375, 0.744415283203125, -1.2347412109375, -0.28551483154296875, 1.119720458984375, 0.4459190368652344, 0.384063720703125, -0.06043243408203125, 0.04296875, -0.05301856994628906, 0.27459716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000022.npy"}
{"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": 0.04329611361026764, "std": 0.9191638231277466, "min": -1.847442626953125, "p10": -1.1380828857421874, "median": -0.009273529052734375, "p90": 1.1880920410156248, "max": 3.0281982421875, "pos_frac": 0.484375, "sample": [0.1494579315185547, 0.0883941650390625, -1.1431884765625, 1.8382415771484375, 2.592559814453125, 2.979248046875, 0.734527587890625, -1.0430145263671875, -1.467681884765625, 1.1650390625, 1.24188232421875, 0.830963134765625, -0.29189300537109375, -0.6568603515625, -1.291717529296875, -0.12886428833007812, -0.06756591796875, 1.46875, -0.4155006408691406, -0.9263153076171875, 0.45259857177734375, -1.38482666015625, -0.5192413330078125, 0.53692626953125, 0.541290283203125, 0.6943817138671875, -0.396697998046875, 3.0281982421875, -0.7793045043945312, -0.8428497314453125, 0.0, 0.5071258544921875, -0.078857421875, -0.32434844970703125, -0.54119873046875, 1.136199951171875, -0.3009033203125, -1.531402587890625, 0.552642822265625, -0.49761962890625, -0.49787139892578125, -1.847442626953125, -0.3905792236328125, -0.6636924743652344, -0.1788330078125, 0.2888946533203125, 0.163848876953125, -0.7738876342773438, 1.94622802734375, 0.02166748046875, 0.6781997680664062, -0.0616302490234375, -0.6855010986328125, 0.876495361328125, 0.09521484375, 0.782257080078125, 0.22484588623046875, -0.1116180419921875, 0.23836517333984375, -0.7129669189453125, -0.273956298828125, 0.03589630126953125, 0.26031494140625, 0.23427581787109375, 0.324859619140625, -0.8296432495117188, -0.9447021484375, 1.313232421875, 1.036346435546875, -0.277099609375, 1.1356964111328125, 0.09370040893554688, -1.26995849609375, -0.0010986328125, 1.275726318359375, 1.5498199462890625, 1.376739501953125, 0.5279045104980469, 1.0977783203125, -0.4032135009765625, -1.5214004516601562, -0.4829998016357422, 0.5600090026855469, 0.9846420288085938, 0.272705078125, -1.4200439453125, -1.2156982421875, -0.2262115478515625, 0.243896484375, 1.595428466796875, -0.837371826171875, -0.44579124450683594, 0.88836669921875, 0.12722015380859375, -0.5981063842773438, -0.47747802734375, 1.38092041015625, 0.283843994140625, 0.2669525146484375, 0.930908203125, -0.343017578125, 0.1568450927734375, -0.01744842529296875, -0.33797454833984375, -0.2633819580078125, -0.3228759765625, -1.135894775390625, 0.32540130615234375, 0.1236419677734375, 0.506866455078125, -0.0399169921875, -0.619781494140625, 0.4582176208496094, -0.8573760986328125, -0.20183563232421875, -1.093719482421875, 1.0545654296875, -0.240509033203125, 0.246612548828125, -1.1925048828125, 0.410797119140625, -1.2828903198242188, 1.06378173828125, -0.4195098876953125, 0.661529541015625, -1.256622314453125, -1.06158447265625, -0.6524925231933594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000023.npy"}
{"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": 0.0021340101957321167, "std": 0.9482470154762268, "min": -2.646087646484375, "p10": -1.0555274963378904, "median": 0.046295166015625, "p90": 0.9777381896972656, "max": 2.826171875, "pos_frac": 0.5234375, "sample": [0.3435707092285156, -1.2887115478515625, -0.2930908203125, 1.9503631591796875, -1.154998779296875, 0.89794921875, 0.59576416015625, 0.930389404296875, 0.31479644775390625, 0.41082763671875, 0.2776031494140625, -0.42407989501953125, -1.0101318359375, -0.49817657470703125, -0.12172698974609375, -0.416259765625, -0.23748779296875, 0.519012451171875, 1.1302490234375, 0.35272979736328125, -1.24334716796875, 0.6282958984375, -0.4952392578125, 0.0, -0.162261962890625, 0.25762176513671875, -1.020263671875, -0.7154388427734375, 0.33724212646484375, -0.78277587890625, 0.79827880859375, 0.09517288208007812, 0.970794677734375, 1.547119140625, 0.7833251953125, 0.64654541015625, -1.36712646484375, 0.1037445068359375, -0.8899002075195312, 0.058685302734375, -0.6418304443359375, 0.9827423095703125, 0.15810394287109375, -0.375579833984375, 2.10968017578125, 0.02953338623046875, 0.0, 0.30657958984375, -0.549072265625, 0.9015350341796875, 0.528076171875, 0.25749969482421875, 0.4599761962890625, 0.6083908081054688, -0.86859130859375, 0.07474517822265625, -1.25946044921875, -0.113739013671875, -0.521484375, 2.72271728515625, -1.9549560546875, 0.0743408203125, 0.7763671875, -0.10952377319335938, -1.755859375, 0.9755935668945312, 0.39617156982421875, -0.57806396484375, -0.65740966796875, 0.934814453125, -0.767120361328125, -0.73834228515625, 0.00234222412109375, -0.706298828125, -1.7954559326171875, 0.790008544921875, 0.7296905517578125, -2.1381683349609375, 1.436981201171875, -0.151763916015625, -0.88287353515625, -0.2457427978515625, 0.24749755859375, -0.5632781982421875, 0.033905029296875, -0.6500244140625, -0.19219970703125, 1.204010009765625, -0.824981689453125, 0.5847396850585938, 0.4312591552734375, 0.68719482421875, -2.278564453125, -1.1378097534179688, 0.16387939453125, -0.14740943908691406, -0.78155517578125, 0.4522705078125, -0.235992431640625, 0.47784423828125, -0.3262939453125, -0.52349853515625, 1.0789031982421875, 2.0577392578125, -0.812286376953125, 1.16827392578125, 0.063262939453125, -0.4765625, 1.64794921875, -0.3470458984375, 0.36505126953125, -0.6807861328125, 0.5126953125, 0.624847412109375, 0.0655517578125, 0.328338623046875, -0.014251708984375, -0.10663604736328125, -2.5963592529296875, -0.4819793701171875, 2.826171875, 0.303680419921875, 0.3260231018066406, -2.646087646484375, -0.4317626953125, 0.897430419921875, -0.88958740234375, 0.5959701538085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000024.npy"}
{"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.09480218589305878, "std": 0.7292216420173645, "min": -2.046234130859375, "p10": -0.7209335327148437, "median": -0.00572967529296875, "p90": 0.9444030761718749, "max": 2.584228515625, "pos_frac": 0.4765625, "sample": [-0.347503662109375, -0.206298828125, -0.272979736328125, 0.8248138427734375, -0.154998779296875, 1.77874755859375, 1.692352294921875, 0.931732177734375, 1.7193603515625, 1.19561767578125, -0.1703338623046875, 1.342071533203125, 0.0509490966796875, 1.945648193359375, -0.0053558349609375, -1.591156005859375, 0.25684356689453125, -0.5401382446289062, -0.80291748046875, 0.661285400390625, -0.028533935546875, -0.1648845672607422, -1.003204345703125, -0.3414306640625, -0.1898345947265625, 0.8403167724609375, -0.5194091796875, -0.6376419067382812, 0.04510498046875, -0.134521484375, 0.4540557861328125, 0.8353271484375, 0.0, -0.17598724365234375, 1.0894775390625, -0.28530311584472656, 0.6243209838867188, 0.37732696533203125, -0.801177978515625, -0.603118896484375, 0.0060577392578125, 0.0, 0.07177734375, -0.740234375, -2.046234130859375, -0.535308837890625, 0.32431793212890625, 0.90435791015625, 0.2089385986328125, -0.09271240234375, -0.04444122314453125, -0.1341552734375, 0.973968505859375, 0.11467742919921875, 0.468475341796875, -0.6726837158203125, 0.5681381225585938, -0.7126617431640625, -0.006103515625, -0.52056884765625, 0.53448486328125, -0.047882080078125, -0.923828125, -0.09210205078125, -0.272735595703125, 0.46151161193847656, 0.6810302734375, -1.2571258544921875, -1.0908660888671875, -0.01055908203125, 0.7770538330078125, 0.5286712646484375, 0.09103775024414062, 0.48388671875, -0.38446044921875, 0.66802978515625, -0.3853759765625, 0.2574806213378906, -0.251007080078125, -0.143829345703125, -0.7796478271484375, 0.59185791015625, 0.13043212890625, 2.584228515625, -0.413116455078125, 0.0523681640625, -0.4454460144042969, 0.29425811767578125, 0.18032455444335938, -0.111114501953125, 0.861724853515625, -0.0801544189453125, 0.380340576171875, -0.13140869140625, 1.37939453125, -0.33551025390625, -0.03313446044921875, 1.0607452392578125, 0.6387481689453125, 0.0780181884765625, 0.23358535766601562, 0.0976409912109375, -0.15899658203125, -0.22263717651367188, -0.0897216796875, 0.69793701171875, 2.0689697265625, -0.097381591796875, -0.687530517578125, -0.21588134765625, -1.2031402587890625, 0.6222991943359375, -0.31494140625, 0.07879638671875, 0.46551513671875, -0.061279296875, -0.681732177734375, -0.027286529541015625, 0.4675750732421875, 0.0907745361328125, -0.020290374755859375, 0.25834083557128906, 1.315185546875, -1.2598648071289062, 0.149200439453125, 0.44683837890625, -1.040283203125, -0.1275615692138672], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000025.npy"}
{"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": 0.13580012321472168, "std": 0.8162452578544617, "min": -2.2113037109375, "p10": -0.6148880004882812, "median": 0.036849021911621094, "p90": 1.121204376220703, "max": 3.193267822265625, "pos_frac": 0.5234375, "sample": [0.7113037109375, 0.3895263671875, 2.650390625, -0.199127197265625, 0.21343994140625, -0.19479751586914062, 0.5958328247070312, 0.2595710754394531, 0.256866455078125, 0.12827682495117188, -0.3529930114746094, -0.40142822265625, 0.0, 0.14044189453125, 1.184783935546875, -0.228057861328125, 1.064788818359375, -0.13213729858398438, 2.13177490234375, 0.7517547607421875, 3.193267822265625, -0.3761138916015625, -0.17638015747070312, 0.778106689453125, 0.509002685546875, -0.4297828674316406, -0.6523895263671875, -0.3155364990234375, 0.303314208984375, -0.40032196044921875, -0.06423187255859375, -0.459442138671875, 0.05530548095703125, 0.321868896484375, 0.3199462890625, 0.47503662109375, 0.7848968505859375, 0.7570953369140625, -1.3769989013671875, 0.0102081298828125, -0.397186279296875, -0.3636970520019531, -0.3061981201171875, 0.157623291015625, -0.43218994140625, -2.2113037109375, 0.1354827880859375, 0.194732666015625, 0.0760955810546875, -0.2498321533203125, -0.0186767578125, 1.193695068359375, -0.5500030517578125, -0.80780029296875, -0.39031982421875, -0.18572616577148438, -0.1504364013671875, -0.1290283203125, -0.0919189453125, -0.948974609375, 0.6422195434570312, -1.68115234375, 1.43353271484375, 0.7283782958984375, -0.94549560546875, -0.28470611572265625, 0.764862060546875, 1.66448974609375, 2.03485107421875, 0.3321990966796875, -0.10931396484375, -0.226318359375, -1.00689697265625, 1.60894775390625, -0.18576812744140625, -0.06266403198242188, 0.3648529052734375, 0.31627655029296875, 1.4202880859375, 0.01168060302734375, 0.7650146484375, 0.949371337890625, 0.10421371459960938, -0.52655029296875, -0.04901123046875, 0.1300811767578125, 0.5303192138671875, -0.55126953125, -0.25080108642578125, -0.1410064697265625, 0.50567626953125, 0.510589599609375, 0.2711296081542969, -0.3232421875, 1.623779296875, -1.47320556640625, 0.0781707763671875, 0.04479789733886719, -0.2148456573486328, 0.2161712646484375, -0.0878448486328125, -0.16747283935546875, 0.22576904296875, 0.028900146484375, -0.32686614990234375, -0.385162353515625, -0.59881591796875, 0.0, -0.9547119140625, 0.19449996948242188, -0.424041748046875, -0.08644866943359375, 0.89215087890625, -0.0060882568359375, 1.0939559936523438, 0.16424560546875, -1.940185546875, 0.670684814453125, 0.900115966796875, 1.3165283203125, 1.0096435546875, -1.618011474609375, -0.19805908203125, 1.357696533203125, 0.6478424072265625, 0.498199462890625, 0.35333251953125, -0.948486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000026.npy"}
{"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": -0.012070447206497192, "std": 1.0115138292312622, "min": -2.81805419921875, "p10": -1.36259765625, "median": 0.02215576171875, "p90": 1.1143737792968749, "max": 3.23321533203125, "pos_frac": 0.515625, "sample": [1.16448974609375, 1.47454833984375, -0.3066558837890625, 0.000885009765625, -0.065643310546875, -0.44287109375, -0.8843307495117188, 0.2005157470703125, -0.0963134765625, 0.22512054443359375, -0.10107421875, -1.564605712890625, 0.1971435546875, -0.26450347900390625, -2.22674560546875, -0.5860595703125, -0.183563232421875, 0.0571746826171875, 0.78656005859375, 1.069427490234375, 0.7685432434082031, 0.1690216064453125, 1.252044677734375, -1.4763641357421875, 2.14215087890625, 3.23321533203125, 0.2783851623535156, 2.92108154296875, -0.1439056396484375, 0.18813323974609375, 0.0189208984375, -0.3733367919921875, 0.5009765625, -0.6435546875, 0.6046142578125, -1.0732421875, -1.41766357421875, 2.0225830078125, 0.0793914794921875, -0.6470184326171875, 2.05059814453125, 1.01751708984375, -2.31414794921875, 0.2730560302734375, 0.3639984130859375, -0.02154541015625, -0.4510688781738281, -1.04388427734375, 0.377471923828125, 0.1198883056640625, -0.18255615234375, 0.5874481201171875, -1.18365478515625, 0.99200439453125, 1.2677001953125, 0.89898681640625, 0.21918869018554688, -0.773529052734375, 0.09951019287109375, 1.8443603515625, -0.44317626953125, -0.5927581787109375, -0.6116485595703125, -1.0592041015625, -0.2554931640625, -0.4317474365234375, 1.765045166015625, 0.5120010375976562, 1.0928955078125, -0.07806396484375, -2.81805419921875, -1.783111572265625, -2.01629638671875, -0.1678466796875, 0.8424530029296875, 0.980987548828125, -1.364776611328125, 1.3729248046875, 2.31585693359375, 0.860076904296875, -0.01621246337890625, 0.67510986328125, -1.531707763671875, 0.55145263671875, -1.361663818359375, 0.18878173828125, 0.797454833984375, -0.28322410583496094, 0.24202728271484375, 0.6552963256835938, -0.915496826171875, -0.5016326904296875, -2.3990478515625, -0.59454345703125, -1.3686370849609375, -1.183837890625, -1.311065673828125, -0.2080078125, 0.29119873046875, -0.7716064453125, 0.2567901611328125, 0.2258625030517578, 0.0557708740234375, 0.07659912109375, 0.18507003784179688, 0.025390625, -0.057281494140625, -0.30970001220703125, -0.36425018310546875, -0.2096099853515625, 0.36669921875, 0.4782543182373047, 0.16116905212402344, 0.194488525390625, -0.50091552734375, -0.544464111328125, 0.02857208251953125, -0.11806869506835938, -0.064453125, -0.179443359375, 0.110015869140625, 0.16255950927734375, 0.094818115234375, -1.64044189453125, 0.453948974609375, 0.46845245361328125, -0.9159393310546875, -0.058429718017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000027.npy"}
{"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.21156001091003418, "std": 0.7359578609466553, "min": -1.85601806640625, "p10": -0.6674671173095703, "median": 0.18730640411376953, "p90": 1.1405364990234375, "max": 2.28997802734375, "pos_frac": 0.6171875, "sample": [0.274078369140625, -0.88037109375, -0.236236572265625, 0.013580322265625, 0.2799835205078125, 0.051753997802734375, 0.48415374755859375, 0.9206352233886719, 0.2229595184326172, 0.4169921875, 0.19669532775878906, -0.181884765625, 1.536895751953125, -0.02227020263671875, 0.325408935546875, -0.1008148193359375, 0.515960693359375, -1.1725006103515625, 1.963623046875, 1.370361328125, 0.4907073974609375, -0.24794387817382812, 0.0605621337890625, -0.061767578125, -0.2705078125, -0.19043731689453125, 1.016845703125, -0.11813926696777344, 0.17791748046875, -1.85601806640625, 1.070465087890625, -0.264923095703125, 0.043914794921875, 0.2807769775390625, -0.46605682373046875, -1.1598052978515625, 0.273956298828125, 1.261383056640625, 1.830780029296875, 1.1220703125, -0.568603515625, -1.0653076171875, -0.36590576171875, -1.14605712890625, 0.587860107421875, 2.28997802734375, 0.1622314453125, 1.150482177734375, -0.11614990234375, 0.135345458984375, 0.04705810546875, 0.6489028930664062, -0.085174560546875, -1.17071533203125, -0.49627685546875, -0.7598724365234375, 0.38388824462890625, 0.95147705078125, 0.165374755859375, 1.14581298828125, -0.5326080322265625, -0.00341796875, 0.2721710205078125, 0.16763687133789062, -0.55853271484375, 1.6710357666015625, -0.6669464111328125, 0.4675445556640625, 0.316650390625, 0.319244384765625, -0.2470703125, 0.206817626953125, -0.31009674072265625, -0.4215278625488281, 0.0100555419921875, 0.83294677734375, -0.77667236328125, -0.4569129943847656, 0.02420806884765625, 0.427978515625, -0.49794769287109375, 0.0554656982421875, 0.457977294921875, 1.0608978271484375, 0.0731048583984375, 0.25504302978515625, 0.9014739990234375, 0.264556884765625, -0.04535675048828125, 0.2088336944580078, 0.10272216796875, 0.8796844482421875, 0.216522216796875, -0.0447998046875, -0.18895912170410156, -0.31235504150390625, 1.138275146484375, 0.403900146484375, 1.09259033203125, 1.11297607421875, -1.355712890625, -0.026523590087890625, 0.52001953125, 1.7523956298828125, 0.4518890380859375, 0.22296142578125, 0.6751480102539062, -1.0558624267578125, 1.9151611328125, 0.2860603332519531, 1.37896728515625, -0.07208251953125, 1.11468505859375, -0.6686820983886719, -0.20855712890625, 0.7000503540039062, 0.6834259033203125, 0.291595458984375, 0.885955810546875, -0.02115631103515625, 0.202056884765625, -0.96588134765625, 1.3624267578125, 0.30275726318359375, 0.73577880859375, -0.4802532196044922, 0.0, -0.2911834716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000028.npy"}
{"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.08448109030723572, "std": 0.886737585067749, "min": -2.150543212890625, "p10": -1.0566390991210937, "median": 0.07337188720703125, "p90": 1.2246551513671875, "max": 3.26995849609375, "pos_frac": 0.5625, "sample": [0.1499462127685547, -0.2796630859375, 2.07220458984375, -0.68682861328125, -0.3142852783203125, -1.95831298828125, 0.010040283203125, 3.26995849609375, -1.76800537109375, 1.332122802734375, 0.175628662109375, 1.70458984375, 1.1864013671875, 0.210052490234375, 0.2554779052734375, 0.0, -1.42181396484375, 0.06343841552734375, -0.50347900390625, 0.016788482666015625, -1.111236572265625, 1.369476318359375, -0.5827789306640625, -1.51470947265625, 0.6376495361328125, 1.2369384765625, 0.498992919921875, 0.3576202392578125, -0.43737030029296875, 1.2086639404296875, -0.44268798828125, -0.949554443359375, -1.0363006591796875, 0.107818603515625, -0.1913604736328125, 0.82098388671875, 1.0576171875, 0.801422119140625, -0.8436279296875, -0.5422439575195312, -0.50921630859375, 0.0057964324951171875, 0.274505615234375, 0.8857574462890625, -0.232513427734375, 0.32306671142578125, 0.0910491943359375, -0.40411376953125, 0.0515594482421875, -0.5458221435546875, -0.598663330078125, -1.237548828125, 0.825927734375, -0.3011016845703125, 0.3179473876953125, 1.835418701171875, 1.681884765625, -1.104095458984375, -1.20257568359375, 0.8053970336914062, 0.27069091796875, -1.2318115234375, 1.5281982421875, 0.9500732421875, -0.809844970703125, 1.4254226684570312, 0.19677734375, -0.3195343017578125, 0.0, -0.06755828857421875, 0.373382568359375, 0.59271240234375, -0.217529296875, -1.033599853515625, 0.3160667419433594, -0.656463623046875, -0.3531494140625, -0.1514892578125, -0.620025634765625, 0.3373565673828125, 0.915191650390625, 0.238525390625, -2.150543212890625, 1.1044540405273438, 0.7531280517578125, -0.3185882568359375, 0.41478729248046875, 0.79742431640625, -1.6405029296875, 1.906768798828125, 0.19467926025390625, 1.35107421875, 0.3872528076171875, 1.285400390625, -0.65985107421875, -0.0411529541015625, -0.49798583984375, 0.0127105712890625, 0.15116119384765625, 1.1693267822265625, -0.929901123046875, -0.13587188720703125, 0.778472900390625, -1.2432708740234375, 0.510467529296875, 1.219390869140625, 0.43829345703125, 0.0203857421875, -0.27325439453125, -0.549896240234375, 0.0919189453125, 0.1868133544921875, -1.249267578125, 0.8942413330078125, 0.947479248046875, -0.35614013671875, 0.16396331787109375, -0.2469482421875, 0.22821044921875, -0.5453033447265625, 0.0685882568359375, 0.0784454345703125, 0.45977783203125, -0.5747756958007812, -0.50152587890625, -0.09332275390625, 0.078155517578125, 0.523284912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000029.npy"}
{"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": 0.0842069536447525, "std": 0.9094097018241882, "min": -2.337738037109375, "p10": -0.9496673583984374, "median": -0.02191162109375, "p90": 1.1378631591796875, "max": 2.8717041015625, "pos_frac": 0.46875, "sample": [2.02313232421875, 1.906524658203125, -0.56707763671875, -0.39752197265625, -0.2502708435058594, 1.280914306640625, -0.1039886474609375, -0.409271240234375, 0.894439697265625, -0.13411712646484375, 0.18706512451171875, 0.5467987060546875, 0.155670166015625, 0.18511962890625, 0.170196533203125, -1.1670989990234375, -1.0013427734375, 2.786468505859375, -0.31479644775390625, -0.2911224365234375, -2.167816162109375, 0.79876708984375, -0.1533355712890625, 0.4877052307128906, 0.217010498046875, -1.4796142578125, -0.27862548828125, 0.22061538696289062, -0.40130615234375, 1.137451171875, 0.7486572265625, -0.33235931396484375, -0.1781005859375, 0.315155029296875, -0.0526275634765625, -0.120208740234375, 0.521240234375, -0.083831787109375, -0.056243896484375, -0.4581298828125, -0.5791473388671875, -0.00731658935546875, -1.1201171875, 1.79541015625, -1.82025146484375, 0.704132080078125, -0.6071434020996094, -0.7183685302734375, 0.8064727783203125, -0.927520751953125, 0.7844467163085938, 1.4520263671875, -0.804840087890625, -0.3789215087890625, 1.911773681640625, -0.0257568359375, -0.2243194580078125, 0.6334762573242188, 0.6385498046875, -0.02838134765625, 0.7923583984375, 0.06048583984375, 1.3130950927734375, -1.4797439575195312, 0.0839996337890625, 0.2012176513671875, 0.4010162353515625, -0.67987060546875, -0.08637619018554688, 1.0958251953125, 0.25217628479003906, -0.696075439453125, -0.17326736450195312, -0.23059463500976562, -0.03839111328125, 1.458740234375, -0.8745269775390625, -0.715087890625, -0.5458831787109375, -1.968719482421875, 0.892059326171875, -0.489959716796875, -0.20611572265625, 0.6409149169921875, 0.2652873992919922, 0.89752197265625, 0.57977294921875, -0.439453125, 1.6008148193359375, -1.0081634521484375, -0.6554946899414062, 1.0850982666015625, 0.88995361328125, -0.62127685546875, -0.562774658203125, 0.1478271484375, -0.01806640625, -5.14984130859375e-05, -0.117706298828125, 2.8717041015625, -0.11846923828125, 0.616180419921875, 0.34627532958984375, 0.1949615478515625, 0.6053466796875, -0.3663330078125, -0.277557373046875, 1.06890869140625, -0.488677978515625, -1.51336669921875, 1.138824462890625, 0.0880279541015625, 1.462921142578125, -1.00225830078125, -0.43850135803222656, -0.01642608642578125, 0.608184814453125, 0.5894927978515625, -0.5623741149902344, -2.337738037109375, -0.490447998046875, 1.1005630493164062, 0.98138427734375, -1.7117919921875, -0.35422325134277344, 0.7370147705078125, 0.565948486328125, 0.76202392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000030.npy"}
{"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.14825668931007385, "std": 0.8822535276412964, "min": -2.2530517578125, "p10": -0.94371337890625, "median": 0.112823486328125, "p90": 1.2469299316406248, "max": 3.848419189453125, "pos_frac": 0.59375, "sample": [0.25864410400390625, -0.7950439453125, -1.17144775390625, -1.17181396484375, 0.048187255859375, 0.6620025634765625, 0.5260467529296875, 1.55438232421875, 0.4533843994140625, 1.747955322265625, 1.79833984375, -0.863067626953125, 0.230010986328125, 0.29449462890625, 0.657379150390625, -0.6463775634765625, 0.7006626129150391, 0.0948944091796875, 0.3954925537109375, 0.4339447021484375, -1.12591552734375, 0.02685546875, -1.2407073974609375, 1.588165283203125, 0.876800537109375, 1.734588623046875, -0.964141845703125, 1.59954833984375, -0.05313873291015625, 0.23664093017578125, 0.19077301025390625, 1.5955047607421875, -0.24616241455078125, -0.6060371398925781, 0.9185791015625, -0.8718719482421875, -0.6958770751953125, 0.0211181640625, -2.060028076171875, 0.0, 0.8310089111328125, 0.3899993896484375, 0.0, 0.76348876953125, 0.0678558349609375, -0.766845703125, 0.4761199951171875, 0.9513816833496094, 0.14288330078125, 1.098297119140625, 0.913238525390625, 0.24700927734375, 0.253265380859375, 0.55792236328125, -0.29961395263671875, -0.512451171875, 0.756744384765625, 3.848419189453125, -0.0206146240234375, -0.4547271728515625, 0.0, 0.11700439453125, -1.369842529296875, -0.55621337890625, -1.4814453125, 0.084625244140625, 0.7154541015625, -0.445587158203125, -0.19964599609375, 0.25193023681640625, -2.2530517578125, -0.4404296875, 0.08852005004882812, -0.47796630859375, -0.7719573974609375, 1.14874267578125, -0.0994873046875, 0.6554832458496094, 0.108642578125, -0.7822113037109375, 0.3527336120605469, -0.937744140625, 1.291107177734375, 0.07220268249511719, 0.41887664794921875, -0.4452056884765625, 1.101806640625, -0.062255859375, 0.10144805908203125, 1.044921875, -0.398284912109375, -0.11357688903808594, -0.9576416015625, 1.3812255859375, 1.227996826171875, -1.1036148071289062, 0.27459716796875, -0.533416748046875, 0.210357666015625, 0.0, -1.3375244140625, -0.52545166015625, 0.32269287109375, 0.7737693786621094, 0.14588546752929688, 1.10601806640625, -1.596710205078125, -0.07092666625976562, 1.3468017578125, 0.5514678955078125, 0.631317138671875, -0.03207206726074219, -0.2486419677734375, 1.819549560546875, 0.82745361328125, 0.5133895874023438, -0.38068389892578125, 0.6369171142578125, 1.408203125, 0.50927734375, 0.250946044921875, 0.9191131591796875, 0.03691291809082031, -0.279632568359375, 0.2046794891357422, -0.3581085205078125, -0.8086395263671875, 0.01660919189453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000031.npy"}
{"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": 0.001748502254486084, "std": 0.970076322555542, "min": -2.5963134765625, "p10": -1.158209228515625, "median": -0.04138946533203125, "p90": 1.225537109375, "max": 3.5072021484375, "pos_frac": 0.484375, "sample": [-2.28033447265625, -0.259674072265625, -1.2713623046875, -0.554229736328125, -0.3850250244140625, -0.9981460571289062, 0.06707382202148438, 0.1176300048828125, -0.6804656982421875, 0.434600830078125, -1.2303314208984375, -0.5745315551757812, 1.3160324096679688, 0.7780075073242188, -0.9385223388671875, -1.4096183776855469, -0.355621337890625, -0.44049072265625, 0.253875732421875, 2.083831787109375, 0.27101898193359375, 0.19671630859375, 0.412567138671875, 1.1796875, -0.138641357421875, 0.91668701171875, 0.56585693359375, -0.3701953887939453, 0.1062774658203125, -0.23956298828125, -0.11993408203125, 0.3462371826171875, -0.9501953125, -1.31732177734375, 0.789306640625, -0.870941162109375, -0.460540771484375, -0.81024169921875, -0.863525390625, -0.063568115234375, 0.9501953125, -1.18597412109375, 0.2060546875, -0.0883636474609375, 0.563018798828125, 0.0962371826171875, 1.694732666015625, 1.2393798828125, 1.11279296875, 0.901153564453125, 0.843841552734375, -1.14422607421875, -0.81689453125, -1.18115234375, -0.570556640625, -0.78369140625, 3.5072021484375, 1.2196044921875, 1.535430908203125, -2.5963134765625, -0.34371185302734375, -1.1298370361328125, -0.357879638671875, -0.6463165283203125, 0.926788330078125, -1.0941009521484375, -0.3030242919921875, 0.7538909912109375, 0.90869140625, 0.690185546875, 1.77685546875, -0.0192108154296875, 0.43772125244140625, -2.1741943359375, 0.423065185546875, -0.957550048828125, -0.63165283203125, 1.35595703125, -0.241119384765625, 1.277069091796875, 0.30401611328125, -1.213623046875, 0.5298309326171875, 0.1274127960205078, 1.0076141357421875, -1.29034423828125, 1.143402099609375, -0.0920867919921875, -0.0866851806640625, -0.013916015625, 0.29001617431640625, -0.25913238525390625, -0.6202545166015625, 0.36785888671875, -0.45745849609375, 0.121124267578125, -1.13385009765625, -0.187408447265625, 0.515167236328125, -1.14837646484375, 0.00516510009765625, -0.3201904296875, -0.602691650390625, -0.16704368591308594, 0.29217529296875, -0.7785797119140625, -0.3439464569091797, -0.16873931884765625, -0.06842041015625, 0.19739532470703125, 1.59197998046875, 0.06424713134765625, 0.2773590087890625, 0.681182861328125, -0.219879150390625, -1.5928802490234375, -1.87481689453125, 1.364959716796875, -0.109283447265625, -0.9178924560546875, 0.14703369140625, 0.3928565979003906, 2.987335205078125, 1.31573486328125, 0.255401611328125, 0.567047119140625, 0.213226318359375, 0.72528076171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000032.npy"}
{"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.17780190706253052, "std": 1.0506521463394165, "min": -4.36505126953125, "p10": -0.9822814941406249, "median": 0.17222976684570312, "p90": 1.5585205078125, "max": 2.222076416015625, "pos_frac": 0.5546875, "sample": [0.20839309692382812, 0.898834228515625, 1.1882781982421875, 0.9447479248046875, -0.7162857055664062, -0.002838134765625, 0.704986572265625, -0.157623291015625, 1.1695556640625, -0.46749114990234375, -0.0916748046875, -0.4329833984375, 1.477020263671875, -1.1531982421875, 0.0, -0.3555908203125, -0.0914459228515625, 1.077545166015625, -1.5264053344726562, -0.559722900390625, -0.7119140625, -0.40540313720703125, 0.12047004699707031, -1.3468017578125, 0.866851806640625, 0.0907745361328125, -0.264617919921875, -0.5882110595703125, 0.069366455078125, 1.6448974609375, 1.482635498046875, -0.819091796875, -0.484527587890625, -0.889892578125, 0.31340789794921875, 2.053741455078125, 1.2834930419921875, 0.1988677978515625, -0.6438140869140625, -0.96600341796875, 0.38260650634765625, -0.7232437133789062, 0.17914581298828125, 1.03594970703125, -0.51287841796875, 0.0590362548828125, 0.955780029296875, -0.151519775390625, -0.28554534912109375, -1.253570556640625, 0.438812255859375, 0.7672271728515625, -2.2912750244140625, 0.0, -1.057861328125, -0.4946136474609375, -0.17246246337890625, 1.94189453125, 0.264801025390625, 1.0147857666015625, 1.64630126953125, -0.81463623046875, -0.795623779296875, 0.5397262573242188, -0.3714790344238281, -1.12860107421875, -0.2181835174560547, 0.284271240234375, -0.8909912109375, 0.8469696044921875, 0.505340576171875, 0.9225006103515625, -1.020263671875, 1.5660400390625, 0.281951904296875, -0.281524658203125, -0.654632568359375, 1.23175048828125, -0.08631134033203125, 2.0125732421875, 0.0860595703125, -1.4080047607421875, 1.7900390625, 0.25803375244140625, 1.5552978515625, -0.119110107421875, -0.4256591796875, 0.649169921875, 1.968994140625, 2.222076416015625, 1.6452484130859375, 0.3844718933105469, 1.1177597045898438, 0.5151290893554688, 0.3345527648925781, -1.39404296875, 1.2144927978515625, 0.955169677734375, -4.36505126953125, -2.7503662109375, 0.298065185546875, -0.3460540771484375, 0.7009124755859375, 1.283355712890625, -0.2353973388671875, -0.33045196533203125, -0.3766021728515625, 1.43902587890625, 0.3686676025390625, -1.798095703125, 1.72357177734375, 1.165069580078125, 1.584228515625, 1.146484375, 1.3006210327148438, -0.74688720703125, 0.6087646484375, 0.7429962158203125, 2.2039794921875, 0.37038421630859375, 0.120819091796875, 0.6300849914550781, -0.92071533203125, 0.165313720703125, -0.41461181640625, 0.4282875061035156, -0.16810989379882812, 0.79010009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000033.npy"}
{"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.1953568309545517, "std": 0.9558647871017456, "min": -2.3743896484375, "p10": -0.9490737915039061, "median": 0.1862812042236328, "p90": 1.3614990234374997, "max": 3.115966796875, "pos_frac": 0.5859375, "sample": [-0.09323883056640625, 1.5915069580078125, 1.89208984375, -0.19869232177734375, -0.488555908203125, 0.0683441162109375, -0.355133056640625, 0.2593650817871094, -0.309661865234375, 1.22796630859375, -2.3743896484375, -1.695709228515625, 1.927154541015625, -1.22259521484375, -1.2141876220703125, 1.27618408203125, 2.207275390625, 0.42050933837890625, 0.20962905883789062, -0.5141220092773438, 0.7271842956542969, -1.97674560546875, 0.462799072265625, 1.0728759765625, 1.04852294921875, 0.8101348876953125, 1.0902099609375, -0.01131439208984375, 0.062255859375, 0.29827880859375, 0.0173187255859375, 0.7405242919921875, 1.340606689453125, -0.811248779296875, 0.02740478515625, -1.2459259033203125, -1.0380859375, -0.6947784423828125, -1.047027587890625, 1.1367034912109375, 0.53857421875, -0.4680023193359375, 0.928192138671875, 2.374755859375, -0.687164306640625, -0.06206512451171875, -1.420623779296875, -0.46551513671875, 0.28192138671875, -0.606231689453125, -0.2815093994140625, 0.6315765380859375, -0.054595947265625, 1.90338134765625, -0.1337890625, 0.1514129638671875, 0.4043426513671875, -0.3458251953125, 0.76470947265625, -0.62030029296875, -0.483734130859375, 0.19391250610351562, 0.2344818115234375, 0.624664306640625, 0.542144775390625, 0.778076171875, 0.4327239990234375, -0.1043853759765625, 1.0273590087890625, 0.2010498046875, 0.17864990234375, 0.043975830078125, -1.0048828125, -0.8142242431640625, -0.3466796875, -0.5308303833007812, 0.16258621215820312, 0.5433807373046875, 0.7082977294921875, -0.3722400665283203, 3.115966796875, 0.811859130859375, 0.39923095703125, -2.2050933837890625, 0.934600830078125, -0.2943115234375, -0.05438232421875, 0.598175048828125, -0.35224151611328125, 1.410247802734375, 0.639068603515625, 0.39501953125, 1.09600830078125, -1.30084228515625, -0.9251556396484375, -2.36370849609375, -0.091156005859375, -0.4607391357421875, -0.08746337890625, 0.045307159423828125, -0.79144287109375, 0.226593017578125, 1.504119873046875, 0.0513916015625, 0.96075439453125, 0.30925750732421875, 0.5497283935546875, -0.05841827392578125, -0.40447235107421875, 1.82818603515625, 0.5510711669921875, -0.74822998046875, 0.267791748046875, -0.67431640625, 1.561553955078125, -0.37917327880859375, -0.088623046875, 1.64501953125, -0.08831787109375, 1.18328857421875, 1.278472900390625, 1.6654052734375, 0.1005859375, 0.850860595703125, 0.62646484375, 0.32535552978515625, 0.70660400390625, 1.2647705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000034.npy"}
{"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": 0.04421742260456085, "std": 1.0081171989440918, "min": -3.6048583984375, "p10": -1.1899765014648438, "median": 0.11786842346191406, "p90": 1.1218200683593749, "max": 2.824798583984375, "pos_frac": 0.5390625, "sample": [-0.39705657958984375, -0.33819580078125, -0.968231201171875, 1.117919921875, 0.40838623046875, 0.800048828125, 1.13092041015625, 0.2352752685546875, -0.350860595703125, -1.2141265869140625, 1.79412841796875, 0.16222381591796875, -0.16156005859375, 0.3026885986328125, 0.7891693115234375, 0.2528648376464844, -2.723876953125, -1.23553466796875, 0.300018310546875, -0.17750167846679688, 0.88751220703125, 0.441680908203125, -0.45758056640625, 0.7880401611328125, -0.232513427734375, 0.720794677734375, 0.80517578125, -0.121368408203125, 0.9235076904296875, -1.893310546875, 0.3524131774902344, -0.5278167724609375, -0.349273681640625, 0.3007965087890625, 1.099273681640625, 2.50299072265625, 0.16516876220703125, 0.450103759765625, 0.6291255950927734, 0.0, 0.297943115234375, 0.95306396484375, 0.16129302978515625, 0.273773193359375, 0.6383132934570312, 0.15069580078125, 2.824798583984375, -0.8698577880859375, 0.269866943359375, -1.687957763671875, -0.672943115234375, -0.16584014892578125, 0.2286376953125, -1.41326904296875, 0.94384765625, -0.29785919189453125, 1.5609130859375, -0.552337646484375, 0.5520401000976562, -0.2512950897216797, -0.78155517578125, 1.18511962890625, 1.620941162109375, 0.44068145751953125, 1.2342529296875, 0.67822265625, -0.15625, 0.2691936492919922, 1.176300048828125, -0.13393211364746094, -1.243072509765625, -1.03021240234375, -0.9377899169921875, 0.57122802734375, -0.3226585388183594, -0.272491455078125, -0.6659698486328125, 0.4052276611328125, -0.318450927734375, -1.7008056640625, 1.0622329711914062, 0.01145172119140625, -1.17962646484375, -0.00843048095703125, 0.8334503173828125, 0.06146240234375, 0.029664993286132812, -0.9629135131835938, 0.87139892578125, 0.3823738098144531, -2.1865234375, 0.069671630859375, -0.10975074768066406, 0.7579116821289062, -0.011156082153320312, -0.3897705078125, 0.4503173828125, -2.3953857421875, 0.1301727294921875, 2.4429931640625, -1.821319580078125, 1.626922607421875, 0.0, 0.7270355224609375, -0.289520263671875, 0.5372467041015625, 0.4746551513671875, 1.32373046875, 1.080718994140625, -0.161834716796875, -0.245880126953125, -1.038238525390625, -0.845184326171875, -0.325592041015625, 0.50994873046875, -0.7833251953125, -0.0160675048828125, -0.25469970703125, -0.796630859375, 0.10556411743164062, 0.8421630859375, 1.606536865234375, -0.8445587158203125, -1.76104736328125, 0.7680625915527344, -3.6048583984375, -1.169036865234375, 0.982269287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000035.npy"}
{"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.4041725695133209, "std": 1.097532868385315, "min": -2.41741943359375, "p10": -0.6654098510742188, "median": 0.2816123962402344, "p90": 1.5002532958984374, "max": 4.13861083984375, "pos_frac": 0.65625, "sample": [1.3836669921875, -1.50543212890625, 0.264007568359375, -0.4392852783203125, -0.610198974609375, 1.2613525390625, 1.9224853515625, 0.1226959228515625, 1.025543212890625, -0.4467582702636719, 0.266937255859375, 0.2234954833984375, 2.22369384765625, -0.633270263671875, -0.2939910888671875, 0.770599365234375, -0.46448516845703125, 0.209136962890625, 0.5413818359375, 0.36783599853515625, 0.7206649780273438, 0.29693603515625, 0.20291900634765625, 0.0, 0.1468505859375, 0.50970458984375, 0.383514404296875, -0.29583740234375, 0.3899078369140625, 0.14065170288085938, 2.87725830078125, -0.1587066650390625, 1.001556396484375, 1.05810546875, 1.538726806640625, 0.035552978515625, 0.06313323974609375, -0.397430419921875, 0.2099132537841797, 0.735382080078125, 3.800537109375, -0.12448883056640625, 0.29628753662109375, -0.2207489013671875, -0.97869873046875, -0.266143798828125, 0.86944580078125, 2.9577789306640625, -0.85888671875, 0.87042236328125, 1.1815185546875, 0.314483642578125, 1.6588134765625, -0.42668914794921875, -1.068267822265625, 0.78546142578125, 0.526641845703125, 0.2148895263671875, 4.13861083984375, 0.674591064453125, 0.0, -1.73687744140625, 1.034912109375, -0.11163330078125, 2.3037109375, 1.4102783203125, -0.009449005126953125, 0.732940673828125, -0.49616241455078125, -0.0878448486328125, 1.43341064453125, 3.930633544921875, 1.013580322265625, 0.59417724609375, -0.6648101806640625, 0.23302841186523438, -0.20182037353515625, 0.386199951171875, -2.332366943359375, -0.818603515625, 1.4837646484375, -0.0914154052734375, 2.64697265625, 1.0272979736328125, -2.41741943359375, 0.013738632202148438, 0.3223876953125, -0.5303497314453125, -0.58648681640625, -0.00206756591796875, 2.04595947265625, 0.50616455078125, 0.8810272216796875, 0.31903076171875, -1.350189208984375, -1.95697021484375, -0.2933483123779297, 0.69427490234375, -0.4227752685546875, 0.9183349609375, 1.1094970703125, 1.449310302734375, 0.9881591796875, -0.0012969970703125, -0.9468994140625, 0.414825439453125, 0.779693603515625, 0.0048828125, 0.441650390625, 0.4749755859375, 0.038372039794921875, 0.13426971435546875, -0.1244354248046875, 0.63177490234375, -0.5541458129882812, -0.138946533203125, 0.121063232421875, 0.405181884765625, 1.4383544921875, -0.9697723388671875, 0.3896484375, 3.06231689453125, 1.1978302001953125, 0.17418670654296875, 0.2542724609375, 0.39572906494140625, 1.4153900146484375, -0.66680908203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000036.npy"}
{"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.16348886489868164, "std": 0.9148265719413757, "min": -2.4455718994140625, "p10": -0.9608367919921875, "median": 0.1529526710510254, "p90": 1.1579383850097653, "max": 3.47393798828125, "pos_frac": 0.5625, "sample": [-0.4454803466796875, 0.0742034912109375, 1.923187255859375, 0.7196044921875, 1.386993408203125, -0.08869743347167969, 1.5992431640625, 0.16219711303710938, 0.740631103515625, -0.719024658203125, 1.595703125, 0.1536426544189453, 0.46722412109375, 0.9477767944335938, 0.0806884765625, 1.691162109375, -0.0689849853515625, 0.3953132629394531, 1.011566162109375, -0.92987060546875, 0.5269927978515625, -0.32391357421875, -0.2934417724609375, -0.956085205078125, -0.32342529296875, 0.5970611572265625, -0.20265579223632812, -1.10235595703125, 0.6442947387695312, 0.10223388671875, 1.098785400390625, -0.6746177673339844, -0.2673835754394531, 0.523681640625, 0.4676513671875, 1.1229248046875, -1.576690673828125, 3.31793212890625, 0.32623291015625, -0.0762176513671875, -0.3643531799316406, -0.12860107421875, 0.03961944580078125, 0.55291748046875, 0.6328277587890625, 1.211456298828125, 0.15226268768310547, 0.211883544921875, 0.7983245849609375, -0.465606689453125, 1.4453125, 0.7780914306640625, -1.6353759765625, 1.04180908203125, -0.53387451171875, -0.38616943359375, 0.3857269287109375, 0.99267578125, -0.2290191650390625, -0.971923828125, 0.46337890625, 0.71832275390625, 0.43984222412109375, -1.182586669921875, 0.2220458984375, -0.17725372314453125, -1.7564697265625, 0.41448974609375, 0.417236328125, 1.49237060546875, 1.1060791015625, 0.7061004638671875, 0.256195068359375, 1.1321563720703125, 0.282196044921875, 0.772552490234375, 0.0503997802734375, 0.614013671875, -0.159942626953125, -1.2369232177734375, 0.49481964111328125, 1.56280517578125, 1.1350021362304688, -1.2119598388671875, -0.4326171875, -0.264862060546875, -0.502349853515625, -0.1466217041015625, 1.459136962890625, -1.45074462890625, 0.085968017578125, 0.4222412109375, -0.622589111328125, -0.6826171875, -0.1095123291015625, -0.061489105224609375, 0.5843505859375, -0.021724700927734375, -0.6648941040039062, 0.416015625, -0.768524169921875, -0.09116744995117188, -1.324676513671875, -0.3197669982910156, 0.3261451721191406, -0.13568115234375, 3.47393798828125, -0.378448486328125, 0.2436676025390625, 0.11093521118164062, -0.517822265625, -0.89697265625, -0.0029296875, -0.630157470703125, 1.04296875, 0.175048828125, 0.26324462890625, -1.1658477783203125, -0.39568328857421875, -0.8095703125, -0.14105224609375, -2.4455718994140625, 0.886138916015625, 0.5108642578125, 0.55303955078125, 0.8765411376953125, -1.193572998046875, 1.962860107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000037.npy"}
{"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.13182410597801208, "std": 1.0418591499328613, "min": -3.47705078125, "p10": -1.0541648864746094, "median": 0.16418838500976562, "p90": 1.3795700073242185, "max": 2.742950439453125, "pos_frac": 0.5546875, "sample": [0.9781494140625, 0.93280029296875, -0.4338550567626953, -0.0193634033203125, -0.23731231689453125, -0.032623291015625, -0.38901519775390625, 0.404815673828125, 0.795074462890625, 1.0859375, 1.904571533203125, 0.37762451171875, -0.031494140625, -0.340576171875, 0.0953369140625, 0.742584228515625, 1.2449188232421875, -0.8606338500976562, -0.9083709716796875, 0.3201904296875, -0.307586669921875, -0.9609375, 0.5986957550048828, 0.22682952880859375, -0.22467803955078125, -1.7320709228515625, -1.243927001953125, 0.0699462890625, -0.0901336669921875, -0.12248420715332031, -0.4740753173828125, 0.94110107421875, 0.577484130859375, -3.47705078125, -0.3757781982421875, 0.3710784912109375, 0.5794830322265625, -0.11379814147949219, 0.362823486328125, -1.741668701171875, -0.429718017578125, -0.1956787109375, 0.755157470703125, -0.254608154296875, 0.4053955078125, 0.6871719360351562, -0.428253173828125, -0.097259521484375, 0.164520263671875, 0.7682952880859375, 0.1517333984375, 0.21239471435546875, -1.309326171875, -0.377166748046875, 0.3700447082519531, -0.3542938232421875, 0.40399169921875, 0.4547119140625, -0.285919189453125, -0.02978515625, 0.376068115234375, -2.20672607421875, 1.67529296875, -0.402587890625, 0.26702880859375, 0.22050857543945312, 1.2008819580078125, -2.7262115478515625, 1.2322845458984375, -0.022308349609375, 1.51953125, 0.258453369140625, 0.346435546875, -0.11346435546875, -0.09716796875, -0.879150390625, 2.5526123046875, -0.19752883911132812, -1.0996932983398438, -3.241943359375, 0.6474151611328125, -0.06107330322265625, -0.485137939453125, 1.75311279296875, 0.0895233154296875, 1.201202392578125, 1.987548828125, -0.2859649658203125, 0.54254150390625, 0.16385650634765625, 0.9231185913085938, -1.0346527099609375, 1.82025146484375, 0.3778839111328125, 1.236053466796875, 0.0218963623046875, 0.3648529052734375, 1.4634857177734375, 1.324371337890625, 0.684783935546875, 0.5505599975585938, 0.0, -0.0876007080078125, -2.1546630859375, 0.071624755859375, 0.741790771484375, 2.742950439453125, 1.566375732421875, 1.7271728515625, -1.714813232421875, 0.947265625, -1.224151611328125, 0.5813751220703125, -0.241424560546875, 1.521728515625, -0.3382568359375, 1.4410400390625, 1.196807861328125, 0.19195556640625, -0.464599609375, -0.44482421875, -0.175445556640625, -2.5787353515625, 0.1898040771484375, 0.8275146484375, 1.3532257080078125, -0.19202423095703125, 0.3380279541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000038.npy"}
{"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 0.3903595805168152, "std": 0.959074079990387, "min": -3.268096923828125, "p10": -0.7176071166992187, "median": 0.2699432373046875, "p90": 1.5787353515625, "max": 3.4337158203125, "pos_frac": 0.671875, "sample": [-0.223052978515625, -0.5916748046875, -1.646942138671875, -0.7739791870117188, 0.395233154296875, 0.166839599609375, -0.2598724365234375, 0.1798248291015625, -3.268096923828125, 0.9112548828125, 0.852752685546875, 1.687652587890625, 2.11407470703125, 0.578277587890625, 1.99346923828125, 0.7080078125, -0.2529296875, 1.577911376953125, -0.5132293701171875, 0.0439453125, -0.9610443115234375, 0.98541259765625, 0.19535064697265625, 0.06915283203125, 2.96673583984375, 0.65374755859375, 0.02347564697265625, -0.6663818359375, -0.9453277587890625, -0.51519775390625, 3.4337158203125, 1.037445068359375, 0.214874267578125, 0.0992431640625, 0.3271484375, 1.15155029296875, -0.20398330688476562, 0.41364288330078125, 1.412078857421875, 0.5297622680664062, 0.3380126953125, 0.20206069946289062, 0.4189300537109375, 0.7950859069824219, -0.08565139770507812, 0.51123046875, 2.23211669921875, -0.061305999755859375, 0.7198638916015625, 1.74749755859375, -0.253326416015625, -0.953948974609375, 1.0208282470703125, 0.54205322265625, 1.4434356689453125, -0.2424468994140625, -0.683380126953125, 1.416656494140625, 1.674072265625, 0.4931640625, -0.5153846740722656, -0.4342498779296875, -0.746826171875, 1.580657958984375, 1.6480331420898438, 1.1568756103515625, -0.0661468505859375, 0.38137245178222656, 1.3986358642578125, -0.6650543212890625, 0.25384521484375, 0.427734375, 0.023965835571289062, -1.0760498046875, 0.286041259765625, 1.158477783203125, 0.04571533203125, 0.560089111328125, 0.8930892944335938, -0.402801513671875, 0.686676025390625, 0.10369110107421875, -0.096099853515625, 1.19921875, 0.739990234375, 0.009063720703125, 2.358917236328125, 1.996337890625, 0.2023468017578125, 0.014682769775390625, 0.324798583984375, -0.7162017822265625, -0.72088623046875, -0.3615875244140625, 0.0, -0.295379638671875, 0.3369636535644531, -0.737213134765625, 1.2413330078125, -0.235626220703125, 1.1763763427734375, 0.207183837890625, 1.140472412109375, -0.7872161865234375, 0.889617919921875, 1.00927734375, -0.862640380859375, 0.6260833740234375, 0.468475341796875, -0.6888427734375, 0.38555908203125, -0.08737945556640625, 0.029466629028320312, 0.548370361328125, -0.02703857421875, 0.10922622680664062, 1.339599609375, 1.308990478515625, 0.966796875, 0.24868202209472656, -0.7265472412109375, 0.038860321044921875, 1.5006866455078125, 3.35516357421875, 0.140106201171875, -0.2156524658203125, -0.10684776306152344, 0.8443374633789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000039.npy"}
{"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.42327409982681274, "std": 1.0472114086151123, "min": -2.517486572265625, "p10": -0.6923210144042968, "median": 0.2962493896484375, "p90": 1.6028717041015625, "max": 4.05462646484375, "pos_frac": 0.6484375, "sample": [0.07295417785644531, -0.963653564453125, -0.70867919921875, 1.265167236328125, -0.14789581298828125, 0.7048492431640625, -2.517486572265625, 1.5986328125, -0.0541839599609375, 2.45281982421875, 0.579498291015625, 0.6793594360351562, -0.162750244140625, 0.99798583984375, 0.429718017578125, 0.7311630249023438, -0.038211822509765625, 0.371551513671875, 0.1569805145263672, 0.674530029296875, 0.17267608642578125, 1.4840087890625, -0.18885040283203125, 0.13528060913085938, 0.2769622802734375, 1.93878173828125, 0.58380126953125, 0.9679107666015625, -0.205078125, 0.6528472900390625, -0.9266128540039062, 0.2751312255859375, -0.36597442626953125, 1.11688232421875, 0.4705810546875, 1.238433837890625, 0.2143096923828125, -0.680145263671875, 2.9183349609375, 2.57666015625, 3.85028076171875, 1.43182373046875, -1.10296630859375, -0.56414794921875, -0.034969329833984375, -0.3061027526855469, 1.383148193359375, 0.3551788330078125, 1.74017333984375, 1.3723907470703125, 0.13628387451171875, 0.4864501953125, 1.28045654296875, -0.0252685546875, 2.046417236328125, -0.1851806640625, 0.16668701171875, 0.5428085327148438, 4.05462646484375, -0.271881103515625, 0.818450927734375, 0.2600555419921875, -0.103118896484375, -1.143829345703125, 0.44793701171875, 1.296783447265625, -0.32757568359375, 0.789398193359375, 1.612762451171875, -0.10214614868164062, -0.6853103637695312, -0.35965728759765625, 0.7033538818359375, 1.523681640625, 0.732513427734375, -0.6224174499511719, 1.4166259765625, 0.5039482116699219, -0.0287017822265625, 1.423828125, -0.3476409912109375, -0.1102294921875, 0.19793701171875, 0.007572174072265625, 0.047637939453125, 0.8419380187988281, -1.0234832763671875, 2.214691162109375, 0.2173004150390625, -1.3542938232421875, -0.14251708984375, 0.280029296875, -0.427703857421875, 1.023040771484375, -1.764801025390625, 0.9586334228515625, -0.06195068359375, -1.3524169921875, 0.638336181640625, 0.5783462524414062, 1.0296478271484375, 1.0164947509765625, -0.255218505859375, 0.1075592041015625, -0.1510772705078125, 1.45648193359375, 0.515594482421875, 0.04952049255371094, -0.101165771484375, 0.617767333984375, -2.1641845703125, 0.866302490234375, 2.2659912109375, 0.257354736328125, -0.074920654296875, -0.63836669921875, -0.79644775390625, 0.5892486572265625, 1.65069580078125, -0.2069091796875, 0.985626220703125, 0.2627239227294922, 0.410552978515625, 0.312469482421875, 1.279449462890625, 2.0638885498046875, -2.134765625, 1.253265380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000040.npy"}
{"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.20153623819351196, "std": 1.096157431602478, "min": -3.53082275390625, "p10": -1.2672515869140624, "median": 0.1945343017578125, "p90": 1.5371948242187499, "max": 3.267333984375, "pos_frac": 0.578125, "sample": [2.785919189453125, 0.936187744140625, 0.203582763671875, -0.5227584838867188, -0.020233154296875, -1.496551513671875, -0.243316650390625, 0.1100921630859375, 0.187469482421875, 0.0523834228515625, 0.43003082275390625, 1.727264404296875, 1.0033493041992188, -0.5879669189453125, -0.926788330078125, -0.127349853515625, 0.1716461181640625, 0.486785888671875, 0.653045654296875, -0.26885986328125, 0.0819854736328125, 1.92047119140625, -0.38482666015625, 1.1930694580078125, -0.316375732421875, 0.2578125, 0.204742431640625, -0.30517578125, 0.3939666748046875, -0.10216331481933594, 1.899261474609375, -1.152618408203125, -0.43299102783203125, 1.376312255859375, 0.3859710693359375, -0.23542213439941406, 2.05615234375, -0.594329833984375, 0.523834228515625, 1.110595703125, 0.50054931640625, 0.8559646606445312, 0.2624664306640625, 1.700469970703125, 1.0485382080078125, 0.013214111328125, -2.69842529296875, -0.15106201171875, -1.3338623046875, -2.123809814453125, -0.2229766845703125, -0.068817138671875, 0.341064453125, -0.35794830322265625, 3.267333984375, -1.7041015625, 0.32984161376953125, 0.5384521484375, 1.2166748046875, 0.7847900390625, -0.0780181884765625, -0.183197021484375, 0.8838653564453125, -0.44818115234375, 0.386444091796875, 0.88348388671875, -0.35308837890625, 0.1423187255859375, 0.0, -0.826202392578125, -0.0052032470703125, 0.50958251953125, -1.289154052734375, -0.2469940185546875, -3.53082275390625, -2.2403564453125, 1.2440261840820312, -0.337677001953125, 0.82562255859375, -1.287994384765625, 0.9467010498046875, 1.0642967224121094, 0.9437789916992188, 0.9932632446289062, 0.764984130859375, 0.6092071533203125, 0.9693069458007812, 1.230712890625, 2.2459564208984375, 0.37401580810546875, -0.461822509765625, 0.84283447265625, 1.2410812377929688, -0.1051788330078125, -0.584991455078125, 1.598297119140625, -0.1549072265625, -0.335052490234375, 0.06131458282470703, -1.3994293212890625, 1.552490234375, 1.5306396484375, -0.247222900390625, -0.27695465087890625, -0.9130859375, 0.5336990356445312, 1.0016098022460938, -0.39484405517578125, 1.252105712890625, 0.20159912109375, 0.8449230194091797, 0.038848876953125, -0.468780517578125, -1.831695556640625, 1.9068603515625, -0.49041748046875, 0.53033447265625, -2.10211181640625, 0.470123291015625, -0.341094970703125, -1.25836181640625, 0.04620361328125, 1.64935302734375, 0.67608642578125, 2.4248809814453125, 1.5078125, 0.815155029296875, -2.38690185546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000041.npy"}
{"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 0.32515278458595276, "std": 1.0887075662612915, "min": -2.842803955078125, "p10": -0.8679878234863281, "median": 0.2975502014160156, "p90": 1.6965377807617186, "max": 3.5048828125, "pos_frac": 0.6328125, "sample": [-0.15186309814453125, 0.34747314453125, 1.8304443359375, 1.03704833984375, 0.450469970703125, 0.167724609375, -0.9152679443359375, 3.22430419921875, 0.025115966796875, 0.0425872802734375, 1.1075439453125, 3.3074951171875, 0.09124755859375, 0.31331634521484375, -1.70941162109375, -0.23975753784179688, 0.781890869140625, 0.0986328125, 1.1794509887695312, 2.548919677734375, -0.0050201416015625, -0.06175994873046875, 0.296295166015625, -0.0852203369140625, 1.48602294921875, -0.4668121337890625, -0.23797607421875, 0.3957061767578125, 2.00286865234375, -0.3314361572265625, -1.5792388916015625, 0.8090724945068359, -0.08013916015625, 0.5283203125, 1.396759033203125, 0.9295501708984375, 1.6812896728515625, 0.29880523681640625, -0.374359130859375, 0.452056884765625, -0.37308502197265625, -0.17151641845703125, 1.113067626953125, -0.417999267578125, -0.0379638671875, -0.50921630859375, 0.745147705078125, 0.23406982421875, 0.1875, -0.2628345489501953, 0.6476898193359375, 0.34552574157714844, 1.0310287475585938, 0.8640899658203125, -0.1968841552734375, -0.02243804931640625, -0.292236328125, -0.47039794921875, 0.4018707275390625, 0.49759674072265625, -0.223388671875, -2.842803955078125, -1.1467132568359375, -0.8537216186523438, 0.416168212890625, 3.215576171875, 0.8177642822265625, 0.0925140380859375, -0.46231842041015625, 0.508575439453125, -0.347747802734375, 0.3555908203125, -0.6002197265625, 1.810333251953125, -1.2698974609375, 0.2071075439453125, -1.507476806640625, -0.600860595703125, 0.4530029296875, 0.043682098388671875, 0.868072509765625, -0.04473876953125, 0.237518310546875, -0.2200927734375, 0.441436767578125, 0.98919677734375, 2.555938720703125, 0.9706878662109375, 0.4607219696044922, 0.073822021484375, 0.8203277587890625, -1.528350830078125, 2.18707275390625, 0.2955780029296875, 1.39532470703125, 0.88604736328125, -1.36077880859375, 3.5048828125, 0.34051513671875, 0.6670074462890625, 0.7872161865234375, -0.6213531494140625, 0.0, 0.4189453125, -2.22430419921875, -0.22097396850585938, 1.35955810546875, -0.005401611328125, 0.3899497985839844, 0.4762420654296875, 1.73211669921875, -0.31732177734375, 0.628662109375, 0.2649574279785156, 1.0960693359375, 1.06573486328125, 0.330078125, 0.38793182373046875, 1.8055801391601562, 0.181121826171875, -0.901275634765625, 2.917022705078125, -2.361297607421875, 0.612335205078125, -2.18365478515625, 0.05615234375, -0.071929931640625, 0.5068740844726562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000042.npy"}
{"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 0.5280700922012329, "std": 1.187109112739563, "min": -3.944915771484375, "p10": -0.7764801025390624, "median": 0.4038581848144531, "p90": 2.0736572265625, "max": 3.73614501953125, "pos_frac": 0.6796875, "sample": [0.404296875, 1.5374755859375, 1.411712646484375, -0.69158935546875, 0.8056488037109375, 1.8602447509765625, 2.9322509765625, 0.09874725341796875, 0.47560882568359375, 0.7442626953125, 0.0841827392578125, -0.6773223876953125, 0.255828857421875, 3.15625, -0.99774169921875, -0.75103759765625, -0.110260009765625, -0.089813232421875, 0.473602294921875, 0.34222412109375, 3.06451416015625, 0.39168548583984375, 1.4493408203125, -0.1884918212890625, 0.4459228515625, -1.9908905029296875, -1.02197265625, 1.4635772705078125, 0.12309837341308594, 1.2811279296875, 0.5859375, 0.6016998291015625, -1.880615234375, -3.944915771484375, 1.876220703125, 0.289093017578125, 0.2531280517578125, 0.7196578979492188, 0.3720226287841797, 1.1509933471679688, -0.1509857177734375, 2.2979736328125, -0.08774185180664062, 1.211456298828125, 0.6135711669921875, 0.05648231506347656, 0.076904296875, -0.37306976318359375, -2.8572998046875, 1.0105133056640625, -0.835845947265625, 0.764801025390625, -0.1753253936767578, -0.23903656005859375, 1.97052001953125, 2.27490234375, 2.0743408203125, 0.94342041015625, -0.5882568359375, 0.64501953125, 0.8288726806640625, -0.34979248046875, 2.0733642578125, 1.2552032470703125, 2.14495849609375, -0.932891845703125, 0.666473388671875, -1.3638916015625, 1.4583282470703125, 1.6190338134765625, 0.198944091796875, 1.72332763671875, 2.2150421142578125, 1.235931396484375, 0.832550048828125, 0.15780258178710938, -1.118377685546875, -0.27972412109375, -0.01303863525390625, 2.395538330078125, 0.06089019775390625, 1.5985107421875, -0.0867919921875, -0.491180419921875, 0.2582550048828125, 1.143402099609375, 0.7193603515625, 0.330963134765625, -0.6217041015625, 0.6294479370117188, 2.43585205078125, 1.91033935546875, -0.0975341796875, 1.0753021240234375, 2.1884765625, -0.3847198486328125, 0.05246734619140625, 3.73614501953125, 0.091033935546875, 0.40341949462890625, -1.84027099609375, 1.6023483276367188, 1.85382080078125, -0.037700653076171875, -0.025390625, 0.70867919921875, 1.750457763671875, 1.2708740234375, 2.13897705078125, 1.85272216796875, 0.228118896484375, -1.10894775390625, 0.229248046875, -1.3617172241210938, 0.02783203125, 1.1760711669921875, 1.216339111328125, 0.888458251953125, 0.698211669921875, 1.0653076171875, -0.1696624755859375, -0.06689453125, -0.3253650665283203, 1.638092041015625, -0.197906494140625, -0.555023193359375, 0.2989654541015625, -0.00031280517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000043.npy"}
{"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 0.5000274181365967, "std": 1.203985333442688, "min": -4.162933349609375, "p10": -0.7942718505859375, "median": 0.45242881774902344, "p90": 1.838191223144531, "max": 3.68878173828125, "pos_frac": 0.734375, "sample": [0.34375, 0.135589599609375, 1.248291015625, 0.3371734619140625, -0.3359527587890625, 2.5953369140625, 0.3505401611328125, 0.4735832214355469, 1.021820068359375, 0.12122917175292969, 0.5513153076171875, 1.145477294921875, 0.774688720703125, -0.35626983642578125, 0.2930908203125, 1.1531219482421875, 0.0980072021484375, 0.31987762451171875, -0.511962890625, 0.90130615234375, 1.53857421875, 1.354248046875, 0.43988037109375, 0.5534782409667969, 0.5391368865966797, 0.27103424072265625, 0.97821044921875, 0.24860191345214844, -4.162933349609375, -0.25156402587890625, 1.8217620849609375, -0.70849609375, -0.920257568359375, -1.6871337890625, -0.06569290161132812, 0.3658447265625, 1.526458740234375, -0.4689159393310547, 3.42681884765625, 1.190765380859375, 3.68878173828125, -0.19519424438476562, 0.1256256103515625, 2.1605224609375, 1.414947509765625, 1.636016845703125, 1.289794921875, 2.40338134765625, 1.2133026123046875, -0.20806884765625, 0.03329658508300781, 0.468017578125, -1.640045166015625, -0.563720703125, 0.6078262329101562, 0.60345458984375, -1.7032470703125, 1.2752685546875, 0.4649772644042969, 0.34539794921875, 0.6118011474609375, 1.422149658203125, 0.6183967590332031, 0.869140625, 1.24017333984375, 0.9224700927734375, -0.33251953125, -1.303558349609375, 1.87652587890625, 3.115509033203125, -0.811126708984375, 0.7442245483398438, -0.215850830078125, 0.089080810546875, 1.4755859375, 0.9776763916015625, 1.79168701171875, -2.107086181640625, 0.049713134765625, -1.864044189453125, 1.081634521484375, 0.2105865478515625, 2.655792236328125, 0.4391803741455078, 0.1402587890625, 0.5758209228515625, 0.070343017578125, 0.8370361328125, 1.9010009765625, -0.121246337890625, -0.6668701171875, 2.3018798828125, -0.03631591796875, 2.1339111328125, 0.269989013671875, 1.175262451171875, 0.013336181640625, -1.350830078125, 0.0698394775390625, -0.78704833984375, 0.27685546875, 0.0023479461669921875, -1.5621185302734375, 1.43157958984375, 0.32564544677734375, 0.93951416015625, -0.0330810546875, 1.42340087890625, -0.41576385498046875, -0.186309814453125, 3.008758544921875, 1.450469970703125, 0.11655426025390625, 3.5618896484375, 0.7426910400390625, 1.53009033203125, 0.164520263671875, 0.112091064453125, -2.6341552734375, -0.379150390625, -1.4607391357421875, 0.7644882202148438, 0.6583251953125, 1.480316162109375, 0.677001953125, -0.6102142333984375, 0.9113235473632812, 1.553497314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000044.npy"}
{"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 0.4916006624698639, "std": 1.4031749963760376, "min": -3.892822265625, "p10": -1.0104614257812499, "median": 0.34033203125, "p90": 2.403822326660156, "max": 6.2027587890625, "pos_frac": 0.6328125, "sample": [-1.52899169921875, 0.3789825439453125, 1.056304931640625, -0.3922271728515625, -0.43064308166503906, 0.903656005859375, 0.1503276824951172, -0.060150146484375, 0.7076187133789062, 1.73980712890625, 1.47015380859375, 2.931121826171875, 2.61102294921875, -0.16828155517578125, 0.788330078125, 0.5787277221679688, -0.20825576782226562, 1.14794921875, 0.0, -2.60302734375, 1.695709228515625, 0.3374786376953125, 0.7483062744140625, 0.6296005249023438, -0.544219970703125, 3.68310546875, -0.7990493774414062, 0.15728759765625, 1.4395904541015625, 0.2842864990234375, -0.537109375, -0.3111000061035156, 0.32720947265625, -0.1025390625, 0.13372802734375, 0.8887939453125, 0.155517578125, -0.008998870849609375, 2.779876708984375, 1.61553955078125, 2.072845458984375, -1.7330322265625, 0.17070770263671875, 0.2718658447265625, 2.3822021484375, 0.5216331481933594, 0.640167236328125, 0.3431854248046875, -0.71453857421875, 1.3488693237304688, 0.5555267333984375, 0.37225341796875, -1.155029296875, 0.2147979736328125, -0.12569046020507812, 0.1282958984375, -0.03287506103515625, 0.7870635986328125, -0.30075836181640625, -3.892822265625, -3.05810546875, 1.852691650390625, -0.6345672607421875, -0.8038177490234375, -1.19427490234375, 0.370819091796875, 0.933013916015625, -0.6030654907226562, 0.991485595703125, 0.3203697204589844, 0.522216796875, 3.2405853271484375, -0.5029296875, 0.6531600952148438, 2.564483642578125, 1.0411834716796875, 0.796600341796875, 1.334747314453125, 1.49835205078125, 0.6240234375, 0.7506866455078125, 3.55078125, 0.4839134216308594, 2.4542694091796875, 0.29522705078125, -1.552703857421875, -1.245513916015625, -0.29035186767578125, -1.0863723754882812, -1.547821044921875, 0.3123321533203125, 3.483673095703125, -0.2925872802734375, -1.5843963623046875, 3.6942138671875, -0.16934967041015625, 1.1436767578125, -0.008453369140625, 1.8551025390625, -0.91290283203125, -0.17095947265625, -0.171783447265625, -0.21018218994140625, -0.5789947509765625, 1.28302001953125, 0.7269287109375, 1.0268707275390625, -1.7768096923828125, 0.17351531982421875, 0.6895942687988281, -0.050872802734375, 0.8264007568359375, 1.2258758544921875, 0.3148193359375, -0.2503471374511719, 1.014892578125, 1.7292022705078125, 0.0518798828125, 6.2027587890625, 1.4420166015625, 0.61785888671875, 3.59857177734375, 0.5373878479003906, -0.2929840087890625, 3.044921875, -0.9779281616210938, -0.39849853515625, 0.5172271728515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000045.npy"}
{"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 0.7308738827705383, "std": 1.3830276727676392, "min": -3.302734375, "p10": -0.6024993896484373, "median": 0.525238037109375, "p90": 2.5790191650390626, "max": 4.60821533203125, "pos_frac": 0.6953125, "sample": [2.20953369140625, 4.2633056640625, 1.333831787109375, 0.30987548828125, 0.05279541015625, 0.792694091796875, 2.875030517578125, 1.942138671875, -0.333831787109375, -0.724029541015625, -1.0657196044921875, 0.1258544921875, 1.85357666015625, -0.5504150390625, 1.2740631103515625, -0.20294189453125, -0.1375732421875, -0.10063552856445312, 1.618377685546875, 0.5584716796875, 1.728363037109375, 1.2430267333984375, -0.47021484375, 1.0694580078125, -1.200164794921875, 0.154632568359375, -0.2515869140625, -0.3406486511230469, 4.58587646484375, -0.23533058166503906, -0.037952423095703125, 0.5456771850585938, 0.3731536865234375, 0.912353515625, 0.927886962890625, 3.06146240234375, 2.42767333984375, 0.1229095458984375, -0.0844268798828125, 0.229644775390625, -0.265594482421875, 0.7166748046875, 0.8631362915039062, 0.910675048828125, 2.61187744140625, -3.01251220703125, 1.261077880859375, -0.3824348449707031, 0.3682403564453125, 0.5039825439453125, 3.458160400390625, 1.4568634033203125, 0.6471099853515625, 0.00048828125, -0.4103546142578125, 2.492828369140625, 1.82330322265625, 4.4788818359375, 0.16904449462890625, 1.43939208984375, -0.05895233154296875, 0.7591686248779297, -0.22369384765625, -0.2026519775390625, 2.10089111328125, -0.22553253173828125, 0.08936119079589844, -1.508819580078125, -0.449737548828125, 0.69635009765625, 0.545135498046875, 0.5665283203125, 2.3487548828125, -0.3387908935546875, -1.2479400634765625, 1.0655517578125, -1.0298690795898438, 0.0970916748046875, 0.38288307189941406, -0.834014892578125, 0.6718521118164062, 0.584381103515625, 0.715576171875, -1.17919921875, 0.670318603515625, 0.4653472900390625, -3.302734375, 0.16953277587890625, 0.59906005859375, 1.3770751953125, -0.35150146484375, 1.605255126953125, 2.57415771484375, 0.436431884765625, -0.014209747314453125, 3.5325927734375, 0.14976119995117188, -0.9869537353515625, 0.3023681640625, 0.9076995849609375, -0.2973003387451172, 1.5295257568359375, 1.72247314453125, -0.129730224609375, 2.863525390625, 1.1378097534179688, 0.0677947998046875, 0.510101318359375, 2.590362548828125, 0.540374755859375, 3.105010986328125, -1.07440185546875, 0.39476776123046875, 0.7008724212646484, -0.2986030578613281, 4.308837890625, 0.2432861328125, 0.9391250610351562, 1.83221435546875, 2.35247802734375, 0.0, 2.376007080078125, 0.07530975341796875, -0.8176727294921875, 0.13983154296875, 4.60821533203125, 1.56494140625, 1.1211395263671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000046.npy"}
{"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 0.3561379313468933, "std": 1.5416926145553589, "min": -4.767303466796875, "p10": -1.1912933349609374, "median": 0.2872123718261719, "p90": 2.2428131103515625, "max": 4.842620849609375, "pos_frac": 0.625, "sample": [-0.5465087890625, 0.10918807983398438, 1.5318603515625, -0.8517303466796875, 4.0855712890625, -4.16448974609375, -2.229705810546875, 0.3497314453125, 2.255950927734375, 0.036407470703125, 1.32659912109375, 0.48676300048828125, -0.513031005859375, -0.2522430419921875, -1.08758544921875, -0.9560546875, 0.5105743408203125, 0.5065765380859375, 1.94873046875, -1.2392578125, 2.586212158203125, -2.432830810546875, -0.04268646240234375, -1.170867919921875, -1.43218994140625, 1.07122802734375, 0.1948699951171875, 0.182037353515625, -4.767303466796875, -0.6085357666015625, 1.0887298583984375, 0.0, 1.1954498291015625, -0.6177978515625, -1.83551025390625, 1.0802764892578125, 2.675506591796875, 0.2169189453125, -0.3929443359375, 0.7506389617919922, 0.0107421875, 0.04805755615234375, 2.826324462890625, 2.855224609375, 2.6929473876953125, 1.2960205078125, -1.5648193359375, 0.67779541015625, 0.6915130615234375, 0.07369804382324219, 0.465423583984375, 0.047061920166015625, 2.2371826171875, -0.032135009765625, -1.092071533203125, 3.62921142578125, -2.87615966796875, 4.842620849609375, 2.07904052734375, 0.27187347412109375, -2.0158233642578125, 1.814208984375, -1.0932121276855469, 1.24871826171875, 1.2085494995117188, -0.6659317016601562, -0.875396728515625, -0.16342926025390625, 0.7270278930664062, 2.169158935546875, 0.791717529296875, 0.07505035400390625, -0.23444366455078125, -0.442901611328125, 1.2618865966796875, 1.026702880859375, 0.0, 1.46881103515625, 2.78936767578125, 0.49932861328125, 0.3962249755859375, -0.15648651123046875, -0.218292236328125, -1.23895263671875, 1.061065673828125, -0.7037353515625, 0.16788291931152344, -2.2735595703125, -0.1953887939453125, 1.143402099609375, 0.9182968139648438, 0.30255126953125, -0.2696990966796875, 3.87652587890625, 0.5346298217773438, -0.297393798828125, 1.514892578125, 1.091064453125, 0.3387451171875, 0.9431838989257812, 0.0303497314453125, 1.0461807250976562, 0.0777130126953125, 0.303375244140625, 1.362091064453125, -0.149200439453125, 0.9030914306640625, -0.7517852783203125, 0.5970306396484375, -0.02991485595703125, 0.04547119140625, 0.56634521484375, 2.75030517578125, 0.3400745391845703, 1.3699951171875, 4.74755859375, -0.6295700073242188, 0.9328289031982422, 0.43768310546875, -1.056396484375, -0.7933349609375, 1.9213714599609375, 0.9668121337890625, -0.559783935546875, 0.8837890625, -3.62384033203125, -0.86480712890625, 0.00977325439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000047.npy"}
{"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 0.6197762489318848, "std": 1.4220972061157227, "min": -3.028076171875, "p10": -0.907958984375, "median": 0.3495035171508789, "p90": 2.3759994506835938, "max": 7.1102294921875, "pos_frac": 0.6640625, "sample": [0.41033935546875, 1.4948577880859375, 0.8660888671875, 0.5053787231445312, 1.106109619140625, 1.40380859375, 0.60205078125, 2.26873779296875, 2.35748291015625, 0.38304901123046875, -0.6182479858398438, -0.5630416870117188, -0.6417160034179688, -1.6492767333984375, -0.6689987182617188, 7.1102294921875, -0.2420654296875, -0.907867431640625, -1.94012451171875, 0.5016937255859375, 1.631622314453125, 0.32745361328125, 0.22479248046875, 0.958404541015625, 1.1974334716796875, -0.5729293823242188, -1.111572265625, 1.3187026977539062, -0.7412109375, 0.7336273193359375, 0.8922805786132812, 1.859375, 1.58245849609375, -1.214874267578125, 2.498992919921875, 0.5975570678710938, -0.00327301025390625, 1.97943115234375, 1.3885650634765625, 2.4192047119140625, 0.32366943359375, 2.872955322265625, 0.0882415771484375, 0.258392333984375, 0.1045074462890625, 0.31211090087890625, -3.028076171875, 0.150726318359375, -1.57513427734375, 0.4584808349609375, 4.1519775390625, 0.6107177734375, 0.3242645263671875, -0.7503280639648438, 0.909210205078125, 0.7979583740234375, -0.03022003173828125, -0.034027099609375, 0.32969093322753906, 1.08251953125, 1.486572265625, 0.734100341796875, 0.7514190673828125, 1.1732406616210938, 0.281951904296875, 0.75115966796875, -1.016387939453125, -1.0732421875, 1.5236129760742188, 1.04840087890625, -1.802490234375, 1.263671875, 0.0, -0.2648468017578125, 0.06060791015625, 0.11868858337402344, 3.13897705078125, 0.3162078857421875, 0.36931610107421875, 2.51239013671875, 0.0754547119140625, 2.0955352783203125, -0.1305694580078125, 3.688720703125, -1.0469970703125, 2.02392578125, -0.114990234375, 1.2135238647460938, -0.6215553283691406, 0.6634521484375, -1.3756256103515625, 0.5814666748046875, -0.1765289306640625, -0.04563140869140625, 2.94384765625, 2.921875, -0.908172607421875, 0.8860626220703125, -0.914947509765625, 0.8499832153320312, 0.02046966552734375, 2.152008056640625, 0.0, 1.1998291015625, 2.5570068359375, 0.1466064453125, 5.8099365234375, -0.0011138916015625, -0.21075439453125, 1.3171234130859375, 0.1419830322265625, -0.08199310302734375, 0.09967041015625, -0.499267578125, -0.3201332092285156, -0.307220458984375, 0.1899261474609375, -0.6287288665771484, 0.15380096435546875, 0.4127540588378906, 1.0970458984375, 1.805328369140625, -0.431732177734375, 1.406707763671875, -0.4602813720703125, -0.76123046875, 2.00909423828125, 3.43017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000048.npy"}
{"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 0.6483792066574097, "std": 1.7574748992919922, "min": -6.051300048828125, "p10": -1.4076629638671876, "median": 0.4374732971191406, "p90": 2.8222091674804686, "max": 6.59356689453125, "pos_frac": 0.703125, "sample": [1.0105819702148438, 2.0592193603515625, 0.7736282348632812, 0.8492889404296875, -1.843109130859375, -0.2519683837890625, -0.210479736328125, 5.15435791015625, 2.142425537109375, 1.1590957641601562, 0.489471435546875, 2.28314208984375, -1.51336669921875, 0.9515838623046875, 0.141876220703125, 3.689361572265625, 1.8031005859375, 1.0959129333496094, 0.43243408203125, 4.0126953125, -2.45391845703125, -0.5962333679199219, 1.89984130859375, 1.13494873046875, 0.731842041015625, -0.16070556640625, -1.40582275390625, 6.59356689453125, 0.23839569091796875, 2.5501708984375, 0.7016754150390625, 3.0057525634765625, 2.32781982421875, 0.4049530029296875, 2.645843505859375, 0.5725307464599609, -0.098785400390625, 0.08514404296875, -0.15798187255859375, -0.425384521484375, 0.3086280822753906, 1.153564453125, -1.6220245361328125, 1.3254928588867188, 1.13934326171875, 1.3381805419921875, 1.4298248291015625, 2.9078826904296875, 2.785491943359375, -1.411956787109375, -0.2001953125, -1.3813934326171875, -0.1739349365234375, 0.3752593994140625, 0.13352394104003906, 2.7315673828125, 0.391571044921875, 0.6225852966308594, -0.2140655517578125, 0.7115097045898438, 0.283416748046875, 2.90789794921875, -2.00128173828125, 3.801513671875, 0.18965530395507812, 0.276763916015625, 0.860931396484375, -1.9237899780273438, 2.298095703125, 1.88092041015625, 0.17694091796875, 1.7567596435546875, 0.02704620361328125, -0.6752510070800781, 0.5407161712646484, 2.46343994140625, -2.72247314453125, 0.44268798828125, 2.4591522216796875, -2.683990478515625, 0.4076042175292969, 3.63787841796875, 1.045562744140625, -0.25992584228515625, 0.136260986328125, 0.8509521484375, -2.49298095703125, 2.126708984375, 3.66571044921875, 3.0223388671875, 0.0811767578125, -0.178466796875, 1.3903656005859375, 1.289947509765625, -0.41310882568359375, -0.465911865234375, -0.0827789306640625, 0.707183837890625, 0.31339263916015625, -3.30316162109375, -0.96563720703125, -0.95770263671875, -2.683441162109375, 0.8577117919921875, 3.529327392578125, -0.8390655517578125, 1.07196044921875, 1.97247314453125, 1.1984405517578125, 0.14573097229003906, 5.043487548828125, 1.0375137329101562, 0.44251251220703125, -0.13534927368164062, 0.4320831298828125, 0.26598358154296875, 0.5883865356445312, 2.2911529541015625, 0.860382080078125, 0.0639801025390625, -6.051300048828125, 0.08001327514648438, 0.279754638671875, 0.322662353515625, 0.416473388671875, -1.108184814453125, -0.9194488525390625, -0.2590179443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000049.npy"}
{"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 0.2618473172187805, "std": 1.6272436380386353, "min": -4.968475341796875, "p10": -1.9197921752929685, "median": 0.24190807342529297, "p90": 2.251685333251953, "max": 5.46661376953125, "pos_frac": 0.59375, "sample": [0.509185791015625, -0.12708282470703125, 0.27642822265625, 0.2028656005859375, -1.428863525390625, -0.9779243469238281, 1.371185302734375, 0.057903289794921875, -0.662750244140625, 4.9095458984375, -0.4613037109375, 0.05805683135986328, 0.32666015625, 1.53314208984375, 3.6414794921875, 1.0849533081054688, 1.2349090576171875, -4.213775634765625, 0.4873504638671875, -0.13967132568359375, -0.35833740234375, -0.29876708984375, -3.3498382568359375, 0.7305450439453125, 1.403961181640625, 2.50775146484375, 1.1936721801757812, 0.91424560546875, 1.72235107421875, -1.0123291015625, -1.494171142578125, 4.326904296875, 0.8891754150390625, -1.555145263671875, 0.5138702392578125, -1.8834686279296875, -0.2864227294921875, -1.0352783203125, 0.364990234375, -0.096649169921875, -0.9257583618164062, -1.5128173828125, 2.173919677734375, 1.55499267578125, -0.1863861083984375, -0.25592041015625, 1.0638427734375, 0.019775390625, -0.534393310546875, -0.2327423095703125, 0.33587646484375, -0.4037933349609375, 0.38653564453125, 0.038021087646484375, 1.0583953857421875, 1.13568115234375, 2.340118408203125, -2.004547119140625, 0.7427825927734375, 1.09375, 1.3942718505859375, 0.928558349609375, -4.968475341796875, 0.06018829345703125, 0.235504150390625, -2.042022705078125, -1.27386474609375, 0.8044052124023438, 0.9866104125976562, -0.28073883056640625, 1.042083740234375, -0.5047492980957031, -0.235931396484375, 0.497222900390625, 0.34613037109375, -2.3045654296875, 0.46063232421875, 0.3310546875, 1.632904052734375, -0.1922321319580078, -0.7751922607421875, -0.098602294921875, -2.71728515625, -0.927154541015625, 0.6325340270996094, 2.262969970703125, -2.292999267578125, -2.007354736328125, 1.34271240234375, -2.5869140625, -0.69439697265625, -1.61285400390625, 2.3638916015625, -0.458587646484375, 0.9506072998046875, 0.096588134765625, 0.03040313720703125, 0.9798126220703125, 1.1666717529296875, 0.3869171142578125, 2.2410888671875, 3.912078857421875, 0.77239990234375, 1.0037994384765625, -0.4619140625, 0.3487396240234375, 0.9555854797363281, 1.045501708984375, -0.178497314453125, -2.263580322265625, -0.050048828125, 2.2468490600585938, 3.141845703125, -2.252410888671875, -0.342041015625, -2.037200927734375, 5.46661376953125, 1.6712646484375, 0.057159423828125, 0.17888259887695312, 2.8123779296875, 0.24831199645996094, 2.28289794921875, -0.54095458984375, 0.927001953125, -0.4094696044921875, 0.002193450927734375, 3.012542724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000050.npy"}
{"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 0.6590977311134338, "std": 1.8907021284103394, "min": -4.9385986328125, "p10": -1.6514831542968749, "median": 0.5449028015136719, "p90": 2.82872314453125, "max": 6.2177734375, "pos_frac": 0.6796875, "sample": [2.252197265625, 0.254913330078125, 1.7342758178710938, 0.9051742553710938, 1.2976341247558594, -1.8330612182617188, 1.793701171875, 4.97552490234375, -0.442138671875, 0.524169921875, 1.7136383056640625, 1.965667724609375, 0.55694580078125, 2.8341064453125, -0.7168521881103516, -0.3455047607421875, 0.495849609375, -0.29693603515625, 0.0205535888671875, 2.2243804931640625, -4.9385986328125, 0.2418975830078125, 0.255340576171875, -0.7311248779296875, 1.5404815673828125, 5.21923828125, -0.4566802978515625, -2.058349609375, 3.3895721435546875, -0.240478515625, -2.4803466796875, 3.12847900390625, 3.433868408203125, 1.1717681884765625, 0.5328598022460938, 2.4632568359375, 2.25067138671875, 1.7374801635742188, 0.8203125, 0.14258956909179688, 0.8788070678710938, 1.6212005615234375, -0.9299774169921875, 1.796234130859375, -0.059719085693359375, 2.826416015625, 0.9989013671875, 0.3304481506347656, 0.38330078125, -4.66180419921875, 2.913604736328125, 2.080810546875, -2.580596923828125, 0.7786865234375, 0.77459716796875, 2.056610107421875, 0.13286590576171875, 0.7779312133789062, -1.36834716796875, 2.9220123291015625, -1.263458251953125, 2.37451171875, -0.99114990234375, 1.7266387939453125, -2.77203369140625, -1.479217529296875, 0.15094947814941406, -0.2597503662109375, -0.1982879638671875, 1.708343505859375, 0.1395416259765625, 0.333404541015625, 0.36285400390625, 2.966461181640625, -0.856842041015625, 0.0, 5.491241455078125, -1.704376220703125, 1.8931503295898438, 1.028564453125, 0.372406005859375, -0.9119834899902344, 0.7981948852539062, 0.77020263671875, -3.284881591796875, 0.379425048828125, -0.8321685791015625, -0.5418930053710938, -2.0474853515625, 2.63262939453125, 0.24547195434570312, 0.594757080078125, 0.5050048828125, -0.8859176635742188, -0.5128097534179688, 0.587127685546875, -0.7902984619140625, 0.510589599609375, 1.2912139892578125, -0.804046630859375, -1.628814697265625, -1.923309326171875, 0.183929443359375, 5.43438720703125, 4.94525146484375, 6.2177734375, 2.0590362548828125, 0.9512939453125, 1.69482421875, 2.2127685546875, 2.392913818359375, 0.1517963409423828, -0.571441650390625, 1.0409965515136719, 1.9871673583984375, 2.332183837890625, -2.0126190185546875, -3.349395751953125, 1.84295654296875, 1.86224365234375, 1.323974609375, 0.7449951171875, 0.20069122314453125, -0.4105224609375, 1.5938720703125, 1.3176116943359375, -0.6204833984375, 0.6558837890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000051.npy"}
{"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 0.9455058574676514, "std": 1.941490650177002, "min": -4.04400634765625, "p10": -0.9965896606445312, "median": 0.7400665283203125, "p90": 3.653390502929687, "max": 8.0499267578125, "pos_frac": 0.7109375, "sample": [1.21185302734375, 1.7012939453125, 0.43212890625, -2.8172607421875, 0.28376007080078125, -1.4000701904296875, 2.8408203125, 0.7456817626953125, 0.27259063720703125, 6.055938720703125, -0.057373046875, 1.04541015625, -0.47650146484375, -3.1708984375, 4.148406982421875, 5.07000732421875, 0.345062255859375, 0.8160400390625, -0.2802581787109375, -0.9918060302734375, 0.961761474609375, 0.25078582763671875, 1.1884078979492188, 1.1001129150390625, 1.2457656860351562, -0.33556365966796875, 0.87896728515625, 2.522308349609375, 1.0987396240234375, -2.243255615234375, 0.64019775390625, -0.1618194580078125, -0.2018585205078125, 1.3430023193359375, -2.44189453125, 0.6913909912109375, -0.3708343505859375, -1.00775146484375, 0.6051292419433594, 0.38824462890625, 5.98724365234375, 0.24177169799804688, -1.6353759765625, 1.177154541015625, 0.3133087158203125, 1.6231002807617188, 0.8109588623046875, 3.54443359375, 0.17205810546875, 3.8919677734375, 0.6238250732421875, 2.650238037109375, 1.8727569580078125, 0.9950714111328125, -0.2416095733642578, 2.297882080078125, 0.11463737487792969, -0.12139892578125, -0.38921356201171875, 0.0567626953125, 1.14404296875, 0.1168060302734375, 2.2398834228515625, -1.3358154296875, 3.726898193359375, 1.7828369140625, -2.1712646484375, 0.15217971801757812, 1.7678451538085938, -0.04082202911376953, -0.40997314453125, 1.1526641845703125, 1.2053680419921875, 1.316986083984375, 0.313720703125, -2.96624755859375, 1.208181381225586, 1.98583984375, 1.3720550537109375, 1.362548828125, 3.920928955078125, 6.11236572265625, 0.7344512939453125, 0.0755157470703125, 2.500274658203125, -2.6585693359375, -0.11857986450195312, 3.126739501953125, 4.150299072265625, 3.04730224609375, 1.3744583129882812, -0.18094635009765625, -0.26467132568359375, 4.389892578125, -0.90838623046875, 0.5576934814453125, -0.154296875, 4.491546630859375, 1.5296401977539062, -0.7045822143554688, 1.0511474609375, 0.855255126953125, 1.157745361328125, -0.0252685546875, 4.100555419921875, 2.039459228515625, -1.8570556640625, -0.15338134765625, 8.0499267578125, 2.7255859375, 1.22186279296875, 3.038055419921875, 0.8534698486328125, -0.5199050903320312, 3.62188720703125, 0.18347930908203125, 0.6155853271484375, 0.4363994598388672, 0.7513961791992188, 0.570709228515625, 1.103485107421875, 0.92608642578125, -4.04400634765625, -0.4582977294921875, 0.6764373779296875, 0.59698486328125, 3.56512451171875, -0.9150238037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000052.npy"}
{"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 0.718916654586792, "std": 2.098438262939453, "min": -5.92547607421875, "p10": -1.4218460083007811, "median": 0.5205917358398438, "p90": 3.4603286743164063, "max": 6.94854736328125, "pos_frac": 0.640625, "sample": [1.4771728515625, -0.4566497802734375, 0.9819488525390625, -2.9987106323242188, 1.5992889404296875, -2.0670166015625, 5.052154541015625, -1.4646453857421875, -0.991455078125, 0.138519287109375, 1.88189697265625, -0.651519775390625, -1.4813690185546875, -4.12548828125, 0.039306640625, 3.25274658203125, 1.05120849609375, -0.773712158203125, 3.2371826171875, -0.005096435546875, -0.646453857421875, -0.5841522216796875, 0.069366455078125, 0.856201171875, 5.48907470703125, 5.48846435546875, 2.686492919921875, 4.0697021484375, 1.6944580078125, 3.452423095703125, 1.2360687255859375, 4.790557861328125, -0.2569580078125, 0.5421295166015625, 1.129903793334961, -0.1998443603515625, 1.654937744140625, -5.92547607421875, -0.28606414794921875, 1.8104324340820312, 1.797821044921875, 0.09381866455078125, 0.09117889404296875, -1.8049163818359375, -0.07045364379882812, -1.40350341796875, -1.5091552734375, 0.06267547607421875, 2.356719970703125, 0.44671630859375, 1.176361083984375, 4.312591552734375, 1.495697021484375, -2.143646240234375, 0.09705352783203125, -0.7344436645507812, 1.902801513671875, 0.2635078430175781, -0.14247894287109375, 2.976165771484375, 3.8310394287109375, 0.6457290649414062, 0.499053955078125, 1.082794189453125, 1.3898773193359375, -0.0809783935546875, -1.103759765625, -0.465911865234375, 0.3951377868652344, 0.7224540710449219, 0.2536888122558594, 0.873199462890625, 4.149017333984375, 2.0702667236328125, 2.790863037109375, 4.33251953125, 0.32537841796875, 1.160064697265625, 0.16943359375, -0.16357421875, 4.423797607421875, 0.5656890869140625, 0.941375732421875, -0.402801513671875, -0.9211273193359375, 3.37872314453125, -4.73944091796875, 1.3402099609375, 0.2741241455078125, -1.3311614990234375, 1.0034408569335938, -1.1514892578125, 3.361083984375, -1.1297073364257812, -0.15032958984375, 6.94854736328125, -0.0304412841796875, -2.01165771484375, 0.00316619873046875, -0.718048095703125, -0.1762237548828125, 1.3351974487304688, -0.4546661376953125, -2.5709228515625, 0.161773681640625, -1.0669631958007812, -0.603973388671875, 1.592529296875, 0.6103439331054688, -5.831268310546875, 0.6539230346679688, -0.20220947265625, 3.4787750244140625, 2.751373291015625, 0.89898681640625, 1.30792236328125, 4.29986572265625, 0.0, 1.604217529296875, 1.02349853515625, 1.152191162109375, 1.158538818359375, 1.1085586547851562, 1.202117919921875, 2.207794189453125, 0.41674041748046875, -0.003570556640625, 3.404998779296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000053.npy"}
{"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 0.7083349227905273, "std": 2.253709316253662, "min": -7.115814208984375, "p10": -1.6310272216796875, "median": 0.4914360046386719, "p90": 3.0752868652343746, "max": 8.77734375, "pos_frac": 0.640625, "sample": [2.3623046875, 4.391754150390625, 2.70159912109375, 1.6219482421875, 1.968658447265625, 1.4239921569824219, -0.061920166015625, 2.90899658203125, 0.00658416748046875, -1.630157470703125, 1.003448486328125, -0.7741851806640625, 0.29808807373046875, -1.7742919921875, 4.453369140625, -0.004146575927734375, 3.17803955078125, 0.5611190795898438, 1.01922607421875, 0.49398040771484375, 1.4621734619140625, 4.24383544921875, 0.1561431884765625, 1.86126708984375, -1.0960845947265625, 0.199615478515625, 2.79315185546875, -0.64556884765625, 1.42437744140625, 0.3713035583496094, 0.18141937255859375, -0.328521728515625, -3.215484619140625, 0.7025604248046875, 0.4888916015625, 1.40509033203125, 5.10302734375, -3.9990234375, 3.03125, 0.3924560546875, 0.5934219360351562, 1.0156898498535156, 7.50653076171875, -0.254638671875, 6.660552978515625, 0.16638946533203125, -3.6536712646484375, 0.14398193359375, -0.9367828369140625, -0.75189208984375, 2.655426025390625, 6.4464111328125, 1.0915985107421875, 0.5969772338867188, -0.0735931396484375, 0.8025398254394531, -1.696136474609375, 1.4253921508789062, 0.4717559814453125, 0.6256141662597656, 2.0419845581054688, 2.183258056640625, -1.274932861328125, -0.9427947998046875, 1.4860076904296875, 1.5766983032226562, -0.1745452880859375, 0.50933837890625, 1.36688232421875, 1.26568603515625, 0.4146995544433594, 1.6746902465820312, 0.122650146484375, 1.8300323486328125, -1.58343505859375, -0.01914215087890625, 0.857177734375, -0.463470458984375, 0.88494873046875, 0.18080711364746094, 0.03891754150390625, -1.1668014526367188, 1.19866943359375, 0.8937301635742188, -0.763427734375, 1.278167724609375, -0.39349365234375, -1.0778350830078125, 1.8721580505371094, 0.4492950439453125, -1.489501953125, 2.470458984375, -3.06256103515625, 4.237762451171875, 0.59588623046875, -0.45899200439453125, -0.660369873046875, 7.30126953125, 8.77734375, 1.534942626953125, 1.74749755859375, 1.665191650390625, 2.962890625, -0.1002197265625, 3.5882110595703125, -0.16022491455078125, -3.57110595703125, -0.4721221923828125, -7.115814208984375, 1.93817138671875, -1.054931640625, -2.1489105224609375, -1.633056640625, -0.1295166015625, 4.0286865234375, 0.023193359375, -1.6626968383789062, -1.650238037109375, 0.36724853515625, -0.3639984130859375, 0.849761962890625, -0.69366455078125, 0.8776378631591797, -1.125732421875, -3.07745361328125, -0.324981689453125, 1.2780494689941406, 1.59698486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000054.npy"}
{"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 0.6486636400222778, "std": 1.8880870342254639, "min": -5.7650146484375, "p10": -1.2505462646484375, "median": 0.3759183883666992, "p90": 3.053733825683593, "max": 6.77386474609375, "pos_frac": 0.609375, "sample": [-0.5706787109375, 1.870391845703125, 0.257476806640625, 0.883392333984375, -0.5167999267578125, -0.81793212890625, 0.37756919860839844, 0.64892578125, 2.507720947265625, -1.3913116455078125, 1.5914764404296875, 2.0849151611328125, 2.796142578125, 1.356689453125, -0.495269775390625, 0.2582263946533203, 0.9063644409179688, 0.532196044921875, 0.7451553344726562, 0.7804412841796875, -0.63446044921875, 1.1018524169921875, 1.01409912109375, 1.823822021484375, 1.5583343505859375, -0.7276611328125, 0.7756118774414062, 2.318023681640625, -0.00836181640625, -0.1046142578125, -1.430999755859375, -0.14862823486328125, -1.5092315673828125, 3.5010528564453125, -0.1414031982421875, 1.6887435913085938, 0.374267578125, 1.5330810546875, 0.01300811767578125, 2.9722900390625, -1.175750732421875, 0.2133026123046875, -0.31768798828125, 0.617645263671875, 0.82745361328125, 1.2041168212890625, 0.3893585205078125, 0.4155731201171875, 0.105194091796875, 5.96649169921875, 0.5153656005859375, -0.6335296630859375, -2.77587890625, 0.392181396484375, -0.599395751953125, -2.1366195678710938, -2.35150146484375, 3.7200927734375, -1.46795654296875, 1.017791748046875, 0.3504180908203125, 1.514068603515625, 5.765167236328125, -0.3292045593261719, 1.14984130859375, 0.06207275390625, -1.327301025390625, 3.842041015625, 0.9321517944335938, -1.4353866577148438, 0.7021026611328125, 0.2580375671386719, -0.2693328857421875, -1.0931243896484375, -0.7975311279296875, 0.1704425811767578, -1.2540283203125, 2.8109588623046875, -0.096771240234375, 1.3721237182617188, -0.9985733032226562, 2.57489013671875, -3.5589599609375, 0.3585700988769531, -0.096343994140625, -0.0474395751953125, -0.6862926483154297, 2.732421875, 3.456939697265625, 0.8482208251953125, -0.20476531982421875, 1.857421875, 0.3282470703125, -0.046722412109375, 6.47607421875, 0.10546875, 3.390655517578125, -0.9652175903320312, 0.946441650390625, 3.3526153564453125, -0.8384552001953125, 1.7111053466796875, 0.5175819396972656, 1.4119873046875, 0.11296844482421875, -3.4286041259765625, 4.553619384765625, 1.0255126953125, 2.4697189331054688, -0.5998306274414062, -5.7650146484375, -0.07706069946289062, 2.0397491455078125, -0.6507949829101562, -0.888641357421875, -0.42038726806640625, 2.9725494384765625, -1.249053955078125, 1.6137924194335938, -0.733184814453125, 1.302459716796875, -0.045562744140625, -0.7037353515625, 3.666168212890625, -0.52081298828125, 6.77386474609375, 3.2431640625, 1.6852798461914062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000055.npy"}
{"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 1.0971472263336182, "std": 2.3451735973358154, "min": -6.117431640625, "p10": -1.2831085205078123, "median": 0.899078369140625, "p90": 4.22430419921875, "max": 8.345947265625, "pos_frac": 0.6953125, "sample": [-0.145172119140625, -3.165313720703125, 2.611297607421875, 5.62677001953125, 1.2567596435546875, 1.140289306640625, 1.1822509765625, 1.35479736328125, 0.8433303833007812, 4.93365478515625, 1.5261688232421875, 0.790435791015625, -0.21269989013671875, -1.0139236450195312, 1.4106292724609375, 0.508392333984375, -0.08652114868164062, -0.297149658203125, 4.1749267578125, -1.44708251953125, 1.870849609375, 0.91375732421875, 2.566986083984375, -2.456817626953125, -0.5090484619140625, 5.046905517578125, 0.8843994140625, 4.31488037109375, 3.4930419921875, 0.3531341552734375, 2.4217529296875, 2.7967681884765625, 1.665008544921875, 0.2963409423828125, 6.079986572265625, 2.0075531005859375, -2.66571044921875, 0.8570785522460938, 0.965972900390625, -0.7795982360839844, 6.757171630859375, 1.8730926513671875, 0.5560379028320312, -0.011074066162109375, 0.2694873809814453, 8.345947265625, 2.4109649658203125, -0.04638671875, -1.761962890625, -0.30202484130859375, -0.032379150390625, -1.3606109619140625, 1.08837890625, -5.09210205078125, -0.4667510986328125, -5.052642822265625, 0.2578544616699219, -0.7429828643798828, 1.849945068359375, 0.25170135498046875, 2.0690155029296875, 5.83355712890625, 2.89666748046875, 2.437347412109375, 0.590240478515625, -1.5895538330078125, 2.54913330078125, 1.3806610107421875, 5.365570068359375, 0.680389404296875, -0.28911590576171875, -0.4787616729736328, -4.0975341796875, 0.6201534271240234, 0.3212127685546875, -0.0956878662109375, -1.99969482421875, 4.18548583984375, -0.1715087890625, 1.7219390869140625, -6.117431640625, 1.11700439453125, 2.502166748046875, 0.28240966796875, 1.0398178100585938, 6.47186279296875, 1.211883544921875, 3.834320068359375, -1.0723114013671875, 2.2181625366210938, 0.4875946044921875, 6.614227294921875, 1.303314208984375, 1.5626983642578125, 1.158935546875, 4.54315185546875, 1.6612625122070312, 0.1896514892578125, 0.4537239074707031, 1.3672714233398438, 2.345916748046875, -0.398040771484375, 1.3791427612304688, -1.2498931884765625, -1.420166015625, 6.832550048828125, -0.5557861328125, 2.03369140625, -0.3750457763671875, 1.402374267578125, 1.0920257568359375, 0.5540313720703125, 2.05645751953125, 0.203643798828125, -0.0436859130859375, 0.555084228515625, -0.541015625, 0.2646675109863281, 1.1711273193359375, 0.45880126953125, 2.56634521484375, -0.5511474609375, 1.8625946044921875, 1.907440185546875, -0.1136474609375, 3.3465576171875, 2.372650146484375, 0.6121902465820312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000056.npy"}
{"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 1.478427529335022, "std": 2.3378560543060303, "min": -2.983245849609375, "p10": -0.6661779403686523, "median": 0.91619873046875, "p90": 4.292214965820312, "max": 15.07452392578125, "pos_frac": 0.75, "sample": [-0.9781875610351562, 3.19073486328125, -2.4906005859375, 3.330596923828125, -0.0877838134765625, -2.43328857421875, 4.22314453125, 0.16847991943359375, -1.14617919921875, -0.10943603515625, 0.955780029296875, 0.155731201171875, 3.570953369140625, 0.37686920166015625, 3.5464706420898438, 0.3330097198486328, 2.022857666015625, 2.219696044921875, 2.284820556640625, 0.80316162109375, 0.9231090545654297, 3.6793975830078125, 0.2456035614013672, -0.0937957763671875, -0.083465576171875, 2.160186767578125, -2.983245849609375, -1.79388427734375, -0.5460205078125, -1.1149673461914062, 2.530792236328125, 0.89306640625, 0.494171142578125, 1.1046142578125, 7.562744140625, 15.07452392578125, 0.4316558837890625, 1.1102294921875, -1.6244354248046875, 3.1795806884765625, 4.50958251953125, 0.9067039489746094, -0.386932373046875, 4.29864501953125, 0.8307876586914062, 1.0872344970703125, -0.2763824462890625, 0.075653076171875, 0.919677734375, 1.2982177734375, 1.270294189453125, 0.2133922576904297, 0.9127197265625, 6.78662109375, 0.89654541015625, 3.49847412109375, 1.2504081726074219, 1.1020736694335938, 3.152923583984375, 0.30218505859375, 3.212310791015625, -0.23480224609375, 0.7775840759277344, 1.23760986328125, 1.332122802734375, 3.861083984375, 5.84521484375, 0.89825439453125, -2.1111602783203125, 0.90625, 3.4710693359375, 0.02374267578125, -0.9260940551757812, -0.6444644927978516, 4.591796875, 2.61968994140625, 1.087310791015625, 4.444427490234375, 1.0692596435546875, 2.968017578125, -0.11334228515625, 0.0023193359375, 6.1026611328125, 5.236724853515625, 1.23956298828125, -0.2704925537109375, -0.6140670776367188, 1.7708740234375, -1.815765380859375, -1.1129608154296875, 2.2873458862304688, 2.637908935546875, 0.42139244079589844, 0.24372482299804688, 0.653076171875, -0.09887313842773438, 3.18280029296875, 0.17177581787109375, 1.04608154296875, -0.627899169921875, -0.4178466796875, -0.288238525390625, 5.6605224609375, 2.072998046875, -0.043849945068359375, 4.724212646484375, 0.298126220703125, 3.876678466796875, 0.1717376708984375, 0.45475006103515625, 0.864990234375, 1.508575439453125, 4.665252685546875, 0.44812774658203125, 2.015430450439453, 2.7723388671875, 2.88360595703125, -0.3450641632080078, 0.35291481018066406, -0.11655235290527344, 3.60223388671875, -0.7168426513671875, 2.6385498046875, 2.7396240234375, 3.3366775512695312, 0.46538543701171875, 4.289459228515625, 2.8173370361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000057.npy"}
{"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 1.0186914205551147, "std": 2.186539888381958, "min": -10.709259033203125, "p10": -1.294573974609375, "median": 0.8364715576171875, "p90": 3.6487976074218746, "max": 7.2547607421875, "pos_frac": 0.7265625, "sample": [0.6031494140625, 3.373992919921875, 0.821990966796875, 2.601959228515625, 2.0804977416992188, -0.6323280334472656, 2.3056716918945312, -1.0941162109375, 0.101470947265625, 4.140869140625, 1.4607696533203125, 3.725830078125, 1.39935302734375, 5.07818603515625, 1.431243896484375, -0.9983901977539062, 4.1019744873046875, 3.61578369140625, 2.3266448974609375, -1.6239013671875, 5.46014404296875, 2.57244873046875, 0.319732666015625, -2.025146484375, 1.562896728515625, 1.7043304443359375, 2.1576995849609375, 1.40966796875, 1.0093994140625, -1.25830078125, 0.0231170654296875, 2.41217041015625, -0.0347900390625, -0.2528095245361328, 2.978973388671875, 0.116943359375, 0.0828857421875, 1.3479232788085938, -0.3381175994873047, 0.5081787109375, -2.4427490234375, 0.574371337890625, 0.37641143798828125, 0.844970703125, 5.52703857421875, -1.4704208374023438, 0.7824306488037109, 4.82427978515625, 0.13350868225097656, -10.709259033203125, -0.31919097900390625, 6.1976318359375, -0.91827392578125, -1.5029487609863281, 1.55023193359375, 0.8105621337890625, -0.29571533203125, 1.1036243438720703, -1.182281494140625, 3.3529052734375, 0.701751708984375, -0.8608474731445312, 3.0035400390625, -1.6828155517578125, 1.411041259765625, 2.4493408203125, 0.58355712890625, 1.330596923828125, 0.062957763671875, 2.777618408203125, -0.040863037109375, -1.8758659362792969, -0.8488693237304688, 2.1481475830078125, 1.5584144592285156, 0.7779541015625, 6.482208251953125, 0.8596038818359375, 0.76702880859375, 0.3577423095703125, 2.1314620971679688, -0.63153076171875, 7.2547607421875, -2.1416015625, 1.2534637451171875, 0.427703857421875, -0.12540817260742188, 0.9064598083496094, 0.435821533203125, 3.563995361328125, -0.5662002563476562, 2.8969268798828125, -0.57196044921875, 1.6547393798828125, 1.737548828125, -1.601654052734375, 4.5510711669921875, -0.95257568359375, 1.78338623046875, 1.678466796875, 1.4463958740234375, 0.1445789337158203, 0.22260284423828125, -1.37921142578125, 2.2510528564453125, 0.827972412109375, 0.4153327941894531, 2.5970916748046875, 3.9679031372070312, 2.5709991455078125, -2.072021484375, 1.76141357421875, 0.250030517578125, 4.262054443359375, 0.9207611083984375, 0.18779754638671875, 0.056854248046875, 0.5816650390625, 2.70648193359375, 1.5056304931640625, 1.7286834716796875, 1.1773681640625, -1.02593994140625, 1.853179931640625, -0.6950759887695312, -2.206878662109375, 1.87774658203125, -1.00421142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000058.npy"}
{"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 0.9600266814231873, "std": 2.5845513343811035, "min": -7.36944580078125, "p10": -1.8207962036132812, "median": 0.751983642578125, "p90": 3.9180480957031243, "max": 8.00335693359375, "pos_frac": 0.6484375, "sample": [-0.400543212890625, 2.41046142578125, 0.26873779296875, -7.36944580078125, 1.1215591430664062, 0.9464569091796875, -0.84014892578125, 2.68499755859375, 0.6825714111328125, -0.12554931640625, 0.78924560546875, 0.6876678466796875, 2.6258087158203125, 3.27496337890625, -2.703094482421875, 0.8158721923828125, 0.06668853759765625, 0.14235305786132812, -1.7803802490234375, 1.3869476318359375, 0.664093017578125, 0.9335556030273438, 0.05929756164550781, -0.314453125, 2.5308990478515625, 1.253021240234375, 0.819122314453125, 3.5467529296875, -3.9724273681640625, 2.0301513671875, -0.3287353515625, 1.1193122863769531, 0.8533172607421875, 1.8198089599609375, 0.0, 4.30364990234375, 0.06183052062988281, 1.507843017578125, -0.6016769409179688, -2.400238037109375, 0.14464569091796875, 3.068115234375, 6.4141845703125, 7.3092041015625, -0.4546661376953125, -4.6256103515625, 2.4656600952148438, 8.00335693359375, 2.364044189453125, 6.872802734375, 1.229522705078125, -1.04840087890625, 5.521728515625, 5.947906494140625, 0.379791259765625, 1.0091705322265625, -0.14507293701171875, 3.2481536865234375, -5.506439208984375, -0.71820068359375, -0.248748779296875, 2.2779541015625, 0.7147216796875, -2.010223388671875, 1.201904296875, 4.363983154296875, 5.05987548828125, -0.2236328125, 2.9885711669921875, 2.5972747802734375, -1.0387725830078125, 2.333404541015625, 0.99493408203125, 5.999420166015625, -2.0536041259765625, -4.2867431640625, 0.5449934005737305, -3.57965087890625, -0.2290191650390625, 0.15862655639648438, 3.112274169921875, -0.22637939453125, 2.7591400146484375, 0.285614013671875, 0.9659652709960938, -1.91510009765625, 3.173980712890625, -0.199676513671875, -0.9513397216796875, -0.8464202880859375, 3.848907470703125, 2.461181640625, -0.3518562316894531, -0.6228141784667969, -1.0235023498535156, 4.079376220703125, 2.78143310546875, -1.5472412109375, -0.3882293701171875, 1.4581298828125, -0.00244140625, 0.5766525268554688, 1.148040771484375, 1.579010009765625, 0.07620620727539062, 1.504302978515625, -1.2218017578125, -4.96710205078125, -1.1924667358398438, 1.4456787109375, 3.8018798828125, 1.99859619140625, 5.61639404296875, 0.04095458984375, -0.5793914794921875, -0.13843536376953125, 7.324066162109375, 1.643157958984375, 3.354278564453125, 0.23020172119140625, -3.170562744140625, 1.106842041015625, 3.830596923828125, 3.004150390625, -0.09765434265136719, -0.0890960693359375, 3.232757568359375, 0.3596954345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000059.npy"}
{"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 1.3049015998840332, "std": 2.410815715789795, "min": -7.41802978515625, "p10": -0.9558685302734374, "median": 0.9844703674316406, "p90": 4.4044189453125, "max": 10.104461669921875, "pos_frac": 0.7265625, "sample": [-4.2528076171875, 2.109527587890625, 4.910800933837891, -0.1134490966796875, 3.54766845703125, 3.738800048828125, -0.3406410217285156, 2.1109695434570312, 2.382843017578125, 0.7337646484375, 0.8064193725585938, -0.5075588226318359, -0.2744560241699219, 1.7939910888671875, 0.761138916015625, 0.816650390625, 0.9477310180664062, -0.0355224609375, 2.1173248291015625, 3.54693603515625, 0.866180419921875, -0.74407958984375, 4.878021240234375, 0.5398712158203125, -1.6029891967773438, 0.3667449951171875, 0.1841278076171875, 0.0, -0.8431587219238281, 4.430908203125, -0.12017822265625, 5.35760498046875, 1.552978515625, 2.75042724609375, 0.3531951904296875, -0.175048828125, 3.65936279296875, -2.890838623046875, 3.1864166259765625, 0.39836692810058594, -0.14719772338867188, 2.272674560546875, 1.630889892578125, 3.3663330078125, 1.4005851745605469, -0.333404541015625, -1.2928409576416016, -0.241546630859375, 0.6218109130859375, 0.295257568359375, -0.14659881591796875, -0.982666015625, 1.41680908203125, 4.784088134765625, 0.12579917907714844, -2.0643310546875, 3.7428741455078125, -0.9678955078125, 1.3036956787109375, 4.39306640625, 1.0366897583007812, 3.37689208984375, 2.5872955322265625, 0.18758010864257812, -7.41802978515625, 2.4147186279296875, -0.18017578125, 0.5731658935546875, 0.1150054931640625, -0.950714111328125, -0.7245330810546875, 4.752532958984375, 3.470306396484375, 5.22332763671875, 3.41473388671875, 0.476531982421875, 0.25206756591796875, 2.619537353515625, 1.2854461669921875, 5.9239044189453125, 2.7607574462890625, 0.194305419921875, -0.225341796875, 1.021209716796875, 5.6746673583984375, 0.582061767578125, 1.35455322265625, -4.160369873046875, 1.5868263244628906, 2.47235107421875, 1.77001953125, 0.7910881042480469, -5.190185546875, 0.28835296630859375, 2.2197265625, 3.40777587890625, 0.22808837890625, 5.66552734375, 0.06804656982421875, -0.588409423828125, 2.534912109375, 3.65777587890625, -0.0019378662109375, -1.1796112060546875, 6.1268463134765625, 2.9386749267578125, 0.1815185546875, 3.5771255493164062, 10.104461669921875, 0.0, 6.7122802734375, -2.200927734375, 2.2662353515625, 0.260101318359375, -0.8777961730957031, 1.093658447265625, 1.727508544921875, 0.71429443359375, 2.416046142578125, -3.61529541015625, 1.56256103515625, 2.51025390625, 1.7487411499023438, 2.451812744140625, 2.13153076171875, 0.73431396484375, 2.859771728515625, 2.1097564697265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000060.npy"}
{"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 1.1489825248718262, "std": 2.5483646392822266, "min": -5.596771240234375, "p10": -1.525543212890625, "median": 0.8577117919921875, "p90": 4.510563659667969, "max": 9.97021484375, "pos_frac": 0.6953125, "sample": [-1.52618408203125, -2.352447509765625, 0.5927753448486328, 0.007110595703125, 4.05267333984375, 4.69976806640625, -1.1459503173828125, 0.8971710205078125, 0.1266345977783203, 0.7853851318359375, 2.0467529296875, -4.51190185546875, -3.03814697265625, 0.6270751953125, 1.1135101318359375, 0.7889003753662109, -1.7978515625, -0.654541015625, -1.5252685546875, -0.9177932739257812, 1.1703720092773438, 1.1389617919921875, 7.51641845703125, 3.73895263671875, 4.10003662109375, 1.355926513671875, 3.91845703125, 0.44769287109375, 4.083251953125, 2.6571807861328125, 0.5519351959228516, 0.32599639892578125, 0.5407276153564453, -0.9862518310546875, 5.128875732421875, -0.0190277099609375, -0.5350112915039062, 4.5500946044921875, 0.87786865234375, 0.6144905090332031, 0.28107452392578125, 1.61773681640625, 1.9734039306640625, -0.096588134765625, 0.1917724609375, 0.3084259033203125, 4.481781005859375, 3.390472412109375, 1.7167205810546875, 2.2218170166015625, -0.0906982421875, 2.250030517578125, 7.087005615234375, 5.2933349609375, 4.408905029296875, 2.1749267578125, 2.4763031005859375, 1.5113964080810547, 6.234954833984375, -0.2112579345703125, 0.888458251953125, 2.597209930419922, 2.1781005859375, -1.181854248046875, 0.366912841796875, 5.66748046875, 2.69512939453125, 5.832611083984375, 3.940216064453125, 1.0516815185546875, 1.7571640014648438, 0.546539306640625, 2.0127410888671875, -1.412567138671875, -0.5900306701660156, 0.49908447265625, -0.12732315063476562, 1.4570159912109375, 2.0831756591796875, -0.655670166015625, 6.033599853515625, 2.1516876220703125, 3.0531005859375, 2.818450927734375, 1.5430755615234375, 2.8797454833984375, 0.08370208740234375, -1.4576034545898438, -0.5903472900390625, -1.858367919921875, -0.36624908447265625, -3.585540771484375, -3.25213623046875, 0.22845458984375, -0.08922576904296875, 0.6745376586914062, 1.0699214935302734, 1.63409423828125, -1.4095458984375, 4.70941162109375, -0.599365234375, 9.97021484375, 1.39385986328125, 0.0, 0.025543212890625, 0.837554931640625, 4.681976318359375, 1.1272125244140625, 1.81982421875, -5.596771240234375, 4.493621826171875, -2.151947021484375, -0.07037353515625, 0.6295013427734375, -4.837371826171875, 2.2246246337890625, -4.8970947265625, -0.3606109619140625, -1.12200927734375, 1.089202880859375, 0.348876953125, 0.834503173828125, -1.0244827270507812, 1.0606842041015625, 1.5152053833007812, 4.4765625, 2.7796478271484375, -2.1218032836914062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000061.npy"}
{"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 1.132265567779541, "std": 2.604006290435791, "min": -5.745849609375, "p10": -1.673359680175781, "median": 0.871246337890625, "p90": 4.606503295898436, "max": 9.328643798828125, "pos_frac": 0.671875, "sample": [5.976409912109375, 0.3799285888671875, 9.328643798828125, 2.09423828125, 3.085052490234375, 0.6491622924804688, 1.43560791015625, 1.15228271484375, -3.2327880859375, -0.4052276611328125, -1.256866455078125, -0.9639511108398438, -0.9025192260742188, 3.249481201171875, 1.503753662109375, 0.8436737060546875, 1.31378173828125, 2.2409210205078125, -0.174285888671875, -1.6296844482421875, 1.487213134765625, 0.9205589294433594, 4.022499084472656, 1.5133895874023438, -3.7364959716796875, 6.42327880859375, -3.376190185546875, -0.760711669921875, 0.9515876770019531, 1.21466064453125, -0.736328125, 1.49639892578125, 2.3778610229492188, 1.0719451904296875, 0.27115821838378906, 0.2072906494140625, 1.500274658203125, -0.8625030517578125, 2.0649642944335938, -1.22808837890625, 2.2890625, 3.92608642578125, 3.5127716064453125, -0.551513671875, -0.8420562744140625, 0.988525390625, 0.6195449829101562, 1.35040283203125, 2.614166259765625, 6.708251953125, -5.745849609375, 3.30133056640625, -1.3177490234375, 3.47857666015625, 3.495849609375, -2.9098663330078125, 6.253662109375, 0.0, 4.0216522216796875, 0.08148956298828125, -3.804229736328125, -1.54034423828125, 5.122711181640625, 0.23983001708984375, 0.666412353515625, -0.1658477783203125, 0.13864898681640625, -1.7752685546875, 0.6065826416015625, 0.8453369140625, 4.424224853515625, 0.5874176025390625, 0.35265350341796875, -2.5920791625976562, 2.19732666015625, 5.3734588623046875, 1.4877166748046875, -2.906219482421875, 0.428375244140625, 2.1574764251708984, 0.4934539794921875, -1.4977645874023438, 2.0014190673828125, 2.61175537109375, -1.19403076171875, 1.750518798828125, 8.49786376953125, 2.3199462890625, -0.03118896484375, 4.50531005859375, 0.89715576171875, 0.325836181640625, 8.138336181640625, 2.1679534912109375, 5.8175048828125, 3.1689605712890625, -2.37548828125, 5.0986328125, -1.591339111328125, -1.90240478515625, 4.1162109375, 0.022918701171875, 1.2847442626953125, 0.835906982421875, 5.4786224365234375, 3.085113525390625, 1.072540283203125, 1.01220703125, -0.3292961120605469, 4.842620849609375, 1.43035888671875, -0.8349761962890625, -0.236480712890625, -1.171417236328125, 0.77392578125, -0.29584503173828125, -2.31414794921875, -0.458404541015625, 2.0143280029296875, 0.009218215942382812, 0.5882568359375, -1.17388916015625, 4.41278076171875, 1.0344696044921875, -0.302581787109375, 1.498077392578125, -1.85693359375, -1.4396514892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000062.npy"}
{"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 1.2779576778411865, "std": 2.9534902572631836, "min": -6.41375732421875, "p10": -2.577613067626953, "median": 1.024637222290039, "p90": 4.526989746093749, "max": 10.55810546875, "pos_frac": 0.6484375, "sample": [3.3433837890625, 3.37847900390625, 0.2695770263671875, -5.9886474609375, -5.4910888671875, -0.5737495422363281, 4.0300140380859375, 0.663360595703125, -0.9408493041992188, 1.6470947265625, -4.9862060546875, -2.8150100708007812, -3.5359649658203125, -2.5055770874023438, 4.090423583984375, -0.8455963134765625, 3.189300537109375, 3.22265625, 7.65203857421875, 8.472930908203125, -2.745697021484375, -3.6302490234375, 9.031494140625, -1.445831298828125, 2.21905517578125, 2.14239501953125, 2.302276611328125, -2.8267059326171875, -0.067352294921875, 3.5124359130859375, 0.06296348571777344, 3.387908935546875, -0.4218292236328125, 0.7919425964355469, 1.5771617889404297, 9.1314697265625, -1.181396484375, 2.3931121826171875, -0.8604736328125, 3.866241455078125, -1.1385498046875, 0.9624614715576172, 4.71942138671875, -2.871551513671875, 0.9763107299804688, 3.2313385009765625, 10.55810546875, -6.41375732421875, -1.19140625, 3.6872711181640625, 0.2470531463623047, 5.066253662109375, 2.317901611328125, 0.0, -0.47602272033691406, 1.709228515625, 2.55657958984375, -1.3633575439453125, -0.3859729766845703, 0.8078193664550781, 0.9587364196777344, 0.068023681640625, 0.234375, 5.48590087890625, -2.837249755859375, 0.9882125854492188, 3.12744140625, -0.5966796875, -1.55718994140625, 3.8121337890625, 7.18280029296875, -0.182708740234375, 3.234893798828125, 0.651611328125, 2.30853271484375, -2.8658447265625, 3.9582366943359375, 1.0610618591308594, 1.8948974609375, -1.0516204833984375, -0.2109375, 7.5806732177734375, 2.081390380859375, -1.0732421875, 2.0764312744140625, 0.5172576904296875, 2.2836227416992188, 0.15631484985351562, 2.3473892211914062, 1.5648574829101562, 3.224853515625, -0.08184814453125, -1.7559967041015625, 4.44451904296875, 0.35754966735839844, 3.892364501953125, -0.10080718994140625, 2.635772705078125, 2.11968994140625, 4.15289306640625, 3.06951904296875, 1.86309814453125, 3.944366455078125, -0.3665313720703125, 1.460662841796875, 0.4214630126953125, -0.73150634765625, -0.00492095947265625, 3.269805908203125, 1.2229461669921875, -0.0831298828125, -0.4544830322265625, 3.0023193359375, 0.7947998046875, 1.958709716796875, 4.91510009765625, 0.0845489501953125, 1.7037506103515625, 1.30975341796875, 1.7996826171875, -3.333892822265625, -0.0416107177734375, 2.25537109375, 2.168212890625, -0.612823486328125, -0.3575706481933594, 4.894195556640625, 4.823822021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000063.npy"}
{"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 1.4951345920562744, "std": 3.2235267162323, "min": -7.511505126953125, "p10": -1.8170089721679688, "median": 1.2017059326171875, "p90": 6.1357666015625, "max": 10.55072021484375, "pos_frac": 0.6796875, "sample": [3.6253662109375, -0.762939453125, -7.511505126953125, 6.812347412109375, -2.49542236328125, 3.39532470703125, 2.5775146484375, 1.4571533203125, 7.68603515625, 0.9002532958984375, 2.654083251953125, 1.8531494140625, 8.42071533203125, 2.997894287109375, 1.4887771606445312, -3.87420654296875, 8.94891357421875, 1.6164016723632812, 0.9061279296875, -2.7835693359375, 1.2138214111328125, 1.0418930053710938, -1.9919586181640625, -1.8627777099609375, 1.3638458251953125, -5.08612060546875, 2.242328643798828, 0.8994293212890625, 3.606048583984375, 5.101470947265625, 1.5499267578125, 2.625518798828125, 2.22052001953125, 1.4581031799316406, 1.28070068359375, -0.48381805419921875, 0.385650634765625, 2.299713134765625, 1.9755172729492188, 0.66876220703125, 2.590545654296875, -0.20259857177734375, -0.29400634765625, 2.8245086669921875, -0.482208251953125, 9.624603271484375, 7.07830810546875, -1.342620849609375, 6.471710205078125, 1.12274169921875, 3.744873046875, -1.0679779052734375, 0.07317161560058594, -0.113922119140625, 0.7365798950195312, -1.4477691650390625, -4.4492645263671875, 2.3406143188476562, 4.7547454833984375, 2.4156646728515625, 2.46240234375, 3.0383148193359375, -1.18182373046875, 4.7413330078125, 1.72674560546875, 1.384796142578125, -0.5291366577148438, -3.654144287109375, -1.16448974609375, 1.16143798828125, -0.6884765625, 1.871368408203125, 7.90765380859375, 1.8573989868164062, 0.5071945190429688, 0.7126922607421875, 2.27587890625, -0.2686004638671875, -0.6298294067382812, 1.0016632080078125, -1.6825065612792969, 5.1910400390625, -0.15139389038085938, -0.0073089599609375, 4.23883056640625, 5.36834716796875, 1.0784530639648438, 1.1836776733398438, 1.1163825988769531, -1.054443359375, 1.849700927734375, 10.55072021484375, 6.17669677734375, 5.82855224609375, 0.15720367431640625, -1.797393798828125, -6.51568603515625, 0.73870849609375, -2.4325408935546875, -0.7815704345703125, 1.848114013671875, 5.87200927734375, 2.8922958374023438, 0.0, 0.36757659912109375, 0.77642822265625, -0.63287353515625, 1.1895904541015625, 4.427947998046875, -1.648773193359375, 6.203338623046875, 1.6976165771484375, -0.6202011108398438, 7.2176971435546875, 1.151947021484375, -0.9225044250488281, 0.1174468994140625, 8.753753662109375, -5.0467529296875, 1.4957809448242188, 0.0, 1.9576416015625, 6.11822509765625, 3.0488739013671875, 3.9885177612304688, 2.20367431640625, -5.26104736328125, -0.2056427001953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000064.npy"}
{"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 1.075089454650879, "std": 2.846367120742798, "min": -6.9703369140625, "p10": -2.3562652587890627, "median": 0.9766826629638672, "p90": 5.001800537109375, "max": 8.101318359375, "pos_frac": 0.6328125, "sample": [-0.019763946533203125, -2.544219970703125, 1.08172607421875, -0.04334831237792969, 6.6015777587890625, 1.3145751953125, 0.8489837646484375, 1.7859649658203125, 3.050018310546875, -6.310638427734375, -3.862518310546875, -6.529144287109375, 1.28179931640625, 3.625030517578125, -0.9452056884765625, 5.8294677734375, 4.146942138671875, -4.7113037109375, 0.4570045471191406, -1.595672607421875, -2.379058837890625, 0.3394432067871094, 3.4809112548828125, 5.123046875, 2.2938709259033203, 2.973052978515625, 4.5306396484375, 1.6448822021484375, -2.9207763671875, -0.0047149658203125, 0.8663787841796875, -0.6721420288085938, -1.089813232421875, 3.05316162109375, 8.101318359375, -0.8455352783203125, -0.06654930114746094, 6.251220703125, 0.33374786376953125, 0.2835578918457031, 1.983062744140625, -3.120513916015625, 3.12017822265625, 1.7540054321289062, 5.100830078125, -1.1999359130859375, 4.3562774658203125, 2.2505645751953125, 5.28717041015625, 0.57720947265625, 2.1587448120117188, 2.0252532958984375, -1.226318359375, 1.943328857421875, 2.787506103515625, -0.280120849609375, 4.074981689453125, 0.9804344177246094, -0.07272720336914062, 3.3791046142578125, -3.26922607421875, 0.27707672119140625, 0.972930908203125, -0.9990043640136719, 1.7321929931640625, 0.2443695068359375, 1.606903076171875, -0.353118896484375, -2.22869873046875, -0.781982421875, 3.1739044189453125, 3.78839111328125, -0.25982666015625, 0.290008544921875, 2.8586883544921875, 0.300079345703125, -2.0672149658203125, -0.019847869873046875, -0.348358154296875, -0.8867378234863281, -3.772430419921875, -0.8280029296875, 0.0, 1.631988525390625, 3.3919219970703125, 2.3650894165039062, -1.526123046875, 1.0438308715820312, 5.9583282470703125, 3.43572998046875, 0.473724365234375, 3.0821380615234375, 6.160400390625, 2.0670318603515625, 5.094512939453125, -2.4803466796875, 3.9701385498046875, 2.9702911376953125, -0.1625957489013672, 7.159912109375, 1.214996337890625, 4.208251953125, -6.9703369140625, 0.9592514038085938, 1.3170089721679688, 4.2503662109375, -0.20501708984375, 6.574462890625, 2.535186767578125, 5.048583984375, 0.18010711669921875, -0.12540435791015625, 2.786163330078125, 1.8605194091796875, -5.6065673828125, -1.0274772644042969, 2.8514556884765625, -2.34649658203125, 0.3238983154296875, 1.1171417236328125, -1.602203369140625, -1.0178375244140625, 1.9373779296875, -1.399749755859375, 4.98175048828125, 0.2599601745605469, 1.028533935546875, -0.2255096435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000065.npy"}
{"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 1.4045995473861694, "std": 3.767148971557617, "min": -10.47271728515625, "p10": -2.4048126220703123, "median": 1.3086090087890625, "p90": 6.241310119628905, "max": 11.92755126953125, "pos_frac": 0.6640625, "sample": [0.49961090087890625, 1.285552978515625, 5.983795166015625, -3.5654296875, -1.0895919799804688, 4.654541015625, -0.498748779296875, 1.3918685913085938, 1.3324127197265625, 1.363037109375, -0.2839698791503906, 0.12138748168945312, 2.0026206970214844, 1.6279983520507812, 1.1897659301757812, -0.697998046875, -0.14748191833496094, 1.5892333984375, 8.316497802734375, 2.513275146484375, 0.77734375, -6.29425048828125, 3.8249969482421875, 3.1511688232421875, 2.655792236328125, 1.6732940673828125, -0.41583251953125, 1.652099609375, -0.4763526916503906, 5.2873992919921875, 2.392608642578125, -0.0414276123046875, 11.40533447265625, -4.99078369140625, 1.1672286987304688, -0.7560615539550781, 1.3316650390625, 11.173736572265625, 4.123046875, -1.9832916259765625, -4.83367919921875, 2.160816192626953, 6.5655364990234375, 1.444427490234375, 0.7743377685546875, 4.1493988037109375, -0.7366943359375, 8.746337890625, 6.7950439453125, -1.8785247802734375, -6.87152099609375, 0.8806228637695312, 2.4456024169921875, 11.92755126953125, 10.0242919921875, 0.46697998046875, 1.10540771484375, 3.3376922607421875, -10.47271728515625, 1.093841552734375, -1.1502685546875, -0.0421600341796875, 7.781982421875, 0.9504852294921875, 4.20001220703125, -7.30035400390625, 0.5911102294921875, -1.4615478515625, 2.81219482421875, -1.4307632446289062, -0.218505859375, 1.27825927734375, 7.916046142578125, -4.133636474609375, 2.745819091796875, 0.8471145629882812, 2.8841094970703125, -0.8490924835205078, 3.06939697265625, -6.221099853515625, 2.8384780883789062, 1.5779457092285156, -2.4579620361328125, 3.55206298828125, -0.34433746337890625, 1.9190483093261719, 0.12960052490234375, 0.6314620971679688, 5.42852783203125, 2.1957244873046875, 0.8797454833984375, 6.10235595703125, 5.778076171875, -7.9761962890625, -0.368072509765625, -2.3569488525390625, -3.068115234375, -2.3820343017578125, 3.9294586181640625, 1.558135986328125, 8.277130126953125, 2.4656982421875, 0.9228515625, 0.0, 3.2588882446289062, -0.8268356323242188, -0.70440673828125, 7.1644287109375, 1.5102691650390625, 3.266571044921875, 1.389251708984375, 2.4683837890625, -0.12578582763671875, -4.632110595703125, -0.360809326171875, 3.218841552734375, -1.6106109619140625, 4.40606689453125, 3.19964599609375, 3.5124664306640625, -1.7239990234375, 0.66265869140625, 1.9020538330078125, -2.2464866638183594, 3.043548583984375, 0.9932594299316406, 6.66046142578125, 3.49041748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000066.npy"}
{"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 1.193260669708252, "std": 3.418285608291626, "min": -8.72747802734375, "p10": -2.5903060913085936, "median": 1.0667152404785156, "p90": 5.449685668945312, "max": 10.24627685546875, "pos_frac": 0.703125, "sample": [-1.0027694702148438, 4.2779541015625, -0.580535888671875, 9.32861328125, 0.28937721252441406, -5.74505615234375, 0.5364761352539062, 3.92205810546875, 4.2459716796875, -0.8032150268554688, 2.1494827270507812, 8.31854248046875, 0.7855453491210938, 2.0860595703125, 2.258087158203125, 5.975341796875, 5.419921875, 4.255615234375, 1.192291259765625, -1.6100921630859375, 1.433807373046875, -1.156646728515625, 3.6415863037109375, 0.2111053466796875, 0.882049560546875, 5.108856201171875, -2.5054931640625, 0.95556640625, 1.8772048950195312, 0.103240966796875, -2.399871826171875, 0.2785205841064453, -1.396942138671875, 1.2865219116210938, 0.873291015625, -4.2824554443359375, 0.0, 0.9380035400390625, 5.519134521484375, 2.0980777740478516, -2.5129852294921875, 2.598968505859375, 3.4601211547851562, 5.645294189453125, 1.2720794677734375, -1.3185396194458008, 0.5106964111328125, -3.17132568359375, -5.324951171875, 0.029064178466796875, -4.53472900390625, 5.81304931640625, -0.57696533203125, 7.53314208984375, 0.039306640625, 2.1025161743164062, 7.5955810546875, -2.770721435546875, 1.844146728515625, 0.641387939453125, 2.46539306640625, 2.8305931091308594, 4.39697265625, -2.9079132080078125, 2.0872344970703125, -0.19652938842773438, 0.7113227844238281, -8.55242919921875, -1.8033447265625, 0.06303977966308594, 4.96148681640625, 2.17803955078125, 0.7618560791015625, 4.105072021484375, -7.486602783203125, 10.24627685546875, 7.752349853515625, 3.7364501953125, -0.1316375732421875, 1.7051544189453125, 1.6851348876953125, -1.6020565032958984, 1.588653564453125, 0.11834716796875, -4.83526611328125, -0.06434345245361328, 6.453254699707031, 0.43242645263671875, 1.2916336059570312, 3.7217864990234375, 6.884063720703125, -8.72747802734375, 5.6402587890625, 3.0533676147460938, -5.5350341796875, 4.14532470703125, 1.76934814453125, 2.0859527587890625, 2.5980377197265625, 4.61065673828125, 4.799468994140625, 1.0843734741210938, -0.4407501220703125, 0.83831787109375, 0.384979248046875, 0.6285858154296875, -0.8321990966796875, -8.221435546875, -1.8249130249023438, 1.5221710205078125, 2.925140380859375, 2.91461181640625, 0.4376678466796875, -0.7204399108886719, -0.7214164733886719, 4.5321044921875, 2.854736328125, 0.4428443908691406, 3.913543701171875, 1.0490570068359375, 2.2738037109375, 1.2004165649414062, 3.38983154296875, 0.9562606811523438, -0.07648468017578125, -0.9681243896484375, 1.305572509765625, -0.7875823974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000067.npy"}
{"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 1.807869553565979, "std": 4.112421035766602, "min": -8.09234619140625, "p10": -3.222084045410156, "median": 1.32366943359375, "p90": 6.7207901000976555, "max": 15.948272705078125, "pos_frac": 0.7265625, "sample": [7.80352783203125, 0.06475067138671875, 1.201171875, -0.02362060546875, -0.7897262573242188, 0.22418212890625, 1.3753662109375, 4.061859130859375, -0.6352481842041016, 1.8602066040039062, 0.8358001708984375, 1.1795654296875, 1.32696533203125, 6.172760009765625, 3.8678436279296875, -5.076263427734375, -0.03610992431640625, 0.05350494384765625, 3.492431640625, 2.3316001892089844, 6.6542816162109375, 1.09002685546875, -1.761810302734375, 1.5064849853515625, 15.948272705078125, 0.7500495910644531, 1.449319839477539, 2.72235107421875, 0.438812255859375, -5.61480712890625, 5.1026611328125, 4.3837890625, 10.92401123046875, 8.455963134765625, 0.17586135864257812, 6.1534423828125, 0.0174713134765625, 0.058380126953125, -3.683563232421875, -5.6402587890625, 1.271148681640625, -0.54541015625, 3.5552902221679688, 4.4324951171875, 0.790740966796875, -5.283294677734375, -0.16101837158203125, 2.9274520874023438, 5.25030517578125, 5.460601806640625, 2.455780029296875, -3.1249237060546875, 1.8863525390625, 5.5919189453125, 11.115692138671875, -0.8595123291015625, 3.776397705078125, 9.78619384765625, -0.8803501129150391, 1.68048095703125, 4.030815124511719, 1.463409423828125, -0.4525489807128906, 1.89453125, 0.3241462707519531, -0.3643150329589844, -7.73858642578125, 4.36669921875, 1.375274658203125, 12.114990234375, 0.06603240966796875, -3.44879150390625, 1.32037353515625, 0.20391845703125, -8.09234619140625, 2.955810546875, 0.22723388671875, -3.8990325927734375, 10.3768310546875, 0.13235855102539062, -2.33941650390625, 2.7590179443359375, -0.07068252563476562, 3.423583984375, 2.9317626953125, 0.256591796875, -1.886962890625, 6.06939697265625, -7.47357177734375, 9.464340209960938, 5.27569580078125, 0.6993942260742188, 0.27203369140625, -2.638397216796875, 1.6812744140625, 5.772613525390625, -4.6154327392578125, -4.6005706787109375, 5.582672119140625, 1.3671112060546875, 4.167694091796875, 0.8967971801757812, 1.835906982421875, 8.539398193359375, 8.594085693359375, 0.9752197265625, 5.6504669189453125, 6.8759765625, 2.4438018798828125, 3.663848876953125, 2.521484375, -3.5603713989257812, 2.415679931640625, 1.755706787109375, -0.5868759155273438, 1.2544212341308594, 4.28118896484375, 0.2953338623046875, -0.2378387451171875, 0.42485809326171875, -0.6850357055664062, -2.5182952880859375, 2.7301788330078125, 11.699981689453125, 3.7369384765625, -2.1127166748046875, -0.5797119140625, 0.5942726135253906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000068.npy"}
{"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 2.1361329555511475, "std": 3.8942792415618896, "min": -8.69097900390625, "p10": -1.7827674865722654, "median": 1.6435089111328125, "p90": 7.063696289062499, "max": 14.2572021484375, "pos_frac": 0.7265625, "sample": [5.0904388427734375, -0.349334716796875, 2.336078643798828, 5.2733612060546875, 1.023681640625, -8.6220703125, 3.974700927734375, 1.8026885986328125, -0.6689453125, -0.2062702178955078, 4.008766174316406, 0.9855918884277344, 11.043121337890625, 0.2409515380859375, 8.251068115234375, -0.43035888671875, 7.9030914306640625, 0.4353199005126953, 3.1517333984375, 1.05621337890625, -0.8039398193359375, 0.5535888671875, -3.1903762817382812, -5.2127685546875, 6.173065185546875, 10.090972900390625, 5.3831787109375, 4.310791015625, 0.7595977783203125, 2.135894775390625, -8.69097900390625, 3.5905914306640625, -0.39513397216796875, 1.5749053955078125, 2.4513931274414062, -0.5664138793945312, 1.2223129272460938, 1.0047073364257812, 5.885467529296875, 2.4898681640625, 9.16534423828125, -5.720306396484375, 0.0076351165771484375, 2.97137451171875, 0.6356048583984375, 1.22515869140625, -3.53289794921875, 9.055816650390625, 2.8775253295898438, 0.4676055908203125, 5.25994873046875, -0.070159912109375, 6.92523193359375, 4.625640869140625, 5.6702880859375, -5.73907470703125, -5.296051025390625, 0.8661422729492188, 1.60723876953125, -0.08386611938476562, 0.646331787109375, 5.783111572265625, 1.3724174499511719, 4.057044982910156, 0.222625732421875, 2.9857330322265625, 0.3681182861328125, 4.71600341796875, 5.1190185546875, 6.6158447265625, 9.45733642578125, 2.131378173828125, 1.45660400390625, 0.698455810546875, 2.711456298828125, -0.2876129150390625, 6.58392333984375, 1.3814315795898438, 2.5902557373046875, 1.5833892822265625, -1.3092975616455078, -0.69732666015625, 1.956277847290039, 1.2117958068847656, -1.0954437255859375, -0.3101043701171875, 2.76318359375, 0.9920501708984375, 4.1179656982421875, 9.91949462890625, 7.38677978515625, 8.474761962890625, 1.6996078491210938, 5.3386077880859375, -2.0275955200195312, -2.8084564208984375, 0.08971405029296875, 2.0669708251953125, 8.670318603515625, -0.06769180297851562, 3.714508056640625, 2.179351806640625, 14.2572021484375, 1.18084716796875, 5.4528350830078125, 3.5962371826171875, -3.1505126953125, 3.274139404296875, -1.205535888671875, 2.2120819091796875, -0.57635498046875, -2.43194580078125, -0.09687042236328125, 6.69244384765625, 6.235443115234375, -7.269683837890625, -1.6778411865234375, -0.5533447265625, 3.2136192321777344, 8.93792724609375, 2.948974609375, 3.35870361328125, 0.648895263671875, -0.521392822265625, -0.5652122497558594, 3.267120361328125, 6.074371337890625, 1.679779052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000069.npy"}
{"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 2.066391944885254, "std": 3.8042795658111572, "min": -13.076171875, "p10": -1.8079448699951173, "median": 1.4648985862731934, "p90": 6.9713928222656225, "max": 13.1705322265625, "pos_frac": 0.7109375, "sample": [-1.630889892578125, 2.6920013427734375, 2.3406448364257812, -1.8827667236328125, -0.524383544921875, 0.6836090087890625, 4.4622039794921875, 3.3962020874023438, -4.6793212890625, 8.069366455078125, -1.8142814636230469, -0.07217788696289062, -1.014556884765625, 4.392822265625, 5.56243896484375, 4.3730621337890625, 2.3314666748046875, 4.14300537109375, 0.6658096313476562, 3.4998779296875, 1.3201980590820312, 8.321319580078125, -0.044097900390625, 2.1497135162353516, -4.65576171875, 3.7121124267578125, 0.8730087280273438, 2.1683502197265625, 2.4375534057617188, -1.8052291870117188, 6.74774169921875, 1.0430068969726562, 0.5840606689453125, -0.04180908203125, 1.47265625, 0.722503662109375, 7.994903564453125, -2.7733154296875, -4.3684539794921875, -0.063018798828125, 1.5819854736328125, 0.24781036376953125, 9.9197998046875, 1.2821044921875, 1.8708305358886719, -0.734130859375, 0.1517181396484375, 2.030487060546875, -0.32605743408203125, -1.8438720703125, 1.3201904296875, 0.6937484741210938, -1.623260498046875, -0.10102462768554688, -0.8792304992675781, 3.682464599609375, 8.620880126953125, 1.65106201171875, 5.31396484375, 1.4571409225463867, 2.7183380126953125, 5.136993408203125, -0.5745773315429688, 1.39532470703125, 1.290496826171875, 3.094329833984375, 0.30875396728515625, 4.551544189453125, 1.3142452239990234, 1.138519287109375, -3.62152099609375, 1.054107666015625, -0.08710289001464844, 2.76361083984375, 0.6012649536132812, 7.69134521484375, -0.750335693359375, 2.0974349975585938, 9.18609619140625, 1.6424560546875, 13.0369873046875, 6.249969482421875, 2.1427764892578125, 2.974365234375, 6.6248931884765625, 1.54736328125, -2.0357513427734375, 2.178314208984375, 0.25365447998046875, 11.25531005859375, 5.353874206542969, 6.113250732421875, 0.0, 0.26065826416015625, 4.50408935546875, 7.483917236328125, 6.581085205078125, 5.655029296875, 0.7800750732421875, 11.55023193359375, 5.63055419921875, 5.79290771484375, 0.323883056640625, -2.0310096740722656, 2.6907730102539062, -0.724700927734375, 7.88629150390625, 1.002288818359375, 4.864349365234375, -1.8463134765625, -0.6261749267578125, -1.0886077880859375, 0.5937271118164062, 3.0950698852539062, -6.9957275390625, -13.076171875, 2.84466552734375, 0.42911529541015625, 1.829986572265625, 6.751739501953125, 2.3374481201171875, 2.3216552734375, 1.813079833984375, -0.093505859375, 13.1705322265625, 0.0, -0.34720611572265625, -0.6180877685546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000070.npy"}
{"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 2.0003890991210938, "std": 4.624166488647461, "min": -12.9241943359375, "p10": -4.092145538330078, "median": 2.077608108520508, "p90": 8.044889831542967, "max": 12.5438232421875, "pos_frac": 0.7265625, "sample": [-0.53338623046875, 12.074676513671875, -0.7881240844726562, -6.10546875, -2.58880615234375, -8.27008056640625, -0.9083709716796875, 10.1805419921875, -7.81365966796875, 4.923095703125, 1.2435646057128906, 3.6527671813964844, -4.654083251953125, 3.974365234375, 2.179454803466797, 6.4354095458984375, 4.883331298828125, 0.808624267578125, 3.194671630859375, 9.59869384765625, 2.653900146484375, -1.208343505859375, 2.37841796875, 12.5438232421875, 1.6877899169921875, -0.8881759643554688, 1.0748672485351562, 7.9761199951171875, 3.1319808959960938, 3.1090087890625, -3.8442535400390625, 5.060859680175781, 2.3211898803710938, 10.415863037109375, -1.065673828125, 3.26153564453125, 7.3082275390625, 0.5744113922119141, 6.90008544921875, -5.6200714111328125, -1.3978271484375, -1.0373077392578125, -5.04132080078125, 2.7670135498046875, 2.2767333984375, 0.01187896728515625, 1.0067138671875, -12.9241943359375, -1.096923828125, -0.7945747375488281, 2.1832427978515625, 6.0565185546875, -1.05816650390625, 4.522239685058594, 1.6756515502929688, 5.6372222900390625, 12.07318115234375, 1.8457489013671875, -0.16397476196289062, 1.70709228515625, 3.827606201171875, -2.07464599609375, 7.5634765625, 7.37506103515625, 0.3287029266357422, 1.11767578125, 0.1592578887939453, -1.7183837890625, 3.79876708984375, -0.50494384765625, 1.77490234375, 10.5518798828125, 4.931884765625, -5.72900390625, -4.452850341796875, 3.646270751953125, 9.6868896484375, -7.471527099609375, -3.1536407470703125, 0.6718330383300781, -4.041290283203125, -2.3612060546875, -1.134552001953125, 2.3134384155273438, 6.655670166015625, 8.205352783203125, 0.3798065185546875, 4.570556640625, 6.4482421875, 2.8310546875, 1.3835372924804688, 4.30633544921875, 2.859619140625, 0.6778068542480469, 1.1049575805664062, 9.634689331054688, -0.6018829345703125, 8.804885864257812, 2.9667091369628906, 5.58612060546875, 7.06591796875, -4.210807800292969, 0.3781719207763672, 5.2498779296875, 9.873443603515625, 0.21559906005859375, 1.72625732421875, 1.3432044982910156, -4.272430419921875, 1.856231689453125, 3.17010498046875, 0.8677291870117188, 4.0279541015625, 2.836334228515625, 2.880523681640625, 2.4491615295410156, 2.214599609375, 4.142608642578125, 2.2998580932617188, 1.8005523681640625, 2.552520751953125, 1.9757614135742188, 2.97186279296875, 0.6111679077148438, 0.09814453125, 11.264739990234375, -11.41680908203125, 3.6507415771484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000071.npy"}
{"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 2.0026025772094727, "std": 4.335275650024414, "min": -10.383636474609375, "p10": -3.0280426025390623, "median": 1.877762794494629, "p90": 8.280729675292969, "max": 14.619354248046875, "pos_frac": 0.671875, "sample": [3.4262542724609375, 4.086151123046875, -2.217041015625, 0.6341552734375, 10.6500244140625, -2.73284912109375, 1.577117919921875, 4.773193359375, 1.7017669677734375, -0.4652824401855469, 0.152801513671875, 1.891845703125, 1.53240966796875, 3.1852264404296875, 4.12286376953125, 8.37762451171875, -2.3826370239257812, 2.483489990234375, -3.8726119995117188, 1.2791748046875, 3.4725875854492188, 7.6666259765625, 2.608245849609375, 11.69793701171875, 2.7146453857421875, 0.42147064208984375, -1.4487838745117188, 2.477081298828125, -4.586799621582031, 0.8314590454101562, 4.75579833984375, -10.383636474609375, -2.932342529296875, -6.020347595214844, 1.878570556640625, 2.38531494140625, 1.1264190673828125, -1.1132965087890625, 2.2641754150390625, 0.6781082153320312, 5.5435791015625, -2.395263671875, 8.345916748046875, 5.367828369140625, 9.387283325195312, 2.7219505310058594, 2.556732177734375, -0.3215179443359375, -6.27618408203125, -2.038665771484375, -2.6995925903320312, 3.969207763671875, -1.543792724609375, 0.0, 7.969970703125, 6.798126220703125, 8.252792358398438, -0.10939788818359375, 4.050811767578125, 8.4375, 0.5255889892578125, -3.724456787109375, 3.7240753173828125, 5.9660186767578125, 2.5274734497070312, 5.259307861328125, 1.0885543823242188, 1.0446319580078125, -0.5576896667480469, 5.233612060546875, -4.3977508544921875, 2.276264190673828, 0.4950408935546875, 4.1410980224609375, 8.669525146484375, 1.9397125244140625, 0.2168750762939453, 5.81866455078125, -1.075592041015625, 9.020011901855469, 1.58612060546875, 1.8769550323486328, -1.4696426391601562, -1.126922607421875, -2.582122802734375, -3.2513427734375, 1.4275970458984375, -0.9343109130859375, -2.414651870727539, 8.833221435546875, -6.874267578125, 9.593536376953125, -0.023681640625, 3.642242431640625, 1.8185272216796875, 6.363441467285156, 5.80120849609375, -0.562255859375, 1.6130218505859375, 4.082550048828125, 2.2117919921875, 7.330474853515625, 0.0, 6.4936370849609375, -3.7013511657714844, 0.48693084716796875, 8.929946899414062, 14.619354248046875, 5.6735382080078125, 7.215057373046875, 3.4242324829101562, -2.7733917236328125, -2.087310791015625, -7.970611572265625, 9.9478759765625, -1.2383880615234375, -2.021106719970703, 0.0, 1.3021736145019531, 3.81292724609375, 2.805206298828125, 2.9691696166992188, -6.02093505859375, 4.614105224609375, 2.0612640380859375, -3.379547119140625, 3.98944091796875, 7.334259033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000072.npy"}
{"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 1.9413487911224365, "std": 4.584399223327637, "min": -17.246551513671875, "p10": -2.927046966552734, "median": 1.6615791320800781, "p90": 7.1511688232421875, "max": 14.43975830078125, "pos_frac": 0.7265625, "sample": [1.3608970642089844, 2.29986572265625, 8.53814697265625, 4.265625, -1.915252685546875, 2.2140960693359375, 1.560760498046875, -0.494415283203125, 0.6900787353515625, 5.2317962646484375, 3.618896484375, 3.494140625, 0.4813385009765625, 9.87847900390625, 0.02434539794921875, 0.09717559814453125, -8.80548095703125, -9.70635986328125, 0.5338592529296875, 0.028900146484375, 2.397216796875, 0.2220916748046875, -0.099822998046875, 2.839385986328125, 1.538818359375, 0.5794334411621094, 2.3774871826171875, 1.3012428283691406, 1.439697265625, 4.7560577392578125, -6.255767822265625, -2.50567626953125, 4.07318115234375, 9.4605712890625, -3.005035400390625, 2.396484375, -4.154815673828125, 10.531005859375, 4.569854736328125, 6.4671478271484375, 4.52264404296875, 0.7652320861816406, 5.055755615234375, -2.8936233520507812, 14.43975830078125, -1.853515625, -0.678497314453125, -1.92816162109375, 2.5878753662109375, 11.712188720703125, -5.417999267578125, 0.7816162109375, 9.29669189453125, -17.246551513671875, 4.0506439208984375, 1.2949409484863281, 3.91912841796875, -4.7871551513671875, 6.09954833984375, -2.5030975341796875, 3.344970703125, 2.41119384765625, 3.521636962890625, -0.517822265625, -2.63970947265625, 6.93463134765625, 4.782871246337891, -2.62255859375, 6.36077880859375, 0.4168548583984375, -1.0903472900390625, -1.9150390625, 4.352935791015625, 3.9786376953125, -0.03375244140625, 1.5577335357666016, -5.95867919921875, -5.18896484375, -1.65533447265625, 1.4397964477539062, 7.104705810546875, 4.4644927978515625, 4.5374603271484375, 1.030670166015625, -2.603729248046875, 3.4029006958007812, 0.37305450439453125, 2.9242401123046875, -0.5833358764648438, 3.6488037109375, 1.1641464233398438, 1.6742095947265625, 7.429412841796875, 3.222747802734375, 6.0283203125, 6.58514404296875, 2.5425682067871094, 4.176513671875, -7.692718505859375, 6.431732177734375, 0.1904449462890625, 4.5799102783203125, 3.6211166381835938, 3.735595703125, 7.25958251953125, 6.7510986328125, 1.1896896362304688, -2.1917724609375, 12.578369140625, 7.0533447265625, 4.941650390625, -2.701324462890625, 2.0550537109375, -0.3307991027832031, 9.405158996582031, -3.9237060546875, 0.43802452087402344, 7.468658447265625, -3.0839157104492188, 0.47271728515625, 7.46636962890625, 1.5728645324707031, 6.687164306640625, 3.6121864318847656, 1.6489486694335938, 1.1931228637695312, 6.4654541015625, -2.546417236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000073.npy"}
{"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 1.9151005744934082, "std": 4.841545104980469, "min": -11.566192626953125, "p10": -4.1999664306640625, "median": 1.694305419921875, "p90": 7.664276123046873, "max": 14.658447265625, "pos_frac": 0.7109375, "sample": [3.31695556640625, 7.45086669921875, 3.40142822265625, 1.183349609375, 8.576904296875, -0.629638671875, -4.443817138671875, 1.99945068359375, 6.8419036865234375, 8.1622314453125, -1.54644775390625, 7.0324249267578125, 0.0, 2.38494873046875, -0.9318389892578125, 8.3597412109375, 5.1865234375, 12.059814453125, -1.7545928955078125, 1.5654449462890625, 2.857696533203125, 3.8645782470703125, 3.32757568359375, 2.2580795288085938, 8.490432739257812, -5.83355712890625, -6.037628173828125, 0.431304931640625, -2.1875, -3.9597625732421875, -4.095458984375, 1.0601806640625, -3.78106689453125, -0.7684974670410156, 5.2129058837890625, 1.726165771484375, 0.303466796875, 6.479572296142578, -0.14700698852539062, 5.9284515380859375, -0.17023849487304688, -2.3546142578125, -11.566192626953125, -7.44970703125, 0.1542205810546875, 5.040679931640625, 0.40838623046875, -2.35986328125, -1.683990478515625, 4.36865234375, -9.82257080078125, -8.171630859375, 0.0078125, 2.55694580078125, 1.566558837890625, -0.9709968566894531, 1.3645286560058594, 1.2895355224609375, -8.2672119140625, 6.026824951171875, 4.299407958984375, 3.607269287109375, 0.30462646484375, 10.0037841796875, 6.455322265625, 1.662445068359375, 10.54852294921875, 4.508026123046875, 1.8107452392578125, 14.658447265625, 2.324798583984375, -0.13372802734375, -1.7904434204101562, 5.637359619140625, 0.14398956298828125, 4.058380126953125, 1.4100341796875, 3.666656494140625, 0.7809906005859375, 4.8039703369140625, 4.499626159667969, 0.30612945556640625, 1.35443115234375, -10.82122802734375, 11.9813232421875, 1.519012451171875, 0.599884033203125, 3.040313720703125, 1.838470458984375, -1.181854248046875, -0.5440139770507812, -4.8500823974609375, 1.59857177734375, 5.4248504638671875, 2.8870697021484375, 3.8822021484375, 5.787872314453125, -7.70489501953125, -7.72088623046875, 6.653228759765625, 5.2430572509765625, -5.1519012451171875, 2.27581787109375, 9.02679443359375, 1.338226318359375, 5.6032562255859375, 3.469024658203125, -3.489593505859375, 6.641754150390625, -1.04193115234375, -0.5535545349121094, 1.2000579833984375, 3.128936767578125, 2.5307235717773438, 12.5042724609375, -0.433837890625, 1.1838607788085938, 4.065673828125, 3.8779449462890625, 0.505767822265625, 5.58917236328125, 2.7195816040039062, 0.9142684936523438, 8.268821716308594, 13.30609130859375, 4.705535888671875, 0.3303985595703125, 6.751312255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000074.npy"}
{"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 2.7109434604644775, "std": 4.487781047821045, "min": -10.03363037109375, "p10": -2.240325927734375, "median": 2.4613494873046875, "p90": 8.064123535156249, "max": 15.29669189453125, "pos_frac": 0.765625, "sample": [1.7286834716796875, 3.1117477416992188, 2.2425460815429688, 2.0557098388671875, -1.863555908203125, 7.041259765625, 6.408294677734375, -2.2486572265625, 4.44268798828125, 3.4492111206054688, 0.7594261169433594, -4.9503936767578125, 1.985107421875, -0.5464611053466797, 8.76318359375, -0.24279212951660156, 15.29669189453125, 2.77716064453125, 5.1025390625, 4.241729736328125, 7.8289794921875, 6.4095916748046875, 6.282806396484375, -3.992279052734375, 0.4900360107421875, 12.26019287109375, -2.79754638671875, 5.592376708984375, 0.256378173828125, 1.46917724609375, 3.0985107421875, 6.470947265625, 3.94854736328125, 3.165435791015625, 0.9368133544921875, -2.069437026977539, 3.413330078125, 2.79656982421875, 4.316719055175781, 1.69305419921875, 12.413970947265625, 6.290218353271484, 1.625762939453125, -10.03363037109375, 2.44647216796875, 0.5797119140625, 0.6973171234130859, -0.572296142578125, 9.68829345703125, 0.44014739990234375, 3.6180419921875, 4.550506591796875, 2.6472320556640625, 0.8143157958984375, 5.4794921875, 4.9078216552734375, -3.33319091796875, 9.12384033203125, -0.0735321044921875, 6.7398681640625, 4.213623046875, 2.493927001953125, 1.1947402954101562, 6.694580078125, 1.4914131164550781, 1.52728271484375, 6.46612548828125, -1.7530746459960938, 1.8784446716308594, 2.4168701171875, 1.84588623046875, -1.113555908203125, 1.5363597869873047, 11.4207763671875, 11.67547607421875, 0.42319488525390625, -9.451995849609375, -7.4400177001953125, -2.23675537109375, 3.5458641052246094, 11.307037353515625, 5.945159912109375, 4.365447998046875, -4.775398254394531, -0.7420425415039062, 8.782745361328125, 0.5716209411621094, -0.023834228515625, 2.476226806640625, 5.850006103515625, 5.552764892578125, -1.2133026123046875, 4.433349609375, 3.6893310546875, 6.323974609375, 12.12359619140625, 0.28607940673828125, 4.47613525390625, -0.264404296875, 6.721092224121094, -8.365509033203125, 0.547210693359375, 3.70648193359375, 8.61279296875, 6.885986328125, 4.63580322265625, -0.3306694030761719, 5.1493072509765625, 1.37127685546875, -5.09844970703125, 5.9188232421875, 0.001857757568359375, 1.11968994140625, 5.630046844482422, 4.716024398803711, 7.53131103515625, -0.19866943359375, 1.6924667358398438, 4.3018798828125, -1.08544921875, 1.69891357421875, 1.6248779296875, 5.0290069580078125, 11.21502685546875, -2.0076751708984375, -2.8643951416015625, -6.405326843261719, 0.0786895751953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000075.npy"}
{"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 3.016417980194092, "std": 5.045590877532959, "min": -8.3529052734375, "p10": -1.835700225830078, "median": 2.0490341186523438, "p90": 10.496844482421874, "max": 23.224899291992188, "pos_frac": 0.75, "sample": [6.160858154296875, -2.4196014404296875, 1.8363685607910156, 1.94329833984375, 2.5150680541992188, 4.7132568359375, 1.780731201171875, 0.9923725128173828, 2.4195499420166016, 4.135986328125, 9.710693359375, -0.049655914306640625, 0.789306640625, -1.5714111328125, 14.3306884765625, 5.301666259765625, 0.20977020263671875, -3.0042877197265625, 1.4866180419921875, -0.9934844970703125, 11.9246826171875, 0.874176025390625, 5.901397705078125, 2.4977874755859375, 2.786285400390625, 13.135498046875, 1.761505126953125, 3.3464622497558594, 1.8847999572753906, 0.80816650390625, 3.3870849609375, -0.056346893310546875, 4.4544525146484375, 8.379180908203125, -8.223388671875, 14.95404052734375, 2.9797821044921875, 5.1118316650390625, 5.2586669921875, 1.5950927734375, 10.63702392578125, 5.5169677734375, 6.226470947265625, 6.672698974609375, -5.42889404296875, 1.387716293334961, 3.7707290649414062, 4.0578155517578125, 0.6241455078125, 8.14599609375, 0.0, 4.4562835693359375, 0.3694000244140625, 0.5208053588867188, 7.725502014160156, -4.06207275390625, 11.00335693359375, -7.21820068359375, -1.1007080078125, -3.234893798828125, 7.1142578125, 12.687530517578125, -1.66644287109375, 1.4656982421875, 2.8592681884765625, 1.059173583984375, 3.07427978515625, 4.301170349121094, 5.624019622802734, -1.8059158325195312, -1.100128173828125, 10.436767578125, -2.91961669921875, 0.891204833984375, -3.13946533203125, 0.0, 11.074127197265625, 3.7128524780273438, 4.29229736328125, 13.049041748046875, 2.490478515625, -0.1800823211669922, 1.809661865234375, 1.1116867065429688, 0.7222518920898438, 2.2817916870117188, 2.779327392578125, 8.0211181640625, 2.2957916259765625, 3.2274322509765625, 4.0699005126953125, 2.5013275146484375, -0.0043792724609375, 0.8900604248046875, -0.2945709228515625, -1.5162353515625, -4.691009521484375, 0.0823974609375, 2.0017547607421875, 0.366424560546875, 4.3192138671875, 1.6107406616210938, 6.519500732421875, 12.502899169921875, 8.610992431640625, 18.28765869140625, 11.670364379882812, 0.43634033203125, 1.62896728515625, -3.7826995849609375, -0.1446075439453125, -1.7102279663085938, 23.224899291992188, -1.06298828125, -8.3529052734375, 0.2553558349609375, 2.0963134765625, 2.76220703125, 10.25396728515625, 1.33953857421875, 2.46099853515625, 7.1298828125, 2.5567626953125, 1.83270263671875, -1.3160400390625, 3.1111679077148438, -1.9051971435546875, -0.328643798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000076.npy"}
{"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 2.412130832672119, "std": 5.317079067230225, "min": -15.38616943359375, "p10": -3.369667053222656, "median": 1.7953567504882812, "p90": 9.434059143066406, "max": 20.03619384765625, "pos_frac": 0.6640625, "sample": [-0.4293670654296875, 2.1358489990234375, 0.700469970703125, 2.9771270751953125, 1.5751953125, 0.19720458984375, 8.5989990234375, 5.33099365234375, -0.29048728942871094, 4.420661926269531, -3.024444580078125, -0.06672096252441406, 10.310211181640625, 5.061767578125, 11.322784423828125, 2.0155181884765625, 1.4107818603515625, 3.3308181762695312, -15.38616943359375, 6.025787353515625, 0.1331939697265625, -0.55926513671875, -1.37884521484375, -1.657806396484375, 5.9365234375, 9.436996459960938, 4.9007568359375, 2.2917709350585938, 4.0346527099609375, 4.39007568359375, -3.937255859375, -1.376251220703125, 1.04205322265625, 7.520843505859375, -1.63458251953125, 0.7691383361816406, 0.9650421142578125, 3.83612060546875, 8.429443359375, -0.9009475708007812, 0.465179443359375, 0.9969100952148438, 6.868011474609375, 2.4693603515625, 4.096397399902344, 3.0155029296875, 0.30008697509765625, 18.2841796875, -1.0883560180664062, 1.018829345703125, 7.05963134765625, -0.28021240234375, 3.20062255859375, 9.15936279296875, 8.611328125, 5.782112121582031, 4.4576568603515625, -3.2946624755859375, 2.1013946533203125, 0.2083911895751953, -0.0309906005859375, -2.4739608764648438, 8.53228759765625, 20.03619384765625, -2.840179443359375, 6.623779296875, 3.7449111938476562, -0.980316162109375, 10.161865234375, 2.7441635131835938, 10.498443603515625, 2.75469970703125, -7.75665283203125, -3.544677734375, -3.00640869140625, -1.9410552978515625, 4.853424072265625, 1.0066604614257812, 1.1103973388671875, 1.090484619140625, -2.005157470703125, 9.6468505859375, -2.4444732666015625, -2.75042724609375, 1.0515823364257812, 4.94207763671875, 9.43280029296875, -9.7431640625, -4.536834716796875, 4.071014404296875, 2.215179443359375, 2.9477005004882812, -1.9447441101074219, 4.1470947265625, 5.1044769287109375, -3.9188232421875, -0.758209228515625, -5.9935302734375, 10.943328857421875, 5.738494873046875, -5.16259765625, 6.6909027099609375, -0.7093009948730469, 1.351470947265625, 5.304443359375, 1.18182373046875, -0.04248046875, 10.59375, 12.0762939453125, -0.8082427978515625, 2.3517799377441406, -4.9798583984375, -5.95245361328125, 15.42523193359375, -0.2786102294921875, 4.372222900390625, -5.3374481201171875, 1.474212646484375, -2.4276866912841797, -1.25732421875, 8.141632080078125, 2.4001617431640625, 4.6274566650390625, 5.5189208984375, 4.67431640625, 14.401702880859375, -3.8147430419921875, 0.3184776306152344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000077.npy"}
{"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 3.007789134979248, "std": 5.2760419845581055, "min": -12.056655883789062, "p10": -3.1706253051757813, "median": 2.5725879669189453, "p90": 9.428723144531249, "max": 23.9454345703125, "pos_frac": 0.7109375, "sample": [-1.3798370361328125, 7.62567138671875, -4.34820556640625, 5.1163177490234375, 9.27655029296875, 1.71185302734375, 3.029327392578125, 4.118377685546875, -3.1712875366210938, 5.1598358154296875, -0.3149261474609375, 1.4835472106933594, 4.586635589599609, 5.0693359375, -1.0332813262939453, 2.76470947265625, 3.8487396240234375, 4.6976165771484375, 10.436599731445312, -0.4459228515625, 0.0, 9.2999267578125, -3.2864532470703125, 3.102081298828125, 1.0358123779296875, 2.3497276306152344, 6.265960693359375, 2.8715267181396484, -2.8202133178710938, 5.369476318359375, -3.3853378295898438, 4.425140380859375, 8.101669311523438, 0.8597259521484375, 7.547637939453125, -3.0963134765625, 1.5103607177734375, 12.607208251953125, 2.353271484375, -4.772552490234375, 4.143501281738281, 5.440185546875, -0.8084564208984375, 23.9454345703125, 12.2794189453125, -2.513336181640625, 3.9712371826171875, 4.6515655517578125, 1.1434326171875, 3.7573089599609375, 4.845703125, -12.056655883789062, 2.7325592041015625, 3.01385498046875, 0.32147216796875, -0.6365547180175781, 5.6000213623046875, 4.383209228515625, 2.7334671020507812, -0.7707195281982422, 9.729248046875, -1.449005126953125, 5.4294586181640625, 5.4422760009765625, 2.1646690368652344, -3.1703414916992188, 0.62969970703125, 3.30242919921875, 5.007080078125, -9.41473388671875, 15.264007568359375, 2.1473236083984375, 2.4989013671875, 0.6494674682617188, 10.3543701171875, 7.48699951171875, -0.29109954833984375, 2.8705787658691406, -1.6245536804199219, 1.1230316162109375, -3.48895263671875, -1.0896720886230469, 8.848846435546875, 16.8695068359375, 0.56866455078125, 1.4422454833984375, -0.701629638671875, 3.6400146484375, -5.509490966796875, 6.416168212890625, 3.1208724975585938, -2.256256103515625, 7.7315826416015625, -0.8833770751953125, -6.645965576171875, 2.1870651245117188, 1.501739501953125, -3.6721954345703125, 11.63897705078125, -0.15032958984375, 0.6979408264160156, 1.1871223449707031, -1.3177566528320312, 10.116058349609375, 5.980621337890625, 2.88958740234375, 6.986419677734375, -0.15948486328125, 1.5749588012695312, 2.6131324768066406, 13.12457275390625, 1.259796142578125, 2.53204345703125, 5.438629150390625, 3.969390869140625, 2.471752166748047, 6.761993408203125, 12.50244140625, 8.868316650390625, -3.9021759033203125, -3.45672607421875, -0.045658111572265625, 18.455230712890625, 1.2294425964355469, 2.40777587890625, 3.2457656860351562, -1.3236846923828125, 8.4249267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000078.npy"}
{"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 3.0994439125061035, "std": 6.668955326080322, "min": -37.19287109375, "p10": -3.0876571655273435, "median": 2.4136581420898438, "p90": 10.595098876953125, "max": 24.09368896484375, "pos_frac": 0.6640625, "sample": [-0.4032745361328125, 5.311729431152344, 9.241165161132812, -2.319976806640625, 5.14996337890625, -1.521392822265625, 0.879913330078125, -0.559600830078125, 2.4618682861328125, -3.2154998779296875, 12.567840576171875, -1.024688720703125, 0.0, 7.51959228515625, -0.04803657531738281, -0.9849205017089844, -0.338775634765625, -2.8342056274414062, -5.860954284667969, -1.4798583984375, 3.10675048828125, -0.455963134765625, -3.032867431640625, -1.9978389739990234, -3.43719482421875, 4.3125457763671875, 2.8795013427734375, 0.0, 10.46990966796875, 0.6993484497070312, -7.5020751953125, 10.59112548828125, 4.552619934082031, 6.2696075439453125, 4.0275115966796875, 1.8413314819335938, 2.8594970703125, 1.2825164794921875, 4.012229919433594, 0.606201171875, 6.2824859619140625, -7.872833251953125, 10.232757568359375, 0.4135761260986328, -0.356109619140625, 5.3595733642578125, 7.00132942199707, 15.156280517578125, 2.365447998046875, 2.70947265625, 2.041107177734375, -3.6331939697265625, -37.19287109375, -5.100494384765625, 2.12518310546875, 7.776702880859375, -0.32073974609375, 15.493560791015625, 9.118057250976562, 9.732666015625, -0.4864616394042969, 6.001220703125, 2.9702911376953125, 7.0148162841796875, 11.1578369140625, 4.3970947265625, 3.6824951171875, 2.681842803955078, -1.1356048583984375, 3.642364501953125, -1.294342041015625, 3.9094772338867188, 0.6115779876708984, 0.874603271484375, -2.0843505859375, -4.389190673828125, 8.18438720703125, -0.70050048828125, 8.684356689453125, 1.38604736328125, -3.93109130859375, 18.136383056640625, 10.6256103515625, 6.059295654296875, 2.913421630859375, 8.9210205078125, 4.4765472412109375, 0.7156791687011719, 0.388702392578125, 1.143798828125, 0.6428871154785156, 5.617160797119141, -0.3425750732421875, 0.08319091796875, 2.121532440185547, 24.09368896484375, -0.034336090087890625, 5.0698699951171875, -3.7303466796875, 7.884735107421875, 13.29541015625, 10.693817138671875, 3.75787353515625, 6.82537841796875, -3.8702392578125, 8.431350708007812, 7.488800048828125, -0.4983062744140625, 8.182708740234375, -0.3765106201171875, -4.2225341796875, 17.41571044921875, 0.2099456787109375, -0.7338790893554688, 4.849811553955078, 6.9835357666015625, 7.665618896484375, 3.78338623046875, 0.19228363037109375, 4.97650146484375, -0.3723106384277344, 8.4775390625, 10.6043701171875, 19.98529052734375, -0.44129180908203125, -2.7003326416015625, 1.6913833618164062, 13.51275634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000079.npy"}
{"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 2.576151132583618, "std": 6.2065935134887695, "min": -23.684051513671875, "p10": -3.590467834472656, "median": 2.308987617492676, "p90": 10.302230834960938, "max": 25.343505859375, "pos_frac": 0.6875, "sample": [2.1933536529541016, 4.633018493652344, -1.9752960205078125, 1.5600852966308594, 3.1016845703125, -2.4044189453125, 1.0823822021484375, -4.931732177734375, 7.0684814453125, -3.92462158203125, 0.2058563232421875, -3.1520233154296875, 5.1474609375, 4.2647705078125, 3.987213134765625, -2.4028472900390625, -0.560089111328125, 10.363006591796875, 3.6421890258789062, 3.77880859375, 16.944305419921875, -0.7910537719726562, 2.9796295166015625, 5.6955413818359375, 0.25177001953125, 10.093963623046875, 5.55865478515625, -3.63226318359375, 3.0610294342041016, 10.27618408203125, 3.116729736328125, 0.7915992736816406, 2.42462158203125, -1.6428031921386719, -12.463577270507812, 1.7499618530273438, -23.684051513671875, 5.0380859375, 0.0, 12.09759521484375, -1.1901092529296875, 9.374053955078125, -2.5356903076171875, -1.5057373046875, 2.788604736328125, -9.301025390625, -0.93206787109375, 6.295166015625, -1.86968994140625, -2.170166015625, 1.3610343933105469, 2.95709228515625, 2.4254608154296875, 5.4202880859375, 9.03466796875, 0.22403526306152344, 0.38373374938964844, 0.0957794189453125, -1.066925048828125, 11.413360595703125, 3.2752685546875, 1.4600372314453125, 5.828441619873047, 5.0916748046875, -4.523712158203125, 1.52099609375, 4.676513671875, 0.7079238891601562, 6.155181884765625, 2.9466896057128906, -2.7003936767578125, 6.352569580078125, 3.0702104568481445, -4.178443908691406, 17.131134033203125, -1.053314208984375, -1.4526138305664062, -0.11089324951171875, 6.4573211669921875, -7.67462158203125, 12.25238037109375, 3.7349853515625, 0.0186767578125, 11.08123779296875, 5.94354248046875, 3.4089698791503906, -6.0586700439453125, 1.211334228515625, -1.7457122802734375, 16.604949951171875, 6.262611389160156, 2.043121337890625, 12.923873901367188, 4.419921875, 4.543363571166992, 11.725860595703125, 7.6861114501953125, 1.60137939453125, -10.841323852539062, 0.43949127197265625, -6.8338623046875, 10.574493408203125, -3.5725555419921875, 3.0184326171875, 2.85565185546875, 9.2044677734375, 0.34320068359375, -0.8205184936523438, 4.144012451171875, 7.764190673828125, -1.718475341796875, 0.1449127197265625, 0.45638275146484375, -0.8401336669921875, 1.356048583984375, -4.5789642333984375, 3.6072731018066406, -0.9471931457519531, 0.9753322601318359, -2.23590087890625, 6.6201629638671875, 10.09210205078125, 9.9146728515625, 4.954071044921875, 25.343505859375, -0.12404060363769531, 3.0367279052734375, 16.0322265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000080.npy"}
{"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 2.870744228363037, "std": 6.568058013916016, "min": -19.4385986328125, "p10": -3.283319854736328, "median": 2.2659454345703125, "p90": 11.426962280273436, "max": 16.5347900390625, "pos_frac": 0.671875, "sample": [-0.2830982208251953, -0.0002040863037109375, 8.524765014648438, -9.828598022460938, 3.891387939453125, -0.041473388671875, -8.825897216796875, 2.3513870239257812, 16.5347900390625, 8.671234130859375, 15.669265747070312, 12.93377685546875, -2.3251190185546875, -3.4077072143554688, 8.190887451171875, -0.2066974639892578, 4.4779052734375, 7.9832611083984375, 5.17974853515625, 4.2175750732421875, 0.8150367736816406, 11.309326171875, -0.1397552490234375, -2.3898162841796875, 5.506492614746094, 3.4465484619140625, 2.165313720703125, -0.5672607421875, 9.062408447265625, 6.646240234375, 12.84100341796875, 1.8608856201171875, 7.255455017089844, 5.2784423828125, -1.5592041015625, 2.6027984619140625, -4.586029052734375, 15.86700439453125, 14.7978515625, 2.6742095947265625, -5.0167236328125, 1.38482666015625, 10.680511474609375, 0.0433197021484375, -0.80322265625, 1.2030181884765625, -14.71575927734375, -2.41815185546875, 7.86700439453125, 10.24603271484375, 1.56964111328125, 2.6776885986328125, -6.9974822998046875, 8.507904052734375, -5.18310546875, -0.8237190246582031, 4.38214111328125, 4.8645477294921875, 1.9280853271484375, -14.906402587890625, -5.215057373046875, 2.8314285278320312, 3.136077880859375, 3.197357177734375, -0.005023956298828125, 3.3079605102539062, 0.480224609375, 0.34696197509765625, 3.459686279296875, 4.1312408447265625, -0.013275146484375, -1.678009033203125, -0.1329803466796875, 11.566009521484375, 13.881744384765625, 9.8453369140625, -1.3948974609375, 5.6771392822265625, 2.1805038452148438, 7.01751708984375, -13.30853271484375, 7.0978240966796875, 10.9053955078125, -2.8286895751953125, 0.82373046875, 2.0928726196289062, -0.9259757995605469, -3.230010986328125, -0.828460693359375, 2.545440673828125, -19.4385986328125, -1.04815673828125, -13.67840576171875, 1.62628173828125, 1.5613555908203125, 1.6903610229492188, -0.7000732421875, 0.05763816833496094, 12.74029541015625, 12.6558837890625, 11.36737060546875, 12.69854736328125, 5.3535003662109375, -2.5728759765625, 10.68585205078125, 2.4739761352539062, -2.0137939453125, 1.5000534057617188, 0.3890972137451172, 6.799591064453125, -3.1836776733398438, 6.215858459472656, -0.06001091003417969, 0.292144775390625, 2.537841796875, 10.186508178710938, 8.919418334960938, -1.4028778076171875, 9.69915771484375, 6.64739990234375, 5.493133544921875, 13.6917724609375, 14.12030029296875, 9.366683959960938, 0.4248046875, 0.6980438232421875, 6.446285247802734, -0.8332366943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000081.npy"}
{"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 2.9909470081329346, "std": 5.935222148895264, "min": -15.76641845703125, "p10": -4.006354522705078, "median": 2.9362945556640625, "p90": 10.564408874511717, "max": 20.6416015625, "pos_frac": 0.734375, "sample": [2.3895721435546875, 6.389801025390625, 1.2496414184570312, -6.1482696533203125, -2.19708251953125, 4.22503662109375, 3.4039306640625, 11.993499755859375, 7.6194915771484375, 1.072723388671875, 0.8130340576171875, 6.91949462890625, -4.860939025878906, -1.616241455078125, -2.0845947265625, 5.2581634521484375, 5.341766357421875, 2.4041061401367188, 12.663772583007812, 5.824470520019531, 0.8378524780273438, -3.189849853515625, 6.805931091308594, -2.121002197265625, 0.3826141357421875, 11.24078369140625, -10.24310302734375, 4.975067138671875, 3.431671142578125, 5.645263671875, 17.89434814453125, 2.5897369384765625, 10.395095825195312, 3.939849853515625, 10.11456298828125, 11.5770263671875, -1.1429443359375, 4.6353759765625, 6.037261962890625, -15.76641845703125, -4.7687835693359375, -0.2662353515625, 3.995513916015625, 4.442390441894531, -2.8991241455078125, 9.557220458984375, -0.724456787109375, 8.87384033203125, 17.265869140625, -2.206451416015625, -2.337799072265625, 18.471633911132812, 5.1104736328125, 4.4457244873046875, 2.588531494140625, 0.34409523010253906, 10.95947265625, -3.907470703125, -0.8860321044921875, 3.1017913818359375, 3.5007171630859375, -5.0748748779296875, -13.038177490234375, 3.3808441162109375, -4.704833984375, 2.0726089477539062, -4.3732147216796875, 2.015411376953125, 13.368804931640625, 2.592987060546875, -0.19191741943359375, 7.1375885009765625, 1.157379150390625, 4.22265625, 0.075439453125, 3.2285919189453125, -0.133819580078125, 1.208456039428711, 5.135478973388672, 2.10107421875, 1.7013435363769531, 20.6416015625, -3.9499969482421875, 5.6097412109375, -0.4547576904296875, 2.7707977294921875, 6.344329833984375, 10.301658630371094, 1.6285247802734375, 2.4279022216796875, 3.2389984130859375, -2.921051025390625, 2.7255287170410156, 5.15277099609375, 6.722745895385742, 6.2305908203125, -5.2911376953125, 2.4907073974609375, 5.466339111328125, 10.35595703125, 5.344024658203125, 0.2068328857421875, 0.48628997802734375, 4.4950714111328125, 6.558685302734375, 3.488035202026367, 0.11839485168457031, -2.019134521484375, -6.60235595703125, 2.1199951171875, -4.137855529785156, 4.4043426513671875, 6.7096710205078125, 6.518226623535156, -3.6002349853515625, 11.97918701171875, 13.003143310546875, 2.6863555908203125, 3.7386627197265625, 6.008941650390625, 4.85858154296875, 4.695587158203125, 13.2183837890625, 6.3105926513671875, 0.7330951690673828, 0.6519317626953125, -9.97784423828125, -3.889892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000082.npy"}
{"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 3.3800435066223145, "std": 6.819630146026611, "min": -15.1282958984375, "p10": -3.444300842285156, "median": 2.7335281372070312, "p90": 12.030741119384764, "max": 27.75341796875, "pos_frac": 0.7109375, "sample": [6.136962890625, 0.0, 10.4114990234375, 3.7261962890625, 1.466156005859375, 0.35992431640625, -1.381866455078125, 7.4476318359375, 3.11767578125, 0.27783966064453125, -2.1444091796875, -15.1282958984375, -0.5007400512695312, -11.644866943359375, 0.798370361328125, 0.80499267578125, 3.8677520751953125, 10.72271728515625, 7.8438873291015625, 2.5763587951660156, 8.9390869140625, 6.630409240722656, -2.8458709716796875, 4.8505706787109375, 9.6640625, -4.11376953125, -1.958740234375, -0.8923110961914062, -1.443603515625, 12.18658447265625, 5.88525390625, 1.24554443359375, 4.361602783203125, 1.412872314453125, -6.234375, 14.70220947265625, 3.4102325439453125, 3.367156982421875, 19.344497680664062, 1.6354827880859375, 4.553520202636719, -1.9423828125, 3.07769775390625, 3.29351806640625, -4.134765625, 11.39892578125, 5.414459228515625, 3.8367691040039062, -2.550933837890625, -2.8312911987304688, 5.449798583984375, 5.903778076171875, 18.4854736328125, 6.9522705078125, -11.426361083984375, 0.9017181396484375, 10.20001220703125, 4.61273193359375, 27.75341796875, 0.25982666015625, -3.123443603515625, 11.963951110839844, 4.570343017578125, 6.6524505615234375, 2.0080184936523438, -1.602447509765625, 2.8394775390625, 0.47174835205078125, -3.4201812744140625, 8.682159423828125, -9.200439453125, 5.92730712890625, 0.17978668212890625, -1.29107666015625, -2.5504150390625, 3.4969863891601562, 15.62615966796875, -3.500579833984375, 2.23919677734375, 8.702430725097656, -3.3178787231445312, 13.561126708984375, 4.58935546875, 4.6038818359375, -13.0440673828125, -5.7808837890625, 6.882946014404297, 14.4854736328125, -9.475982666015625, 11.7852783203125, 0.06168365478515625, -1.6463699340820312, 0.662628173828125, -3.177703857421875, 5.121402740478516, 16.396392822265625, 13.54388427734375, 1.684722900390625, 2.474609375, 1.7639236450195312, 15.76397705078125, -1.04669189453125, 6.09405517578125, 2.6212158203125, -2.2625732421875, 0.6236038208007812, 7.6882781982421875, 5.144439697265625, 2.6275787353515625, -1.22235107421875, 9.449310302734375, -3.9644927978515625, 9.7457275390625, 9.258056640625, 1.046630859375, 3.9373321533203125, 1.875274658203125, 16.790283203125, 19.069183349609375, -3.82940673828125, 1.7252159118652344, -1.1179046630859375, 8.203887939453125, -1.6819076538085938, 4.1630706787109375, 6.46514892578125, 5.048095703125, 2.4718170166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000083.npy"}
{"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 2.77628231048584, "std": 7.115567684173584, "min": -22.0458984375, "p10": -3.822842407226562, "median": 3.0348119735717773, "p90": 11.523545837402342, "max": 19.23089599609375, "pos_frac": 0.6640625, "sample": [-1.3940696716308594, 7.28582763671875, 15.736785888671875, -0.38608551025390625, -2.853179931640625, -3.53265380859375, 6.72296142578125, 3.7982406616210938, 5.361625671386719, -4.5999755859375, 0.6927261352539062, 8.40496826171875, 1.7421722412109375, -2.2237014770507812, 4.818115234375, 3.3559722900390625, 6.245491027832031, 0.5365753173828125, -0.5410041809082031, 9.690200805664062, -3.391918182373047, 8.0201416015625, 3.112945556640625, 3.87713623046875, 8.06683349609375, -0.55535888671875, -9.08687973022461, 5.17413330078125, 8.00579833984375, 4.7193145751953125, -1.32305908203125, -19.4581298828125, -22.0458984375, 14.4244384765625, 2.41094970703125, -0.846893310546875, 7.42919921875, 2.0934219360351562, 1.073150634765625, -0.5730361938476562, 4.589874267578125, 2.1607666015625, 12.383026123046875, -3.23773193359375, 12.6248779296875, -2.264129638671875, 1.0222320556640625, -3.2554702758789062, 13.6748046875, 4.7285614013671875, 18.849456787109375, -17.18994140625, 9.9322509765625, 2.35467529296875, 4.160377502441406, -3.6514739990234375, 3.5251617431640625, -3.45648193359375, 7.3716888427734375, 7.956443786621094, 5.883697509765625, 3.8365936279296875, -0.4215087890625, 4.0302581787109375, 11.205322265625, 0.889129638671875, 4.057830810546875, -0.1033172607421875, -0.549468994140625, 3.0074615478515625, -4.1695098876953125, 4.311126708984375, 2.459136962890625, -0.4432525634765625, 2.7960281372070312, 14.003448486328125, 7.661838531494141, 9.181175231933594, -13.03546142578125, 4.19158935546875, 3.3834457397460938, 1.322235107421875, -3.4754867553710938, 2.349821090698242, -4.21783447265625, -0.3283710479736328, -1.570068359375, 7.936492919921875, 17.05133056640625, -0.48614501953125, 9.76678466796875, 4.242805480957031, -5.207489013671875, 5.680816650390625, -0.14849090576171875, -0.8479423522949219, 4.0526580810546875, 3.062162399291992, 3.200714111328125, 2.824859619140625, 5.174896240234375, -0.8731918334960938, 2.526763916015625, 2.1704788208007812, 9.800155639648438, 6.042804718017578, -4.2841796875, 19.23089599609375, -13.191650390625, 9.81003189086914, 17.288360595703125, 1.858734130859375, 6.733482360839844, 4.81512451171875, 17.929443359375, 9.77984619140625, -7.320556640625, -16.22801971435547, 4.205078125, -0.15966796875, 0.3775806427001953, 12.266067504882812, 13.71319580078125, -3.2945480346679688, 3.4534454345703125, 5.7724456787109375, 1.790740966796875, -3.6742706298828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000084.npy"}
{"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 1.955026388168335, "std": 5.742084503173828, "min": -14.5870361328125, "p10": -5.778296661376952, "median": 2.497142791748047, "p90": 8.485348510742188, "max": 16.335601806640625, "pos_frac": 0.6640625, "sample": [2.441497802734375, -2.6721954345703125, -1.5260124206542969, 15.231109619140625, 6.00714111328125, 4.9530792236328125, -1.67803955078125, -0.530914306640625, 6.513153076171875, 3.0904541015625, 2.0389041900634766, -3.180694580078125, 1.9225120544433594, 8.069808959960938, 5.401641845703125, 3.209564208984375, 6.321502685546875, 5.7955780029296875, 6.195159912109375, 2.6211929321289062, 1.227386474609375, -0.2105712890625, 9.170974731445312, 14.265213012695312, 1.3110389709472656, -0.676605224609375, -1.351470947265625, -4.5013580322265625, -8.206085205078125, 5.9378662109375, 5.13055419921875, 4.0370941162109375, -1.4076995849609375, 3.5343017578125, 10.262451171875, 10.66827392578125, -3.6197128295898438, 5.45648193359375, 0.31052398681640625, -1.51556396484375, 10.607131958007812, 4.2623291015625, -1.0672607421875, 5.643592834472656, 11.149078369140625, -14.5870361328125, 2.3577957153320312, 2.8826065063476562, 3.0930709838867188, 13.87310791015625, -0.99114990234375, -6.459442138671875, 4.880247116088867, 1.33209228515625, 2.6236343383789062, -4.945198059082031, -0.2537841796875, 1.3264236450195312, 5.6955108642578125, 2.5527877807617188, -4.41748046875, 4.5184783935546875, 3.9281845092773438, -2.668182373046875, 7.350196838378906, 5.224639892578125, 4.00506591796875, 4.78399658203125, -4.45361328125, -9.31719970703125, -3.229461669921875, 6.395866394042969, -8.248931884765625, -1.825286865234375, 0.05660247802734375, -6.35418701171875, 10.802490234375, 8.5234375, 0.6143684387207031, 4.2903289794921875, 2.869140625, 8.469024658203125, 3.8810882568359375, 6.869415283203125, 16.335601806640625, -12.008697509765625, 0.05377197265625, 8.37225341796875, 3.270893096923828, -4.97186279296875, -9.751747131347656, 1.77587890625, 4.178436279296875, 2.911548614501953, 10.291717529296875, 2.352863311767578, -1.7187957763671875, 3.4480743408203125, -6.353912353515625, 3.7856521606445312, -0.4537544250488281, 6.885589599609375, 1.0709228515625, -0.832244873046875, 0.6340179443359375, 1.82891845703125, -7.3648223876953125, 0.4420928955078125, 7.649871826171875, 3.7016754150390625, -6.20123291015625, -12.324951171875, -0.5942535400390625, 6.9471282958984375, -3.292236328125, 2.298095703125, 6.3282012939453125, 1.3314437866210938, 13.757186889648438, -9.14141845703125, -2.9589385986328125, 4.652923583984375, -5.597038269042969, 2.33587646484375, 4.8000640869140625, -2.396352767944336, 3.1587066650390625, 5.515192031860352], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000085.npy"}
{"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 3.675945997238159, "std": 7.590166091918945, "min": -16.4664306640625, "p10": -4.493408203125, "median": 3.6366119384765625, "p90": 12.682342529296873, "max": 25.85687255859375, "pos_frac": 0.6953125, "sample": [4.620941162109375, -10.56219482421875, 4.49822998046875, -6.11480712890625, 5.1645660400390625, 3.6218719482421875, 3.925567626953125, 3.7043380737304688, 6.91064453125, 5.615325927734375, 25.85687255859375, -3.701995849609375, 1.5335273742675781, 1.7493133544921875, 7.206287384033203, -9.935028076171875, -12.5406494140625, -4.0647125244140625, 0.8145332336425781, -1.5519790649414062, 2.5567169189453125, 5.7538299560546875, -1.7737998962402344, -0.44305419921875, 16.006103515625, -1.45281982421875, -13.53460693359375, 0.94512939453125, 19.2369384765625, 13.182647705078125, -7.926849365234375, 11.199615478515625, -0.5220565795898438, 10.2974853515625, 20.55279541015625, 7.92999267578125, 12.467926025390625, -4.51348876953125, 3.1788330078125, 11.918792724609375, 3.6513519287109375, -3.391845703125, 8.351577758789062, -1.370269775390625, 3.8669662475585938, 1.5459365844726562, 8.27096939086914, 12.169952392578125, 10.146820068359375, 5.583343505859375, -16.31414794921875, 12.10333251953125, 1.591064453125, 7.3713226318359375, 5.819305419921875, -2.1536712646484375, 12.33892822265625, -16.4664306640625, 1.7295379638671875, 9.67120361328125, 19.691940307617188, 4.16156005859375, 3.08685302734375, 11.5731201171875, 8.695526123046875, -2.48358154296875, 0.0, -1.815887451171875, 3.0213661193847656, 6.080291748046875, 4.581756591796875, 13.73095703125, -11.8953857421875, 4.503578186035156, 2.046468734741211, 1.72589111328125, -0.4488792419433594, -0.4752197265625, 11.750091552734375, 8.925369262695312, -2.0817413330078125, 7.73052978515625, 10.903076171875, 0.1583099365234375, 5.231109619140625, -8.16815185546875, -11.611236572265625, 7.883514404296875, -4.221923828125, 0.6183929443359375, -0.9274330139160156, 6.0873565673828125, 1.452667236328125, 1.6119232177734375, 2.51470947265625, -5.1346435546875, 4.499053955078125, 6.17242431640625, 2.791595458984375, 3.579458236694336, 0.033935546875, -4.48480224609375, -4.136421203613281, 13.70880126953125, 14.90728759765625, 8.862518310546875, -1.0380325317382812, 22.864990234375, 2.120502471923828, 5.126800537109375, -2.0750160217285156, 5.738739013671875, 13.518203735351562, 4.690765380859375, 11.73931884765625, -0.45121002197265625, 6.1645965576171875, 15.19732666015625, -0.0625762939453125, 5.66668701171875, 6.582977294921875, 10.51534652709961, 9.157318115234375, 13.433624267578125, -1.0380859375, -2.377655029296875, 1.225433349609375, 3.0588836669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000086.npy"}
{"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 4.105223178863525, "std": 6.792147636413574, "min": -18.851959228515625, "p10": -2.5716869354248044, "median": 3.3288650512695312, "p90": 13.92198486328125, "max": 24.563873291015625, "pos_frac": 0.71875, "sample": [4.9061279296875, 8.731346130371094, 5.783843994140625, 9.6756591796875, 16.874237060546875, 0.0, -0.3388214111328125, -2.4998512268066406, 8.989727020263672, 15.2603759765625, -3.99713134765625, 7.622894287109375, -3.044677734375, 10.081823348999023, 24.563873291015625, -0.44620323181152344, 0.8018760681152344, 8.136962890625, 15.511123657226562, 3.9368515014648438, -2.0264892578125, -7.410552978515625, 0.71832275390625, 5.9008026123046875, 2.996856689453125, -1.86767578125, -0.9748382568359375, 5.618408203125, 16.141387939453125, 0.958648681640625, 10.73486328125, -6.6229248046875, 4.91729736328125, 16.02728271484375, 6.732879638671875, 4.0517730712890625, 4.578819274902344, 3.44000244140625, 5.35504150390625, 0.93170166015625, -0.6890220642089844, -0.853424072265625, 7.8839111328125, -2.7393035888671875, -2.2066497802734375, -1.062713623046875, 10.43804931640625, -4.590667724609375, 10.972991943359375, 3.38739013671875, 8.131904602050781, 0.166259765625, 14.22772216796875, 16.1951904296875, 12.29217529296875, 9.336944580078125, 6.7296142578125, 9.392578125, 11.1182861328125, 18.300506591796875, 2.36566162109375, 0.753631591796875, 2.0079421997070312, 10.19891357421875, 3.1791839599609375, 21.58416748046875, -0.6875, 0.0, 3.831756591796875, 3.89923095703125, 3.3866729736328125, 5.643096923828125, 13.20404052734375, 5.14697265625, 14.422103881835938, 3.7912750244140625, 3.502899169921875, -0.7346115112304688, 9.8016357421875, -0.0264892578125, -8.24822998046875, -3.1795501708984375, -1.5203094482421875, 1.0064697265625, -0.9308929443359375, 11.99761962890625, -3.68463134765625, 6.579160690307617, 3.313690185546875, 7.1463623046875, 3.65716552734375, 1.51611328125, 2.9109649658203125, -1.23773193359375, 0.78955078125, -13.84295654296875, 1.5100173950195312, 1.3460063934326172, -6.313507080078125, 13.79095458984375, -0.25621795654296875, 3.3440399169921875, -0.5941810607910156, 0.93487548828125, 12.57763671875, -18.851959228515625, 4.0668792724609375, -6.8768310546875, 0.06230926513671875, 2.2063140869140625, 13.47821044921875, 15.672027587890625, -1.028717041015625, 7.053070068359375, -0.154327392578125, 0.42620849609375, 9.137985229492188, 2.4888076782226562, 2.50341796875, 1.0799026489257812, 4.363861083984375, 1.8690338134765625, -2.3831024169921875, 2.386180877685547, 6.1225433349609375, 1.0789794921875, 14.7166748046875, 0.9846992492675781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000087.npy"}
{"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 3.250743865966797, "std": 7.431678295135498, "min": -26.794036865234375, "p10": -4.294695281982421, "median": 2.3234901428222656, "p90": 13.281439208984374, "max": 26.58538818359375, "pos_frac": 0.6953125, "sample": [8.578567504882812, 14.936492919921875, 10.70294189453125, 0.2223644256591797, 17.941436767578125, 2.3153152465820312, -2.309234619140625, 0.71722412109375, 7.306800842285156, 3.516071319580078, 1.558197021484375, -4.2417755126953125, 2.60064697265625, 21.440216064453125, -0.665191650390625, -2.00762939453125, 0.06024742126464844, 7.508392333984375, 1.1269912719726562, -2.34588623046875, 13.48126220703125, -4.5672149658203125, 10.279266357421875, -3.5369949340820312, -1.0467529296875, 5.908477783203125, -5.7747650146484375, 1.13482666015625, 13.55712890625, -6.843902587890625, -8.348876953125, 9.552360534667969, 3.509429931640625, 6.552490234375, 2.4792098999023438, -1.5279197692871094, 19.533538818359375, 0.0, 0.15239334106445312, 2.419464111328125, 6.703125, 4.8278350830078125, 10.657577514648438, -2.2301025390625, 15.297317504882812, 13.19580078125, 2.5245361328125, -6.3262939453125, 2.2341041564941406, 3.6021575927734375, 0.0, 4.034423828125, 3.218475341796875, 6.65960693359375, 14.022186279296875, 0.839935302734375, -0.8916282653808594, 1.3726959228515625, 0.161102294921875, 5.026397705078125, 10.151596069335938, -0.9002132415771484, 0.484222412109375, -6.043647766113281, 5.078582763671875, 2.3316650390625, -4.418174743652344, 10.71429443359375, 0.47296905517578125, 10.31622314453125, 2.9945335388183594, -5.44036865234375, 12.2918701171875, 0.38549041748046875, -0.952545166015625, -21.373565673828125, 3.6916046142578125, 1.3304519653320312, -0.0757293701171875, -2.4003524780273438, 7.2378082275390625, 7.6324462890625, 7.91943359375, -1.6444034576416016, -0.31561279296875, -4.4580078125, -2.2407073974609375, -6.6728515625, 6.608909606933594, 7.611320495605469, 0.14465904235839844, 3.579925537109375, 2.677703857421875, 1.9283447265625, 5.4105987548828125, 15.428619384765625, 0.63848876953125, -2.35467529296875, -2.1830596923828125, 18.770751953125, 5.3404541015625, 3.5656890869140625, -1.3779830932617188, 9.209259033203125, -1.8194732666015625, 26.58538818359375, 13.69720458984375, -0.27604103088378906, 11.30682373046875, -9.899322509765625, 8.43414306640625, 1.1493301391601562, 2.723419189453125, 2.651763916015625, 21.242431640625, 4.19135856628418, -1.2304134368896484, 0.19864845275878906, 1.6352081298828125, 0.9322509765625, -26.794036865234375, 7.689788818359375, -2.6976318359375, 7.864891052246094, 1.3726959228515625, 9.831939697265625, 5.26226806640625, 0.14173126220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000088.npy"}
{"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 3.8826417922973633, "std": 6.812313079833984, "min": -18.1171875, "p10": -2.739274406433105, "median": 3.1840152740478516, "p90": 11.81005859375, "max": 26.00982666015625, "pos_frac": 0.7890625, "sample": [0.7013702392578125, -11.4150390625, 16.620635986328125, 2.186920166015625, 4.6443939208984375, 17.47808837890625, -1.453592300415039, 1.6829071044921875, 0.917236328125, 8.6800537109375, 2.79443359375, 3.6414947509765625, 3.66387939453125, 10.738525390625, 2.909332275390625, 11.63037109375, 1.2928524017333984, -3.8504791259765625, 5.6622467041015625, 2.007781982421875, 5.550384521484375, 13.434814453125, -3.3599472045898438, 0.7033061981201172, 1.6646957397460938, 1.4399261474609375, 4.456939697265625, -1.091705322265625, -1.200897216796875, 2.8973541259765625, 2.024566650390625, 0.6993560791015625, -0.917633056640625, 13.65869140625, 14.00860595703125, 26.00982666015625, 11.731689453125, -8.11651611328125, 0.8997039794921875, 2.532440185546875, -18.1171875, -2.151641845703125, 3.7923583984375, -8.427581787109375, -0.9312744140625, 4.391357421875, 8.18939208984375, -11.993362426757812, 0.250244140625, -2.3463134765625, -0.08817100524902344, 8.55029296875, 3.68292236328125, 0.40216064453125, 9.154739379882812, 9.30926513671875, 6.0945281982421875, -3.0136051177978516, 7.6252593994140625, 9.45318603515625, 2.182598114013672, 15.81903076171875, 8.24346923828125, 7.631629943847656, 7.13946533203125, 1.6645660400390625, -6.483978271484375, 2.3370361328125, 1.1928024291992188, -3.5426177978515625, 3.888031005859375, 6.052520751953125, 4.466621398925781, 6.987213134765625, -5.4832611083984375, 0.14679718017578125, 3.489208221435547, 8.50830078125, 21.9776611328125, -4.7755126953125, 1.84039306640625, 6.125823974609375, 3.7243804931640625, 8.463836669921875, 1.215301513671875, 11.73297119140625, 3.1028709411621094, 14.028106689453125, 9.532684326171875, 19.983428955078125, 5.094856262207031, -2.39886474609375, 13.097251892089844, 2.361541748046875, 2.504291534423828, 3.2651596069335938, 25.0848388671875, -0.9992904663085938, 1.5598678588867188, 3.7273788452148438, 7.814727783203125, 1.6506309509277344, 2.665008544921875, -0.696685791015625, 3.8116073608398438, 5.500480651855469, -1.91595458984375, -0.971221923828125, 0.789398193359375, 4.6393585205078125, 4.3253021240234375, 3.3057727813720703, 10.043647766113281, 11.98992919921875, 6.4327392578125, 4.381080627441406, 5.0367584228515625, 4.043914794921875, 8.099639892578125, -14.784835815429688, 2.3197383880615234, 11.282470703125, 5.3848876953125, 2.053985595703125, 1.7544059753417969, 2.908447265625, 1.8606109619140625, -2.6217041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000089.npy"}
{"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 3.4542462825775146, "std": 7.377211570739746, "min": -21.262451171875, "p10": -4.787701416015625, "median": 2.900043487548828, "p90": 12.3369384765625, "max": 25.665496826171875, "pos_frac": 0.703125, "sample": [9.2811279296875, 4.800079345703125, 0.7205429077148438, 18.15142822265625, -10.847091674804688, -0.23333740234375, -3.87420654296875, 0.9591064453125, -3.12249755859375, 11.32476806640625, 2.0278549194335938, 2.8761978149414062, -7.2755889892578125, -0.32653045654296875, -2.12841796875, 6.757080078125, 22.783401489257812, 18.017303466796875, 11.06011962890625, 0.5214614868164062, 5.947052001953125, -0.294677734375, 0.7862091064453125, 2.92388916015625, -12.583816528320312, -6.4460906982421875, 12.496917724609375, 7.454345703125, 7.83624267578125, 4.1377410888671875, -1.8876800537109375, 5.688514709472656, 8.75469970703125, 3.61053466796875, 8.04248046875, 2.1919708251953125, -5.537445068359375, -3.96490478515625, 0.086669921875, 6.0529327392578125, -5.044883728027344, 2.243743896484375, -0.343475341796875, -10.86895751953125, 7.507232666015625, 10.1103515625, 5.44403076171875, 8.3448486328125, 5.675567626953125, 1.73492431640625, 8.455474853515625, -2.2613372802734375, 0.9669189453125, 1.5189647674560547, 0.37408447265625, 1.1837615966796875, -11.373077392578125, 0.1423797607421875, 0.42812347412109375, 6.337497711181641, 0.2653656005859375, 0.0, 12.92999267578125, 6.751800537109375, 10.325496673583984, -0.4557037353515625, 8.608642578125, 2.9279098510742188, -1.74102783203125, 11.482086181640625, 1.6420135498046875, 2.196319580078125, 5.23046875, 15.4151611328125, 11.2049560546875, 0.760009765625, 0.0652475357055664, 3.4235153198242188, 3.2304458618164062, 10.14166259765625, 12.314453125, -21.262451171875, 11.73248291015625, -10.50537109375, 0.0742950439453125, 3.997802734375, 4.964973449707031, 2.7525634765625, -0.3472099304199219, -3.4834518432617188, -1.0274658203125, -4.498779296875, 10.625885009765625, 12.389404296875, -3.45648193359375, 1.91827392578125, -4.886138916015625, 5.8579559326171875, 8.902862548828125, 10.26214599609375, -1.21136474609375, 11.881778717041016, 6.948516845703125, 0.15798187255859375, -2.2264251708984375, 25.665496826171875, -1.9724807739257812, 4.4240570068359375, -5.492218017578125, -11.103057861328125, 5.2596282958984375, 1.04046630859375, -4.745513916015625, 12.21588134765625, 4.202056884765625, -0.3197460174560547, 14.17303466796875, 3.532623291015625, 14.33154296875, 3.227142333984375, -0.8306884765625, -2.2277584075927734, 21.03131103515625, 14.05206298828125, 3.2025146484375, 12.226486206054688, 3.14642333984375, 13.479095458984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000090.npy"}
{"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 3.438483238220215, "std": 7.675367832183838, "min": -13.186614990234375, "p10": -6.1161254882812495, "median": 3.6446762084960938, "p90": 13.269223022460936, "max": 25.241607666015625, "pos_frac": 0.6484375, "sample": [5.957366943359375, 1.6912078857421875, -0.9048843383789062, 16.491989135742188, 9.6295166015625, 13.961898803710938, 0.6418857574462891, 3.9934616088867188, 4.99298095703125, -2.826507568359375, 1.287567138671875, 6.2631683349609375, 6.417854309082031, 6.5680694580078125, 6.100221633911133, 7.836669921875, 2.0494613647460938, -13.186614990234375, -4.76422119140625, -0.01934814453125, 7.689521789550781, 5.351715087890625, 6.0141448974609375, 7.5068359375, -1.5238533020019531, 6.459320068359375, 0.4511299133300781, -0.7425079345703125, -0.524444580078125, -7.8718414306640625, 12.220458984375, -0.3450965881347656, -0.4467926025390625, -5.7701416015625, -3.792510986328125, 8.858413696289062, -5.5419769287109375, 6.04107666015625, 5.819221496582031, 6.2665252685546875, -7.902984619140625, -0.8505325317382812, 4.717926025390625, -3.545196533203125, 5.11474609375, 20.05694580078125, -6.055572509765625, -5.02227783203125, 0.0, -11.599945068359375, -6.295036315917969, -4.236900329589844, -12.1605224609375, 2.560699462890625, 7.318511962890625, 5.749267578125, 0.80816650390625, -6.257415771484375, 13.477142333984375, 5.5743560791015625, 9.40789794921875, -8.7847900390625, 3.5048980712890625, 0.4642467498779297, -1.9964752197265625, -2.382537841796875, 22.972091674804688, -1.321441650390625, -0.2365875244140625, 11.587814331054688, -7.4089508056640625, -5.1915283203125, 25.241607666015625, 3.7858829498291016, 15.01751708984375, 1.0600967407226562, 4.5859832763671875, 0.6823043823242188, 4.0399017333984375, 12.160858154296875, -8.195098876953125, -4.578826904296875, 21.586151123046875, 9.60772705078125, 13.66064453125, 0.0, 8.802558898925781, 3.784454345703125, 6.64251708984375, 13.18011474609375, 4.075950622558594, 8.14996337890625, 4.001716613769531, -8.60430908203125, 0.990203857421875, 12.41265869140625, -8.601951599121094, 2.5400161743164062, 4.1684722900390625, 19.66455078125, 6.911529541015625, 8.118942260742188, 9.29608154296875, 3.8609619140625, 3.89373779296875, -9.565948486328125, -2.1014404296875, 12.231986999511719, -0.80126953125, 19.534423828125, -2.45697021484375, 18.347793579101562, 21.392730712890625, -0.1951885223388672, -1.474517822265625, 10.039276123046875, 2.0948104858398438, -4.128448486328125, 7.349700927734375, 1.2571945190429688, 9.491943359375, -4.2329559326171875, 1.67022705078125, 1.7164230346679688, 5.280853271484375, 2.94561767578125, 7.7225799560546875, 1.6971511840820312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000091.npy"}
{"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 3.778657913208008, "std": 7.912725925445557, "min": -29.0321044921875, "p10": -5.459805297851562, "median": 3.3975601196289062, "p90": 13.419631958007812, "max": 21.456024169921875, "pos_frac": 0.71875, "sample": [-5.510711669921875, 0.22766876220703125, -2.1280517578125, 3.7686767578125, 2.4899368286132812, 5.344398498535156, 4.796241760253906, 3.589111328125, 11.984100341796875, 6.564605712890625, 1.0899124145507812, 2.2811203002929688, 3.740203857421875, -0.096282958984375, 20.189422607421875, 19.5194091796875, 19.0355224609375, -7.86920166015625, 6.5811309814453125, 1.3649253845214844, -7.376068115234375, 8.832122802734375, 15.771438598632812, 12.4742431640625, 2.51092529296875, -2.4191131591796875, 3.2060089111328125, 12.4248046875, 2.903472900390625, -1.807952880859375, 4.88116455078125, 1.2993316650390625, 12.401763916015625, 8.29437255859375, 0.738433837890625, 1.0910491943359375, -1.099151611328125, 3.0939178466796875, 13.369857788085938, 10.77972412109375, 1.2031631469726562, 2.42645263671875, 0.2030658721923828, -29.0321044921875, 10.11907958984375, 3.14837646484375, 3.942962646484375, 3.024139404296875, 0.149566650390625, 1.3632049560546875, -4.36102294921875, -8.164459228515625, -8.8018798828125, -0.324249267578125, 17.468017578125, -5.88348388671875, -10.608505249023438, -0.0194549560546875, -7.002471923828125, 6.9693603515625, 5.88800048828125, 5.9192047119140625, 8.42730712890625, 5.379096984863281, 5.982818603515625, 4.457206726074219, 10.020191192626953, 0.0, -3.119873046875, 13.4859619140625, -2.26678466796875, 3.88665771484375, 2.119964599609375, -1.0381126403808594, -2.6628799438476562, 8.346878051757812, -0.7573204040527344, 1.2204055786132812, 10.306884765625, -1.0182523727416992, -6.4238433837890625, 14.658447265625, 4.5281524658203125, -22.26434326171875, -2.1854934692382812, 0.12939453125, 6.1477813720703125, 19.3109130859375, 19.910980224609375, 4.737358093261719, 7.589130401611328, 7.067352294921875, -0.4520759582519531, -0.2243804931640625, 0.16786766052246094, 21.456024169921875, -1.241668701171875, -5.43798828125, 19.652984619140625, 5.61297607421875, 11.3480224609375, 10.865005493164062, -11.1724853515625, 13.391204833984375, 8.9195556640625, 3.81060791015625, 2.903900146484375, 4.5528564453125, 3.0403213500976562, 4.500030517578125, 8.897857666015625, 4.13409423828125, 9.348251342773438, -9.64849853515625, 6.2046051025390625, 7.987060546875, 0.3769817352294922, 2.401214599609375, -1.345794677734375, 8.32415771484375, -3.366546630859375, 8.819509506225586, 3.7745513916015625, 2.4374771118164062, 16.19281005859375, 17.401611328125, 9.23919677734375, -1.1385345458984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000092.npy"}
{"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 3.501368999481201, "std": 7.500576972961426, "min": -17.677886962890625, "p10": -4.159017181396484, "median": 2.482696533203125, "p90": 14.425430297851562, "max": 22.161773681640625, "pos_frac": 0.6796875, "sample": [0.7202301025390625, 13.4488525390625, 0.6044464111328125, 21.0836181640625, -6.139312744140625, 8.194198608398438, -1.6563339233398438, 11.298873901367188, 2.43853759765625, 4.903158187866211, 3.8280029296875, -9.430831909179688, 1.0379638671875, 3.743621826171875, 4.400604248046875, 12.4691162109375, 2.032562255859375, 16.763290405273438, -2.7926559448242188, 4.9515380859375, 5.9644317626953125, 4.9969635009765625, 3.149433135986328, 5.014595031738281, 14.541534423828125, 1.29107666015625, -0.100799560546875, 2.01312255859375, 11.21026611328125, 11.41357421875, 2.4705352783203125, -9.199676513671875, 0.13446807861328125, -1.921051025390625, 2.8275909423828125, 5.4564971923828125, 4.995941162109375, 4.536346435546875, -4.5149078369140625, 13.812835693359375, 2.4948577880859375, 5.888786315917969, 3.7686004638671875, -3.9751129150390625, -0.14049530029296875, 4.6341705322265625, 0.20458984375, 14.659820556640625, 1.27154541015625, -1.3327789306640625, 1.8298225402832031, 6.7119598388671875, 17.7418212890625, -6.187255859375, 8.32037353515625, 1.40606689453125, -1.6471786499023438, -0.8197975158691406, 10.272430419921875, 0.996337890625, 8.859466552734375, 5.46875, 16.75933837890625, -2.2889862060546875, -2.9473724365234375, -8.471435546875, 10.128814697265625, 16.59063720703125, -6.90899658203125, -1.2695770263671875, 3.4813709259033203, 3.0502071380615234, -3.8986129760742188, -1.587371826171875, 6.8486175537109375, -2.45135498046875, 14.37567138671875, -0.255767822265625, 20.6651611328125, -17.677886962890625, 20.9251708984375, 8.25640869140625, 16.39605712890625, -6.93304443359375, 5.9144134521484375, 1.6500129699707031, 0.0, -10.166748046875, 2.2831554412841797, 4.545036315917969, 17.27618408203125, 12.578857421875, -3.4898681640625, 2.5742645263671875, 6.197978973388672, -15.33978271484375, 7.2047576904296875, -0.8483676910400391, -0.710601806640625, -6.443792343139648, 9.795623779296875, 7.010009765625, 2.4595947265625, 2.153778076171875, 5.06011962890625, 0.0, 1.7193336486816406, -1.2694244384765625, -4.006492614746094, -0.6745796203613281, -1.6841888427734375, 0.2508411407470703, 6.811439514160156, 2.982391357421875, -11.869049072265625, 22.161773681640625, 21.6463623046875, -2.1488037109375, -0.7767410278320312, -1.2941207885742188, 3.6853179931640625, 3.811279296875, 1.2320556640625, 14.06268310546875, 0.9372978210449219, 3.3160552978515625, 0.4515419006347656, 5.919525146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000093.npy"}
{"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 3.9316585063934326, "std": 9.196497917175293, "min": -21.185760498046875, "p10": -7.011383819580078, "median": 3.118694305419922, "p90": 15.092288208007808, "max": 31.43341064453125, "pos_frac": 0.7109375, "sample": [-1.759918212890625, 10.309974670410156, -6.2903900146484375, 16.19708251953125, 5.5193634033203125, 9.01959228515625, 19.83477783203125, 3.4689903259277344, 4.450439453125, 8.647811889648438, 2.4427490234375, -5.905876159667969, -13.759307861328125, -0.10030746459960938, 12.077606201171875, -0.154937744140625, 13.2398681640625, 0.9685478210449219, 0.108367919921875, 0.1727447509765625, 13.28045654296875, 13.444015502929688, 4.8957061767578125, -1.6778335571289062, 19.27764892578125, 14.181884765625, 2.021221160888672, 4.272918701171875, -12.8431396484375, -2.375091552734375, 11.720306396484375, 20.8587646484375, -12.381500244140625, 23.52423095703125, 6.883647918701172, 6.615478515625, 6.6387939453125, 2.203144073486328, -7.111289978027344, 20.7225341796875, 4.7356719970703125, 31.43341064453125, 14.015380859375, 0.2398681640625, 14.5411376953125, 3.1727752685546875, 13.617416381835938, -6.96856689453125, 7.238311767578125, -2.686368942260742, 1.7343063354492188, 0.6703910827636719, -0.7162723541259766, 7.273162841796875, 3.2461395263671875, -5.6124114990234375, 7.9043426513671875, -4.289703369140625, 8.0589599609375, -8.1497802734375, -0.8778076171875, -6.16656494140625, 3.1648178100585938, -4.441230773925781, 1.40802001953125, 14.618804931640625, 16.37933349609375, 1.0064010620117188, -6.9336090087890625, 13.361587524414062, -0.763671875, 1.147705078125, 7.6515960693359375, 13.500778198242188, 17.57550048828125, -6.933074951171875, 5.07879638671875, -1.7483978271484375, 1.28277587890625, 4.601226806640625, 3.925933837890625, 0.50897216796875, -5.326622009277344, -4.8122100830078125, -11.150558471679688, 20.5352783203125, 12.771774291992188, -9.21783447265625, 17.330291748046875, -13.96063232421875, -21.185760498046875, 4.6694793701171875, 2.615327835083008, 8.657051086425781, 4.617889404296875, 13.347511291503906, 9.53216552734375, 7.526214599609375, 18.009109497070312, -7.5083160400390625, 7.407417297363281, -6.649200439453125, 12.418975830078125, 1.3991622924804688, 2.4649200439453125, 7.6008453369140625, 2.0532379150390625, 10.032798767089844, 1.940155029296875, 1.6470184326171875, 3.240020751953125, 5.0333251953125, 1.430450439453125, -20.10418701171875, 3.07257080078125, 1.23870849609375, 8.858642578125, 1.304962158203125, 1.26190185546875, -7.83856201171875, 9.081741333007812, 0.8339996337890625, 26.4842529296875, 12.710113525390625, -10.162811279296875, 1.38275146484375, -0.05252552032470703, -0.733673095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000094.npy"}
{"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 4.491640090942383, "std": 9.039749145507812, "min": -23.41253662109375, "p10": -6.403250122070312, "median": 3.994354248046875, "p90": 16.592184448242183, "max": 27.169219970703125, "pos_frac": 0.7265625, "sample": [-2.8262691497802734, -1.171722412109375, 3.6183395385742188, 24.615814208984375, 7.787353515625, 2.0008544921875, 1.7951431274414062, 7.722991943359375, 23.99273681640625, 8.708770751953125, 0.243804931640625, -0.4747161865234375, 3.1037445068359375, 2.569774627685547, -8.966384887695312, 6.45916748046875, 2.129791259765625, 6.2395172119140625, 11.29779052734375, 2.7549362182617188, 13.118865966796875, 7.462890625, -11.98626708984375, 6.4227294921875, -0.46722412109375, 1.3742523193359375, 27.169219970703125, -4.213409423828125, 13.79840087890625, -23.41253662109375, 21.03973388671875, -1.5252685546875, 9.365644454956055, -6.9100494384765625, 23.5069580078125, 10.12713623046875, 0.67822265625, 6.8237152099609375, 2.4794769287109375, 10.1893310546875, 2.9197616577148438, 5.33551025390625, -4.13427734375, -0.12248039245605469, 0.327056884765625, 9.718276977539062, 4.44244384765625, 9.613739013671875, 22.13751220703125, 2.24200439453125, -1.30242919921875, 9.639114379882812, 5.3268585205078125, -0.20684242248535156, 6.578369140625, 7.026763916015625, 4.713531494140625, 3.9496917724609375, 5.802215576171875, 10.752410888671875, 8.0189208984375, -2.146484375, 2.3285064697265625, 1.7249298095703125, 3.5452499389648438, 7.46917724609375, -4.6933441162109375, 12.62847900390625, 5.998451232910156, 4.0390167236328125, -5.57305908203125, 20.6422119140625, 9.075164794921875, 18.104080200195312, 0.36627197265625, -1.2917709350585938, 0.09765434265136719, 15.944229125976562, 3.0989990234375, 5.9832763671875, -10.003326416015625, 9.699996948242188, 3.0998687744140625, 1.481790542602539, 4.9602813720703125, 22.9107666015625, 6.376808166503906, 8.738388061523438, -1.781890869140625, -15.379791259765625, 4.9347991943359375, 7.29534912109375, -2.97894287109375, 1.2727508544921875, -13.80694580078125, 2.1162109375, 8.35235595703125, 6.678680419921875, -9.54046630859375, 4.7407989501953125, -2.9579620361328125, -2.9270801544189453, -6.1860504150390625, 2.3486099243164062, 11.532936096191406, -8.150016784667969, 5.681640625, -0.5382766723632812, 25.4200439453125, -9.627288818359375, 3.8808517456054688, 6.48175048828125, -0.6844482421875, 8.96197509765625, 1.7504119873046875, 14.558990478515625, 3.561614990234375, 9.204093933105469, 6.53790283203125, -11.9136962890625, -9.485931396484375, -0.6490592956542969, 5.942535400390625, 20.998565673828125, 4.86578369140625, 25.74737548828125, -10.816162109375, 25.46087646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000095.npy"}
{"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 4.334616661071777, "std": 8.418305397033691, "min": -20.41943359375, "p10": -4.504640960693359, "median": 3.501619338989258, "p90": 16.573052978515623, "max": 28.02740478515625, "pos_frac": 0.7109375, "sample": [-0.3050537109375, -9.90069580078125, -7.75164794921875, 1.9769935607910156, 16.984619140625, 9.062004089355469, -13.931873321533203, -4.66021728515625, 1.99639892578125, 1.530609130859375, 3.6184635162353516, -4.437965393066406, -5.73870849609375, -2.431884765625, 8.436912536621094, 1.4680023193359375, 5.051544189453125, -6.0915069580078125, 1.6569366455078125, 4.462677001953125, 9.678253173828125, 1.634735107421875, 8.33734130859375, 6.8053436279296875, -2.79266357421875, 3.5683746337890625, 18.720458984375, 10.1163330078125, 2.0634918212890625, 4.76318359375, 27.996246337890625, 5.3322296142578125, 11.0130615234375, 10.853828430175781, 6.3524017333984375, 3.7071990966796875, 3.5002479553222656, 3.5378875732421875, 1.13970947265625, 1.7966842651367188, 6.300567626953125, 4.67706298828125, -3.012969970703125, 19.74139404296875, -5.094970703125, 5.60443115234375, -7.738960266113281, -0.5329532623291016, 4.658294677734375, -3.27471923828125, 5.6735076904296875, -0.3687744140625, 8.737228393554688, -4.1715850830078125, 2.27691650390625, -3.9852294921875, 5.516716003417969, 13.69476318359375, 3.50299072265625, 14.169921875, -1.7677001953125, 3.0775909423828125, 12.98516845703125, 9.314453125, 7.3824462890625, 19.43206787109375, 28.02740478515625, 3.183929443359375, -0.8866043090820312, 16.39666748046875, 4.67608642578125, 15.782188415527344, 2.4940719604492188, -10.674560546875, 23.57855224609375, 1.627410888671875, 9.242218017578125, -1.699798583984375, 18.1815185546875, 0.1138458251953125, 9.23944091796875, 0.64508056640625, 5.558197021484375, -6.227447509765625, 0.5552978515625, -1.0368003845214844, 0.393951416015625, 11.935867309570312, 18.256683349609375, -0.948760986328125, -0.8331203460693359, -0.5530166625976562, 4.695793151855469, 18.086212158203125, 21.863525390625, -6.5482330322265625, 3.53936767578125, 7.4917144775390625, 10.859439849853516, 13.679443359375, -14.692596435546875, 20.5499267578125, 2.6347198486328125, -1.9662723541259766, -0.001373291015625, 9.314468383789062, 0.826324462890625, 27.32373046875, 0.9026718139648438, 5.52386474609375, 9.8521728515625, 0.0, 0.8091888427734375, 2.5424976348876953, 0.6792240142822266, 4.541206359863281, -3.348175048828125, 9.30499267578125, 4.27947998046875, 10.397125244140625, -3.7933311462402344, 7.864742279052734, -2.7038650512695312, -2.90216064453125, 2.6428451538085938, 7.9466552734375, 0.109161376953125, -20.41943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000096.npy"}
{"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 4.717471599578857, "std": 9.897416114807129, "min": -17.320556640625, "p10": -7.570498657226563, "median": 4.0987548828125, "p90": 16.955999755859374, "max": 32.40985107421875, "pos_frac": 0.6875, "sample": [5.439579010009766, 7.410247802734375, 6.2746124267578125, -7.7488861083984375, 7.568351745605469, 2.94586181640625, -1.4659156799316406, 5.383392333984375, 8.95074462890625, 8.18377685546875, -1.3364276885986328, 32.40985107421875, 11.616973876953125, 16.428695678710938, 0.00131988525390625, 3.821746826171875, 16.791168212890625, -1.62615966796875, -4.06048583984375, 11.4771728515625, 3.2982025146484375, 2.3728179931640625, -1.832916259765625, -1.488006591796875, 8.266021728515625, -7.913848876953125, 4.16754150390625, -5.6572418212890625, 9.012035369873047, 8.886405944824219, -6.7129974365234375, 1.3233795166015625, -3.2103271484375, -7.5374755859375, 7.13116455078125, 26.912200927734375, 10.55179214477539, -2.982879638671875, -5.9877471923828125, -10.27545166015625, 3.059906005859375, 3.4935302734375, 4.1963653564453125, 24.111328125, -9.482635498046875, 7.4630126953125, 4.02996826171875, 9.347827911376953, 2.9908447265625, 9.337203979492188, 12.268112182617188, -6.603851318359375, 25.91162109375, 0.471221923828125, 7.48974609375, 11.184280395507812, -7.0933837890625, 6.9034423828125, 1.742767333984375, -1.8780517578125, 9.264595031738281, 6.0940093994140625, -1.7498245239257812, 7.941972732543945, 4.6685333251953125, 0.0, 27.571533203125, 8.179153442382812, 9.15875244140625, 30.664459228515625, 1.0656890869140625, 13.080322265625, 7.91986083984375, 7.4223175048828125, -0.6870594024658203, 26.55963134765625, 3.7923583984375, -13.6517333984375, -7.470977783203125, 18.422607421875, 6.7744598388671875, 1.3239250183105469, -5.875946044921875, 4.295875549316406, 13.4771728515625, 6.005184173583984, -0.017887115478515625, 27.342987060546875, 4.7692413330078125, -2.8648529052734375, 7.635913848876953, -8.64007568359375, 5.5830078125, 2.5013351440429688, 7.485260009765625, -17.320556640625, 0.0, 7.695098876953125, 13.39373779296875, 1.5590057373046875, 1.6566696166992188, 29.1927490234375, -9.51666259765625, 2.0517501831054688, 3.2660980224609375, 2.22552490234375, -0.50457763671875, -7.647552490234375, 27.3790283203125, 8.0517578125, -3.994873046875, 24.7783203125, -8.471996307373047, -11.14923095703125, 0.454620361328125, 12.9521484375, 3.236602783203125, 6.27825927734375, -0.7206039428710938, -12.567047119140625, 9.879447937011719, -4.656410217285156, 0.7950439453125, 12.719696044921875, 17.340606689453125, -12.9827880859375, 5.622100830078125, 9.0670166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000097.npy"}
{"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 4.760181427001953, "std": 10.36374568939209, "min": -28.615234375, "p10": -5.289419174194336, "median": 3.6015853881835938, "p90": 17.48275146484375, "max": 34.433746337890625, "pos_frac": 0.671875, "sample": [11.99276351928711, 11.711883544921875, 24.47735595703125, 3.2291336059570312, 31.19189453125, 19.033065795898438, 14.208038330078125, -4.00958251953125, 3.3916015625, -1.2402877807617188, 16.359466552734375, -11.980300903320312, -1.815673828125, 5.268310546875, 7.888397216796875, 7.402801513671875, 1.7882537841796875, -12.725341796875, -6.275245666503906, 1.9832191467285156, 12.376617431640625, -4.198944091796875, -5.9915771484375, 5.1676483154296875, 2.3705215454101562, 4.087913513183594, 0.0906219482421875, -5.5305633544921875, -4.86590576171875, -19.103179931640625, 2.358816146850586, -28.615234375, -4.8682861328125, -7.920654296875, 3.984874725341797, -0.17236328125, -5.348499298095703, 14.369903564453125, 2.5146636962890625, 2.5191497802734375, 2.825897216796875, 9.245330810546875, 0.65484619140625, 7.41058349609375, -1.182281494140625, 16.88580322265625, 1.28363037109375, 18.925140380859375, -5.238250732421875, 13.408496856689453, 33.10845947265625, 14.95599365234375, -1.9970703125, 3.384063720703125, -0.05863189697265625, -0.3031005859375, 28.84796142578125, 13.943695068359375, -2.806396484375, 22.8988037109375, -15.912200927734375, -3.6205368041992188, -1.765594482421875, -5.26409912109375, 22.568450927734375, -1.36395263671875, 10.690460205078125, 5.475982666015625, 6.389802932739258, -2.4025115966796875, 7.3048095703125, 17.520263671875, -15.887664794921875, 3.2508087158203125, 20.8553466796875, 14.308578491210938, 6.572601318359375, 12.98089599609375, -0.2359619140625, 15.41339111328125, 5.7625579833984375, 17.4666748046875, -8.607666015625, -1.2812461853027344, 15.83026123046875, -4.488616943359375, -0.5394821166992188, 1.5135040283203125, 3.5995941162109375, -0.4774169921875, 23.078338623046875, 8.824615478515625, 5.814338684082031, -1.7288970947265625, 7.2352447509765625, 7.815673828125, 0.15822601318359375, 5.317268371582031, 2.7047042846679688, 5.4810791015625, 1.6597442626953125, 16.297927856445312, 34.433746337890625, 2.57257080078125, 6.867095947265625, -4.982673645019531, -4.499725341796875, 5.36376953125, 5.7192535400390625, 3.6653900146484375, 0.082061767578125, 3.60357666015625, 3.9205169677734375, 7.448066711425781, -1.4732666015625, 7.8748779296875, 3.134063720703125, -2.1709442138671875, 7.244449615478516, 13.0777587890625, 12.103096008300781, -4.72607421875, 14.58233642578125, 6.87896728515625, 21.46612548828125, 4.1302642822265625, -19.078338623046875, 12.4527587890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000098.npy"}
{"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 5.092775344848633, "std": 10.464632034301758, "min": -24.829193115234375, "p10": -7.718811035156249, "median": 5.005241394042969, "p90": 18.72445068359375, "max": 26.8302001953125, "pos_frac": 0.7421875, "sample": [-18.96112060546875, 0.8951644897460938, -11.579193115234375, 26.8302001953125, 4.7709503173828125, 11.783447265625, -1.0601272583007812, 17.311309814453125, 1.3131103515625, -5.1268310546875, 5.052490234375, -9.969833374023438, 0.913299560546875, 7.126976013183594, -7.490478515625, 14.89385986328125, 18.78802490234375, 8.724464416503906, 8.091400146484375, -1.5655517578125, 5.141571044921875, 3.2500839233398438, 11.18450927734375, -18.03399658203125, 24.59344482421875, 7.570343017578125, 5.116458892822266, -10.59332275390625, 5.940185546875, 16.22650146484375, 12.9815673828125, -4.391998291015625, 18.69720458984375, 2.4858779907226562, 5.606903076171875, 1.461578369140625, 4.02008056640625, 8.01959228515625, 24.693389892578125, 4.497901916503906, 11.720352172851562, 16.7322998046875, 6.384773254394531, 16.57452392578125, 6.702545166015625, -5.5504913330078125, 12.987396240234375, 1.8315277099609375, -0.266143798828125, 1.040618896484375, -13.999313354492188, -17.8963623046875, -18.8260498046875, 10.487037658691406, 11.155349731445312, -3.3845443725585938, 14.172136306762695, -19.327178955078125, 14.49591064453125, 8.587310791015625, 5.51898193359375, 1.22320556640625, 19.6318359375, 1.3386344909667969, -3.831878662109375, 16.025283813476562, -24.829193115234375, 6.048004150390625, 2.72760009765625, 18.890045166015625, 0.369384765625, 17.654510498046875, -2.62353515625, 0.6400184631347656, -3.7911376953125, 8.740447998046875, 5.069080352783203, 18.888214111328125, 5.586212158203125, 3.3190383911132812, 9.156280517578125, 1.0484695434570312, -2.00537109375, 3.7929916381835938, 16.900711059570312, 4.125587463378906, 12.213470458984375, -4.86474609375, 7.6303558349609375, 4.9579925537109375, -8.2515869140625, 2.625732421875, 2.2960357666015625, 7.656486511230469, 19.952423095703125, 1.3371009826660156, 17.78497314453125, -10.654449462890625, 1.4383773803710938, -3.022918701171875, 20.399627685546875, -0.5401229858398438, 0.0, 5.157283782958984, 0.152679443359375, 16.872894287109375, 2.63201904296875, -4.3914947509765625, 16.573455810546875, 24.3214111328125, 22.09747314453125, 13.208877563476562, -20.3392333984375, 15.6927490234375, 2.5763397216796875, 6.402366638183594, 11.3431396484375, 23.14276123046875, -0.96051025390625, 1.9996490478515625, 2.43133544921875, -1.8056640625, 15.575347900390625, -0.7788009643554688, 13.322052001953125, 7.51129150390625, 22.58966064453125, 3.144866943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000099.npy"}
{"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 4.4393839836120605, "std": 9.596574783325195, "min": -18.30029296875, "p10": -6.715933227539061, "median": 3.3702259063720703, "p90": 18.737898254394526, "max": 27.51153564453125, "pos_frac": 0.6953125, "sample": [6.79400634765625, 3.20849609375, 11.449386596679688, 3.6498870849609375, 17.43463134765625, 4.6092529296875, -10.105560302734375, 18.19598388671875, 0.8017578125, 9.02801513671875, -3.3525390625, 1.450714111328125, 0.9096527099609375, 5.77972412109375, 14.177520751953125, 26.6826171875, 0.7455501556396484, 0.5424346923828125, -9.051528930664062, 7.993133544921875, -0.076934814453125, 5.294586181640625, -9.4808349609375, -1.9255790710449219, 0.9745941162109375, 9.96337890625, 2.02587890625, -9.291793823242188, 0.08402252197265625, 16.269134521484375, -13.4102783203125, 1.99267578125, -0.0193328857421875, 4.948219299316406, 2.25823974609375, 3.8706817626953125, -0.511993408203125, 20.002365112304688, 3.4949264526367188, 14.741943359375, -3.635162353515625, 12.212677001953125, 6.9840087890625, 3.40374755859375, -1.5640411376953125, 5.8658447265625, 7.035675048828125, 20.91015625, 22.1383056640625, 13.919647216796875, -1.1010589599609375, -7.442230224609375, -4.893440246582031, 5.642599105834961, 2.9783973693847656, 11.08648681640625, -14.2396240234375, -1.72589111328125, -18.30029296875, 2.5532760620117188, 0.5548477172851562, 26.3121337890625, -5.835296630859375, 5.101997375488281, -1.060272216796875, 6.2070770263671875, -5.8101654052734375, -13.358200073242188, 4.953887939453125, 5.2984619140625, 27.51153564453125, 8.690673828125, 4.368263244628906, 5.856378555297852, -1.1138076782226562, -6.4046630859375, -4.569366455078125, 6.892059326171875, 0.422088623046875, 11.497846603393555, -10.927383422851562, 4.042304992675781, 4.242767333984375, 2.445098876953125, 6.869659423828125, 3.3367042541503906, -3.215972900390625, -0.789306640625, 16.432037353515625, 25.03179931640625, 0.524139404296875, 3.7901153564453125, -4.5953521728515625, 2.46905517578125, 6.554595947265625, 1.7604751586914062, 4.414543151855469, 2.9964065551757812, -0.15760231018066406, 11.574493408203125, 23.649566650390625, 0.17864990234375, 24.5640869140625, 14.3212890625, 4.584197998046875, -3.52777099609375, 16.139976501464844, 25.6676025390625, 1.2500057220458984, 2.4209747314453125, 24.206756591796875, 5.4390869140625, -0.49596405029296875, -4.3438568115234375, -10.450469970703125, 15.754486083984375, 0.005039215087890625, 3.618743896484375, 20.46978759765625, -10.511871337890625, -0.051952362060546875, 8.707366943359375, 16.665863037109375, -3.402294158935547, 8.861026763916016, -0.315338134765625, 21.5977783203125, -13.051803588867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000100.npy"}
{"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 4.510093688964844, "std": 10.243278503417969, "min": -22.5250244140625, "p10": -6.640007781982422, "median": 4.506874084472656, "p90": 16.201307678222655, "max": 34.546142578125, "pos_frac": 0.71875, "sample": [5.841911315917969, -22.5250244140625, 4.8664093017578125, 14.804931640625, 3.4614944458007812, 4.1019287109375, 3.0741729736328125, 5.33184814453125, -3.29400634765625, 1.927825927734375, 16.616729736328125, 10.84716796875, -2.775421142578125, 21.667205810546875, 31.454345703125, 25.667953491210938, 2.2756710052490234, 5.032684326171875, -1.081624984741211, 10.304412841796875, 7.31866455078125, 4.494712829589844, 10.694183349609375, 9.81341552734375, 1.5864429473876953, 20.257965087890625, 5.3697662353515625, 2.175567626953125, -0.212799072265625, 27.32012939453125, -2.6699600219726562, 6.24957275390625, 3.6574935913085938, 5.830841064453125, 14.59930419921875, 8.692169189453125, 20.612579345703125, 16.023269653320312, 12.089500427246094, 0.72491455078125, -9.213104248046875, 4.723876953125, 11.1005859375, 9.824615478515625, 18.98438262939453, 16.873748779296875, 7.22283935546875, -3.9709014892578125, -0.2843170166015625, -11.35009765625, -1.6800689697265625, 7.1205902099609375, -17.14007568359375, 10.930618286132812, 20.33782958984375, 6.9969482421875, 6.2137603759765625, -18.900360107421875, -1.0732593536376953, 4.00775146484375, -4.185523986816406, 9.5059814453125, 3.688629150390625, 12.116783142089844, -1.06524658203125, -4.534284591674805, 7.117317199707031, 10.702362060546875, -1.684906005859375, 14.967269897460938, -0.721832275390625, 3.7885589599609375, 1.52020263671875, 6.074775695800781, 0.9242954254150391, 2.0053863525390625, 20.18280029296875, 1.6803665161132812, 4.204353332519531, -22.16912841796875, -22.23785400390625, 8.610076904296875, -9.64483642578125, 14.112548828125, 2.9669036865234375, 1.8545761108398438, 12.980224609375, 13.976123809814453, 1.0535087585449219, -2.7981414794921875, 9.84103012084961, 5.204933166503906, 8.237083435058594, 4.531402587890625, 8.052154541015625, 3.1623992919921875, -2.393585205078125, 30.003570556640625, 13.56951904296875, 5.2521209716796875, -16.9356689453125, -21.1287841796875, 9.93052864074707, 1.5186614990234375, 4.5146942138671875, -4.5758056640625, 8.668701171875, 4.499053955078125, 34.546142578125, -6.858741760253906, 8.951171875, -0.16778564453125, 11.70440673828125, -2.667236328125, 0.35996246337890625, 6.576103210449219, -2.84783935546875, -6.5462646484375, -2.8236083984375, 7.176582336425781, -7.8325653076171875, -0.94122314453125, 2.5445556640625, 0.08998870849609375, -13.142501831054688, 0.9914522171020508, 8.10260009765625, 10.179759979248047], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000101.npy"}
{"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 6.361630439758301, "std": 9.435796737670898, "min": -19.025543212890625, "p10": -5.664135742187499, "median": 5.767856597900391, "p90": 17.832826232910158, "max": 32.655517578125, "pos_frac": 0.7265625, "sample": [-5.402313232421875, -0.0757293701171875, 13.032318115234375, 5.9388580322265625, 1.262451171875, 17.94898223876953, 9.396133422851562, -0.45281410217285156, 2.541778564453125, 14.633026123046875, 6.530364990234375, 6.262020111083984, 12.188385009765625, 7.73211669921875, -0.9102630615234375, 15.861801147460938, 10.35211181640625, -9.493194580078125, -2.4716339111328125, -0.7730484008789062, -3.77423095703125, 11.896438598632812, 12.260528564453125, 23.1036376953125, 16.40301513671875, -3.141204833984375, 5.13783073425293, -6.66400146484375, 5.66973876953125, -7.5746917724609375, -6.939994812011719, 5.484161376953125, 3.2333641052246094, 9.192718505859375, -6.436676025390625, 19.707550048828125, 14.826263427734375, -1.0504684448242188, 1.2738571166992188, 0.8017826080322266, -1.1524658203125, 1.6157379150390625, -2.0691776275634766, 16.55291748046875, 8.558349609375, 5.846504211425781, 3.24078369140625, -6.275054931640625, -9.16864013671875, -6.7141876220703125, 17.529083251953125, -4.3896331787109375, 5.807411193847656, 6.739990234375, 4.2283172607421875, 16.249664306640625, 6.9802093505859375, 32.655517578125, 11.491943359375, 17.09625244140625, 1.5245132446289062, -1.0628662109375, 18.973388671875, 10.882415771484375, 8.31591796875, -4.36077880859375, -8.135406494140625, 3.2486419677734375, -3.5301170349121094, 5.728302001953125, 10.253734588623047, 16.750572204589844, 21.569114685058594, 14.7626953125, 5.455831527709961, 19.591323852539062, 19.06817626953125, 5.6322021484375, 7.05694580078125, -12.519927978515625, 0.5954113006591797, 9.510284423828125, 0.0, 5.585174560546875, -1.2580413818359375, 17.84197998046875, 0.0, 5.9588165283203125, 12.5860595703125, -0.35495948791503906, 12.65081787109375, 12.776092529296875, -2.182403564453125, 2.41552734375, 16.021697998046875, 4.9730224609375, 13.3836669921875, 0.11296844482421875, 12.328521728515625, 14.763298034667969, 17.725982666015625, 4.507423400878906, 2.4779930114746094, 25.8448486328125, 6.264862060546875, 9.962371826171875, 2.6811161041259766, 18.725975036621094, 1.017364501953125, 17.828903198242188, -19.025543212890625, 11.338851928710938, -2.5763092041015625, 3.1374588012695312, 12.839874267578125, -15.003021240234375, 16.21550750732422, 2.2490692138671875, 31.0032958984375, 2.8106231689453125, 17.523345947265625, 3.31634521484375, 10.429855346679688, 9.393646240234375, 8.88134765625, 28.161529541015625, -0.89093017578125, -11.800155639648438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000102.npy"}
{"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 4.72640323638916, "std": 10.705635070800781, "min": -25.402435302734375, "p10": -6.126161193847654, "median": 2.3903045654296875, "p90": 18.738290405273435, "max": 46.93379211425781, "pos_frac": 0.71875, "sample": [-20.0889892578125, -7.2713623046875, 6.813751220703125, 9.687576293945312, 18.986328125, 15.602973937988281, -0.37353515625, -2.36676025390625, 23.28704833984375, 1.13946533203125, -0.83306884765625, 15.12493896484375, 18.631988525390625, 3.085845947265625, 0.0, -1.48614501953125, -14.10089111328125, -10.070541381835938, -2.956024169921875, 5.00775146484375, -0.8024501800537109, 0.9799575805664062, 10.277130126953125, -25.402435302734375, 2.015655517578125, 5.04644775390625, 1.114013671875, 2.515289306640625, 6.19195556640625, 1.2109832763671875, 11.359130859375, 9.27587890625, 0.4106903076171875, 6.166717529296875, -4.62933349609375, 7.081027984619141, -2.076629638671875, 16.956283569335938, 8.34161376953125, 7.443912506103516, -1.8152971267700195, 14.035369873046875, 0.530975341796875, 30.378997802734375, -1.8583984375, 12.69171142578125, 3.660247802734375, 9.18707275390625, 13.47412109375, 46.93379211425781, 18.074310302734375, 12.803985595703125, 10.5689697265625, -12.4478759765625, 0.635284423828125, 0.1547088623046875, -10.441543579101562, 8.554718017578125, 9.647705078125, 11.110244750976562, 17.809814453125, 6.677093505859375, 1.3983306884765625, 11.826385498046875, 5.588844299316406, 20.267547607421875, 10.657440185546875, 12.300300598144531, 3.2860183715820312, 6.5247802734375, 0.89398193359375, 0.6208648681640625, 0.5939178466796875, 1.918243408203125, 0.162200927734375, 1.1289958953857422, -8.95004653930664, -19.08319091796875, 2.26531982421875, 2.0653533935546875, 22.393341064453125, 14.062423706054688, 0.6289215087890625, 12.005111694335938, -1.0835533142089844, 10.261322021484375, 5.142738342285156, 1.3401947021484375, 27.6075439453125, -2.24859619140625, 2.8153915405273438, 0.859527587890625, 15.090713500976562, 7.3537139892578125, -2.4457664489746094, 1.331146240234375, 23.65594482421875, -2.724029541015625, -5.6353607177734375, -14.130615234375, 0.0, 3.4864349365234375, 3.5525665283203125, 17.837158203125, -2.0911865234375, 19.4454345703125, -11.34161376953125, 2.091552734375, 1.4489822387695312, 12.33392333984375, 12.220993041992188, -5.023384094238281, -4.476184844970703, 20.385772705078125, 1.322052001953125, 0.6791000366210938, 21.554367065429688, 22.37860107421875, 9.70587158203125, 21.385284423828125, -0.808013916015625, 6.078197479248047, -2.1922607421875, -15.40850830078125, 1.4607315063476562, -18.668304443359375, -4.491081237792969, 0.7055511474609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000103.npy"}
{"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 4.3887553215026855, "std": 11.368512153625488, "min": -26.378631591796875, "p10": -8.282865142822265, "median": 3.59149169921875, "p90": 19.53848876953125, "max": 46.53973388671875, "pos_frac": 0.6953125, "sample": [3.6195068359375, -4.007843017578125, 8.468948364257812, 17.72625732421875, -11.14642333984375, 5.2181854248046875, 4.503974914550781, -0.9199256896972656, 18.98394775390625, -5.905792236328125, 17.17047119140625, -7.685268402099609, 4.323486328125, -15.956985473632812, 8.195159912109375, 8.198665618896484, 26.3890380859375, 19.948440551757812, 0.03564453125, -3.199798583984375, 2.2749481201171875, -2.1440792083740234, 6.741668701171875, -23.052520751953125, 26.022430419921875, -3.8682861328125, 3.7928619384765625, -1.862081527709961, 2.937204360961914, 26.063995361328125, 7.068244934082031, 15.97796630859375, -13.953277587890625, 0.5649795532226562, 3.2373046875, -8.127540588378906, 0.9595966339111328, -8.311698913574219, 21.4056396484375, -2.98211669921875, 12.042266845703125, 0.272064208984375, 0.2668952941894531, -9.759765625, -9.256011962890625, 10.05596923828125, 0.2070465087890625, 9.29229736328125, -1.969818115234375, -3.164398193359375, -3.1895751953125, 7.9321136474609375, 7.6953125, 3.985595703125, 26.10772705078125, 5.388866424560547, 19.70050048828125, -5.01873779296875, -2.6894302368164062, 2.0938282012939453, 4.781885147094727, 9.110069274902344, 4.7274169921875, 6.7014312744140625, 6.13348388671875, 3.0627593994140625, -1.2097244262695312, 16.39898681640625, 4.9611663818359375, -14.5235595703125, 10.702392578125, 19.46905517578125, 0.3072013854980469, -26.079315185546875, 46.53973388671875, 5.412986755371094, 22.7017822265625, 9.006271362304688, 11.079780578613281, 22.5587158203125, 5.871337890625, -26.378631591796875, -10.575531005859375, 11.667404174804688, 2.998931884765625, -0.6714096069335938, 1.68463134765625, 6.72943115234375, -8.2705078125, 18.153106689453125, 23.54962158203125, 0.4452972412109375, 4.603363037109375, 8.362335205078125, 0.6177215576171875, 0.38658905029296875, -6.0872802734375, 0.9652214050292969, 24.53021240234375, 11.492095947265625, 15.413490295410156, 28.158782958984375, -5.331718444824219, 15.933837890625, -18.84674072265625, 1.0611724853515625, -4.85028076171875, 3.5634765625, 5.274192810058594, -14.500732421875, 0.0, 10.003448486328125, -1.2204704284667969, 13.144882202148438, 3.0570220947265625, 6.9811248779296875, 9.944122314453125, -6.648765563964844, 8.635231018066406, 0.7496337890625, -0.08966064453125, 15.40264892578125, 3.0246963500976562, 6.9489593505859375, 16.6064453125, -1.916412353515625, 0.16485595703125, 2.4832916259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000104.npy"}
{"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 7.867586135864258, "std": 11.103824615478516, "min": -22.825790405273438, "p10": -4.91966552734375, "median": 7.008720397949219, "p90": 21.902227783203124, "max": 33.6951904296875, "pos_frac": 0.796875, "sample": [4.529510498046875, 3.454681396484375, 13.99957275390625, 1.3841876983642578, -7.5318603515625, 17.5196533203125, 8.29205322265625, 15.532264709472656, 11.001594543457031, 2.328399658203125, 13.488494873046875, 5.998565673828125, 21.778900146484375, 18.651641845703125, -9.55224609375, 22.04180908203125, 5.741180419921875, -1.77899169921875, 25.49920654296875, 7.3448486328125, -0.8316078186035156, -1.525634765625, 8.261947631835938, 6.474853515625, 21.738525390625, 28.855865478515625, -21.594482421875, 0.32202911376953125, -4.978668212890625, 27.2403564453125, 4.6106719970703125, 13.727310180664062, 0.14238739013671875, 17.606597900390625, -1.787994384765625, 21.8424072265625, 23.482162475585938, 3.95965576171875, 18.06829833984375, -10.894500732421875, 2.3309478759765625, 20.41546630859375, 17.03656005859375, 16.4097900390625, 2.84912109375, 18.288055419921875, 10.596923828125, -2.328277587890625, 21.20379638671875, 20.687255859375, 2.7615432739257812, 12.902679443359375, 0.1593780517578125, 12.851531982421875, -1.397796630859375, 11.680145263671875, 10.546844482421875, 23.170135498046875, 8.476455688476562, -12.5084228515625, 1.850982666015625, -11.683197021484375, 0.39263916015625, 3.613433837890625, 5.891265869140625, 7.6505126953125, 14.832916259765625, -0.1295166015625, 2.328775405883789, 2.817901611328125, 19.658447265625, -21.28631591796875, 17.01971435546875, 22.119110107421875, 7.82879638671875, -4.8514404296875, 11.607177734375, 17.96551513671875, 25.9622802734375, 2.1881103515625, 7.13421630859375, 2.6194000244140625, 7.019989013671875, 0.875, 7.575439453125, 19.234893798828125, -14.207763671875, 23.2139892578125, 6.9974517822265625, -22.825790405273438, 2.9914932250976562, 10.538131713867188, 23.577896118164062, -11.3759765625, 6.670097351074219, 4.239044189453125, 8.961057662963867, -6.965972900390625, 12.4932861328125, 0.0, 4.627494812011719, 7.478363037109375, 0.7108097076416016, -0.87158203125, 15.15185546875, -2.306915283203125, 33.491607666015625, 27.025115966796875, 6.3453216552734375, 4.284736633300781, 6.375640869140625, 21.8043212890625, 0.24273681640625, 3.8554763793945312, 18.665679931640625, 11.749099731445312, -4.894378662109375, 12.13525390625, 3.20709228515625, -8.099761962890625, 8.966705322265625, 20.81568145751953, 3.9231109619140625, 4.487861633300781, -0.44146728515625, 33.6951904296875, 19.87120819091797, 9.635986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000105.npy"}
{"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 5.437828063964844, "std": 11.474313735961914, "min": -22.238937377929688, "p10": -8.656402587890623, "median": 4.705909729003906, "p90": 20.92259521484375, "max": 35.729339599609375, "pos_frac": 0.6875, "sample": [-6.591522216796875, -15.314788818359375, 14.202789306640625, -3.317657470703125, -3.613616943359375, 20.0814208984375, -5.844512939453125, 7.7109527587890625, -12.541229248046875, 15.828750610351562, -12.136871337890625, -0.23043060302734375, 4.19140625, 20.990997314453125, 12.640380859375, 30.029052734375, 25.627777099609375, -8.382293701171875, -2.9585113525390625, -3.5361099243164062, 14.409416198730469, 3.4120025634765625, 4.747772216796875, 5.043212890625, 12.7825927734375, 1.7724609375, 3.5115585327148438, 7.3688507080078125, 11.221878051757812, 23.911239624023438, 1.084625244140625, 9.134857177734375, 16.683685302734375, 1.6005706787109375, 0.4271411895751953, -6.105323791503906, 11.29632568359375, 3.2346649169921875, 5.975311279296875, 8.862548828125, 21.171295166015625, 1.09466552734375, 7.2027740478515625, -15.546661376953125, -3.0848236083984375, 13.62579345703125, -7.497406005859375, -0.0291748046875, 13.066780090332031, 4.6640472412109375, 6.07373046875, 7.7199554443359375, -4.34124755859375, 16.12335205078125, -19.139404296875, -22.238937377929688, 9.899627685546875, 31.79034423828125, 27.680496215820312, 20.66314697265625, 8.060791015625, 7.70123291015625, -9.295989990234375, -6.8475189208984375, 11.002410888671875, 0.2352752685546875, -5.2999267578125, -4.4018096923828125, 22.23883056640625, -11.472412109375, 11.383026123046875, 6.641937255859375, -0.9087982177734375, 6.7639312744140625, -12.378997802734375, 1.3601875305175781, -2.6411590576171875, 2.84014892578125, 16.8018798828125, 2.8224029541015625, -0.982086181640625, 35.729339599609375, 8.55670166015625, 2.2394790649414062, -7.214263916015625, 2.14520263671875, 19.7061767578125, -0.05086517333984375, 14.927276611328125, -9.317901611328125, 10.410606384277344, 4.327117919921875, 16.45123291015625, -0.759002685546875, 1.1426162719726562, 4.420372009277344, 12.071182250976562, 19.740478515625, 5.59539794921875, 8.068267822265625, 29.071441650390625, -16.334075927734375, 1.0488319396972656, -1.45318603515625, 6.01129150390625, 12.612346649169922, 13.139312744140625, -11.10400390625, 2.194244384765625, -6.391265869140625, -11.0782470703125, -4.16046142578125, 32.889892578125, 7.318431854248047, -6.345001220703125, 6.1446685791015625, 12.228775024414062, -5.779151916503906, 19.227203369140625, 3.14593505859375, 20.893280029296875, 6.932769775390625, 10.2335205078125, 24.80645751953125, 21.61151123046875, 0.11560821533203125, 9.999267578125, 1.14007568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000106.npy"}
{"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 7.278087139129639, "std": 12.257946968078613, "min": -56.0831298828125, "p10": -5.073626708984375, "median": 5.73663330078125, "p90": 23.577279663085935, "max": 34.264678955078125, "pos_frac": 0.7421875, "sample": [0.324310302734375, -3.3752288818359375, 3.8751983642578125, 5.0016021728515625, 11.741973876953125, -4.3558349609375, 13.79351806640625, 24.05133056640625, 6.495635986328125, 5.798126220703125, -5.0087890625, 4.45831298828125, 1.2720794677734375, 12.998855590820312, 12.306140899658203, -2.15252685546875, 15.844146728515625, -8.526885986328125, 12.0726318359375, 20.7958984375, -2.1878013610839844, 22.18560791015625, 7.383522033691406, 13.136871337890625, -11.00543212890625, 27.994140625, 8.45751953125, -13.583450317382812, 15.660064697265625, -7.0731201171875, 17.498138427734375, 1.3947601318359375, 0.505706787109375, 18.211387634277344, 20.88275146484375, 23.374114990234375, 9.774871826171875, 5.919548034667969, 32.86444091796875, 0.1175079345703125, 27.100479125976562, 7.380615234375, 7.08184814453125, 12.152626037597656, 3.0956802368164062, 1.1797008514404297, -2.909332275390625, 5.5150299072265625, 14.253488540649414, 6.418792724609375, 8.085437774658203, -1.8669586181640625, 4.646888732910156, 3.61505126953125, 20.59539794921875, -5.621955871582031, 3.177154541015625, 1.025390625, 20.4669189453125, 10.298385620117188, 5.675140380859375, 3.25299072265625, -3.76776123046875, 32.970458984375, 0.4726409912109375, -6.1884002685546875, 4.415824890136719, 31.46868896484375, 22.462539672851562, 17.4906005859375, -11.3060302734375, 25.242996215820312, -4.510650634765625, 15.503936767578125, 21.72900390625, 1.8929328918457031, 3.602630615234375, 20.159881591796875, 13.474365234375, 11.847146987915039, 27.335174560546875, 14.331085205078125, -2.29229736328125, -10.597976684570312, -1.7703857421875, 5.507476806640625, 6.717559814453125, -1.4413681030273438, 34.264678955078125, -7.694267272949219, 27.4814453125, 12.773284912109375, -1.4564132690429688, 5.839508056640625, 4.3011474609375, 1.7565250396728516, 15.555145263671875, 1.9248046875, 0.713836669921875, 0.0032806396484375, -56.0831298828125, -10.23785400390625, 10.1026611328125, 27.4803466796875, 16.630508422851562, -3.31781005859375, 26.752105712890625, -2.560333251953125, 7.2905120849609375, 1.6018123626708984, 19.593841552734375, -0.4278564453125, 9.521163940429688, 2.747222900390625, 7.19061279296875, 16.17901611328125, -3.5948028564453125, -0.6256103515625, -5.411102294921875, -0.9151611328125, 2.44573974609375, 9.36846923828125, -5.22491455078125, 6.86083984375, -0.3202095031738281, 30.586776733398438, 4.686347961425781, 17.5224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000107.npy"}
{"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 8.01645565032959, "std": 10.861185073852539, "min": -20.229217529296875, "p10": -4.820976257324218, "median": 6.847217559814453, "p90": 24.253485107421874, "max": 34.27685546875, "pos_frac": 0.7734375, "sample": [24.032882690429688, 17.287490844726562, -1.58563232421875, 6.6139068603515625, 12.808502197265625, -7.63287353515625, 19.077499389648438, 8.711105346679688, 3.5850257873535156, 13.311447143554688, 2.5564117431640625, -5.20416259765625, -4.4716644287109375, 11.742141723632812, 4.537841796875, -8.038902282714844, 27.775482177734375, 6.3933868408203125, -1.8996963500976562, 8.08135986328125, 13.3095703125, 5.7352447509765625, 10.351882934570312, -7.185699462890625, -0.7080535888671875, 8.368759155273438, 25.0728759765625, -0.965728759765625, 0.0, -18.1737060546875, 13.955070495605469, 5.36865234375, 10.223403930664062, 4.7271575927734375, 16.8240966796875, 2.257049560546875, 27.8941650390625, 6.1322021484375, 17.415191650390625, -0.9201698303222656, 2.8261051177978516, 10.33538818359375, 2.936004638671875, -6.281280517578125, 5.094085693359375, 26.908599853515625, 11.1907958984375, -15.71917724609375, 3.75518798828125, 25.563812255859375, 14.641380310058594, 4.2140655517578125, 8.97222900390625, 34.27685546875, 5.53143310546875, 1.86175537109375, 0.0, -4.6567535400390625, 1.6845455169677734, 0.5082435607910156, 2.020265579223633, 22.65533447265625, 14.770233154296875, 9.116729736328125, -20.229217529296875, 8.1146240234375, 18.919281005859375, 11.848640441894531, -0.8975753784179688, 12.60711669921875, 8.84844970703125, 5.754875183105469, -2.045928955078125, 1.790771484375, -4.5428009033203125, 24.19769287109375, 21.058807373046875, 2.430999755859375, 10.27886962890625, 18.14447021484375, 12.0672607421875, -5.937400817871094, 25.80560302734375, 10.605865478515625, 4.2923431396484375, 0.1024169921875, 22.09307861328125, 10.657600402832031, 7.946014404296875, 27.964080810546875, 7.40069580078125, 20.40045166015625, 3.689361572265625, 11.640396118164062, 8.641304016113281, -3.30316162109375, 29.2354736328125, 15.767745971679688, 26.97076416015625, 20.6395263671875, 15.316459655761719, 4.2023773193359375, 8.954727172851562, 0.0, -2.053680419921875, 9.543022155761719, 7.080528259277344, 32.208526611328125, 5.614723205566406, 24.3836669921875, 6.30218505859375, -9.9404296875, 3.9396209716796875, 0.9525985717773438, 16.588211059570312, 27.386962890625, 22.58929443359375, 3.41357421875, 3.3749618530273438, 10.643913269042969, -11.016098022460938, -11.568389892578125, -11.34930419921875, 6.4493865966796875, -0.496124267578125, 9.269287109375, 4.96075439453125, 16.827713012695312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000108.npy"}
{"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 8.28098201751709, "std": 13.377141952514648, "min": -29.43463134765625, "p10": -6.796925544738769, "median": 6.763092041015625, "p90": 24.34248352050781, "max": 57.98699951171875, "pos_frac": 0.75, "sample": [3.893798828125, 11.128448486328125, -9.719276428222656, 23.934844970703125, 4.078399658203125, 7.556793212890625, 15.002243041992188, 15.61358642578125, -9.47930908203125, 7.1887359619140625, 6.093994140625, 19.574462890625, 0.15445518493652344, -3.3837127685546875, -7.307098388671875, -5.1800537109375, -0.246734619140625, 31.593170166015625, 39.4827880859375, 1.21282958984375, 1.1407623291015625, 1.2141265869140625, -3.1456298828125, -12.94854736328125, -6.667158126831055, 10.1912841796875, 1.109039306640625, -0.5300865173339844, -0.4078330993652344, 10.163299560546875, 6.562339782714844, -1.783355712890625, 14.73626708984375, 33.566162109375, 25.29364013671875, 9.928596496582031, 10.972549438476562, 19.268096923828125, 2.2636260986328125, 19.514892578125, 12.55804443359375, 18.7906494140625, 5.794731140136719, -0.08936500549316406, 14.697723388671875, 33.607177734375, -0.8674163818359375, 28.29852294921875, 7.113304138183594, 5.692447662353516, 12.827316284179688, 6.118255615234375, 8.17083740234375, 5.67059326171875, -20.12249755859375, 6.767578125, 6.326324462890625, 6.917213439941406, 12.825355529785156, 7.63104248046875, 15.849365234375, 3.5988998413085938, 0.031494140625, 22.937103271484375, 7.6849365234375, 11.649711608886719, -4.66485595703125, 10.40423583984375, 10.296333312988281, 11.961067199707031, 16.212799072265625, 4.264190673828125, -7.0997161865234375, 18.186248779296875, 10.239761352539062, 23.344757080078125, 47.61273193359375, -4.248138427734375, 0.903228759765625, 29.21527099609375, 5.27166748046875, 5.34027099609375, 18.847434997558594, 13.689414978027344, 0.91351318359375, 1.0651779174804688, 13.728271484375, 0.6285018920898438, -9.780181884765625, 40.12786865234375, 21.926025390625, 19.041656494140625, 6.75860595703125, 57.98699951171875, 9.50823974609375, -0.09710693359375, 1.85577392578125, 35.9134521484375, -10.535614013671875, 3.5206222534179688, -7.317962646484375, 18.656936645507812, 0.0, 1.1099777221679688, 30.8681640625, 9.699188232421875, 13.08221435546875, 9.215591430664062, -0.7623748779296875, 9.587066650390625, -0.70440673828125, 5.811302185058594, 3.08355712890625, -2.0410728454589844, 20.351333618164062, -2.044189453125, 13.136337280273438, 18.3985595703125, -11.72210693359375, -29.43463134765625, -26.700584411621094, 25.71728515625, 4.9788360595703125, 16.459815979003906, 5.23284912109375, -8.17236328125, -5.2724609375, 14.2965087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000109.npy"}
{"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 7.9183149337768555, "std": 13.422061920166016, "min": -36.966796875, "p10": -6.401950073242188, "median": 5.706504821777344, "p90": 26.188053894042966, "max": 44.54052734375, "pos_frac": 0.71875, "sample": [-0.5287704467773438, -36.966796875, -10.889259338378906, 27.6385498046875, -2.1923904418945312, 2.1986312866210938, -2.042633056640625, 1.611114501953125, -1.890380859375, 28.347076416015625, 4.469879150390625, 3.2489166259765625, 2.894725799560547, 23.40765380859375, -5.7379608154296875, -0.3618316650390625, 11.757781982421875, 5.8414306640625, 17.41229248046875, 4.151885986328125, -9.260551452636719, 18.781646728515625, 1.8051681518554688, 23.63946533203125, 8.264862060546875, 3.4269561767578125, -6.455131530761719, 12.871814727783203, 2.8935546875, -2.86590576171875, 28.3218994140625, 13.19891357421875, 25.977783203125, 2.362112045288086, 35.543212890625, -2.667816162109375, 16.345260620117188, 17.6676025390625, 10.7032470703125, -4.26348876953125, 5.7129058837890625, 6.44122314453125, 11.569992065429688, 27.8109130859375, -10.48699951171875, -0.12459564208984375, 17.089462280273438, -8.942733764648438, 10.335540771484375, 0.5313568115234375, 21.576019287109375, -21.544464111328125, 12.1484375, 1.5332794189453125, 30.822235107421875, 11.541900634765625, 3.985687255859375, -31.117095947265625, 15.406524658203125, 18.285400390625, 3.6298065185546875, 6.58880615234375, 3.01416015625, -0.1658935546875, 0.1392822265625, 27.8353271484375, -6.379158020019531, 25.986724853515625, 18.94696044921875, -14.582244873046875, -0.43800926208496094, 18.65814208984375, 4.691497802734375, 3.6302642822265625, 12.104621887207031, 6.9825439453125, 0.5968704223632812, -3.0109786987304688, -2.1066131591796875, 12.438796997070312, 0.0, 3.693878173828125, 21.7772216796875, 14.119232177734375, -10.555328369140625, 15.419647216796875, 0.4361991882324219, -6.852142333984375, 6.67425537109375, 26.657821655273438, 14.157684326171875, 37.833984375, 13.70770263671875, 25.691253662109375, 5.636016845703125, 17.0391845703125, 25.2650146484375, -2.0995330810546875, -1.6900062561035156, 19.670166015625, 27.0640869140625, 6.767391204833984, 1.5070228576660156, 22.9132080078125, -15.300605773925781, 7.909820556640625, 4.676795959472656, -5.3007659912109375, 23.64605712890625, 20.813873291015625, -2.282562255859375, 31.062362670898438, -3.7693862915039062, 9.155776977539062, 10.405975341796875, 26.89556884765625, 44.54052734375, -3.883636474609375, 18.920166015625, 23.960662841796875, 1.1274871826171875, 18.23431396484375, 5.700103759765625, 7.953151702880859, -13.0389404296875, 3.00885009765625, -1.2062530517578125, 5.684600830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000110.npy"}
{"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 8.618398666381836, "std": 13.457062721252441, "min": -22.144027709960938, "p10": -8.561305236816404, "median": 7.074161529541016, "p90": 26.22658538818359, "max": 46.5718994140625, "pos_frac": 0.7578125, "sample": [3.194366455078125, -11.946258544921875, 12.38568115234375, -22.144027709960938, 44.75341796875, -7.875568389892578, 10.342880249023438, 24.9073486328125, 1.4970779418945312, -0.178253173828125, 27.15081787109375, 5.5740966796875, 6.858428955078125, 6.056121826171875, 6.56658935546875, 14.098129272460938, 7.872734069824219, 35.139678955078125, -17.8082275390625, -13.051040649414062, 11.310806274414062, 6.029212951660156, 14.16473388671875, 2.86639404296875, 14.23846435546875, 1.12554931640625, 4.2393035888671875, 26.019424438476562, 0.6610794067382812, 11.970664978027344, 14.800704956054688, 19.505340576171875, -2.2793121337890625, 33.728240966796875, -4.754608154296875, 6.35662841796875, -5.20953369140625, 15.716278076171875, 0.284027099609375, 11.635482788085938, 1.2599334716796875, 10.910377502441406, -8.058090209960938, 22.3238525390625, 11.677047729492188, 1.53045654296875, 16.45806884765625, -1.6440582275390625, 32.0172119140625, 7.289894104003906, 1.5623779296875, 46.5718994140625, -9.956085205078125, -0.79742431640625, 2.2271728515625, 19.1737060546875, 14.3531494140625, 21.940155029296875, 4.629638671875, 30.316131591796875, 4.239013671875, 11.618194580078125, 5.273475646972656, 11.49176025390625, 2.8447952270507812, -7.91131591796875, 3.1663055419921875, 14.336578369140625, 40.7569580078125, 8.9903564453125, 6.56378173828125, -1.3272628784179688, 7.891456604003906, 0.9219512939453125, 5.164215087890625, -0.24097633361816406, -0.98114013671875, 24.7432861328125, 9.655288696289062, -14.866485595703125, 1.12152099609375, 9.939453125, -9.7354736328125, 4.767822265625, -3.6551513671875, -5.1593170166015625, 0.0718994140625, 2.23687744140625, 17.466629028320312, 28.3851318359375, 16.855987548828125, 31.3359375, 18.386627197265625, -2.9553470611572266, 16.08807373046875, 10.613784790039062, 16.23623275756836, -3.1515769958496094, 43.32806396484375, 3.07440185546875, 20.765037536621094, 8.844146728515625, -13.98577880859375, -10.24151611328125, -1.27154541015625, 25.6287841796875, -10.53973388671875, 20.7613525390625, -2.6381072998046875, 19.880020141601562, 14.830886840820312, 8.01959228515625, 16.934326171875, 36.211151123046875, 20.46337890625, 18.13958740234375, 26.7099609375, -16.7344970703125, 11.155426025390625, 14.053356170654297, 6.798469543457031, -10.186187744140625, 12.2496337890625, 7.6683349609375, 19.24969482421875, 1.2812347412109375, -11.043914794921875, 2.981781005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000111.npy"}
{"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 4.241701602935791, "std": 15.17699909210205, "min": -49.020751953125, "p10": -13.017182922363281, "median": 3.3975181579589844, "p90": 24.148242187499992, "max": 45.514251708984375, "pos_frac": 0.59375, "sample": [15.680229187011719, 3.5441970825195312, -3.7946014404296875, 10.159393310546875, -0.089263916015625, 8.079437255859375, -0.81280517578125, -30.322021484375, -4.122341156005859, 1.1953887939453125, -2.3002967834472656, 6.098396301269531, -11.07952880859375, 19.204803466796875, -1.047943115234375, 3.0812911987304688, 13.050689697265625, 16.79583740234375, 7.353340148925781, 23.396484375, -8.46826171875, 19.946929931640625, 12.375778198242188, 15.0126953125, -49.020751953125, -12.95782470703125, -9.273750305175781, 12.167022705078125, 9.831771850585938, 8.670097351074219, -9.791305541992188, 3.1928367614746094, -24.835662841796875, 5.984893798828125, 4.736045837402344, -17.586563110351562, 5.4054107666015625, 5.675243377685547, -0.35504150390625, 3.7492408752441406, 18.1085205078125, -3.0494384765625, 6.300048828125, 15.058609008789062, 12.748420715332031, -3.0261154174804688, 14.166748046875, -0.01214599609375, 16.908477783203125, -6.802337646484375, 8.710113525390625, 4.6164398193359375, 0.6516571044921875, 15.3631591796875, -17.950164794921875, -3.3260135650634766, -15.026687622070312, -8.468276977539062, -35.361419677734375, -8.838462829589844, -1.8478565216064453, -1.0806541442871094, 16.66363525390625, 36.135406494140625, -4.90692138671875, 3.2508392333984375, 30.984130859375, 15.975296020507812, 0.7498931884765625, 6.945381164550781, 7.624664306640625, -5.534767150878906, 18.381561279296875, 15.11785888671875, 6.9540863037109375, 36.825408935546875, -2.945606231689453, -0.102630615234375, 29.14923095703125, -17.65441131591797, -1.75738525390625, 9.191802978515625, -0.11280059814453125, -0.5793476104736328, 19.702392578125, 23.100830078125, 12.120956420898438, -0.2647705078125, -5.735038757324219, 2.3132553100585938, 25.70068359375, -13.155685424804688, 28.036407470703125, 4.743751525878906, -6.02093505859375, 15.732131958007812, -2.182464599609375, 0.4747505187988281, -38.051605224609375, -6.3481597900390625, -9.07989501953125, 26.40057373046875, -3.9139556884765625, 0.1844482421875, 26.20220947265625, 16.903762817382812, -15.35626220703125, 23.48291015625, 5.73223876953125, 18.16510009765625, 36.29296875, -11.120574951171875, 3.782379150390625, 26.33404541015625, 6.0008392333984375, 1.7039031982421875, -17.3018798828125, -4.148590087890625, 1.6512451171875, 1.314788818359375, 10.6033935546875, -5.4898681640625, 27.0992431640625, 26.644180297851562, 45.514251708984375, -17.447265625, -3.36968994140625, 9.259384155273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000112.npy"}
{"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 9.922706604003906, "std": 14.676850318908691, "min": -28.74029541015625, "p10": -7.534024047851561, "median": 8.256317138671875, "p90": 30.609376525878904, "max": 57.94122314453125, "pos_frac": 0.734375, "sample": [5.337989807128906, 16.280738830566406, 8.2376708984375, 0.235107421875, 15.00347900390625, 1.2554779052734375, -5.88714599609375, 15.008331298828125, 25.90118408203125, 4.341522216796875, 6.232818603515625, 8.131820678710938, 20.17608642578125, -10.821643829345703, -1.642822265625, 2.876689910888672, 3.2239952087402344, 2.715240478515625, 4.55908203125, 12.989532470703125, 57.94122314453125, -4.60107421875, -2.731201171875, 40.07429504394531, 12.72711181640625, -28.74029541015625, 5.0591583251953125, 24.00152587890625, 14.71514892578125, -2.6424942016601562, -1.3459930419921875, -9.8748779296875, 23.703216552734375, 31.513702392578125, 37.4859619140625, -1.40789794921875, 14.385665893554688, -6.919921875, 28.761611938476562, -10.990692138671875, 12.433822631835938, 30.419143676757812, -1.0874786376953125, 13.646697998046875, 8.642059326171875, 4.324977874755859, -9.864471435546875, 0.9279823303222656, 0.3714313507080078, 31.053253173828125, 4.5364227294921875, 18.3515625, -23.801315307617188, 26.543670654296875, 32.95263671875, 24.419708251953125, 10.619819641113281, 33.194915771484375, 7.1085968017578125, -10.782058715820312, 8.460655212402344, -0.317474365234375, -4.913200378417969, 23.876739501953125, -0.5766067504882812, 8.27496337890625, 5.911594390869141, 6.355201721191406, -14.12371826171875, 17.5919189453125, 6.34307861328125, 31.170074462890625, -2.0084972381591797, 11.513916015625, -8.207794189453125, 3.4699859619140625, 12.49017333984375, 29.992202758789062, 14.332489013671875, 14.751815795898438, 14.301651000976562, -11.40966796875, 16.485519409179688, 28.11260986328125, 35.8214111328125, 13.155494689941406, 13.196640014648438, -4.057891845703125, -3.2155914306640625, 5.468902587890625, 35.487884521484375, 25.74627685546875, 19.53863525390625, 12.861358642578125, -7.32818603515625, 23.430419921875, 0.101837158203125, 12.827438354492188, 14.876312255859375, 7.86376953125, 4.905792236328125, -1.5208740234375, 11.685699462890625, 3.3647537231445312, 23.2208251953125, 38.8370361328125, 23.693740844726562, 38.632843017578125, -10.439697265625, 3.618560791015625, 22.66497802734375, 17.629119873046875, 12.528045654296875, 40.76690673828125, -5.4128265380859375, -8.014312744140625, 3.082427978515625, 2.5832672119140625, 19.4306640625, 16.098129272460938, -0.44085693359375, 16.2193603515625, -17.128265380859375, 11.3516845703125, 3.5789260864257812, -3.5619964599609375, 21.184967041015625, -5.3854827880859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000113.npy"}
{"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 5.211361885070801, "std": 13.620423316955566, "min": -26.3292236328125, "p10": -12.779486083984375, "median": 4.481662750244141, "p90": 23.300534057617188, "max": 39.826171875, "pos_frac": 0.65625, "sample": [3.396932601928711, 20.519134521484375, 4.0500335693359375, 26.011852264404297, -0.1285400390625, 8.142059326171875, 1.7904529571533203, 4.502815246582031, -0.4846954345703125, -4.06982421875, -1.9953765869140625, 23.39202880859375, -0.453033447265625, 28.911224365234375, 14.838729858398438, 38.77667236328125, -11.39065170288086, 2.52374267578125, -20.83746337890625, 2.27886962890625, 0.9224853515625, -16.540252685546875, 5.2064208984375, 14.686805725097656, 1.122406005859375, 18.874969482421875, 9.844512939453125, 11.336837768554688, 11.52520751953125, 15.17510986328125, 2.870758056640625, 14.258377075195312, 4.665191650390625, 7.76776123046875, 17.4918212890625, -12.708770751953125, 10.326942443847656, -24.7547607421875, 26.627105712890625, 4.46051025390625, -4.8125152587890625, -6.111358642578125, 2.683124542236328, -26.3292236328125, 6.09063720703125, 1.8511581420898438, -0.3783721923828125, -0.296844482421875, 7.262626647949219, -9.188262939453125, 0.44297027587890625, -12.944488525390625, 10.149261474609375, -7.6846466064453125, -14.270172119140625, -18.653045654296875, 11.93109130859375, 39.826171875, 27.539642333984375, -8.2694091796875, 25.959121704101562, -24.0855712890625, 18.783355712890625, -6.7877197265625, -3.716217041015625, 20.87213134765625, 36.93952941894531, -8.651885986328125, -0.01312255859375, 12.75250244140625, -24.172027587890625, 0.7978248596191406, -14.3896484375, 10.22760009765625, 4.185791015625, 32.80010986328125, 1.3751258850097656, -0.859375, -2.0344085693359375, -16.277862548828125, 2.6142120361328125, 9.811744689941406, 6.921722412109375, -18.355010986328125, 1.7404594421386719, -1.4835891723632812, -16.86029052734375, -4.361930847167969, 0.0, 23.261322021484375, 11.525535583496094, -2.5138778686523438, 13.238922119140625, 5.77630615234375, -2.22882080078125, 22.458709716796875, -3.62774658203125, 16.91119384765625, 1.716339111328125, 6.143440246582031, 10.516716003417969, 31.20269775390625, 12.242992401123047, 12.092681884765625, 37.4024658203125, 15.929878234863281, 7.852996826171875, 5.051490783691406, -7.8252410888671875, 0.0, 33.7591552734375, -1.734130859375, -6.73291015625, 13.344635009765625, 7.57666015625, 20.33154296875, 6.962348937988281, 4.57330322265625, 5.410270690917969, 6.385002136230469, 1.89434814453125, 8.420257568359375, 5.863185882568359, -1.649383544921875, 14.5458984375, 7.5718231201171875, 3.6478958129882812, 14.2510986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000114.npy"}
{"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 5.615972995758057, "std": 14.985543251037598, "min": -29.927947998046875, "p10": -11.766989135742188, "median": 4.745277404785156, "p90": 28.329476928710935, "max": 57.50482177734375, "pos_frac": 0.671875, "sample": [29.391876220703125, -15.353683471679688, 12.716583251953125, -10.351554870605469, -1.39739990234375, -13.052818298339844, -8.7974853515625, 3.77764892578125, 3.1390762329101562, 19.932580947875977, -4.820285797119141, 9.349258422851562, 6.454132080078125, 7.969455718994141, 3.95330810546875, -9.688034057617188, 9.469276428222656, -5.23486328125, 15.686248779296875, 2.55804443359375, -11.915416717529297, 0.89971923828125, 6.0956878662109375, 8.222930908203125, 3.9195175170898438, -4.5921630859375, 1.73602294921875, 12.907760620117188, 12.366851806640625, 15.02069091796875, -3.1196136474609375, 11.933700561523438, 11.64990234375, 3.3767471313476562, 3.7485198974609375, 8.07232666015625, -7.612579345703125, -19.830490112304688, -0.8103790283203125, 32.84698486328125, 5.380363464355469, 28.25897216796875, 35.4871826171875, 6.9347076416015625, -20.305023193359375, -18.34259033203125, 23.799156188964844, -10.464363098144531, 26.343643188476562, 41.27227783203125, 4.7360382080078125, 57.50482177734375, -11.768890380859375, -1.535003662109375, -12.87091064453125, 14.289459228515625, 28.493988037109375, 23.774871826171875, 11.25238037109375, -23.5380859375, -17.215545654296875, -6.57373046875, 28.9437255859375, 3.74237060546875, -3.6734161376953125, 12.993927001953125, 6.1673126220703125, 4.0013427734375, 5.276145935058594, 32.6353759765625, 0.9700756072998047, 8.56884765625, 15.031291961669922, 9.796737670898438, -2.873981475830078, 6.814165115356445, 43.09844970703125, 1.20623779296875, -19.05499267578125, -8.101398468017578, 1.64337158203125, 0.1083984375, -11.76617431640625, 35.332061767578125, 21.64891815185547, 4.7545166015625, -26.7919921875, 9.810012817382812, 2.4304428100585938, 30.118072509765625, -4.610980987548828, 3.4466705322265625, -5.730438232421875, 33.794830322265625, 15.58587646484375, -4.148712158203125, 2.3507537841796875, 4.940155029296875, -8.304443359375, 6.003143310546875, 9.333465576171875, 4.733222961425781, 9.284400939941406, 7.929468154907227, 7.89190673828125, 2.036865234375, 40.575439453125, 5.59136962890625, -9.191009521484375, -10.744354248046875, 11.919036865234375, 4.918723106384277, 16.43859100341797, 7.26025390625, 13.090682983398438, -11.4246826171875, -6.047035217285156, 13.212039947509766, -1.3307113647460938, 21.473236083984375, -29.927947998046875, -3.4879150390625, 10.973419189453125, -0.39715576171875, 13.384124755859375, 1.1290283203125, -1.54840087890625, 10.079986572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000115.npy"}
{"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 8.125114440917969, "std": 14.436942100524902, "min": -39.33709716796875, "p10": -8.119629669189454, "median": 6.736347198486328, "p90": 24.58442993164062, "max": 55.039306640625, "pos_frac": 0.734375, "sample": [2.151458740234375, -3.979412078857422, 5.997810363769531, 2.7288589477539062, 6.0058746337890625, 10.393829345703125, 10.344573974609375, 25.22747802734375, 41.565582275390625, 5.79498291015625, 19.774139404296875, 10.1990966796875, 20.737579345703125, 3.284557342529297, 15.26177978515625, 11.129451751708984, -6.2332916259765625, 12.609573364257812, 4.235374450683594, -9.914749145507812, -19.7763671875, -6.536344528198242, 22.787368774414062, 24.213043212890625, 0.8625564575195312, -8.74383544921875, 10.64532470703125, -0.7269287109375, 5.365135192871094, -0.565765380859375, 2.0817127227783203, 24.308837890625, -9.647247314453125, 17.031829833984375, -5.26409912109375, 0.5431976318359375, 22.358184814453125, 12.18072509765625, -8.074203491210938, -4.1976165771484375, -2.947673797607422, 19.19488525390625, 15.08984375, -9.872932434082031, 7.6958770751953125, 4.265510559082031, -2.52825927734375, 5.3656005859375, 21.1373291015625, 31.748779296875, -39.33709716796875, 3.803558349609375, 14.370773315429688, -16.77618408203125, -0.32170867919921875, 27.6575927734375, 12.703399658203125, 24.2740478515625, -3.463775634765625, 0.6677970886230469, 10.299942016601562, 12.163467407226562, 6.758209228515625, 3.03350830078125, 8.977611541748047, 0.0, 8.917091369628906, 35.55743408203125, -2.5574722290039062, -8.459259033203125, 18.90081787109375, 33.772369384765625, 0.589385986328125, -4.6621551513671875, -8.225624084472656, 1.9151229858398438, 1.2416229248046875, 2.423381805419922, 6.522594451904297, 22.7593994140625, 3.884815216064453, 37.6021728515625, 17.20794677734375, 0.0, 12.25857162475586, 6.0026397705078125, 19.81976318359375, 4.573974609375, 34.66619873046875, -1.141815185546875, -1.7558135986328125, 55.039306640625, 13.01226806640625, -34.1241455078125, 8.1358642578125, 30.4715576171875, 19.338134765625, 4.702911376953125, 9.247919082641602, 9.403900146484375, 10.999557495117188, 0.878021240234375, 8.599014282226562, 28.476242065429688, 20.899169921875, -21.439437866210938, 7.83270263671875, 12.012344360351562, 2.9216690063476562, -13.143775939941406, 11.301513671875, 7.5417022705078125, 45.214263916015625, -0.845703125, 23.722061157226562, 17.042892456054688, 7.15924072265625, 17.2899169921875, 22.624832153320312, 12.681861877441406, -6.874053955078125, 6.714485168457031, 0.501556396484375, 29.699005126953125, -14.349822998046875, 6.267646789550781, -1.3076171875, 20.43438720703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000116.npy"}
{"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 8.214824676513672, "std": 13.6837158203125, "min": -24.611968994140625, "p10": -7.0300148010253904, "median": 7.10627555847168, "p90": 28.245816040039063, "max": 42.143463134765625, "pos_frac": 0.71875, "sample": [2.7788963317871094, 7.107585906982422, 13.446575164794922, 16.434036254882812, 5.779701232910156, 14.10711669921875, 1.67767333984375, -1.8900318145751953, 10.4329833984375, -3.6461029052734375, 14.29901123046875, -6.235595703125, -0.668731689453125, 14.300033569335938, 18.40102767944336, 4.554779052734375, 10.083969116210938, 8.761398315429688, 24.350860595703125, 3.638671875, 18.348846435546875, -9.189613342285156, 7.1049652099609375, 1.7817192077636719, 14.288671493530273, 29.2813720703125, 1.246673583984375, -18.634628295898438, 12.38690185546875, 42.143463134765625, 15.835948944091797, 28.735443115234375, 32.42041015625, 14.406547546386719, 9.267614364624023, -10.142333984375, 23.936538696289062, 0.5708770751953125, 1.1317119598388672, -9.46575927734375, 34.63134765625, 7.2093963623046875, 3.127410888671875, 29.91595458984375, 0.0, 23.644317626953125, -5.700408935546875, 6.21514892578125, 36.398681640625, 23.879486083984375, 9.099639892578125, 33.749847412109375, -22.25518798828125, 18.435943603515625, 24.31011962890625, 22.197662353515625, -6.20123291015625, 2.692270278930664, 4.5052337646484375, 7.454376220703125, 17.81402587890625, 17.33984375, 12.38568115234375, 11.614437103271484, 5.5185546875, 11.56121826171875, 14.12652587890625, -0.9616851806640625, -7.198814392089844, -7.740478515625, 8.324676513671875, -21.1842041015625, -0.11373519897460938, 9.973731994628906, 27.21710205078125, -14.38519287109375, 11.360992431640625, 9.767059326171875, 20.960205078125, 2.0718650817871094, -1.5490875244140625, -3.438873291015625, 6.298065185546875, 33.269683837890625, 4.8466796875, -6.957672119140625, -6.542572021484375, -13.11883544921875, -1.0130615234375, 4.561239242553711, -24.611968994140625, 15.5052490234375, 28.006500244140625, 28.1844482421875, 7.832370758056641, 4.3951416015625, -11.716049194335938, -1.5882415771484375, -1.49603271484375, 29.14617919921875, 3.24957275390625, -4.29022216796875, 1.5091018676757812, -4.9275665283203125, -5.52398681640625, 7.3811798095703125, 28.389007568359375, 6.603675842285156, 24.2059326171875, -5.348415374755859, 16.071250915527344, 15.759414672851562, 21.5780029296875, 0.5485382080078125, 0.3530082702636719, -1.791015625, 4.504697799682617, -0.3712615966796875, 5.3111419677734375, -4.328273773193359, -21.235252380371094, 38.3956298828125, 9.855224609375, 14.231945037841797, 2.5915298461914062, 17.01319122314453, 36.621063232421875, 10.1962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000117.npy"}
{"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 8.885705947875977, "std": 14.826265335083008, "min": -24.96721649169922, "p10": -9.180372619628907, "median": 7.854942321777344, "p90": 26.206878662109375, "max": 50.55926513671875, "pos_frac": 0.734375, "sample": [6.929985046386719, 9.57489013671875, 17.96038818359375, 4.7576904296875, 28.440628051757812, 24.482666015625, -21.94482421875, -13.683391571044922, 19.340957641601562, 26.07073974609375, 14.519378662109375, 6.2800750732421875, 6.5349884033203125, 5.6081085205078125, -14.17132568359375, 33.1422119140625, 15.761260986328125, 19.497299194335938, 22.56389617919922, 5.291908264160156, 10.97100830078125, 18.051513671875, 32.123741149902344, 0.50323486328125, -24.026641845703125, -24.96721649169922, 16.93994140625, 3.5802001953125, -19.681869506835938, 22.470916748046875, 18.05194091796875, 23.79566192626953, 43.4691162109375, 15.18621826171875, 3.63800048828125, 6.0931396484375, 2.92578125, 8.061660766601562, 17.9932861328125, 20.791893005371094, -4.52728271484375, 21.25933837890625, 31.435943603515625, -2.6249465942382812, 2.063140869140625, -4.629669189453125, 19.722503662109375, 17.798095703125, 13.2559814453125, -0.2489471435546875, 0.5706787109375, 3.2674560546875, 26.5245361328125, 15.7144775390625, 27.525344848632812, 39.039306640625, -7.8354644775390625, 11.084442138671875, -0.5796852111816406, 20.396224975585938, -5.668052673339844, 0.0, -8.0784912109375, 9.616546630859375, -10.420379638671875, -7.758689880371094, -19.425628662109375, 5.4346466064453125, 15.552452087402344, -9.86444091796875, -0.3500785827636719, 28.976531982421875, -0.39404296875, 17.284896850585938, 7.025390625, 23.562286376953125, 24.74945068359375, 16.604232788085938, -9.2537841796875, 24.8087158203125, 5.1703338623046875, 22.061904907226562, 10.087615966796875, -4.3303070068359375, 12.763877868652344, 24.92596435546875, -18.5238037109375, 2.2081756591796875, -9.073944091796875, -7.575691223144531, 47.6591796875, -4.7037200927734375, 24.409454345703125, 7.648223876953125, -1.2936477661132812, 10.005104064941406, 1.6881179809570312, -15.532035827636719, 10.802978515625, -9.148910522460938, 5.235477447509766, 34.3720703125, 8.29583740234375, 12.299957275390625, 0.563690185546875, 10.416412353515625, 4.748680114746094, 4.299468994140625, 10.414958953857422, 32.348236083984375, -5.944362640380859, -7.075286865234375, 16.492050170898438, 14.762954711914062, 6.63543701171875, 2.9862518310546875, -21.1846923828125, 10.69927978515625, 6.3046875, 20.4205322265625, 1.6750106811523438, 16.6641845703125, 3.690633773803711, 21.915023803710938, 0.0, 5.3181304931640625, 16.665573120117188, 50.55926513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000118.npy"}
{"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 7.537725448608398, "std": 18.146522521972656, "min": -38.70518493652344, "p10": -13.9692626953125, "median": 5.4830474853515625, "p90": 30.230828857421873, "max": 66.66738891601562, "pos_frac": 0.671875, "sample": [-38.70518493652344, -18.741783142089844, 25.7359619140625, 3.598388671875, 1.3630828857421875, 4.571601867675781, -8.401939392089844, -8.013992309570312, 7.3660888671875, -5.05718994140625, 22.786407470703125, -11.039192199707031, -13.6162109375, 30.39288330078125, -14.32080078125, 9.923690795898438, -15.481201171875, 23.1546630859375, 26.2099609375, -2.779693603515625, 8.022186279296875, 17.369354248046875, -1.2792015075683594, -8.385002136230469, 7.1012420654296875, 14.232742309570312, 21.070266723632812, 6.9718017578125, 32.56062316894531, 4.79156494140625, 3.5497913360595703, 2.74029541015625, -3.015228271484375, 0.01776123046875, 13.070663452148438, 11.588821411132812, 12.2803955078125, 3.9623947143554688, 9.017181396484375, 3.504070281982422, 21.09051513671875, -9.822830200195312, -0.96038818359375, 0.2205352783203125, 31.9044189453125, 0.9885311126708984, 3.814117431640625, -34.52203369140625, -12.4263916015625, -18.455291748046875, 50.092010498046875, 11.287261962890625, 3.952484130859375, 29.556671142578125, -6.087120056152344, -1.5482940673828125, 13.950347900390625, 2.3618850708007812, 15.21795654296875, 22.573822021484375, 6.4691162109375, -0.51953125, 24.39422607421875, 15.381195068359375, 19.778961181640625, -8.33636474609375, 39.434814453125, 11.593963623046875, 32.54571533203125, 18.22174072265625, 0.7261714935302734, 8.785797119140625, 13.898910522460938, 0.0, 37.59269714355469, -1.899505615234375, -2.0801467895507812, -24.084320068359375, 18.3690185546875, -23.92449951171875, 21.639251708984375, 12.310012817382812, 5.637542724609375, -16.534423828125, 11.926116943359375, 0.24481201171875, 30.161376953125, -8.013872146606445, 16.6785888671875, 7.7419891357421875, 24.2320556640625, -17.301971435546875, -2.5234375, 58.4879150390625, -4.457427978515625, 66.66738891601562, 7.90802001953125, -1.7783203125, 8.509033203125, -28.69482421875, 24.932373046875, 31.31597900390625, 1.64337158203125, 19.850494384765625, -8.12847900390625, 22.0963134765625, -7.681489944458008, 5.51531982421875, 4.058708190917969, -13.818603515625, 2.8490753173828125, -22.44329833984375, 50.927642822265625, 14.399314880371094, 57.02587890625, 6.9296722412109375, -1.8023529052734375, 14.695709228515625, -0.15728759765625, 19.837188720703125, 7.343902587890625, 3.357147216796875, 26.5650634765625, -6.43671989440918, 39.246917724609375, 5.450775146484375, 4.429962158203125, -17.6669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000119.npy"}
{"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 8.903035163879395, "std": 16.93987274169922, "min": -32.86474609375, "p10": -12.348928070068355, "median": 9.259346008300781, "p90": 32.317736816406246, "max": 56.874053955078125, "pos_frac": 0.671875, "sample": [-10.37432861328125, 24.224456787109375, -5.776390075683594, 10.797607421875, -3.6392822265625, 11.967391967773438, 23.195159912109375, -11.3572998046875, 7.60205078125, 14.39874267578125, -1.9425621032714844, -4.106376647949219, 20.75921630859375, 6.235809326171875, -17.36883544921875, -0.11053466796875, 32.62176513671875, -3.544353485107422, 10.762908935546875, 19.059106826782227, -6.9285888671875, 32.22772216796875, 30.683746337890625, 26.094207763671875, 27.305908203125, 11.524887084960938, 5.0072479248046875, -1.22979736328125, -4.833869934082031, 32.52777099609375, 27.09710693359375, 0.9869384765625, 6.47198486328125, 3.70526123046875, -25.361968994140625, 8.245513916015625, 0.0292205810546875, 14.745681762695312, 20.76715087890625, 19.515289306640625, 1.1079750061035156, -2.1097259521484375, 10.99273681640625, -23.713226318359375, 0.0, -1.1364288330078125, -22.18450927734375, -2.381195068359375, -6.8798980712890625, 4.8608551025390625, 11.68243408203125, 11.338695526123047, 50.991455078125, 20.542877197265625, 13.99786376953125, 9.892257690429688, -14.662727355957031, 15.00677490234375, 8.379180908203125, 3.4345779418945312, -3.7806243896484375, 0.0, 10.29693603515625, 20.51751708984375, 56.874053955078125, 14.25189208984375, 16.485198974609375, 36.08453369140625, -17.64129638671875, 19.46674346923828, 20.525146484375, 8.231155395507812, 3.7619476318359375, 15.0670166015625, 12.900344848632812, -5.9709930419921875, 7.99822998046875, 33.3828125, 20.7288818359375, 21.2159423828125, 11.26129150390625, -16.0830078125, -18.421478271484375, -8.506027221679688, -10.559898376464844, 36.717926025390625, 8.626434326171875, 34.052337646484375, 43.128265380859375, 15.867500305175781, -0.1055450439453125, 25.948486328125, -28.141815185546875, 33.039306640625, 36.79118347167969, 4.1973876953125, -0.3452301025390625, 17.3092041015625, 10.949249267578125, 19.089065551757812, -22.98401641845703, 2.5402164459228516, 0.31948089599609375, 19.23102569580078, 26.235870361328125, -0.8908424377441406, 11.871063232421875, -2.040374755859375, 40.135498046875, -32.86474609375, 22.12225341796875, 22.3475341796875, 19.424453735351562, -17.846771240234375, -23.70404052734375, 17.111785888671875, 13.913581848144531, 20.69219970703125, 6.776390075683594, -0.9959716796875, -4.0589599609375, 14.712882995605469, -2.8284912109375, 48.39886474609375, 3.0604476928710938, 3.58251953125, -4.8537750244140625, 13.824722290039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000120.npy"}
{"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 5.305604934692383, "std": 15.068872451782227, "min": -32.7904052734375, "p10": -11.558734130859373, "median": 4.185138702392578, "p90": 24.52123718261718, "max": 46.077056884765625, "pos_frac": 0.625, "sample": [-1.5261077880859375, -1.770477294921875, 3.5741939544677734, 7.4070892333984375, -0.3692474365234375, 2.5771026611328125, -1.37158203125, 6.09906005859375, 9.022087097167969, 26.294586181640625, 3.0491600036621094, -1.2886505126953125, 33.398895263671875, 4.957210540771484, 16.68439483642578, 0.3234100341796875, -3.9646759033203125, 3.8917236328125, 0.40712547302246094, 23.76123046875, 18.718109130859375, -17.076065063476562, 19.157028198242188, 15.34869384765625, -25.593902587890625, 0.0, 8.226020812988281, 2.6923980712890625, 1.364959716796875, 18.494049072265625, 10.058969497680664, 8.776039123535156, -32.7904052734375, 27.326202392578125, 5.397708892822266, -32.690460205078125, -3.2318344116210938, 5.487480163574219, 21.0859375, 12.652008056640625, -3.8420257568359375, -9.481048583984375, 6.9097442626953125, 1.9040307998657227, 6.294891357421875, 1.3936920166015625, 16.627426147460938, -10.729248046875, -7.949188232421875, 0.0, 7.4085845947265625, -4.14286994934082, -1.3248367309570312, -2.63214111328125, 14.866592407226562, -6.822914123535156, -0.4205322265625, 5.9617462158203125, 18.189666748046875, -13.49420166015625, -3.59161376953125, -18.0391845703125, -2.5453033447265625, 46.02001953125, 13.558982849121094, 9.117378234863281, 42.97080993652344, -5.944091796875, -5.1341552734375, -2.086517333984375, -5.6571197509765625, 16.6043701171875, -2.69873046875, 13.891029357910156, -9.151908874511719, -0.624755859375, 4.41943359375, 8.019508361816406, -0.0924224853515625, 14.993831634521484, 14.111968994140625, -0.8502674102783203, -6.175323486328125, 28.461280822753906, 46.077056884765625, 6.0704193115234375, -25.327590942382812, -0.5855712890625, -19.637237548828125, 44.255584716796875, 16.421035766601562, 18.9881591796875, 16.5081787109375, 2.29168701171875, 33.004180908203125, 2.8695220947265625, -23.408660888671875, 4.6161346435546875, 13.257057189941406, 1.3076019287109375, 2.327249526977539, 7.027374267578125, 12.97256851196289, 7.83843994140625, 5.4804229736328125, 13.7740478515625, -1.39398193359375, 4.299842834472656, -4.053520202636719, -2.5270042419433594, 6.46868896484375, 34.735870361328125, 10.342803955078125, 4.0704345703125, 8.223388671875, 9.556884765625, 28.777984619140625, -25.517715454101562, 27.18408203125, 13.433685302734375, -0.2457427978515625, -14.072952270507812, 41.242462158203125, 13.583602905273438, 0.8965606689453125, -18.219696044921875, -22.503326416015625, 15.853317260742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000121.npy"}
{"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 7.35720157623291, "std": 18.093772888183594, "min": -38.03387451171875, "p10": -15.980528259277344, "median": 7.0319366455078125, "p90": 30.63475646972656, "max": 59.10382080078125, "pos_frac": 0.6484375, "sample": [21.677001953125, -15.516082763671875, 25.417770385742188, -15.932586669921875, 14.914794921875, 21.834564208984375, 8.856719970703125, 3.2131500244140625, -34.9140625, 30.698715209960938, 25.849090576171875, 0.0, 9.130340576171875, -25.905670166015625, 0.01427459716796875, -38.03387451171875, 10.395614624023438, 16.637725830078125, -12.815437316894531, 14.302291870117188, -4.996551513671875, 23.95306396484375, -22.116973876953125, -31.277587890625, 6.7294921875, 33.5198974609375, -12.065704345703125, 10.079536437988281, 46.207733154296875, 2.3858718872070312, 8.341339111328125, -1.99261474609375, 7.334381103515625, 8.164299011230469, 17.039398193359375, -5.3943328857421875, 4.275627136230469, -21.20635986328125, 0.867462158203125, 10.562507629394531, -27.463043212890625, 12.344970703125, 8.054656982421875, 45.2861328125, -2.791933059692383, -23.23004150390625, -11.388198852539062, 8.718132019042969, 38.792633056640625, -7.067352294921875, -6.8970184326171875, 9.157722473144531, 13.956298828125, -21.363677978515625, 12.1806640625, 14.240486145019531, -0.9613590240478516, 22.968971252441406, -28.642303466796875, 10.0390625, 14.66680908203125, 20.98529052734375, 17.966873168945312, 0.359893798828125, -7.451377868652344, 35.210174560546875, 13.56549072265625, 26.53240966796875, 17.87432861328125, 30.204620361328125, 44.14068603515625, 12.677175521850586, 0.8932037353515625, 35.9478759765625, 59.10382080078125, 8.729095458984375, 4.318878173828125, -2.748077392578125, 30.607345581054688, 21.37744140625, -3.561614990234375, -16.092391967773438, 4.29339599609375, -10.73974609375, -5.0616455078125, -3.095794677734375, -0.4847831726074219, 13.572471618652344, 5.955230712890625, -4.8957366943359375, 16.24688720703125, -0.4803619384765625, 2.873037338256836, -18.129806518554688, -3.4401473999023438, -17.82550048828125, -0.666717529296875, 26.820846557617188, 23.4747314453125, -1.26556396484375, 6.390350341796875, -5.9405975341796875, 22.344879150390625, 19.443572998046875, 2.4682846069335938, 25.682632446289062, 21.929412841796875, 28.510658264160156, 4.452301025390625, 1.9272384643554688, -0.3121490478515625, -1.790740966796875, -6.942840576171875, 15.159019470214844, 49.37957763671875, 1.17132568359375, 32.1815185546875, -3.6405487060546875, 2.8629913330078125, 25.671096801757812, 14.291015625, 20.1024169921875, -4.053318023681641, 35.223876953125, -5.2142333984375, 2.5026092529296875, 31.95526123046875, 7.337799072265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000122.npy"}
{"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 10.464584350585938, "std": 18.44901466369629, "min": -48.146728515625, "p10": -10.919158935546875, "median": 9.480300903320312, "p90": 32.597877502441406, "max": 67.2071533203125, "pos_frac": 0.7265625, "sample": [25.88232421875, -11.031097412109375, 31.025894165039062, -10.5770263671875, 7.227325439453125, 6.2682342529296875, 11.0792236328125, 4.29583740234375, 14.1492919921875, 10.87603759765625, -0.4135475158691406, 4.42486572265625, -3.2723541259765625, 3.356231689453125, 6.88551139831543, -10.581085205078125, -10.871185302734375, 17.22064208984375, 43.47100830078125, 7.940826416015625, -21.65509033203125, -22.489044189453125, 8.707672119140625, 17.26814842224121, 49.7625732421875, 3.607208251953125, 18.4718017578125, 28.44818115234375, 30.593505859375, 1.393096923828125, 27.207061767578125, 22.94140625, 21.549957275390625, 24.523544311523438, 9.120437622070312, 25.75531005859375, 22.829559326171875, -24.11517333984375, 0.355133056640625, 28.7625732421875, 16.90771484375, 4.184661865234375, 2.0778656005859375, 32.3963623046875, 11.056854248046875, 56.715911865234375, 0.8138885498046875, 3.073406219482422, 18.9267578125, 29.8446044921875, 3.0412025451660156, 17.813369750976562, 21.12249755859375, 67.2071533203125, 43.839569091796875, 12.572738647460938, -6.459831237792969, -0.351654052734375, 2.69000244140625, 46.42962646484375, -3.4626922607421875, 5.498430252075195, -48.146728515625, 32.70314025878906, 9.931907653808594, -17.61224365234375, 22.2454833984375, 4.76531982421875, 37.120025634765625, 7.281942367553711, 10.6123046875, 0.2666473388671875, 8.621444702148438, -18.610198974609375, 11.595428466796875, 26.839111328125, 22.690155029296875, -4.39959716796875, -20.328277587890625, 42.369110107421875, -3.5361785888671875, 11.422576904296875, 20.283519744873047, -5.12310791015625, 7.98846435546875, 18.7720947265625, -19.20452880859375, 17.083343505859375, 17.3721923828125, -7.1700439453125, 0.0, -4.226043701171875, 37.5426025390625, 11.350082397460938, 9.668792724609375, 0.0, 12.970443725585938, 10.105514526367188, 0.5730495452880859, 19.5477294921875, -6.356964111328125, -15.563232421875, 5.5242919921875, 27.349151611328125, -0.384002685546875, 11.932907104492188, 9.29180908203125, 34.028350830078125, 16.3856201171875, -16.397598266601562, -2.148712158203125, -3.3709716796875, 30.44915771484375, 8.7607421875, -29.956893920898438, 47.81512451171875, 15.294647216796875, -7.29754638671875, 12.91668701171875, -2.2867050170898438, -0.15602493286132812, 12.21795654296875, 5.612953186035156, 32.552764892578125, 44.292877197265625, -18.2998046875, 16.416351318359375, 21.145050048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000123.npy"}
{"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 9.492400169372559, "std": 17.256988525390625, "min": -52.613616943359375, "p10": -7.769679641723632, "median": 6.919187545776367, "p90": 33.240301513671874, "max": 57.51324462890625, "pos_frac": 0.7734375, "sample": [1.10906982421875, 28.47173309326172, -52.613616943359375, 16.860992431640625, -4.793853759765625, 1.9712448120117188, 25.753158569335938, 7.602691650390625, 35.733001708984375, 45.2674560546875, 3.7986907958984375, -28.03533935546875, 5.746490478515625, -7.5403594970703125, 21.921478271484375, 4.8276214599609375, 5.74981689453125, 20.942718505859375, -6.532461166381836, 14.17034912109375, 5.85687255859375, 5.3580780029296875, 0.3194732666015625, 10.971343994140625, 19.9046630859375, 4.158355712890625, -0.78741455078125, -4.425991058349609, 2.5213165283203125, 8.133354187011719, 27.700302124023438, -13.822998046875, 3.9193954467773438, 24.434112548828125, 43.510711669921875, 38.17225646972656, 19.61956024169922, 11.509231567382812, 4.695068359375, 22.7823486328125, 21.683792114257812, 17.90180206298828, 39.716705322265625, 10.670684814453125, 3.463726043701172, -1.8035659790039062, 0.7942657470703125, 13.281326293945312, 46.088531494140625, -8.304759979248047, 7.29522705078125, -18.302410125732422, 6.761451721191406, 0.0, 2.646045684814453, -1.85211181640625, 25.380859375, 5.88665771484375, 3.876800537109375, -10.641410827636719, 33.09918212890625, -38.660797119140625, -25.840240478515625, 41.13714599609375, 13.0772705078125, 22.889190673828125, 8.509521484375, 29.8150634765625, 24.64031982421875, -1.355255126953125, 12.20098876953125, 4.549285888671875, -19.871612548828125, -24.5889892578125, 21.584304809570312, 1.3694267272949219, 4.335563659667969, 12.666046142578125, 33.569580078125, 21.693756103515625, 11.105438232421875, 30.63873291015625, 7.183929443359375, -15.654296875, -4.6015625, 6.385047912597656, 2.8765716552734375, 14.781295776367188, 35.234222412109375, 15.874828338623047, 3.171844482421875, 13.157333374023438, -3.4164581298828125, 11.6033935546875, 10.295761108398438, 7.850122451782227, -2.66400146484375, 57.51324462890625, 11.368936538696289, 20.436798095703125, -3.8388519287109375, 2.52398681640625, 15.74090576171875, 2.1735763549804688, 2.961761474609375, 54.05853271484375, -10.369308471679688, -0.2415771484375, 7.076923370361328, 16.5230712890625, -1.62109375, 0.3424072265625, 6.617889404296875, 10.904266357421875, 6.1467742919921875, 20.723176956176758, 10.359451293945312, 34.889984130859375, 0.34281349182128906, -1.9543228149414062, 5.217140197753906, 15.0338134765625, 22.043304443359375, 44.51513671875, 4.5140380859375, 12.567718505859375, 3.6969757080078125, -13.366706848144531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000124.npy"}
{"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 9.40181827545166, "std": 18.968177795410156, "min": -33.334228515625, "p10": -10.983584976196287, "median": 6.944709777832031, "p90": 35.69768371582031, "max": 63.931396484375, "pos_frac": 0.65625, "sample": [4.7589263916015625, -9.512847900390625, -33.334228515625, 37.7601318359375, 1.6167335510253906, 14.401168823242188, 16.134689331054688, -10.5933837890625, 3.371821403503418, -16.163848876953125, 4.0118408203125, -2.1613388061523438, 33.150604248046875, -0.2410888671875, 13.364532470703125, 24.3544921875, -29.915557861328125, 14.840362548828125, 13.773666381835938, 14.04119873046875, -8.87957763671875, -3.3146705627441406, 19.06121826171875, -4.353324890136719, 15.15032958984375, 13.088851928710938, 11.57135009765625, 14.267181396484375, -4.287017822265625, 47.16162109375, 4.638282775878906, -6.1223602294921875, 12.581146240234375, -1.5004730224609375, 7.388404846191406, 32.013824462890625, 20.798095703125, 35.964263916015625, 14.647018432617188, -4.129753112792969, 0.060821533203125, -22.50213623046875, -10.475784301757812, -6.7378997802734375, -5.132232666015625, 44.0531005859375, 2.6531982421875, -8.34173583984375, -32.240325927734375, 14.989913940429688, 16.50830078125, -16.129638671875, 25.4271240234375, 63.931396484375, 6.6475982666015625, 12.50210952758789, 9.145393371582031, 20.68170166015625, 18.3590087890625, 26.875732421875, -1.746124267578125, 36.896087646484375, 13.9525146484375, -0.9568405151367188, 0.45404052734375, 15.12799072265625, 2.909912109375, 1.3097991943359375, 6.557464599609375, 5.558929443359375, 55.478759765625, 10.718109130859375, 28.755516052246094, 43.8863525390625, -11.894054412841797, -20.251220703125, 8.56341552734375, 23.1632080078125, 35.58343505859375, 24.2977294921875, 31.38458251953125, 4.6577301025390625, 1.1844482421875, 43.62823486328125, -2.048553466796875, 13.834793090820312, -12.510238647460938, -6.798736572265625, 20.92827606201172, 30.370025634765625, 27.08226776123047, 5.8551025390625, -6.4995269775390625, 14.795654296875, -7.777557373046875, -1.6446685791015625, -12.52459716796875, 28.269866943359375, 27.377960205078125, -1.101654052734375, -5.9342498779296875, 0.0, 2.9430179595947266, -4.3139495849609375, 41.564666748046875, -8.703201293945312, 0.3637847900390625, 5.420928955078125, 7.2418212890625, 44.35784912109375, -6.8260650634765625, -10.371795654296875, 20.75799560546875, 30.48291015625, 0.6182823181152344, 26.639312744140625, 51.427520751953125, 11.267608642578125, -15.2369384765625, 16.70703125, 27.54486083984375, -32.99003601074219, -6.608642578125, 47.940155029296875, 13.060867309570312, 19.0399169921875, -2.246185302734375, -17.221160888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000125.npy"}
{"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 10.519754409790039, "std": 18.86772346496582, "min": -44.563262939453125, "p10": -10.520159912109374, "median": 8.6103515625, "p90": 34.81512756347656, "max": 50.45220947265625, "pos_frac": 0.7109375, "sample": [21.910308837890625, 50.45220947265625, 14.087158203125, 17.31060791015625, 27.8843994140625, 39.839599609375, 11.855705261230469, 39.41163635253906, 4.160951614379883, -1.3831024169921875, -42.380462646484375, 24.677574157714844, 2.5522918701171875, 19.247833251953125, 27.496299743652344, 5.125, 25.047332763671875, -5.5998382568359375, -23.04815673828125, -5.318634033203125, -14.462150573730469, 29.348068237304688, -10.443450927734375, -7.166118621826172, 4.97833251953125, -9.465347290039062, -4.195892333984375, -3.3090896606445312, 5.7373046875, -6.675323486328125, 9.05059814453125, -3.236480712890625, 16.44988250732422, 12.316177368164062, 32.364837646484375, -10.608428955078125, -7.940032958984375, 1.1246337890625, -1.7005615234375, 4.760503768920898, 7.709442138671875, 22.848880767822266, 35.085296630859375, -0.6029853820800781, 15.463348388671875, 22.849853515625, 8.040023803710938, 16.21514892578125, -14.539321899414062, -2.734130859375, 33.553070068359375, 0.0435028076171875, 12.330825805664062, 48.73590087890625, 3.699512481689453, 36.89622497558594, 16.020530700683594, 6.966499328613281, 4.146759033203125, 20.952611923217773, 10.801399230957031, 29.011474609375, 17.88720703125, 25.0438232421875, -13.569427490234375, 16.793670654296875, 4.6607208251953125, 24.437820434570312, 21.659454345703125, 46.0537109375, 41.224090576171875, -4.45880126953125, 26.540939331054688, 28.87047576904297, -3.0284957885742188, 7.2446136474609375, 8.17010498046875, -8.078506469726562, 0.10433387756347656, -23.997283935546875, 27.803932189941406, 10.50958251953125, -2.7982940673828125, 45.15631103515625, 20.9486083984375, 2.780231475830078, -28.79193115234375, 6.798919677734375, 21.272705078125, -34.20965576171875, 14.675750732421875, -10.001739501953125, 24.580810546875, 31.748992919921875, -44.563262939453125, -8.930694580078125, 4.354459762573242, 9.201705932617188, 5.094451904296875, -15.11724853515625, 23.658340454101562, 26.30963134765625, 33.82820129394531, 41.53033447265625, 7.012176513671875, 19.31744384765625, -6.25445556640625, 45.89349365234375, 21.978118896484375, 38.102447509765625, -12.18707275390625, 20.5621337890625, 17.291213989257812, 34.6993408203125, 20.27484893798828, 5.878173828125, 4.1491241455078125, -2.5055999755859375, 11.06805419921875, -0.9982757568359375, 14.0438232421875, -10.482330322265625, 2.274169921875, 4.9346923828125, 29.649566650390625, 50.189239501953125, 4.903434753417969, -18.413909912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000126.npy"}
{"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 12.216462135314941, "std": 19.16884422302246, "min": -45.045440673828125, "p10": -10.065970611572265, "median": 8.967277526855469, "p90": 38.80675354003906, "max": 54.8297119140625, "pos_frac": 0.7421875, "sample": [-0.22883224487304688, -1.502899169921875, -8.86956787109375, 25.083892822265625, 9.84149169921875, 2.8471527099609375, 18.197174072265625, 11.669788360595703, -10.7103271484375, -45.045440673828125, 29.8555908203125, 3.017650604248047, 19.735015869140625, 16.42292594909668, 10.199356079101562, -32.81939697265625, -8.697235107421875, 13.567459106445312, -16.845077514648438, 6.48876953125, 6.6717529296875, 6.872711181640625, 20.50885009765625, 3.000885009765625, 18.7801513671875, 47.925048828125, 22.93927001953125, 35.5921630859375, 12.627151489257812, 24.52001190185547, 39.31976318359375, 14.84625244140625, 28.21142578125, 3.655813217163086, 9.60687255859375, 45.57965087890625, -9.789817810058594, -19.36688232421875, -1.2664413452148438, 9.434127807617188, 47.591644287109375, 7.305873870849609, 8.172027587890625, 4.938079833984375, -14.793624877929688, -24.662559509277344, 34.811065673828125, 35.14698791503906, 10.775123596191406, -20.601806640625, 18.707778930664062, 35.773681640625, 19.86553955078125, -6.3933868408203125, 30.831634521484375, 40.066619873046875, 24.69464111328125, 15.0985107421875, -3.34332275390625, 2.605968475341797, 1.624664306640625, 3.48577880859375, 36.966156005859375, 22.978729248046875, 8.311813354492188, 32.8792724609375, -7.4481201171875, 5.3336029052734375, 30.415756225585938, 38.77618408203125, 25.4151611328125, -2.0531539916992188, -11.604927062988281, -2.410615921020508, 10.258636474609375, -11.550033569335938, 0.0601806640625, -14.483745574951172, 7.1833953857421875, -2.813640594482422, 4.042388916015625, 22.59783935546875, 54.8297119140625, -1.6652946472167969, 40.5909423828125, 19.500503540039062, 4.800018310546875, 54.8050537109375, 30.513580322265625, 20.8114013671875, 45.269317626953125, -8.625335693359375, 7.06396484375, 7.755126953125, 4.119136810302734, -1.1283092498779297, -20.6098575592041, 7.69873046875, 9.763252258300781, 31.18115234375, -5.2446746826171875, 8.50042724609375, -8.623855590820312, 30.38147735595703, 22.22911834716797, 5.605720520019531, 39.534271240234375, 48.273834228515625, 1.11004638671875, -3.125701904296875, 6.3748779296875, 3.533843994140625, 23.69293212890625, 34.90948486328125, -17.152931213378906, 4.258140563964844, 38.878082275390625, 38.36669921875, -1.5791549682617188, 10.9266357421875, 28.400665283203125, 22.2664794921875, 3.27789306640625, 31.15887451171875, 1.90875244140625, -8.61517333984375, 22.720138549804688, 48.6351318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000127.npy"}
{"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 13.526945114135742, "std": 20.690589904785156, "min": -42.794921875, "p10": -10.995611572265625, "median": 12.721439361572266, "p90": 37.46295318603515, "max": 72.891357421875, "pos_frac": 0.7265625, "sample": [27.712051391601562, 4.8001708984375, 19.16844940185547, 24.45806884765625, -33.063629150390625, -34.65020751953125, -19.41937255859375, 7.9972381591796875, 25.490020751953125, 27.348052978515625, 23.50243377685547, 3.1787109375, 35.13128662109375, 0.22676849365234375, 38.29316711425781, 0.4048004150390625, 3.4035263061523438, 23.610824584960938, 27.976898193359375, 23.277618408203125, -2.2555694580078125, -1.0862045288085938, 35.525535583496094, 66.8092041015625, 2.689605712890625, 8.034950256347656, -2.640960693359375, 6.305328369140625, 28.933502197265625, 21.104278564453125, 8.720260620117188, 11.744441986083984, -18.1529541015625, 34.20765686035156, -8.710121154785156, 30.484359741210938, -15.175216674804688, 7.0741119384765625, -1.9095611572265625, -21.18096923828125, 24.96324920654297, 5.016632080078125, -3.2627716064453125, 17.98284912109375, 33.479705810546875, -3.989990234375, 0.8171272277832031, 24.731929779052734, 18.649139404296875, 37.107147216796875, 34.4873046875, 27.971099853515625, -3.7218017578125, 34.710693359375, -1.895233154296875, -5.756378173828125, 3.2889404296875, 24.64044189453125, 19.722381591796875, 41.305633544921875, 27.226329803466797, 4.6436920166015625, 13.698436737060547, 0.6695632934570312, 0.5574874877929688, 7.7669525146484375, 39.468170166015625, 28.605316162109375, 26.191009521484375, 13.949443817138672, 15.041748046875, 29.129959106445312, 50.28306579589844, 17.9281005859375, 48.042449951171875, 34.454071044921875, 10.821937561035156, -5.40277099609375, -1.76922607421875, -1.1202163696289062, 40.6268310546875, 32.3172607421875, -15.12713623046875, 18.861907958984375, 29.699066162109375, -42.794921875, 20.005218505859375, -19.574005126953125, 18.47325897216797, 72.891357421875, 8.906341552734375, -10.96435546875, 35.80354309082031, 16.084930419921875, -6.08575439453125, 1.0672740936279297, -17.46697998046875, -7.014617919921875, 6.017578125, 11.127830505371094, 35.682891845703125, -6.664794921875, 5.836669921875, 46.04785919189453, 22.544395446777344, 34.965118408203125, -8.49056625366211, -4.953887939453125, 55.956024169921875, 0.6417617797851562, 9.36566162109375, 0.0, -21.430679321289062, -6.0327606201171875, 26.8018798828125, -12.045913696289062, -11.06854248046875, 14.48468017578125, 23.50714111328125, 42.498199462890625, 3.507049560546875, 19.08624267578125, 11.348800659179688, 40.143585205078125, 62.143280029296875, 26.204147338867188, 19.84149169921875, -1.1275482177734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000128.npy"}
{"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 12.351709365844727, "std": 21.43317413330078, "min": -47.18731689453125, "p10": -12.234698486328123, "median": 11.280723571777344, "p90": 41.88966827392578, "max": 78.561767578125, "pos_frac": 0.765625, "sample": [-47.18731689453125, 27.4609375, -23.710342407226562, -13.626708984375, 11.002578735351562, 44.52860641479492, 27.194244384765625, 37.06517791748047, 30.369338989257812, 2.851165771484375, 30.50006103515625, 2.5545215606689453, 13.52349853515625, -0.762420654296875, -7.0872650146484375, -11.5601806640625, 27.736846923828125, 35.097412109375, 13.183197021484375, -11.533927917480469, 13.016258239746094, 12.155105590820312, 1.33306884765625, 17.8048095703125, 42.056304931640625, -5.883758544921875, 16.435333251953125, -2.5305557250976562, 78.561767578125, 60.6263427734375, 6.948638916015625, 21.980224609375, 51.05241394042969, 15.16937255859375, 1.2963714599609375, -43.426177978515625, 34.1517333984375, 5.444122314453125, 62.67559814453125, 10.804737091064453, -4.511814117431641, 13.920928955078125, 41.81825256347656, 42.952880859375, 11.077346801757812, 42.801605224609375, -25.2100830078125, 11.482292175292969, 0.6480941772460938, 4.035308837890625, 27.348342895507812, 18.064712524414062, 8.3978271484375, 25.37237548828125, -11.21051025390625, 9.01852035522461, 5.4359893798828125, 42.23199462890625, 2.6120452880859375, 27.30664825439453, 1.5206069946289062, 2.8883895874023438, 41.0755615234375, 21.89495849609375, 61.55328369140625, -7.9517822265625, 15.0042724609375, 16.727859497070312, 26.731246948242188, 4.56341552734375, 4.5472412109375, 6.806205749511719, 2.6974029541015625, -4.501262664794922, 1.8516845703125, -14.741668701171875, 25.79485321044922, 20.772483825683594, -2.3846359252929688, 58.598876953125, 26.892311096191406, 8.6346435546875, 18.80645751953125, 3.4346084594726562, 2.5771141052246094, 1.611785888671875, 29.9417724609375, 3.975921630859375, 19.775146484375, 39.53791809082031, -9.881622314453125, -27.8416748046875, -2.8130874633789062, 1.4940185546875, -7.279640197753906, 15.695068359375, -11.63812255859375, -32.951690673828125, 15.886962890625, 13.00439453125, 46.858795166015625, -15.905426025390625, 12.4608154296875, -3.5443477630615234, -19.196853637695312, -21.499755859375, 11.351852416992188, 1.97198486328125, 16.033363342285156, 24.2747802734375, 28.007492065429688, 6.24232292175293, 18.0965576171875, 10.128875732421875, 10.20456314086914, 44.97357177734375, 11.2095947265625, 30.448928833007812, -20.38250732421875, 18.984146118164062, 11.777839660644531, -0.43189430236816406, 2.389678955078125, 29.471710205078125, 12.516883850097656, -15.465461730957031, 24.7845458984375, 32.08160400390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000129.npy"}
{"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 11.562756538391113, "std": 19.5655517578125, "min": -38.590667724609375, "p10": -11.609603881835934, "median": 9.112709045410156, "p90": 36.45964965820312, "max": 55.918060302734375, "pos_frac": 0.734375, "sample": [8.718994140625, -2.7161407470703125, 29.5765380859375, 4.1082305908203125, 5.527427673339844, 41.087890625, 9.24066162109375, 29.360214233398438, 39.38549041748047, -7.46661376953125, -3.5097618103027344, 8.66864013671875, 22.197731018066406, 41.443206787109375, 0.569000244140625, 6.472991943359375, -1.355987548828125, -16.240264892578125, 7.594451904296875, 19.48687744140625, 17.057373046875, 37.72528076171875, -1.0673828125, 10.125076293945312, -2.377490997314453, 55.918060302734375, 1.411651611328125, 20.5673828125, 46.986419677734375, 49.004913330078125, -32.22760009765625, 0.484100341796875, 0.0, -9.124099731445312, 25.456634521484375, 13.752639770507812, 24.52703857421875, -38.590667724609375, 9.345012664794922, 21.9674072265625, 25.80207061767578, 14.070556640625, 48.650665283203125, -10.844589233398438, -37.488037109375, -36.989837646484375, -4.0399322509765625, 33.39753723144531, 2.980224609375, -1.4543609619140625, 35.017669677734375, 0.7504196166992188, 11.039718627929688, 33.41960144042969, -16.60498046875, -1.948333740234375, 13.047046661376953, 19.88861083984375, 35.917236328125, -4.0743408203125, 8.356559753417969, 1.2651863098144531, 23.940673828125, 16.550506591796875, 5.151947021484375, 44.07591247558594, 26.6796875, -16.53363037109375, -13.394638061523438, -1.0337371826171875, 39.64190673828125, 8.0498046875, 29.82684326171875, -23.294952392578125, 5.653022766113281, 15.63348388671875, -15.427177429199219, 30.43865966796875, 8.570602416992188, 15.111457824707031, 29.125289916992188, 18.641357421875, 30.534271240234375, 14.6705322265625, 32.986602783203125, 0.8884677886962891, 31.581268310546875, 8.984756469726562, 28.796417236328125, -3.7076339721679688, 9.93890380859375, 5.81396484375, 5.12408447265625, 2.5860252380371094, 22.845413208007812, -28.4248046875, -4.5436248779296875, 4.867401123046875, 2.046112060546875, 50.5579833984375, 43.943603515625, 34.53851318359375, 19.922454833984375, 0.38832855224609375, -25.96685791015625, 27.9993896484375, 5.6872711181640625, 4.923309326171875, 20.344512939453125, -5.49090576171875, 33.548362731933594, 8.552692413330078, 26.793441772460938, -4.8295135498046875, 20.010162353515625, 19.881305694580078, 39.3349609375, 22.431884765625, -0.8717727661132812, 17.467758178710938, 15.141738891601562, -5.235443115234375, -1.88818359375, 1.7708587646484375, 28.4005126953125, -24.875152587890625, 4.820713043212891, 27.08563232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000130.npy"}
{"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 9.554973602294922, "std": 21.83217430114746, "min": -45.58601379394531, "p10": -19.091271972656248, "median": 8.670059204101562, "p90": 39.06194152832031, "max": 70.68505859375, "pos_frac": 0.640625, "sample": [30.874725341796875, 9.701011657714844, 37.374786376953125, -22.724365234375, 13.401840209960938, 2.3472442626953125, 2.742481231689453, 23.694839477539062, -15.652053833007812, 39.409912109375, 30.847503662109375, 29.00390625, 42.197021484375, 17.34161376953125, -16.029144287109375, 2.1321792602539062, 3.7919921875, -16.8740234375, 10.124404907226562, 46.498565673828125, 33.4898681640625, 9.506561279296875, -5.63568115234375, -6.059898376464844, 29.632949829101562, -6.725982666015625, -11.22113037109375, 42.163360595703125, 10.924224853515625, 14.133682250976562, -1.5192413330078125, 58.90855407714844, -5.7421112060546875, 0.237091064453125, 19.824859619140625, -3.207712173461914, -22.14122772216797, 42.93206787109375, -4.484527587890625, 28.131103515625, 7.494476318359375, 3.9344024658203125, -2.4117088317871094, -18.010589599609375, 26.080322265625, 7.0826873779296875, 57.223785400390625, -18.98419189453125, 21.182113647460938, 25.628265380859375, -1.5957794189453125, -2.11260986328125, -15.788421630859375, 17.27740478515625, 11.61065673828125, 38.912811279296875, -23.599533081054688, -5.104095458984375, 25.394596099853516, 35.21559143066406, -8.357711791992188, -19.34112548828125, -20.01361083984375, 30.586326599121094, 6.967254638671875, 6.215110778808594, 17.0032958984375, 27.7900390625, 70.68505859375, 8.85858154296875, -45.58601379394531, 3.3559722900390625, 12.740509033203125, -9.505903244018555, -3.7323150634765625, 17.815963745117188, -26.6697998046875, 61.703094482421875, -6.472330093383789, 34.64894104003906, 12.3895263671875, -23.63739013671875, 2.0684814453125, -13.1517333984375, 12.11053466796875, 18.5408935546875, -14.4547119140625, 38.4951171875, 40.786285400390625, -22.63494873046875, 0.5097885131835938, -3.7775650024414062, 44.204833984375, -21.85306739807129, 24.7615966796875, 5.204856872558594, 1.868753433227539, 8.481536865234375, -7.4160614013671875, 17.254901885986328, 10.067848205566406, -20.892425537109375, 28.317779541015625, -2.522552490234375, 11.94293212890625, 34.135528564453125, -4.77392578125, -30.1104736328125, 52.266754150390625, 7.19580078125, 32.1046142578125, 41.3675537109375, 13.009552001953125, 19.771392822265625, -12.89013671875, -2.845062255859375, 20.91510009765625, -9.799591064453125, 27.698646545410156, 9.401611328125, 19.575286865234375, 14.7982177734375, 22.94232177734375, 17.2720947265625, -7.138092041015625, -21.102783203125, -10.171417236328125, 1.2757072448730469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000131.npy"}
{"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 12.947770118713379, "std": 24.67292022705078, "min": -44.425994873046875, "p10": -17.85454559326172, "median": 10.49224853515625, "p90": 45.50893096923828, "max": 94.4620361328125, "pos_frac": 0.703125, "sample": [-7.9022216796875, 24.2623291015625, 36.24053955078125, 12.459938049316406, -0.928070068359375, -8.6597900390625, 27.388999938964844, 10.721038818359375, -4.9049072265625, 37.187530517578125, 4.228199005126953, -9.91033935546875, -21.381072998046875, 34.58624267578125, 21.710296630859375, 8.337265014648438, 2.585601806640625, 45.86537170410156, 41.58778381347656, 60.190948486328125, 34.56378173828125, 30.840240478515625, 71.523681640625, 48.03680419921875, 14.485816955566406, 1.709869384765625, -0.6921768188476562, -32.67677307128906, 13.685821533203125, 20.31659698486328, -1.38238525390625, 46.5323486328125, 2.028076171875, 20.05560302734375, -24.63262939453125, -17.928726196289062, -16.520286560058594, 9.759765625, 3.07781982421875, 4.3538818359375, 42.40789794921875, 16.362472534179688, -9.637832641601562, -14.6151123046875, -22.654781341552734, -0.20156097412109375, 6.236053466796875, -17.82275390625, 14.9075927734375, -11.75933837890625, 15.737258911132812, -10.932435989379883, 16.779783248901367, 30.00225830078125, 3.294281005859375, 0.0369873046875, 4.538818359375, 18.617446899414062, 94.4620361328125, 0.16168975830078125, 25.67559814453125, 28.06573486328125, 21.44952392578125, -22.95843505859375, 36.700340270996094, -0.938507080078125, -3.4651756286621094, 12.429763793945312, -19.598052978515625, 0.423126220703125, 17.191925048828125, 6.543182373046875, 13.423736572265625, 35.82954406738281, -2.852020263671875, -0.7399749755859375, 0.0, 35.942047119140625, 13.686569213867188, -19.353347778320312, 8.570816040039062, 36.124755859375, 51.27618408203125, -31.547286987304688, -44.425994873046875, 56.395782470703125, 13.516510009765625, 17.245513916015625, 2.90716552734375, 9.5592041015625, 21.763702392578125, 12.174407958984375, 32.660797119140625, 24.99248504638672, 54.68678283691406, 91.21240234375, -15.552154541015625, 28.214614868164062, 0.92938232421875, 14.606231689453125, 25.120391845703125, 39.565032958984375, 1.581817626953125, 0.6597347259521484, 10.263458251953125, 51.81610107421875, 4.0850830078125, 6.11181640625, 0.0, -18.67919921875, 31.57183837890625, 40.09700012207031, -9.575645446777344, 1.3454818725585938, -9.26300048828125, 10.931396484375, -2.343231201171875, 57.82427978515625, -26.931549072265625, 45.356170654296875, 34.101226806640625, 28.2462158203125, -23.843063354492188, 14.838512420654297, -16.350021362304688, 17.13055419921875, 1.013916015625, 63.1497802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000132.npy"}
{"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 13.395610809326172, "std": 22.293352127075195, "min": -66.0233154296875, "p10": -12.347975158691401, "median": 11.504281997680664, "p90": 41.792425537109374, "max": 70.19522094726562, "pos_frac": 0.7578125, "sample": [22.589874267578125, 1.9041690826416016, -33.299163818359375, 0.0, 0.412139892578125, 31.24627685546875, -0.9472942352294922, 14.097152709960938, 4.648414611816406, 28.438186645507812, 16.986160278320312, 18.32183837890625, 4.697479248046875, 24.680503845214844, 38.390716552734375, 50.17523193359375, -8.825653076171875, -0.073699951171875, -1.1426544189453125, -16.02886962890625, 33.13261413574219, 48.4544677734375, 70.19522094726562, 5.4503173828125, 26.147003173828125, 42.86517333984375, 22.233882904052734, 54.268951416015625, 0.0, -0.9227294921875, 17.622528076171875, 23.610137939453125, -3.559083938598633, 1.320526123046875, 27.486968994140625, 32.20613098144531, 22.954681396484375, 27.614242553710938, -26.37640380859375, 31.1414794921875, 39.059967041015625, 7.137939453125, -5.9668121337890625, -7.28912353515625, 11.420516967773438, 0.70587158203125, -18.39886474609375, 8.216033935546875, 2.849456787109375, 10.8023681640625, 16.215179443359375, 39.44927978515625, 3.7186279296875, 29.438369750976562, 49.572113037109375, 9.801788330078125, 3.3782196044921875, 42.76171875, -4.690460205078125, 29.86260986328125, -21.419769287109375, 11.442249298095703, 35.283905029296875, 0.705078125, 17.970773696899414, 9.039863586425781, 25.261138916015625, 43.69970703125, 30.061492919921875, 26.305519104003906, -36.384765625, -0.889373779296875, 11.566314697265625, 8.082351684570312, 49.5794677734375, 9.131750106811523, -5.494773864746094, -21.15618896484375, 4.00433349609375, 47.0107421875, -6.905208587646484, 1.9627227783203125, 9.876556396484375, -10.6512451171875, 21.890594482421875, -10.902023315429688, 7.600982666015625, 25.90387725830078, 24.1212158203125, 40.40692138671875, -5.378120422363281, 47.429412841796875, 28.743637084960938, 13.639846801757812, 16.437515258789062, 5.14801025390625, 8.42462158203125, 40.21954345703125, 24.645965576171875, 44.6102294921875, 6.717796325683594, 29.649215698242188, 21.45288848876953, 24.67824935913086, -27.843154907226562, 22.161426544189453, 21.042572021484375, 28.408599853515625, -26.291397094726562, 48.3516845703125, 20.252197265625, 8.289352416992188, 4.51678466796875, -15.72186279296875, -50.88360595703125, 19.592758178710938, -66.0233154296875, 3.6134777069091797, 3.0607452392578125, 41.37701416015625, 8.298553466796875, 23.944549560546875, 35.81826400756836, -4.2003021240234375, -25.310333251953125, 2.5942821502685547, 25.883575439453125, 40.02153778076172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000133.npy"}
{"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 9.294631958007812, "std": 20.882354736328125, "min": -41.90576171875, "p10": -14.68097381591797, "median": 7.556632995605469, "p90": 32.34643096923828, "max": 91.43109130859375, "pos_frac": 0.6875, "sample": [40.44212341308594, 25.34637451171875, -16.333663940429688, -5.130706787109375, -25.79803466796875, -14.798263549804688, 6.54376220703125, -14.630706787109375, -4.480987548828125, 25.72216796875, -37.02276611328125, -6.8408660888671875, 10.3360595703125, 15.63726806640625, 26.979080200195312, 0.49463844299316406, -5.785575866699219, 5.759426116943359, 23.651123046875, 8.048030853271484, 14.48846435546875, 61.890899658203125, 0.0, 21.3001708984375, 18.4462890625, 23.400238037109375, 31.85015869140625, -2.43927001953125, 4.88201904296875, 45.933197021484375, -29.564191818237305, -4.9279632568359375, 13.489959716796875, 2.7115135192871094, 28.662269592285156, 1.920684814453125, 26.243270874023438, 14.366828918457031, 16.144012451171875, 4.619293212890625, -10.256668090820312, 26.03730010986328, 14.7301025390625, 26.85565185546875, 91.43109130859375, 34.408203125, 2.579193115234375, -13.04083251953125, 68.732421875, 25.292816162109375, -22.338104248046875, 30.773345947265625, 10.2784423828125, -14.288108825683594, -0.02321624755859375, 1.316162109375, 28.76837158203125, 20.579071044921875, 38.83488464355469, 8.48797607421875, -11.375064849853516, -3.2076568603515625, 10.770050048828125, 7.85650634765625, -27.04656982421875, 41.009918212890625, -2.41180419921875, 15.6583251953125, 24.50408935546875, 10.402336120605469, 2.2777328491210938, 13.691650390625, -17.18441390991211, -3.877574920654297, 2.6658935546875, 41.31158447265625, 10.438629150390625, 28.2144775390625, 6.322263717651367, -33.132110595703125, 17.571044921875, 32.78466796875, 27.789199829101562, -11.78741455078125, -0.9415283203125, -1.7945709228515625, 10.530303955078125, 12.320623397827148, 43.8184814453125, 0.21750259399414062, 0.1824951171875, -5.5689849853515625, 9.89138412475586, 0.088653564453125, -41.90576171875, 27.372695922851562, 22.825164794921875, 14.1273193359375, 6.595863342285156, -9.173629760742188, 5.444248199462891, 11.690032958984375, -14.1591796875, -40.482757568359375, -16.549819946289062, 17.8184814453125, -16.2786865234375, 3.2564697265625, 36.52546691894531, 0.910675048828125, 7.2567596435546875, 32.013671875, 0.9169769287109375, 20.22198486328125, 37.56097412109375, 24.896259307861328, 32.15861511230469, 3.04949951171875, 9.527252197265625, -7.9571990966796875, -0.1292724609375, 1.469879150390625, -3.4116897583007812, -3.574951171875, 32.12158203125, 19.05016326904297, 3.8214035034179688, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000134.npy"}
{"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 13.174522399902344, "std": 20.827495574951172, "min": -51.62745666503906, "p10": -10.038611221313475, "median": 9.562899589538574, "p90": 44.89795837402343, "max": 60.477294921875, "pos_frac": 0.7734375, "sample": [0.4662303924560547, -25.837661743164062, 9.618688583374023, 24.05511474609375, -9.94027328491211, 42.706878662109375, 1.1300153732299805, 12.9322509765625, 29.744903564453125, 3.3420257568359375, 51.949371337890625, 17.690887451171875, -6.398040771484375, -3.2818603515625, 11.1661376953125, 5.71875, 8.37603759765625, 27.30078125, 22.48065185546875, 2.1687088012695312, 1.5596694946289062, 7.380210876464844, -1.1202011108398438, 2.279712677001953, -28.93438720703125, 0.38101768493652344, 33.767974853515625, 25.181991577148438, 19.152305603027344, 46.56256103515625, 31.9830322265625, -0.9892730712890625, 29.70733642578125, 3.04217529296875, -18.059219360351562, 38.3665771484375, 3.0491485595703125, 17.47894287109375, 27.7711181640625, 8.3497314453125, 2.212646484375, 13.633453369140625, 14.03912353515625, 35.51678466796875, 58.3297119140625, 19.320709228515625, 47.30096435546875, 11.098861694335938, 1.96533203125, -2.7490234375, 20.788833618164062, 21.9912109375, 2.3082656860351562, -10.26806640625, -17.49114990234375, -6.44256591796875, 43.906036376953125, 7.4636383056640625, 40.31842041015625, 16.4700927734375, 23.344635009765625, 15.199737548828125, 32.60289001464844, -5.3499603271484375, -3.091522216796875, 9.507110595703125, -3.5677490234375, 20.3109130859375, 17.893463134765625, 3.772918701171875, 16.29199981689453, 6.7871551513671875, 50.28192138671875, 17.931243896484375, 32.959808349609375, 31.50048065185547, 16.2225341796875, 7.4683837890625, -2.846649169921875, 51.984039306640625, 22.605941772460938, -19.983688354492188, -1.5041351318359375, 1.8291015625, 8.628936767578125, 51.23046875, 0.3974609375, 52.92138671875, 36.550506591796875, 49.032318115234375, 19.488494873046875, 39.5343017578125, 2.5433349609375, -40.033843994140625, 44.6328125, 5.8948211669921875, 5.601188659667969, 8.384620666503906, 1.5040283203125, 23.764556884765625, 8.246833801269531, 10.580963134765625, 29.6749267578125, 60.477294921875, 2.768962860107422, 0.041656494140625, 4.172676086425781, 15.308425903320312, 47.224609375, 57.02679443359375, -13.4642333984375, -15.09832763671875, 4.26611328125, 9.869049072265625, 0.0, 19.797210693359375, -18.62371826171875, -51.62745666503906, -0.9482803344726562, 11.058441162109375, -6.4813232421875, 21.067703247070312, 4.8460693359375, 45.516632080078125, 27.409149169921875, -12.811004638671875, -18.1539306640625, -0.0456695556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000135.npy"}
{"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 12.453495025634766, "std": 22.901521682739258, "min": -55.0660400390625, "p10": -9.694569396972655, "median": 10.502715110778809, "p90": 41.176338195800774, "max": 82.24954223632812, "pos_frac": 0.7578125, "sample": [43.35369873046875, 17.530181884765625, 1.801361083984375, 15.46548080444336, 34.04615020751953, 39.126861572265625, 45.962158203125, 11.824661254882812, 33.279518127441406, 58.74322509765625, 6.88189697265625, 5.348812103271484, 44.666900634765625, 6.455970764160156, 25.62514877319336, 39.45445251464844, 13.02032470703125, 39.15068054199219, 9.10980224609375, -0.1342620849609375, -43.527130126953125, 35.38866424560547, 8.28289794921875, 24.172821044921875, -19.09576416015625, 4.625457763671875, -7.2132568359375, -41.004974365234375, 8.47454833984375, 11.773040771484375, 4.730926513671875, -0.7894973754882812, -6.2079620361328125, 8.0233154296875, 24.78595733642578, -26.966873168945312, -8.5484619140625, 30.38420867919922, 13.362373352050781, -4.124237060546875, 27.159759521484375, -20.454177856445312, 58.708740234375, -55.0660400390625, 12.908294677734375, -6.097602844238281, 3.6409912109375, 5.358245849609375, 17.140625, -3.7256088256835938, 15.4423828125, -35.62255859375, 42.65925598144531, 17.601806640625, 30.98876953125, 70.1719970703125, -9.2523193359375, 7.3299407958984375, 12.21929931640625, 30.604385375976562, 44.7918701171875, 18.48504638671875, 35.965301513671875, -12.934654235839844, 10.683612823486328, 0.40511322021484375, -4.308250427246094, 10.193115234375, 33.042694091796875, -3.1700439453125, 3.4888153076171875, 13.8433837890625, 8.933420181274414, 9.38916015625, 4.14874267578125, 51.21551513671875, -0.8649787902832031, 13.056549072265625, 10.094459533691406, -8.12164306640625, -1.04815673828125, -27.18780517578125, 21.04339599609375, 57.768890380859375, 37.10601806640625, 24.307159423828125, 1.6977195739746094, 16.97104263305664, -7.433229446411133, 46.38134765625, 18.0213623046875, 10.725341796875, -10.726486206054688, 12.168861389160156, 5.7276611328125, 0.900726318359375, 56.868438720703125, 12.368232727050781, 29.254867553710938, 22.632125854492188, -0.1782207489013672, 6.939781188964844, 32.953857421875, 0.6948165893554688, 30.8890380859375, -13.760894775390625, -8.830215454101562, 2.2575836181640625, 35.43231201171875, 40.540802001953125, 35.127197265625, 10.321817398071289, -20.147247314453125, -7.568511962890625, 2.802093505859375, 12.40313720703125, 82.24954223632812, 5.047412872314453, 0.389190673828125, 8.66387939453125, 14.244308471679688, 13.300682067871094, 12.397850036621094, 3.163116455078125, 14.294082641601562, -49.39251708984375, 30.778564453125, 4.1937408447265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000136.npy"}
{"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 11.167598724365234, "std": 21.649494171142578, "min": -46.0382080078125, "p10": -14.902126693725583, "median": 8.694133758544922, "p90": 36.5694580078125, "max": 73.87570190429688, "pos_frac": 0.7734375, "sample": [-29.7005615234375, 3.08319091796875, 36.4420166015625, -6.8908538818359375, 12.194549560546875, 19.09588623046875, 38.49900817871094, 11.579803466796875, 2.81536865234375, 3.0784912109375, 18.834693908691406, 14.703018188476562, 26.4654541015625, 56.412567138671875, 15.18597412109375, 3.523681640625, 0.3350982666015625, 31.09881591796875, 1.414794921875, -14.09613037109375, 29.30132293701172, 20.548614501953125, 8.758636474609375, 12.57989501953125, -42.7581787109375, 22.538070678710938, 3.707183837890625, -16.95013427734375, 30.345550537109375, 8.629631042480469, 46.793426513671875, 16.223312377929688, -1.8879241943359375, 39.30029296875, -18.022354125976562, 0.10540771484375, 5.20184326171875, 8.110664367675781, 1.0164794921875, -7.795982360839844, 9.31817626953125, -4.765787124633789, -20.37236785888672, 2.746337890625, -46.0382080078125, 4.99365234375, 7.6624298095703125, 14.293991088867188, -44.40574645996094, 16.428638458251953, 7.94940185546875, 45.107757568359375, 7.883819580078125, 10.5277099609375, 13.659820556640625, 33.533050537109375, -18.2637939453125, 36.8668212890625, -16.7430419921875, 1.3058929443359375, 5.239776611328125, -1.2939682006835938, 27.1302490234375, 0.4692230224609375, 12.524383544921875, 56.7042236328125, -26.86053466796875, 26.361038208007812, 7.958740234375, -2.12249755859375, 2.3350067138671875, 24.63751220703125, 29.929901123046875, 12.03076171875, 34.55495071411133, 1.6936264038085938, 0.9835910797119141, 73.87570190429688, 28.673629760742188, 16.270462036132812, 7.9153594970703125, 57.1724853515625, -1.4684066772460938, 35.08721923828125, 17.54168701171875, 19.1279296875, 57.3837890625, 19.294509887695312, -4.792610168457031, 34.50518798828125, 1.2657394409179688, -11.845726013183594, -1.277557373046875, 7.7349853515625, 1.8183841705322266, 63.38165283203125, 26.217041015625, 3.361896514892578, 17.98961639404297, 14.637420654296875, -36.495513916015625, -31.222030639648438, 8.393646240234375, -32.23774719238281, -12.788192749023438, 51.496124267578125, -14.113162994384766, 27.1142578125, 38.00518798828125, 33.636749267578125, 23.518524169921875, 32.64093017578125, -0.11114501953125, 20.025726318359375, 14.48956298828125, 23.040538787841797, -3.2033538818359375, 6.8772430419921875, 1.267120361328125, 6.426788330078125, 7.361396789550781, 20.67169952392578, 6.512359619140625, 13.141204833984375, 14.845947265625, 18.2774658203125, 28.986297607421875, -4.76055908203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000137.npy"}
{"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 11.174501419067383, "std": 24.39612579345703, "min": -46.346160888671875, "p10": -17.537298583984374, "median": 8.14291000366211, "p90": 41.89575080871582, "max": 80.31588745117188, "pos_frac": 0.6796875, "sample": [69.49679565429688, 41.29443359375, 45.339935302734375, 17.62511444091797, -12.166877746582031, 6.47467041015625, -6.83428955078125, 67.9041748046875, 6.70867919921875, 24.148406982421875, 12.071517944335938, 11.642242431640625, 4.90911865234375, 14.181564331054688, 12.792472839355469, -30.50067901611328, -18.011199951171875, 29.62890625, -11.29193115234375, -28.064178466796875, 21.16162109375, 6.46185302734375, -12.936965942382812, -8.783157348632812, -46.346160888671875, 25.362411499023438, -3.8830108642578125, 35.04498291015625, 1.9176101684570312, 80.31588745117188, -9.12799072265625, 6.431243896484375, 3.55133056640625, 43.298824310302734, 1.3086318969726562, 26.492462158203125, -25.939041137695312, 11.253036499023438, 40.856903076171875, -1.0634307861328125, 11.661697387695312, -2.0489368438720703, 30.756210327148438, 36.295928955078125, 13.41204833984375, 33.04179382324219, 6.669342041015625, 4.1420745849609375, 15.96051025390625, -14.782608032226562, 11.939102172851562, 1.6964492797851562, 37.771759033203125, -16.926956176757812, 43.76885986328125, 11.228515625, 37.29461669921875, -4.1014404296875, 19.506362915039062, -18.1375732421875, -18.258934020996094, 3.3673782348632812, -1.8843994140625, -1.0292816162109375, 21.587615966796875, -28.122833251953125, 66.3883056640625, 30.834579467773438, 27.1448974609375, -2.791748046875, -28.25274658203125, 17.4691162109375, -5.8690948486328125, 54.157745361328125, 14.808990478515625, -9.509246826171875, -3.7139816284179688, 21.47857666015625, 14.008338928222656, 1.8024520874023438, 13.338455200195312, 32.171630859375, 27.978546142578125, 33.68212890625, -39.979156494140625, -10.641143798828125, 14.158859252929688, 0.384307861328125, -14.606529235839844, -16.623687744140625, 28.486312866210938, 56.242462158203125, 43.50947570800781, 4.95208740234375, -2.601358413696289, 27.6156005859375, 28.280029296875, 26.4293212890625, 1.820068359375, -3.5192108154296875, 29.455047607421875, -15.243682861328125, 73.6923828125, -5.345893859863281, 7.462196350097656, 14.4603271484375, -8.750381469726562, -17.334197998046875, 6.2625274658203125, 39.103660583496094, -18.837249755859375, 3.153594970703125, 2.5689773559570312, 19.25830078125, 27.939666748046875, 2.5074539184570312, 53.836944580078125, -33.760009765625, 31.8907470703125, -36.742431640625, 4.999843597412109, 22.383056640625, 3.1621246337890625, 19.649429321289062, 48.2291259765625, 26.359268188476562, 8.823623657226562, -15.415847778320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000138.npy"}
{"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 8.27757453918457, "std": 23.85011100769043, "min": -46.21221923828125, "p10": -21.421119689941406, "median": 5.867824554443359, "p90": 41.10589904785156, "max": 68.01846313476562, "pos_frac": 0.625, "sample": [-36.61322784423828, -3.86248779296875, -31.589202880859375, 5.8721160888671875, -28.3653564453125, 15.416458129882812, 31.425888061523438, 18.1878662109375, -2.3629150390625, -11.36612319946289, 47.038665771484375, 3.4225311279296875, 22.858978271484375, 11.893829345703125, -28.012420654296875, 44.488555908203125, 53.561614990234375, 10.6981201171875, -14.616874694824219, 10.923149108886719, 5.4174652099609375, 25.077651977539062, 32.22764587402344, -11.104522705078125, 13.586212158203125, -12.011260986328125, 6.030906677246094, 34.72541809082031, -6.342529296875, -46.21221923828125, 0.7206954956054688, -7.768707275390625, 35.26585388183594, 40.85205078125, 26.50830078125, 0.22155380249023438, 12.386276245117188, -8.950790405273438, 4.40985107421875, 29.26239013671875, -2.6110000610351562, -20.875228881835938, 0.41046142578125, 21.016571044921875, 12.328887939453125, -4.856689453125, -40.220703125, 0.591217041015625, -11.203174591064453, 40.313629150390625, -8.490509033203125, -0.8743896484375, -22.802947998046875, -21.104965209960938, 7.647224426269531, 42.1494140625, -2.8420257568359375, -0.04937744140625, -22.1588134765625, -18.154632568359375, 39.92816162109375, -13.940895080566406, 41.81915283203125, 23.56670570373535, 27.294219970703125, 15.5570068359375, 0.22847938537597656, 1.4140167236328125, 39.9576416015625, 22.295303344726562, -16.391555786132812, 18.269638061523438, 15.18853759765625, 42.12474060058594, -3.8586273193359375, -9.622528076171875, 3.689910888671875, 66.74136352539062, -9.60699462890625, 41.698211669921875, 24.61072540283203, 21.346588134765625, 14.4383544921875, 52.03089904785156, 5.76629638671875, -14.1507568359375, 54.10430908203125, -37.10235595703125, 17.685562133789062, -4.336977005004883, 27.471893310546875, 46.602813720703125, 18.3465576171875, 15.65020751953125, -17.374969482421875, 16.85369873046875, -0.2144927978515625, -38.103302001953125, -7.1992950439453125, 2.1740646362304688, 23.668441772460938, 8.296173095703125, -8.972259521484375, 3.7940216064453125, 11.24786376953125, -8.682861328125, 27.126129150390625, 68.01846313476562, 9.966880798339844, 57.7626953125, 39.200164794921875, 18.67962646484375, 3.039306640625, -31.851226806640625, 5.446502685546875, -32.82025146484375, 0.0, 7.0907135009765625, 14.51971435546875, 33.016849517822266, -7.988494873046875, -32.5399169921875, 19.523155212402344, -4.85784912109375, 21.09454345703125, -13.135009765625, 36.555877685546875, 5.863533020019531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000139.npy"}
{"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 14.982853889465332, "std": 22.47102928161621, "min": -41.382415771484375, "p10": -9.303962707519531, "median": 11.130599975585938, "p90": 47.528585815429686, "max": 79.6031494140625, "pos_frac": 0.7109375, "sample": [-6.482421875, 24.58123779296875, 32.397857666015625, -7.7728271484375, -10.995491027832031, 49.96063232421875, -5.0609588623046875, 38.839813232421875, -2.065093994140625, 0.592437744140625, -16.927833557128906, 11.401336669921875, 29.15899658203125, 19.275596618652344, 5.958625793457031, 8.393783569335938, 2.1688232421875, 19.820281982421875, 47.51513671875, 10.29541015625, 7.838722229003906, -17.464508056640625, 27.860618591308594, -24.9814453125, 21.228164672851562, 40.323760986328125, -2.46685791015625, 6.522491455078125, 43.22248840332031, 42.622459411621094, 30.2271728515625, 33.38624572753906, 64.19549560546875, 39.85828399658203, -9.330184936523438, 56.813201904296875, -6.119781494140625, 3.508056640625, 48.7999267578125, 49.06573486328125, -1.4153060913085938, -25.614044189453125, 10.957000732421875, -0.23885345458984375, -3.950836181640625, 34.07769775390625, 50.375244140625, 11.490997314453125, -6.9122314453125, 79.6031494140625, 9.792465209960938, 24.772796630859375, 30.808685302734375, 11.615440368652344, 3.3232269287109375, 21.583740234375, 19.314300537109375, 6.017000198364258, -8.219329833984375, 23.5831298828125, 26.570343017578125, 0.3393402099609375, 3.5063934326171875, 21.22039794921875, 7.7765960693359375, 1.270843505859375, -4.1992034912109375, -9.292724609375, 0.0, -41.382415771484375, -1.3769073486328125, -27.995513916015625, 67.68278503417969, 0.7184371948242188, 4.948860168457031, 14.4195556640625, 22.869949340820312, 30.1312255859375, 3.0413894653320312, 47.559967041015625, 10.797401428222656, 3.77703857421875, 1.1121540069580078, 43.644775390625, 52.146820068359375, 35.34038543701172, 31.138336181640625, -3.1169471740722656, 34.297149658203125, 37.394309997558594, 6.091985702514648, 6.714469909667969, 19.795738220214844, -3.34100341796875, 28.687644958496094, 32.32112121582031, 0.0, -6.399463653564453, 19.3189697265625, -1.1695556640625, -8.27227783203125, 47.81695556640625, -16.650146484375, 45.53704833984375, 9.564422607421875, -19.039718627929688, 44.1627197265625, 22.199249267578125, 21.669708251953125, 22.548843383789062, 44.092681884765625, 22.645751953125, 49.3316650390625, 19.30455780029297, 26.754119873046875, -8.196746826171875, -4.23394775390625, -13.80804443359375, 9.224212646484375, -26.2200927734375, 11.326385498046875, 55.14862060546875, 11.30419921875, -3.3640975952148438, 8.13348388671875, -16.36968994140625, 26.547607421875, 23.157470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000140.npy"}
{"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 14.343294143676758, "std": 25.20753288269043, "min": -76.11990356445312, "p10": -10.821463394165038, "median": 13.425251007080078, "p90": 47.16129455566406, "max": 90.11578369140625, "pos_frac": 0.75, "sample": [41.604217529296875, 17.235015869140625, -25.598304748535156, -9.990386962890625, 5.6263427734375, 13.775650024414062, -7.222137451171875, 22.95379638671875, -46.64642333984375, 90.11578369140625, 5.573158264160156, 12.15283203125, 30.248680114746094, 30.860549926757812, 14.338973999023438, 43.0950927734375, 24.640960693359375, 18.504150390625, 17.244537353515625, 24.0428466796875, 57.871429443359375, -5.2701416015625, 0.4614715576171875, 44.855926513671875, 0.24152755737304688, -15.86958122253418, 38.40482711791992, 13.323234558105469, 0.89019775390625, 47.791717529296875, 34.71087646484375, -16.816131591796875, -11.78076171875, 72.01373291015625, 1.13720703125, 17.484466552734375, -24.867172241210938, -10.410335540771484, 11.072723388671875, -17.48565673828125, 43.3970947265625, 36.70391845703125, 13.552534103393555, 19.261245727539062, -7.210670471191406, 9.06585693359375, 2.603485107421875, -7.339500427246094, 59.83648681640625, 7.920135498046875, 13.527267456054688, 24.454071044921875, 3.330280303955078, -9.206512451171875, 29.337615966796875, 23.429229736328125, 1.5049514770507812, 1.0012588500976562, 24.436492919921875, 0.6425018310546875, 32.725128173828125, 34.09093475341797, -12.42791748046875, -41.75151062011719, -6.12481689453125, 69.28289794921875, -0.5828800201416016, 25.200660705566406, -5.014152526855469, 37.531158447265625, 5.8177642822265625, 14.443084716796875, 6.984638214111328, 23.997169494628906, 30.294891357421875, 36.20991516113281, 33.18914794921875, -7.280670166015625, 0.0, -15.967037200927734, 40.9688720703125, -2.414764404296875, 0.03009033203125, 58.113525390625, 6.0764312744140625, 53.69891357421875, -76.11990356445312, 16.03948974609375, 20.55401611328125, -2.39349365234375, 23.295852661132812, 18.404586791992188, 33.45001983642578, -44.37239074707031, -7.094879150390625, 17.784317016601562, 0.6524124145507812, 55.8271484375, -0.300872802734375, 30.329689025878906, -4.7469940185546875, 10.45526123046875, 18.535446166992188, 35.874267578125, 15.8134765625, 4.203485488891602, 24.25213623046875, 48.964324951171875, 42.52825927734375, -10.162246704101562, 7.256343841552734, 12.47552490234375, 56.99615478515625, -36.20216369628906, 1.58868408203125, 10.459381103515625, 10.78228759765625, 12.576202392578125, 20.0618896484375, 19.628692626953125, -9.094810485839844, 46.89111328125, 48.2227783203125, 7.966217041015625, 11.058784484863281, 56.32037353515625, 16.380126953125, 3.1464996337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000141.npy"}
{"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 13.999345779418945, "std": 25.26759147644043, "min": -66.7784423828125, "p10": -14.321574401855461, "median": 9.38804817199707, "p90": 47.31090393066406, "max": 79.06155395507812, "pos_frac": 0.6796875, "sample": [-2.275115966796875, -2.0610580444335938, 9.731075286865234, 27.773773193359375, 2.050935745239258, -19.64117431640625, 10.67083740234375, -1.15081787109375, 72.28256225585938, 39.66375732421875, 42.537017822265625, -3.2866058349609375, -6.982086181640625, 34.3365478515625, 17.35723876953125, 29.2080078125, -0.3273811340332031, -3.86669921875, 54.3463134765625, 25.32855224609375, 0.15771484375, 25.654747009277344, 0.0, 20.8984375, 20.78765869140625, 46.821014404296875, 12.70343017578125, 3.3380279541015625, -2.7209548950195312, -0.625152587890625, -1.156280517578125, 27.098846435546875, -2.4259567260742188, -22.998016357421875, 3.9760589599609375, 1.856231689453125, 24.64544677734375, 13.996482849121094, -24.675445556640625, 26.215225219726562, 20.13555908203125, 50.191864013671875, 44.294189453125, 17.505874633789062, 18.637359619140625, -10.737213134765625, -1.2930793762207031, 44.51078796386719, 20.174766540527344, 69.24929809570312, 14.9564208984375, -6.397462844848633, 4.9557342529296875, 13.403533935546875, -12.311904907226562, 2.9393234252929688, 0.33660888671875, 34.833709716796875, -19.01080322265625, -19.848785400390625, 21.26397705078125, 32.30322265625, -9.572601318359375, -8.99957275390625, 48.4539794921875, 37.430931091308594, 17.629638671875, 0.8087425231933594, 46.164268493652344, -2.8013992309570312, 7.374969482421875, 21.62383270263672, 0.976531982421875, 6.2200927734375, -23.255233764648438, 30.448211669921875, -39.656524658203125, 28.416175842285156, 41.821533203125, -25.895751953125, -4.84051513671875, 69.02200317382812, 79.06155395507812, -1.46343994140625, -1.9140853881835938, 13.792312622070312, 8.366180419921875, 51.07720947265625, 9.045021057128906, 6.395160675048828, -2.8572998046875, -3.629241943359375, 31.1920166015625, 42.67198181152344, 55.0279541015625, 6.561134338378906, 35.76837158203125, 24.471298217773438, 21.925079345703125, 53.658233642578125, -66.7784423828125, 53.558837890625, -29.68798828125, 1.5853557586669922, -0.5608711242675781, -26.988616943359375, 35.27661895751953, 17.112197875976562, 32.461761474609375, 0.41583251953125, 36.98016357421875, -35.788330078125, 3.0762710571289062, 5.7923583984375, 22.737998962402344, -8.0826416015625, 29.1087646484375, -1.4237995147705078, 4.0798187255859375, 2.0540771484375, -23.397735595703125, 63.49200439453125, 0.2917499542236328, -1.780120849609375, 31.11029052734375, 29.54564666748047, 66.48643493652344, 41.38374328613281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000142.npy"}
{"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 16.160198211669922, "std": 27.57181739807129, "min": -76.98345947265625, "p10": -11.176780700683592, "median": 12.544143676757812, "p90": 48.534072875976555, "max": 100.5396728515625, "pos_frac": 0.7421875, "sample": [27.098892211914062, 49.608917236328125, 56.056365966796875, 30.2772216796875, 40.70344543457031, 15.43646240234375, 30.334503173828125, 9.589530944824219, -17.330078125, -8.114322662353516, 28.74163818359375, -0.1378936767578125, 21.46673583984375, 12.447097778320312, 1.8455085754394531, -12.368423461914062, 48.07342529296875, 4.64532470703125, -19.565109252929688, 100.5396728515625, -20.39617919921875, -8.5068359375, 44.55094909667969, 45.52976989746094, 23.328330993652344, 44.3851318359375, 35.364288330078125, 60.7872314453125, 89.13223266601562, 1.43450927734375, -2.68731689453125, 42.03227233886719, 27.862945556640625, 31.188709259033203, 25.954666137695312, -6.3765869140625, -76.98345947265625, -2.7807769775390625, 28.79651641845703, 7.702117919921875, 9.070098876953125, -31.816253662109375, 10.118972778320312, 41.387451171875, 5.873039245605469, 6.083770751953125, 10.173089981079102, 21.126129150390625, 17.717525482177734, -67.36871337890625, 6.434234619140625, 28.66558837890625, 3.0633468627929688, 38.859840393066406, 16.7945556640625, 24.20697021484375, -1.00335693359375, 22.68760108947754, -13.677352905273438, -1.2632560729980469, -8.162994384765625, 0.7746448516845703, 70.29351806640625, 19.7862548828125, 5.539937973022461, -2.1643524169921875, 57.86865234375, 32.04173278808594, 40.15594482421875, 3.678558349609375, 29.960235595703125, 34.84724426269531, -8.82065200805664, 3.8084945678710938, 33.17222595214844, -10.66607666015625, -37.8487548828125, 0.6387557983398438, 10.6229248046875, 12.641189575195312, 4.1107177734375, -1.98736572265625, 8.425216674804688, 17.523147583007812, 11.882568359375, -2.16546630859375, -16.368194580078125, 3.220367431640625, -0.24365234375, 13.17388916015625, 34.562076568603516, 14.691741943359375, -2.9643707275390625, 21.792388916015625, 18.751068115234375, 24.632568359375, 0.27126502990722656, 15.42437744140625, -4.612495422363281, 2.5846099853515625, 82.9910888671875, 31.8282470703125, -4.1159820556640625, 11.902336120605469, -30.954559326171875, 75.26019287109375, 35.16339111328125, 3.3061256408691406, 3.961214065551758, -2.2459869384765625, 27.503402709960938, 57.046173095703125, 28.679344177246094, 67.6617431640625, 25.95074462890625, 63.0189208984375, 43.749298095703125, 44.33477783203125, -40.0252685546875, 26.078327178955078, -17.2315673828125, 50.75830078125, -6.4512939453125, 11.862548828125, 47.492401123046875, 7.0150146484375, 15.08367919921875, 1.210123062133789], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000143.npy"}
{"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 13.564828872680664, "std": 25.16206932067871, "min": -66.06709289550781, "p10": -14.241423034667969, "median": 6.773612976074219, "p90": 45.983526611328124, "max": 90.30520629882812, "pos_frac": 0.7265625, "sample": [-0.9507045745849609, 17.781333923339844, 45.3448486328125, 5.392219543457031, 4.735260009765625, 9.91046142578125, -17.48590087890625, -10.655250549316406, 1.3626289367675781, 90.30520629882812, 2.1781234741210938, -13.601150512695312, 7.292766571044922, 58.236328125, 5.464996337890625, 36.6441650390625, 2.745624542236328, 45.03790283203125, 8.94366455078125, 0.706817626953125, 2.1305694580078125, 20.36358642578125, -14.389549255371094, 2.513519287109375, 4.21142578125, -5.616455078125, -21.36297607421875, -2.7073745727539062, 28.724578857421875, 4.978219985961914, -38.7464599609375, 6.491044998168945, 0.0506744384765625, 24.45953369140625, 50.68132019042969, 39.810791015625, 32.512664794921875, 69.82708740234375, 8.30487060546875, 59.06488037109375, -1.0427017211914062, -8.363662719726562, 6.183341979980469, 19.628570556640625, -0.6438694000244141, -17.65135955810547, 6.4283447265625, 35.1121826171875, -16.605545043945312, 21.27978515625, 13.37255859375, 55.67999267578125, 36.712127685546875, -19.39501953125, -66.06709289550781, -26.7156982421875, 16.625823974609375, -14.177940368652344, -12.746795654296875, 2.3652496337890625, 35.53631591796875, -0.7547607421875, -5.642242431640625, 4.1345062255859375, 26.08795928955078, 13.82037353515625, 67.17041015625, 64.62203979492188, 38.06697082519531, 39.492919921875, 1.2058258056640625, 19.996017456054688, 49.443450927734375, -26.112335205078125, 23.453109741210938, 30.105682373046875, 3.8323974609375, 10.020515441894531, 45.7237548828125, 44.59515380859375, 45.532257080078125, 36.99081802368164, 5.881805419921875, -9.11322021484375, 15.527008056640625, 23.92367172241211, -0.624481201171875, 6.596038818359375, 1.4187660217285156, 33.5516357421875, 71.10952758789062, 33.7591552734375, 13.679092407226562, 0.567352294921875, 12.313404083251953, 48.186187744140625, 2.785369873046875, -3.76300048828125, 7.445709228515625, 34.47412109375, 10.91156005859375, 0.3448524475097656, 0.0, -1.1369781494140625, -13.062149047851562, -11.70770263671875, 3.71630859375, -34.954925537109375, 24.7369384765625, 24.707252502441406, 18.4306640625, -1.14990234375, 43.796356201171875, -6.8820343017578125, 28.558929443359375, 4.313812255859375, 18.082275390625, 5.022918701171875, 18.897857666015625, -23.42706298828125, 13.113616943359375, 35.4842529296875, 6.9511871337890625, 46.58966064453125, -17.01922607421875, -0.0573883056640625, 71.245361328125, 5.0827789306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000144.npy"}
{"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 17.322629928588867, "std": 26.237226486206055, "min": -54.271759033203125, "p10": -13.624105834960936, "median": 15.310516357421875, "p90": 52.219961547851554, "max": 84.62322998046875, "pos_frac": 0.7109375, "sample": [-12.485870361328125, 38.95458984375, -41.127105712890625, 36.8310546875, 13.72479248046875, -26.683349609375, 43.683746337890625, 19.234100341796875, 80.63827514648438, 20.2021484375, 18.046417236328125, 11.235260009765625, 23.951271057128906, 32.491790771484375, 29.133705139160156, 64.449462890625, 2.5386962890625, 31.893646240234375, 7.0829010009765625, 6.051666259765625, 6.33624267578125, -7.2640533447265625, 51.2213134765625, -11.56967544555664, -8.677490234375, 67.75775146484375, -27.56353759765625, 49.929840087890625, -0.1475982666015625, 22.993301391601562, 2.21527099609375, 30.68743133544922, 8.664161682128906, 36.51032257080078, 55.692291259765625, 23.312255859375, 74.16494750976562, -54.271759033203125, 8.14727783203125, 2.3203125, 22.1739501953125, 20.23333740234375, -16.2015380859375, 63.623687744140625, 71.41836547851562, 14.496063232421875, 36.97613525390625, 34.21575927734375, 9.204948425292969, 20.82672119140625, 15.220184326171875, -7.45611572265625, -6.244911193847656, 33.18553161621094, 46.87757873535156, 44.531890869140625, 0.39056396484375, -4.37896728515625, -14.945762634277344, 3.469339370727539, 15.400848388671875, 7.513458251953125, -3.23846435546875, 42.20079040527344, 62.7330322265625, -4.9569549560546875, 55.7054443359375, -8.916305541992188, 54.744781494140625, 31.575681686401367, 54.550140380859375, -9.970382690429688, 6.727542877197266, 35.53666687011719, 49.24456787109375, 10.512908935546875, 1.4959678649902344, 12.613906860351562, 28.852310180664062, -19.41839599609375, -24.622352600097656, 5.722076416015625, 9.651378631591797, 30.78076171875, 84.62322998046875, 22.167423248291016, 5.40625, 22.447174072265625, -23.13782501220703, -13.400726318359375, -2.9358596801757812, 0.0, 6.676063537597656, 19.569305419921875, 9.60552978515625, 3.40435791015625, 40.355712890625, 71.6326904296875, 43.11126708984375, 24.360763549804688, -2.93243408203125, -7.86688232421875, 25.675689697265625, 23.883026123046875, 16.932327270507812, -14.14532470703125, 24.927696228027344, 27.915924072265625, -2.6877784729003906, -0.9014978408813477, -15.41259765625, 48.92852783203125, 26.042343139648438, 39.4129638671875, -2.6092300415039062, -3.487274169921875, 3.8622283935546875, 31.1715145111084, 43.754486083984375, 48.773284912109375, 34.32257080078125, -15.26123046875, -1.5954456329345703, -3.3466873168945312, 26.0989990234375, -22.8590087890625, -7.091217041015625, 19.544204711914062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000145.npy"}
{"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 14.029760360717773, "std": 25.29774284362793, "min": -60.65203857421875, "p10": -14.324851989746092, "median": 8.797430038452148, "p90": 45.334329223632814, "max": 78.8472900390625, "pos_frac": 0.7421875, "sample": [5.313897132873535, 2.4486465454101562, 18.135578155517578, -37.47796630859375, 11.936767578125, 40.60295104980469, -26.715499877929688, 12.51141357421875, 33.99327087402344, -26.352508544921875, 78.8472900390625, 74.39328002929688, 9.5760498046875, -7.758056640625, 17.9620361328125, 3.8404312133789062, 42.03656005859375, 26.178924560546875, 2.7015647888183594, 2.360687255859375, 28.833480834960938, 2.1862640380859375, 9.13614273071289, 15.474517822265625, -19.558380126953125, -8.386962890625, 3.7703895568847656, 13.181289672851562, 15.133995056152344, 22.10614013671875, 6.5825958251953125, 0.6940174102783203, 38.00604248046875, 0.23809814453125, 5.77728271484375, 17.58660888671875, 16.553573608398438, -22.591026306152344, 14.739303588867188, 3.5234298706054688, 7.5062713623046875, 48.96299743652344, 1.2911911010742188, 35.8206787109375, 0.0, 45.806884765625, -4.45806884765625, -1.3331546783447266, 27.101150512695312, 27.138702392578125, 14.0289306640625, -15.58795166015625, 10.041738510131836, -2.684112548828125, 51.824615478515625, 60.1815185546875, -60.65203857421875, 29.083786010742188, 6.3890228271484375, 35.02012634277344, 33.88739013671875, 6.431640625, 4.990791320800781, 47.07049560546875, 35.914642333984375, -13.783523559570312, -7.01434326171875, -13.201690673828125, 6.172210693359375, -18.989898681640625, 51.57923889160156, -23.554092407226562, 4.830596923828125, 42.60394287109375, -20.02093505859375, 63.48321533203125, 9.334075927734375, 8.458717346191406, -10.392974853515625, 70.03729248046875, 8.198211669921875, 33.25128173828125, 3.6901493072509766, -10.4764404296875, 33.8466796875, 1.7868595123291016, -7.113372802734375, 3.331573486328125, 0.0, 38.902244567871094, -2.800201416015625, -4.704475402832031, 29.193435668945312, -21.5640869140625, 2.37701416015625, 59.38201904296875, 21.423919677734375, 40.31535339355469, 7.6937255859375, 52.40022277832031, -21.402603149414062, -12.68243408203125, 4.670978546142578, 12.925018310546875, 43.540985107421875, -9.588119506835938, 31.90863037109375, 6.78680419921875, 52.163818359375, 44.72996520996094, 45.131805419921875, -58.935791015625, 29.78750991821289, 44.50458526611328, 17.1927490234375, 37.6260986328125, -7.6351318359375, 43.47227478027344, 3.408905029296875, 39.68402099609375, 0.61236572265625, 6.695980072021484, 24.110382080078125, 43.97425842285156, 30.318695068359375, -8.758567810058594, 30.984146118164062, -9.393402099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000146.npy"}
{"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 11.809995651245117, "std": 28.58364486694336, "min": -75.05926513671875, "p10": -18.594769287109372, "median": 8.764068603515625, "p90": 49.325766754150386, "max": 88.740478515625, "pos_frac": 0.65625, "sample": [0.83917236328125, -3.60845947265625, 8.494514465332031, 28.299530029296875, -9.730171203613281, -2.6781005859375, 84.20462036132812, -13.2144775390625, 0.0, 69.34552001953125, -15.732177734375, 18.046524047851562, 13.787628173828125, 5.18084716796875, 24.43266487121582, 4.7589263916015625, -19.554107666015625, 47.516571044921875, 50.7017822265625, 3.3735885620117188, -4.25921630859375, 19.34393310546875, -0.6392288208007812, 34.67106628417969, -1.09918212890625, 51.05018615722656, 1.1102180480957031, 44.69293212890625, 9.320343017578125, 1.099853515625, 69.8736572265625, 36.253082275390625, 15.932622909545898, 22.210708618164062, 8.49560546875, -29.72137451171875, -43.931541442871094, -6.477783203125, -16.9017333984375, 17.568965911865234, 12.4085693359375, 35.26462173461914, 31.58099365234375, 6.052211761474609, -31.585063934326172, 28.812652587890625, -29.046722412109375, -5.241157531738281, -0.6399288177490234, 2.2276229858398438, -2.7178497314453125, 51.934295654296875, 10.8463134765625, 13.957992553710938, 26.467994689941406, 27.694732666015625, 81.030029296875, 12.972000122070312, 5.4437103271484375, -5.0121612548828125, 21.05975341796875, 25.736785888671875, 58.76399230957031, 10.47232437133789, 57.2698974609375, 0.9293212890625, 2.763916015625, 63.093353271484375, 39.87333679199219, 2.67474365234375, -45.457733154296875, 12.472686767578125, 0.5750732421875, -1.0098876953125, 30.17388916015625, 5.64276123046875, 15.639270782470703, 3.6383590698242188, -0.09538078308105469, -0.26561737060546875, 28.264205932617188, -9.110519409179688, -0.226898193359375, 36.15251922607422, 17.233261108398438, 35.88824462890625, 9.03253173828125, 32.71246337890625, -13.863861083984375, 32.72254180908203, 14.23895263671875, 43.3643798828125, -24.50933837890625, -75.05926513671875, -38.271400451660156, -1.574920654296875, 10.498016357421875, 34.8743896484375, -17.221771240234375, -39.45942687988281, -8.782089233398438, 3.7778396606445312, 54.0728759765625, 15.68682861328125, -18.183624267578125, -7.7620086669921875, 46.523681640625, -3.495330810546875, -40.08668518066406, -13.829803466796875, -66.02034759521484, 6.374908447265625, 5.9950714111328125, 39.8046875, 48.736045837402344, 33.62384033203125, 13.378791809082031, 53.962371826171875, -6.96136474609375, -7.9837646484375, 18.32646942138672, 43.25390625, 43.36932373046875, 14.528289794921875, 88.740478515625, 12.060935974121094, -4.964290618896484, -27.610931396484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000147.npy"}
{"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 12.378179550170898, "std": 27.142812728881836, "min": -58.727447509765625, "p10": -25.831606292724608, "median": 10.733524322509766, "p90": 48.65820465087891, "max": 92.61074829101562, "pos_frac": 0.640625, "sample": [16.135009765625, 13.162155151367188, 17.099853515625, -14.06500244140625, -1.8551082611083984, 69.11135864257812, -0.10093879699707031, 20.45648193359375, 72.2271728515625, 26.401885986328125, 5.952939987182617, 20.597976684570312, -4.8252410888671875, 55.94059753417969, 11.668182373046875, 37.056243896484375, 8.63311767578125, -18.2030029296875, 15.924346923828125, 27.330059051513672, -21.630340576171875, 40.703857421875, -46.11863708496094, -0.094757080078125, 9.165435791015625, -28.86431884765625, 15.19842529296875, -5.494163513183594, 16.061553955078125, -9.237781524658203, 20.430641174316406, 14.627471923828125, -8.670280456542969, -58.727447509765625, -10.122894287109375, -16.74102783203125, 60.250518798828125, 10.416702270507812, -28.18609619140625, 21.269134521484375, 57.027099609375, 1.625772476196289, -17.348846435546875, -33.0665283203125, 1.3639373779296875, 6.615470886230469, 1.7920074462890625, -14.64251708984375, 21.79559326171875, 31.829315185546875, 12.96181869506836, 8.504737854003906, 7.26519775390625, 17.22662353515625, -32.97163391113281, 71.58682250976562, 9.201248168945312, 37.01625061035156, 11.050346374511719, 48.72332763671875, 21.604736328125, -8.998710632324219, -0.7397651672363281, -1.720672607421875, 0.0, 8.27853012084961, 58.397186279296875, 28.529014587402344, 36.90753936767578, 38.5306396484375, -3.5481529235839844, -30.43353271484375, -4.1222076416015625, -36.50189208984375, 41.78376770019531, 8.043037414550781, -29.048614501953125, 31.119354248046875, -1.5837631225585938, 32.8209228515625, -0.3196697235107422, 21.829971313476562, -26.8126220703125, 27.78021240234375, 56.200592041015625, 20.04212760925293, 2.5013046264648438, 15.799407958984375, -9.663520812988281, -15.035064697265625, 28.097640991210938, 18.480846405029297, -0.9558181762695312, 28.5716552734375, 28.608062744140625, 48.28111267089844, -26.888717651367188, 6.577888488769531, -13.085655212402344, 62.884490966796875, 3.5872802734375, 28.509063720703125, 92.61074829101562, 36.362945556640625, 2.6386871337890625, -8.552556991577148, 32.055179595947266, -27.552154541015625, 29.82080078125, 19.86992645263672, 48.63029479980469, 42.43669891357422, 17.119232177734375, 29.77105712890625, -25.411170959472656, -2.097259521484375, -27.024169921875, 16.34783935546875, 62.22967529296875, 0.0, -8.81396484375, 14.113739013671875, 34.81787109375, -0.4763069152832031, 8.285385131835938, 44.51373291015625, 61.88734436035156, -3.9248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000148.npy"}
{"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 15.681751251220703, "std": 26.6094913482666, "min": -42.512664794921875, "p10": -22.546885681152343, "median": 12.680839538574219, "p90": 48.29830627441406, "max": 99.6488037109375, "pos_frac": 0.765625, "sample": [48.142730712890625, -27.66839599609375, 25.317138671875, 33.34032440185547, 44.989898681640625, 47.83807373046875, 22.69232177734375, 5.969890594482422, 7.47100830078125, 6.64501953125, 17.75762939453125, 3.0045166015625, 44.360321044921875, 15.5535888671875, 29.951690673828125, -3.0554046630859375, 44.22344970703125, 10.062374114990234, -2.982391357421875, -8.397796630859375, 16.636795043945312, 5.08807373046875, 20.9630126953125, 4.136783599853516, 29.454410552978516, 13.5264892578125, -41.785423278808594, 39.24957275390625, 48.66131591796875, 17.034622192382812, 7.7182464599609375, 16.196533203125, 35.57781982421875, 12.92578125, 2.205045700073242, 0.6377639770507812, 51.18150329589844, 3.176647186279297, -21.904144287109375, 0.5358123779296875, 35.311126708984375, 63.7386474609375, -11.961433410644531, 1.4480514526367188, 76.21087646484375, 51.77813720703125, 10.539794921875, -24.421066284179688, 39.797698974609375, 51.047454833984375, 20.778060913085938, -4.759674072265625, 7.21038818359375, 25.25336456298828, 4.6631927490234375, 77.7421875, -28.9285888671875, 15.41241455078125, 20.538116455078125, 13.8878173828125, -6.596641540527344, 27.395057678222656, -25.17578125, 39.815185546875, -25.399017333984375, 4.555206298828125, 8.727371215820312, -6.699432373046875, 25.257537841796875, 23.948740005493164, 40.665924072265625, 14.239883422851562, 34.536590576171875, 12.435897827148438, 42.421783447265625, 14.870651245117188, 53.321319580078125, 6.6710052490234375, 56.366485595703125, 0.05841827392578125, 42.554779052734375, 0.6455841064453125, 31.945518493652344, 27.956298828125, 10.38616943359375, -11.2376708984375, 57.63226318359375, 41.21966552734375, -40.15911865234375, 3.9161148071289062, 11.466827392578125, 38.69879150390625, -5.884033203125, -25.7318115234375, -1.5401153564453125, -4.19476318359375, 66.0338134765625, 4.411041259765625, 45.53700256347656, 3.6705322265625, -8.683853149414062, 3.2142333984375, 32.56013488769531, 1.3486480712890625, 28.808807373046875, -1.662078857421875, -28.903289794921875, -24.046615600585938, 2.706451416015625, 99.6488037109375, -3.097412109375, -5.495941162109375, 8.809326171875, 29.684982299804688, 7.09051513671875, -11.923812866210938, 41.302162170410156, 28.088088989257812, -41.23970031738281, -42.512664794921875, 19.913818359375, 36.733642578125, -25.628936767578125, 74.39224243164062, 39.85802459716797, 24.927688598632812, 1.689208984375, 9.219390869140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000149.npy"}
{"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 14.565540313720703, "std": 25.279369354248047, "min": -52.49542236328125, "p10": -16.018336486816406, "median": 13.64509391784668, "p90": 46.490483093261716, "max": 84.18478393554688, "pos_frac": 0.7734375, "sample": [23.2559814453125, 33.94562530517578, -4.238189697265625, 31.3623046875, 9.771728515625, -30.01104736328125, -3.7104568481445312, 16.907958984375, -31.850189208984375, 29.471923828125, 26.162521362304688, 20.829681396484375, 45.26385498046875, 15.7838134765625, -15.891357421875, 3.212646484375, 8.603286743164062, 29.54578399658203, 6.0325927734375, 36.604522705078125, 25.014694213867188, -16.711868286132812, -1.622467041015625, 4.3147735595703125, -5.932401657104492, 17.9857177734375, 8.035675048828125, 3.2763214111328125, 65.66703796386719, 34.14398956298828, 68.30532836914062, 0.111419677734375, 21.96527099609375, 8.080682754516602, 17.142684936523438, 27.011627197265625, 21.819564819335938, -35.807647705078125, 39.021934509277344, 11.112380981445312, 27.0206298828125, 43.0067138671875, 14.438079833984375, 39.25291442871094, -9.388221740722656, 3.0939788818359375, 2.2337875366210938, 3.9586181640625, 13.88006591796875, 61.26312255859375, 20.08248519897461, 76.5709228515625, 1.572113037109375, 32.165802001953125, 47.07208251953125, 63.336029052734375, 54.13011169433594, 6.116218566894531, -0.8967132568359375, 0.43375205993652344, 0.14483642578125, -0.6630973815917969, -33.9444580078125, 47.385292053222656, 5.051971435546875, 0.19830322265625, 6.323516845703125, -1.233642578125, 42.974609375, 41.726654052734375, 21.02532958984375, 37.77978515625, 3.1767120361328125, 46.24122619628906, 21.813995361328125, -8.985595703125, -44.087646484375, 7.70550537109375, 3.112752914428711, 23.411293029785156, 16.531288146972656, -52.49542236328125, -0.36846923828125, -17.753890991210938, -5.065040588378906, 18.6180419921875, -22.6070556640625, 84.18478393554688, 35.77099609375, 15.877777099609375, 14.314056396484375, -10.865631103515625, 7.0076141357421875, 21.90810775756836, -46.4610595703125, 10.882957458496094, 41.15547180175781, -2.5870361328125, -26.273773193359375, 1.4226570129394531, 16.572853088378906, 29.82733917236328, 8.257659912109375, 53.093536376953125, 11.721168518066406, 19.235626220703125, 33.700439453125, 9.5267333984375, 40.630889892578125, 29.775390625, 3.663055419921875, 14.29925537109375, 5.428924560546875, 51.04638671875, 8.1785888671875, 0.7039031982421875, 13.912117004394531, -9.337905883789062, 55.8238525390625, -16.314620971679688, 0.0, 27.348876953125, 76.8970947265625, 0.412353515625, 13.41012191772461, 16.025039672851562, 15.130851745605469, -30.26435089111328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000150.npy"}
{"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 22.27634620666504, "std": 29.07924461364746, "min": -56.14976501464844, "p10": -16.897601318359374, "median": 24.176010131835938, "p90": 58.613424682617186, "max": 84.651611328125, "pos_frac": 0.7578125, "sample": [38.90940856933594, 18.888397216796875, 1.3655128479003906, -16.68657684326172, -0.10239410400390625, 84.094970703125, 51.89302062988281, -27.479949951171875, 0.862457275390625, 41.11370849609375, -2.773712158203125, -4.120262145996094, 21.764007568359375, 36.99737548828125, -7.439979553222656, -17.389991760253906, -2.835601806640625, 40.29447937011719, 36.92718505859375, 22.418556213378906, -56.14976501464844, 2.204559326171875, -13.001953125, 27.261322021484375, 4.2669677734375, 43.193603515625, 8.939315795898438, -3.3985519409179688, 26.455657958984375, 37.28900146484375, 50.063140869140625, 6.4936676025390625, -4.7589874267578125, 22.113616943359375, 39.65559387207031, 8.415084838867188, 20.68402099609375, 19.028884887695312, -6.124217987060547, 79.23458862304688, 27.881072998046875, 32.36357116699219, 13.971343994140625, 17.43658447265625, 42.9901123046875, -21.251327514648438, 24.940505981445312, 7.7135009765625, 46.57489013671875, 44.635284423828125, 43.630455017089844, -7.976806640625, 4.394626617431641, 0.4291229248046875, 39.08135986328125, 49.60722351074219, 17.40740966796875, 23.76446533203125, -27.480148315429688, 22.330650329589844, -4.4478759765625, 84.651611328125, 51.461151123046875, -15.171012878417969, 34.153594970703125, 48.368927001953125, 24.587554931640625, -25.11480712890625, 43.126007080078125, 59.0115966796875, -37.148956298828125, -27.72991943359375, 44.01190185546875, 32.88331604003906, 31.855575561523438, 28.486534118652344, 58.442779541015625, 39.31574630737305, 76.1348876953125, -16.39385986328125, -25.7410888671875, 45.721649169921875, 70.80062866210938, 47.320404052734375, 79.81298828125, 7.6822967529296875, 37.30781555175781, -0.8964443206787109, 0.0, 8.731521606445312, 23.427215576171875, 43.4720458984375, 34.292633056640625, -10.615402221679688, 36.508758544921875, 30.498382568359375, 51.81494140625, 69.15878295898438, -13.986946105957031, 71.12860107421875, -31.602508544921875, 37.78973388671875, 2.309051513671875, 36.15235137939453, 64.2161865234375, 47.54417419433594, 19.406082153320312, -21.026752471923828, 30.317672729492188, 34.172637939453125, 14.559814453125, 67.7574462890625, 18.37615966796875, -19.630844116210938, 74.36642456054688, 18.621978759765625, 52.120941162109375, 31.844100952148438, 67.08169555664062, 25.39764404296875, 42.982147216796875, 12.463790893554688, 11.914764404296875, 32.084228515625, 40.30958938598633, 0.48284912109375, 8.14105224609375, -34.71379089355469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000151.npy"}
{"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 16.921005249023438, "std": 29.282859802246094, "min": -58.996063232421875, "p10": -13.612596130371093, "median": 15.46114730834961, "p90": 55.719931030273436, "max": 98.44195556640625, "pos_frac": 0.703125, "sample": [34.38311767578125, 48.317047119140625, -4.66485595703125, 13.536003112792969, 7.10302734375, 33.03607177734375, 11.5205078125, 34.08857727050781, 58.86314392089844, 37.396820068359375, 21.468292236328125, 37.23358154296875, 69.95700073242188, 4.48284912109375, 41.16363525390625, -10.597137451171875, 2.693695068359375, 3.294036865234375, 1.209991455078125, -24.12188720703125, -55.820281982421875, 28.1739501953125, -58.996063232421875, 20.266571044921875, -11.260719299316406, 42.838653564453125, 15.247505187988281, -0.4408149719238281, 51.87493896484375, -4.28985595703125, -0.9626026153564453, 46.52223205566406, 56.17877197265625, 32.117767333984375, 25.81829833984375, 35.76695251464844, 50.899810791015625, 35.801239013671875, 57.92120361328125, 29.486732482910156, -42.57512283325195, 98.44195556640625, -29.431884765625, 10.606689453125, 34.45244598388672, -1.5702896118164062, -13.106369018554688, 22.024688720703125, -9.417057037353516, 0.5833053588867188, 2.395233154296875, -7.5145721435546875, 17.551422119140625, -7.6937713623046875, -7.485084533691406, 26.8623046875, 17.758575439453125, -25.129547119140625, -8.748970031738281, -2.31439208984375, 62.30963134765625, 15.674789428710938, -17.378738403320312, 19.216171264648438, 68.2823486328125, -24.900634765625, 15.817230224609375, -17.76750946044922, -13.576583862304688, 45.67364501953125, 8.61646842956543, -8.78680419921875, -7.004261016845703, 11.051239013671875, 3.1339263916015625, 64.79971313476562, 40.39064025878906, 22.6453857421875, 77.22450256347656, 31.846214294433594, 55.523284912109375, -24.0372314453125, 2.3730850219726562, -8.6744384765625, 3.098358154296875, 4.219093322753906, 43.623382568359375, 42.14617919921875, 67.75823974609375, 9.548709869384766, 7.4829559326171875, -56.04144287109375, -10.23968505859375, 65.14668273925781, 32.96556091308594, 35.30998992919922, -11.882568359375, 5.196624755859375, 24.231460571289062, -13.696624755859375, 21.20672607421875, 46.152099609375, 29.1300048828125, 21.26397705078125, -12.34429931640625, 57.6895751953125, 4.122852325439453, 9.04791259765625, -8.746337890625, -5.2850189208984375, 42.86659622192383, 7.7364959716796875, -44.11114501953125, 31.128265380859375, -1.167327880859375, 6.3268585205078125, 36.535362243652344, 1.4990005493164062, 49.497039794921875, 62.55645751953125, 47.320404052734375, 51.007476806640625, 46.960655212402344, 32.445838928222656, 6.4958648681640625, 25.16509246826172, -7.6923370361328125, 44.564109802246094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000152.npy"}
{"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 24.299457550048828, "std": 27.60611915588379, "min": -58.799591064453125, "p10": -3.7336742401123044, "median": 20.25487518310547, "p90": 64.1160675048828, "max": 90.13818359375, "pos_frac": 0.8671875, "sample": [15.883148193359375, 0.6078376770019531, 74.1136474609375, 14.875457763671875, 61.508331298828125, 47.456581115722656, 49.91705322265625, 1.4302215576171875, 90.13818359375, 6.442596435546875, 45.725830078125, 4.42047119140625, -18.496490478515625, 22.86590576171875, 35.92896270751953, 5.779396057128906, -10.164321899414062, 2.1211090087890625, 29.8631591796875, 68.75129699707031, 64.16329956054688, 73.80677795410156, 36.18433380126953, 3.6737060546875, 2.694305419921875, -3.6004867553710938, 33.696983337402344, -4.044445037841797, 1.698699951171875, 4.2818603515625, -14.834701538085938, 5.9905853271484375, 48.5445556640625, 4.933502197265625, 1.93280029296875, 14.89910888671875, 53.30322265625, 44.13459777832031, 28.68793296813965, 0.0, 64.0958251953125, 41.33367919921875, 59.45228576660156, 20.853759765625, 45.33929443359375, 43.42901611328125, 50.39056396484375, -2.574493408203125, 19.12469482421875, 16.462936401367188, 15.08441162109375, 52.27496337890625, -58.799591064453125, 5.3514251708984375, -5.8079833984375, 37.2933349609375, 50.526947021484375, 37.30548095703125, -1.3009490966796875, 70.93087768554688, 16.63275146484375, 9.558822631835938, 22.353805541992188, 62.50201416015625, -5.5048065185546875, 51.409515380859375, 8.9730224609375, 10.87957763671875, 19.303802490234375, 20.28369140625, -4.9687347412109375, 18.136871337890625, 3.7309112548828125, 39.97361755371094, 1.2613067626953125, 0.3697662353515625, 8.914276123046875, 31.031982421875, 14.50152587890625, 38.3492431640625, 36.502166748046875, -14.518661499023438, 51.64404296875, 72.53302001953125, 24.879425048828125, 20.226058959960938, 6.8564910888671875, 15.251007080078125, 41.491729736328125, 0.152557373046875, 2.5075836181640625, 23.55279541015625, 66.75985717773438, 46.349761962890625, 55.71504211425781, 10.616130828857422, 6.171014785766602, 20.685028076171875, 4.62188720703125, 8.013824462890625, 47.70440673828125, 45.535858154296875, 40.25526428222656, 8.555946350097656, 3.1843414306640625, 5.7244873046875, 29.866287231445312, 77.43875122070312, 28.959613800048828, 72.14984130859375, 66.07139587402344, 25.76374053955078, 6.803558349609375, 64.70309448242188, -24.610265731811523, 27.867034912109375, 11.075332641601562, 61.76788330078125, 29.1605224609375, 5.493316650390625, -21.05859375, 69.63972473144531, 60.420684814453125, 14.356231689453125, -48.29163360595703, -39.69464111328125, 32.59649658203125, 31.1065673828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000153.npy"}
{"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 17.585636138916016, "std": 32.39811706542969, "min": -78.66000366210938, "p10": -25.08275146484375, "median": 18.27135467529297, "p90": 60.4552001953125, "max": 86.767822265625, "pos_frac": 0.7109375, "sample": [26.215301513671875, 50.540130615234375, -26.052978515625, 63.51348876953125, 34.167572021484375, 7.593055725097656, -4.738752365112305, 59.692718505859375, 7.7778472900390625, 16.57000732421875, -51.30308532714844, 38.711544036865234, 19.694549560546875, 14.241546630859375, 34.455230712890625, -78.66000366210938, 37.423828125, -26.083709716796875, 34.81871032714844, 1.576446533203125, 12.605819702148438, -0.3067474365234375, 8.34033203125, -25.63629150390625, -3.767059326171875, 5.344474792480469, 36.21917724609375, 7.5823974609375, 2.5985107421875, 73.71817016601562, 12.77264404296875, -6.943559646606445, 59.462677001953125, 67.70632934570312, 4.2708892822265625, 50.389122009277344, 31.04742431640625, 69.775634765625, 39.68182373046875, 63.8092041015625, 3.97515869140625, 59.73492431640625, 68.433349609375, 30.373939514160156, -55.56330871582031, 19.131790161132812, 81.26712036132812, -1.1377029418945312, -8.965118408203125, -27.636993408203125, -5.160072326660156, -7.937164306640625, 18.157913208007812, -24.84552001953125, 25.04364013671875, 42.96234130859375, 31.119781494140625, -6.145538330078125, 4.270599365234375, -22.2647705078125, -2.3161277770996094, 13.767578125, 47.810028076171875, 23.93541717529297, 26.604736328125, -52.70648193359375, -19.15070915222168, -1.175018310546875, -2.5245819091796875, -10.857086181640625, -12.630935668945312, -47.86497497558594, 53.93583679199219, 68.44088745117188, -3.149627685546875, -9.419792175292969, 10.59835433959961, -5.98675537109375, 34.328704833984375, 8.307083129882812, -38.56581115722656, 11.832527160644531, -1.0415878295898438, 23.581039428710938, 38.524200439453125, 10.31707763671875, 59.880279541015625, 32.275299072265625, -5.018585205078125, 32.920005798339844, 22.39849853515625, 21.308563232421875, 61.26055908203125, 49.54547882080078, 36.99566650390625, -16.026214599609375, 21.654541015625, 3.9420928955078125, 86.767822265625, 14.416412353515625, 21.555938720703125, 48.4090576171875, -26.35504150390625, -65.40357971191406, 41.585357666015625, 29.96051025390625, 35.77607727050781, 32.08319091796875, 18.384796142578125, 24.346084594726562, 68.0565185546875, 34.302764892578125, 34.02085876464844, 58.20768737792969, 3.70611572265625, 5.250764846801758, 82.01922607421875, 17.603515625, 17.912399291992188, 24.938705444335938, -45.048553466796875, -13.258529663085938, 60.11004638671875, 32.921173095703125, 3.1129837036132812, 22.956787109375, 38.690879821777344, 64.56230163574219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000154.npy"}
{"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 12.913494110107422, "std": 29.611757278442383, "min": -62.4676513671875, "p10": -22.11088981628418, "median": 9.875738143920898, "p90": 50.47807540893553, "max": 85.97900390625, "pos_frac": 0.6640625, "sample": [67.18878173828125, 26.068397521972656, 14.7193603515625, -6.51934814453125, 37.807559967041016, -0.18358421325683594, 8.579421997070312, -1.211456298828125, 8.23834228515625, 49.09931182861328, 44.420413970947266, 3.5228958129882812, 30.40301513671875, -23.520233154296875, 23.577972412109375, 30.88885498046875, -16.435211181640625, 74.2335205078125, 85.97900390625, -7.01806640625, 11.981231689453125, 29.359588623046875, 48.074615478515625, 85.0325927734375, 6.0051116943359375, 42.030517578125, -43.697265625, 0.0, -33.490631103515625, -9.261138916015625, -21.506885528564453, 56.75823974609375, 23.55596923828125, 48.88811492919922, 2.45294189453125, -15.266845703125, -21.287498474121094, 42.4453125, -10.32550048828125, -0.571197509765625, 8.88079833984375, 29.901748657226562, 20.6104736328125, 2.0235977172851562, 38.529510498046875, 57.6212158203125, 31.736297607421875, 14.4627685546875, 4.192314147949219, 36.201438903808594, -9.632865905761719, -53.05908203125, 8.420907974243164, 8.31097412109375, 3.097675323486328, -38.307037353515625, -1.34051513671875, 37.85711669921875, 75.65435791015625, 54.07135009765625, 3.0448989868164062, 4.637931823730469, 20.938873291015625, -9.072044372558594, -44.27729797363281, -9.734804153442383, 47.6806640625, 59.085693359375, 12.669857025146484, 16.912704467773438, -2.77825927734375, -30.550277709960938, 38.79179382324219, 46.75408935546875, -16.082244873046875, 33.465850830078125, 8.5631103515625, -7.3328399658203125, 38.274078369140625, -27.256256103515625, 34.928497314453125, 55.78340148925781, -26.402008056640625, -14.017135620117188, 1.2487335205078125, 8.167015075683594, 40.221435546875, -42.749359130859375, 3.352325439453125, 42.90472412109375, 17.149078369140625, 27.14874267578125, 25.55010986328125, -0.904052734375, -42.428009033203125, 14.71551513671875, 11.126388549804688, 65.11921691894531, 12.847503662109375, -16.91205596923828, -15.070159912109375, -15.391197204589844, -4.106170654296875, 10.870677947998047, 18.013946533203125, 41.25122833251953, 4.9752960205078125, -5.793998718261719, -62.4676513671875, 1.6660537719726562, -43.03515625, 12.811912536621094, 38.46841812133789, -13.075958251953125, 53.6951904296875, 29.3902587890625, 36.207855224609375, 27.420806884765625, 14.602783203125, 62.169464111328125, -20.743499755859375, 0.475677490234375, -0.7995223999023438, 8.132720947265625, 16.2681884765625, 22.907745361328125, 13.938430786132812, -0.691009521484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000155.npy"}
{"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 22.138381958007812, "std": 33.930152893066406, "min": -80.51608276367188, "p10": -23.624906921386717, "median": 26.129791259765625, "p90": 61.70093383789062, "max": 98.53097534179688, "pos_frac": 0.7578125, "sample": [37.394065856933594, 41.82270050048828, 29.379005432128906, 22.14990234375, -0.52777099609375, 4.018272399902344, 47.99713134765625, 46.863616943359375, 31.833702087402344, 30.77058982849121, 45.36701965332031, 26.923599243164062, 47.793243408203125, 33.5025634765625, 3.247222900390625, -22.345413208007812, -7.017616271972656, 35.772003173828125, 10.792282104492188, 42.50396728515625, 3.77239990234375, 46.480743408203125, -13.544357299804688, -28.327552795410156, 32.949951171875, 8.196273803710938, -3.419219970703125, 21.37099838256836, 0.8623428344726562, 46.075439453125, 71.40798950195312, -8.505401611328125, 75.00445556640625, -5.3134613037109375, 17.06740379333496, -24.387252807617188, -39.154083251953125, 8.278961181640625, 51.4317626953125, 3.3768157958984375, 30.537750244140625, 60.97882080078125, 32.3917236328125, 9.111358642578125, 6.8397979736328125, 2.858367919921875, 46.88990783691406, -37.799591064453125, 47.177005767822266, 30.522369384765625, 3.2761001586914062, 37.388885498046875, 16.87481689453125, 26.187103271484375, 84.7769775390625, -28.15484619140625, -39.978759765625, 89.02835083007812, -2.46002197265625, 9.288322448730469, -7.43218994140625, 85.505859375, 34.12261962890625, 51.63420104980469, 29.252960205078125, 59.427947998046875, 53.51312255859375, 15.591712951660156, 61.06597900390625, 6.662559509277344, 17.470947265625, 0.4053955078125, -4.773887634277344, 67.7083740234375, 34.30912780761719, 33.386199951171875, 98.53097534179688, 20.71624755859375, 0.4000415802001953, -3.4149913787841797, 84.91595458984375, 35.27032470703125, 72.84451293945312, 4.0680084228515625, 8.996421813964844, 31.03728485107422, 2.5488319396972656, 29.96221923828125, 32.861534118652344, -32.598907470703125, 46.57696533203125, 52.60426330566406, 39.94683837890625, 47.568603515625, 28.37762451171875, 11.88104248046875, -23.298187255859375, 63.76124572753906, 26.072479248046875, -27.7606201171875, 63.1824951171875, 15.390228271484375, 18.978973388671875, -25.25738525390625, -80.51608276367188, 54.502227783203125, -0.7927188873291016, 87.9677734375, -1.713165283203125, -42.41581726074219, 4.636322021484375, 34.981842041015625, 90.4962158203125, 12.898529052734375, -75.95364379882812, -39.97574996948242, -2.70501708984375, 46.916709899902344, 53.96405029296875, 58.988739013671875, -0.6726341247558594, 28.583938598632812, 43.499908447265625, 60.15826416015625, -19.474945068359375, 59.80242919921875, -7.591728210449219, 8.514907836914062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000156.npy"}
{"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 15.462820053100586, "std": 30.315139770507812, "min": -60.77027893066406, "p10": -22.957958984374997, "median": 12.493057250976562, "p90": 54.135337066650386, "max": 98.59100341796875, "pos_frac": 0.6953125, "sample": [33.588897705078125, -0.6625137329101562, 24.106536865234375, 65.53164672851562, -1.986663818359375, -20.803550720214844, 13.906585693359375, 23.122711181640625, 56.21875, -7.76373291015625, 28.217926025390625, -0.031036376953125, -44.90557861328125, 31.463134765625, 1.1114044189453125, 40.03094482421875, 10.872802734375, -0.7390670776367188, 3.0285720825195312, -13.539321899414062, -23.7989501953125, 71.77569580078125, 31.475143432617188, -11.336166381835938, 48.440181732177734, -0.3996753692626953, -32.83770751953125, -1.865020751953125, 4.3184661865234375, 19.079345703125, 13.102447509765625, -5.484100341796875, 13.951644897460938, -47.53717041015625, 37.77039337158203, 51.33314514160156, 22.906646728515625, 33.45802307128906, -45.851837158203125, -60.77027893066406, 82.005859375, 14.652046203613281, 35.8143310546875, 19.45196533203125, 83.20761108398438, 40.66059875488281, 12.200157165527344, -4.472892761230469, 14.395599365234375, 0.0, 17.426864624023438, 50.50126647949219, 8.774520874023438, -7.07476806640625, 66.69708251953125, 12.6490478515625, 39.87849426269531, -8.6702880859375, -4.625396728515625, 1.4594669342041016, 50.1058349609375, 24.6787109375, 55.37938690185547, 11.814178466796875, 33.227394104003906, -2.8325061798095703, 51.6163330078125, -4.963043212890625, -53.3968505859375, 39.35215759277344, 2.24285888671875, -16.166336059570312, -43.710418701171875, 10.84161376953125, -12.40570068359375, 16.835723876953125, -34.2037353515625, 41.772705078125, 6.437049865722656, 56.233734130859375, 21.36016845703125, 47.361236572265625, 0.6342849731445312, -7.20294189453125, 2.3625335693359375, 12.337066650390625, 7.57196044921875, 14.656051635742188, 26.711811065673828, 0.0, -28.658355712890625, 61.25439453125, 22.191329956054688, 17.023239135742188, 6.151947021484375, -3.7900390625, 28.96837043762207, 56.797401428222656, 97.92868041992188, 53.6021728515625, 10.796417236328125, -22.5975341796875, 34.581817626953125, 6.450162887573242, 5.309326171875, 3.2046966552734375, 34.97480773925781, 34.829044342041016, 25.73699951171875, -9.222381591796875, -31.237152099609375, 46.455413818359375, 4.78363037109375, 46.7435302734375, 31.713333129882812, 35.57861328125, -5.320953369140625, 2.0760059356689453, -26.13427734375, 47.173851013183594, -28.447296142578125, 8.351799011230469, 5.940370559692383, 30.661035537719727, 5.4535064697265625, 18.42919158935547, 98.59100341796875, 60.81536865234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000157.npy"}
{"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 16.806304931640625, "std": 30.43543243408203, "min": -54.71142578125, "p10": -24.256170654296874, "median": 16.921424865722656, "p90": 54.25279998779297, "max": 87.05532836914062, "pos_frac": 0.71875, "sample": [5.39390754699707, -1.665618896484375, 22.0762939453125, 54.79449462890625, -34.147186279296875, 4.5417938232421875, -5.9849090576171875, 27.369110107421875, -10.675491333007812, 55.559539794921875, -18.541595458984375, 1.8896484375, -43.854827880859375, -42.3653564453125, 23.537353515625, 19.657745361328125, 34.7562370300293, 30.737991333007812, 12.100250244140625, 31.267105102539062, 14.456802368164062, 63.24371337890625, 40.48797607421875, 4.238182067871094, 20.28282928466797, 48.438568115234375, 10.411109924316406, -54.71142578125, 38.802459716796875, 17.6337890625, -9.107025146484375, 38.470062255859375, -16.754425048828125, 55.6395263671875, 16.79736328125, 6.8795013427734375, 40.748321533203125, -31.966785430908203, 15.900680541992188, -46.31108093261719, 42.56330871582031, 6.9718017578125, 9.753746032714844, -26.180801391601562, 9.501953125, -3.0447845458984375, -8.037460327148438, 17.190887451171875, 40.36726379394531, 21.52825164794922, 68.7286376953125, -4.065919876098633, 1.4707508087158203, 7.073314666748047, 35.95271301269531, 48.682769775390625, -38.481163024902344, -23.9881591796875, 65.306884765625, 21.063018798828125, 63.790863037109375, 6.936248779296875, -6.4556884765625, 9.464996337890625, 36.4825439453125, 13.904937744140625, -22.3162841796875, 0.0, 32.09149932861328, 15.440338134765625, 70.3770751953125, -6.9991455078125, -1.546600341796875, 50.154296875, 28.8076171875, 53.545562744140625, 0.2574920654296875, -24.898536682128906, -10.089691162109375, -2.090402603149414, 52.35865783691406, 7.553802490234375, -46.47662353515625, 17.045486450195312, 32.362762451171875, 37.67138671875, 40.98614501953125, -0.5631294250488281, 44.435943603515625, 39.456787109375, 47.6778564453125, 27.9678955078125, 14.557197570800781, 19.463226318359375, 69.57666015625, 38.144561767578125, 87.05532836914062, -8.587505340576172, -24.88153076171875, -22.652603149414062, 17.843170166015625, 86.27545166015625, -17.082855224609375, 43.62736511230469, 50.607566833496094, 47.761688232421875, 55.11932373046875, 38.48187255859375, 33.843902587890625, 1.4500312805175781, 17.518310546875, 33.5625, 38.38507843017578, 1.8232269287109375, 30.858291625976562, -14.798019409179688, 54.02064514160156, 3.303668975830078, -35.649810791015625, -7.92236328125, 28.63177490234375, 43.144073486328125, -51.1849365234375, 11.470184326171875, 83.4100341796875, 14.197372436523438, 30.067955017089844, 0.04834747314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000158.npy"}
{"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 13.09582233428955, "std": 35.16963577270508, "min": -90.01708984375, "p10": -32.69352722167969, "median": 11.028488159179688, "p90": 60.888293457031246, "max": 99.9844970703125, "pos_frac": 0.6328125, "sample": [-18.55853271484375, 58.47871398925781, 27.2796630859375, 29.5113525390625, -2.3600387573242188, -17.09063720703125, -34.91876220703125, 44.3123779296875, -4.133941650390625, 61.74853515625, -55.54664611816406, 3.2657012939453125, -90.01708984375, -21.473411560058594, 3.4994354248046875, 42.352577209472656, 12.490699768066406, -5.1566925048828125, -33.671966552734375, 46.51524353027344, 19.793914794921875, 4.22698974609375, -13.892471313476562, -28.7171630859375, 21.23223876953125, 59.733551025390625, 19.386795043945312, 60.710296630859375, 32.868194580078125, -8.761764526367188, 20.85642433166504, 2.22821044921875, 34.375, 15.220245361328125, 22.391441345214844, -20.581512451171875, 4.746612548828125, -27.894287109375, -4.243404388427734, -33.979248046875, 48.79841613769531, 15.865463256835938, 22.750473022460938, 41.502342224121094, -7.805938720703125, 19.74505615234375, -31.817184448242188, 3.551799774169922, -46.05999755859375, -10.191360473632812, 1.3656768798828125, -7.7905426025390625, 52.8360595703125, 11.080657958984375, 17.73565673828125, 51.891448974609375, 1.984161376953125, 50.27159118652344, -17.735580444335938, -2.425933837890625, 99.9844970703125, 11.098342895507812, -34.602394104003906, 10.976318359375, 81.44610595703125, 13.441070556640625, 52.7288818359375, 0.27978515625, -9.470733642578125, 12.686065673828125, 0.0586090087890625, 35.52197265625, -6.985103607177734, 49.85900115966797, 39.66656494140625, -32.32891845703125, 3.3080902099609375, 64.07546997070312, -17.935577392578125, 16.7918701171875, -20.007232666015625, -36.9375, 38.370384216308594, 74.90573120117188, 81.062255859375, -9.344520568847656, -10.026321411132812, -33.544281005859375, -9.732513427734375, 64.16586303710938, 26.285545349121094, 17.172958374023438, -19.73175048828125, 32.502220153808594, 12.677871704101562, 31.18865966796875, -34.711700439453125, 63.42236328125, -19.98412322998047, 40.69980239868164, 33.82014465332031, -29.211557388305664, -44.14015197753906, 6.1313934326171875, 87.666015625, 80.49810791015625, 23.208953857421875, -56.61332702636719, 51.89747619628906, 9.91763687133789, -2.8826141357421875, -15.2100830078125, 61.303619384765625, -45.60516357421875, 78.2518310546875, -6.561126708984375, 52.3515625, 38.936431884765625, 56.71807861328125, 39.98358154296875, 9.66217041015625, 9.923534393310547, -4.5015869140625, 71.301025390625, 1.0989990234375, 12.457168579101562, -10.18585205078125, 45.23641586303711], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000159.npy"}
{"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 20.541229248046875, "std": 31.246671676635742, "min": -48.471343994140625, "p10": -14.07395324707031, "median": 17.578895568847656, "p90": 61.5118293762207, "max": 110.64910888671875, "pos_frac": 0.75, "sample": [-5.6396942138671875, 47.08866882324219, 0.0, 89.32040405273438, 24.283340454101562, 24.70245361328125, 28.643829345703125, 13.401123046875, 85.07058715820312, -1.74017333984375, 56.88946533203125, 69.75942993164062, 26.267377853393555, 75.48565673828125, 15.627548217773438, -3.090972900390625, 75.29254150390625, 26.01019287109375, 8.719017028808594, 19.326797485351562, 17.698379516601562, -0.18603515625, 36.51530456542969, 34.42787170410156, 0.8744049072265625, -6.452178955078125, -8.641128540039062, 81.58406066894531, 1.2700862884521484, -46.322357177734375, 11.470733642578125, 109.14190673828125, -42.20490264892578, 61.848899841308594, 19.84618377685547, 2.3711318969726562, 13.969123840332031, 3.6044445037841797, 36.20831298828125, 5.15373420715332, 36.330162048339844, 53.826171875, 10.3812255859375, 0.0, 34.7381591796875, -13.71478271484375, 32.247901916503906, -24.365081787109375, -17.573471069335938, 16.688201904296875, 27.07806396484375, 36.39878845214844, 110.64910888671875, 71.00277709960938, -27.522293090820312, 20.121383666992188, 27.0487060546875, -14.912017822265625, 9.552474975585938, -13.582061767578125, -11.61328125, 54.82878112792969, 8.9251708984375, -44.25128173828125, -10.7425537109375, 59.376983642578125, 61.36737060546875, 57.41737365722656, 5.101049423217773, 17.45941162109375, 39.373260498046875, 2.091522216796875, -0.2932300567626953, 9.309173583984375, 35.53521728515625, 28.48828125, 66.52587890625, 51.47309112548828, 36.081146240234375, 18.200439453125, -26.345962524414062, -6.0309600830078125, 43.29957580566406, -31.669151306152344, -11.374542236328125, 57.43940734863281, 30.201560974121094, 19.28497314453125, 44.387603759765625, -16.0306396484375, 9.192581176757812, 1.7646636962890625, 3.9871978759765625, -29.93359375, 35.129364013671875, -48.471343994140625, -0.24896240234375, 30.617828369140625, 62.30743408203125, 0.9553985595703125, 13.914798736572266, 7.42547607421875, -13.417898178100586, 22.554580688476562, 19.67303466796875, 6.022834777832031, 6.275783538818359, -8.717880249023438, 37.44189453125, 36.93061828613281, 59.803070068359375, 2.6216583251953125, 12.7088623046875, 46.38072204589844, -6.40863037109375, 26.338531494140625, 55.94439697265625, 11.39556884765625, 41.01960754394531, 47.53828430175781, 62.1849365234375, 31.462554931640625, -31.96697998046875, 36.0513916015625, 8.166374206542969, 52.842132568359375, 5.4590911865234375, 4.527458190917969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000160.npy"}
{"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 21.31352996826172, "std": 33.697593688964844, "min": -71.36550903320312, "p10": -15.629969787597656, "median": 15.858104705810547, "p90": 72.06614990234375, "max": 93.728271484375, "pos_frac": 0.703125, "sample": [-3.8001556396484375, 1.2311859130859375, -2.305938720703125, -69.26800537109375, 1.786886215209961, 19.239898681640625, 2.98236083984375, 6.923553466796875, 7.8953857421875, -0.17169189453125, 3.925037384033203, 80.17892456054688, 30.975677490234375, -27.679107666015625, 4.7661895751953125, 2.4994850158691406, 78.41317749023438, 16.518218994140625, 30.439411163330078, -6.295623779296875, 45.644256591796875, 13.820968627929688, 59.98872375488281, 79.45001220703125, -14.675701141357422, 3.78997802734375, 6.358375549316406, 78.38369750976562, 6.84912109375, 31.746170043945312, 8.123443603515625, 21.319921493530273, -6.91424560546875, -15.801498413085938, 21.460098266601562, 23.750640869140625, 32.4134521484375, 51.604705810546875, -1.3452186584472656, 13.104576110839844, -3.4473876953125, -45.03462219238281, -30.2181396484375, -15.017097473144531, 15.839256286621094, 46.2686767578125, -14.98529052734375, -5.48944091796875, 0.0, 41.70500946044922, 15.876953125, 36.5914306640625, 10.129470825195312, 0.6582126617431641, 11.816162109375, -4.27203369140625, 54.929595947265625, 57.23804473876953, 42.25940704345703, 24.49616241455078, 57.253875732421875, 67.51547241210938, -19.68157958984375, 19.15362548828125, -10.953033447265625, 69.12509155273438, 56.0374755859375, 77.86959838867188, -17.951446533203125, 15.072433471679688, 77.81097412109375, 27.716510772705078, 39.42547607421875, 72.60281372070312, -30.896270751953125, 73.42941284179688, 40.293739318847656, -1.7582321166992188, 1.2779541015625, 41.37489318847656, -18.12334442138672, -0.7266082763671875, 93.728271484375, 72.29049682617188, 32.4630126953125, 75.57781982421875, 2.6866607666015625, 55.413909912109375, 86.18087768554688, -1.8025703430175781, -5.094451904296875, 38.958587646484375, -17.279449462890625, 39.6917724609375, -4.130126953125, 30.60015869140625, 69.18414306640625, -5.235176086425781, 38.8321533203125, -9.49639892578125, 21.84093475341797, 62.47715759277344, 23.186737060546875, -15.50244140625, 9.193000793457031, 44.96272277832031, -15.55645751953125, 15.49090576171875, 40.820823669433594, 53.35845947265625, 61.596893310546875, 67.698486328125, -19.09856414794922, 37.760345458984375, 28.847883224487305, -53.54168701171875, 71.97000122070312, 9.122360229492188, 11.503135681152344, 0.0, -4.745880126953125, 72.72702026367188, 62.70768737792969, -71.36550903320312, 31.12890625, 22.323898315429688, 1.0689239501953125, 53.046630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000161.npy"}
{"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 19.47369384765625, "std": 32.418212890625, "min": -78.3275146484375, "p10": -22.33901214599609, "median": 18.33917236328125, "p90": 62.17110900878906, "max": 90.65530395507812, "pos_frac": 0.7421875, "sample": [-9.361083984375, 6.611030578613281, 66.89013671875, 43.38346862792969, 4.606681823730469, 16.815216064453125, -27.897964477539062, 22.24298095703125, 5.4362945556640625, 29.613052368164062, 55.284812927246094, 77.1854248046875, 22.99053955078125, 26.84313201904297, -4.849365234375, -0.950775146484375, 25.11163330078125, 29.346221923828125, 1.4454498291015625, 3.9269332885742188, -32.32489013671875, 36.356658935546875, -35.74437713623047, 29.73023223876953, 16.586883544921875, 35.003150939941406, 60.3338623046875, 11.02801513671875, -42.829742431640625, 6.35089111328125, -0.7694320678710938, 19.21197509765625, 47.45263671875, -33.52351379394531, 71.50241088867188, 13.254684448242188, 18.056640625, -63.051116943359375, -8.316841125488281, 40.45100402832031, 3.9776573181152344, 54.969635009765625, 2.7830562591552734, 66.9246826171875, 65.67547607421875, -78.3275146484375, 61.84130859375, 27.47960662841797, 28.63433837890625, 64.9273681640625, 4.320621490478516, 37.753387451171875, 45.65484619140625, 16.85723876953125, 29.4000244140625, -4.357723236083984, -19.63971710205078, 46.01469421386719, 74.0052490234375, 28.967742919921875, 65.16900634765625, 90.65530395507812, -0.522705078125, -0.338043212890625, 43.38261413574219, 58.673065185546875, 59.96720886230469, 70.98749542236328, -20.663330078125, 51.82286071777344, -0.4327831268310547, 9.021915435791016, -27.889892578125, 19.398193359375, 90.18524169921875, -30.470260620117188, -29.247756958007812, 3.225555419921875, 14.636154174804688, 53.1318359375, -5.940095901489258, 29.4556884765625, 45.777687072753906, 21.695770263671875, 62.940643310546875, 61.542022705078125, 7.187902450561523, -4.35028076171875, 29.951278686523438, 27.926849365234375, 11.855560302734375, 9.789306640625, 7.15995979309082, -59.90870666503906, -1.25872802734375, 24.896240234375, 60.487449645996094, 37.48968505859375, 18.6217041015625, 61.8302001953125, 44.68226623535156, -42.57257080078125, 50.881500244140625, 9.191848754882812, 2.2469482421875, 60.2738037109375, -10.099960327148438, 1.251495361328125, -4.195087432861328, 50.04570007324219, -25.461380004882812, 5.149391174316406, -9.396881103515625, 42.036773681640625, 29.51116943359375, 37.299591064453125, 11.810569763183594, -2.1508522033691406, 26.36090087890625, 20.329193115234375, 6.707731246948242, -21.0008544921875, 1.8487396240234375, -1.160247802734375, 5.01275634765625, 63.07273864746094, 27.2530517578125, 4.5676422119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000162.npy"}
{"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 23.355548858642578, "std": 28.831661224365234, "min": -50.5469970703125, "p10": -13.194119262695311, "median": 21.375022888183594, "p90": 63.04934997558593, "max": 89.10504150390625, "pos_frac": 0.7734375, "sample": [22.93145751953125, -2.4581756591796875, 41.61902618408203, -13.653045654296875, 16.70819091796875, -14.498695373535156, 25.045867919921875, 18.115158081054688, 35.74131774902344, 16.420867919921875, 6.0449981689453125, 65.60478973388672, 44.93658447265625, 14.0880126953125, 75.36166381835938, 24.203445434570312, 12.420852661132812, -6.24908447265625, 8.369522094726562, -7.198020935058594, -15.269256591796875, 38.29122543334961, 25.82611083984375, 0.23980712890625, 16.52880859375, 52.68540954589844, -7.745935440063477, 15.52581787109375, 50.64056396484375, -2.801910400390625, -6.5428619384765625, 62.67071533203125, -2.6346435546875, 45.898284912109375, 28.317626953125, 27.402099609375, 11.400238037109375, 49.651947021484375, -7.35772705078125, -6.7928619384765625, 47.725830078125, 32.796295166015625, -18.760787963867188, 18.662757873535156, 4.9152984619140625, 21.322219848632812, -50.5469970703125, 63.932830810546875, 47.567665100097656, -32.383270263671875, 37.581459045410156, 8.1854248046875, 41.6650390625, 42.462158203125, -1.7945709228515625, 61.11961364746094, 19.17333984375, 65.77044677734375, 74.02728271484375, 42.39453125, 3.725372314453125, -2.05401611328125, -11.4625244140625, 26.542808532714844, -9.9027099609375, -12.9974365234375, 7.589282989501953, 51.776092529296875, 7.119476318359375, 42.12892150878906, -41.77008056640625, 64.22662353515625, 20.25847625732422, 28.728302001953125, 13.89483642578125, -17.634918212890625, 75.68096923828125, -16.74017333984375, 68.06057739257812, 58.65625, 35.967857360839844, 46.470733642578125, 22.80084228515625, 19.97146987915039, 44.821327209472656, 41.367454528808594, 35.239227294921875, 1.542999267578125, 64.98858642578125, 50.695068359375, 8.832269668579102, 7.653778076171875, 22.290916442871094, 89.10504150390625, 39.468597412109375, -25.69061279296875, -29.606346130371094, 18.71630096435547, 79.35772705078125, 34.211395263671875, 56.92015075683594, 51.4952392578125, -18.715713500976562, 65.8558349609375, -6.3390350341796875, 18.991851806640625, 59.9547119140625, 13.400146484375, 29.83575439453125, 32.05280303955078, 60.18011474609375, 39.02227783203125, 18.80841064453125, 0.9322910308837891, 9.178314208984375, 21.427825927734375, 36.46905517578125, 59.08094787597656, 29.558380126953125, 66.46613311767578, -3.398427963256836, 7.452545166015625, -43.89532470703125, 57.30729675292969, 47.09141540527344, 20.03264617919922, 9.571601867675781, 1.437530517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000163.npy"}
{"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 19.673667907714844, "std": 31.68404769897461, "min": -57.60491943359375, "p10": -15.122299194335938, "median": 15.740829467773438, "p90": 62.43864135742187, "max": 99.1290283203125, "pos_frac": 0.734375, "sample": [-15.4896240234375, 16.104263305664062, -2.537881851196289, 55.161865234375, 0.6174049377441406, -1.25445556640625, -6.224517822265625, -6.6513824462890625, 35.949737548828125, 31.3189697265625, 39.642059326171875, -13.769088745117188, 44.103424072265625, 35.748931884765625, 26.8387451171875, -57.60491943359375, -1.861083984375, 56.087493896484375, 10.602619171142578, 68.58282470703125, -0.697479248046875, -47.1336669921875, 49.88470458984375, 15.6080322265625, 12.253677368164062, -20.208786010742188, 29.215484619140625, 3.8460235595703125, -20.52191162109375, 5.1233978271484375, 35.993927001953125, -1.104095458984375, 3.61090087890625, 61.95347595214844, -5.7955474853515625, 51.779327392578125, -42.23719787597656, 3.8826751708984375, 1.8237075805664062, -44.49064636230469, -6.1126556396484375, 72.4952392578125, 43.37432861328125, 19.533935546875, 75.49923706054688, 8.301773071289062, 55.14996337890625, 2.36883544921875, 0.691131591796875, 4.11834716796875, -27.94849395751953, 11.381736755371094, 18.773284912109375, 37.908355712890625, -14.964874267578125, 64.9822998046875, -1.1794281005859375, 43.54948425292969, 33.91087341308594, 35.545013427734375, -0.566986083984375, 8.13397216796875, 64.23992919921875, 63.57069396972656, -13.615066528320312, 65.88031005859375, 27.22515106201172, -2.7587890625, 47.0655517578125, 15.873626708984375, 83.39581298828125, 31.97087860107422, 69.56246948242188, 3.5970458984375, -8.149850845336914, 1.0080337524414062, 24.27375030517578, 5.335580825805664, -48.107025146484375, 15.157402038574219, 30.72412109375, 56.407989501953125, -16.256175994873047, 3.7856979370117188, 0.43072509765625, 42.8145751953125, 56.13470458984375, 55.701416015625, 41.58038330078125, 27.855438232421875, -7.154209136962891, 48.689117431640625, 0.432891845703125, 45.529937744140625, 10.483154296875, 56.082183837890625, 50.57178497314453, 65.52362060546875, -9.551544189453125, 1.1171493530273438, -38.955810546875, 28.91473388671875, 1.1252365112304688, 30.216087341308594, 13.976303100585938, -7.542999267578125, 57.870697021484375, -12.5528564453125, 49.22101593017578, 7.9882659912109375, 20.15032958984375, 50.492431640625, 13.476470947265625, 18.16949462890625, -53.78460693359375, -23.185272216796875, 8.310981750488281, -8.792938232421875, 99.1290283203125, 20.0247802734375, 21.289722442626953, 47.501068115234375, 21.444061279296875, 31.9307861328125, 46.4761962890625, 64.40444946289062, 98.20388793945312, 13.202856063842773], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000164.npy"}
{"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 17.21072006225586, "std": 30.182281494140625, "min": -63.18553161621094, "p10": -16.573405456542968, "median": 10.128440856933594, "p90": 56.454553222656244, "max": 92.97146606445312, "pos_frac": 0.6640625, "sample": [6.1056060791015625, 55.8831787109375, -6.277565002441406, 50.0264892578125, -13.41387939453125, 4.332664489746094, 32.44154357910156, 4.435699462890625, 56.9833984375, 45.938873291015625, 66.289306640625, 4.155399322509766, -61.065673828125, 50.30134582519531, 62.103851318359375, 14.0731201171875, -0.16790771484375, 68.42877197265625, -6.0460052490234375, 46.98004150390625, 43.13130187988281, -47.417327880859375, 55.587860107421875, 38.61393737792969, 7.4338226318359375, -26.690845489501953, -23.7510986328125, -12.138397216796875, 17.486114501953125, 35.814056396484375, 6.59637451171875, -8.14141845703125, -2.0931396484375, 2.6985931396484375, 0.9331893920898438, 13.880859375, 30.018951416015625, -4.5933990478515625, -6.300989151000977, -10.9107666015625, 15.85595703125, 13.229133605957031, 56.227386474609375, -19.255966186523438, 15.2186279296875, 55.90167236328125, -1.2681808471679688, -0.42635345458984375, -15.118576049804688, 12.269058227539062, 87.83587646484375, 9.768325805664062, 4.978660583496094, 0.7203521728515625, 63.064605712890625, 32.275146484375, -7.302276611328125, 4.916563034057617, -5.051944732666016, 7.644927978515625, -3.484771728515625, 67.1651611328125, -11.374008178710938, 44.72412109375, 28.694610595703125, -16.720672607421875, 25.716201782226562, -5.1949920654296875, -5.76605224609375, -6.361907958984375, 2.0732421875, -1.9639511108398438, 24.6278076171875, 14.371543884277344, 31.757049560546875, 4.586963653564453, 39.649932861328125, -25.50762939453125, 45.756561279296875, 14.6451416015625, -29.5631103515625, 33.30906677246094, 68.50875854492188, -16.69921875, 7.767274856567383, 42.260162353515625, -0.175689697265625, 64.84417724609375, 56.2279052734375, 19.79833984375, 14.933197021484375, -0.808441162109375, 41.8236083984375, 44.52389144897461, 55.25947570800781, -19.554473876953125, -16.519485473632812, 24.028213500976562, 47.201271057128906, 5.95758056640625, 92.97146606445312, -2.7097015380859375, 39.21885681152344, 42.33146667480469, 61.27610778808594, -24.182212829589844, -4.861118316650391, 15.743942260742188, 55.73481750488281, 10.488555908203125, 0.6733264923095703, 4.692909240722656, 69.91726684570312, 32.7650146484375, 53.1102294921875, 49.891387939453125, 6.444488525390625, 15.678237915039062, 7.7318878173828125, 49.561241149902344, -1.6656036376953125, 48.29779052734375, -63.18553161621094, -30.4547119140625, -2.6039505004882812, 58.38410949707031, -8.873062133789062, -7.070648193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000165.npy"}
{"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 15.431792259216309, "std": 36.47883605957031, "min": -79.913818359375, "p10": -27.597836303710935, "median": 12.410015106201172, "p90": 63.77384033203125, "max": 107.5062255859375, "pos_frac": 0.6640625, "sample": [15.0960693359375, 5.359588623046875, -28.666015625, 15.463455200195312, -2.657470703125, 10.498558044433594, 64.28619384765625, 60.9935302734375, -3.3468265533447266, 3.0889110565185547, 10.733863830566406, -0.8913650512695312, 107.5062255859375, -15.224899291992188, 3.423675537109375, -7.70599365234375, 24.162826538085938, 27.01019287109375, 12.436759948730469, 1.4780960083007812, -1.5675468444824219, -31.606292724609375, 25.693740844726562, 68.65425109863281, -60.683502197265625, -27.140045166015625, 1.5394649505615234, 32.93463134765625, -56.81510925292969, -17.494659423828125, -8.388031005859375, 19.92828369140625, 29.075515747070312, 77.3363037109375, 62.80274963378906, 79.22412109375, 3.6446266174316406, -11.069961547851562, -17.425094604492188, -76.815185546875, 6.351047515869141, -29.5203857421875, 5.497688293457031, -2.815521240234375, -11.259757995605469, 71.83810424804688, 29.770198822021484, 19.80323028564453, -16.872344970703125, 34.3037109375, -3.2911376953125, 91.40765380859375, -50.467987060546875, 71.04486083984375, 95.10879516601562, -53.51472473144531, -36.487518310546875, 94.7122802734375, 36.299896240234375, 44.398406982421875, -31.821914672851562, -11.04803466796875, -17.84991455078125, 37.544097900390625, 51.38665771484375, -24.97784423828125, 54.95574951171875, 37.7698974609375, 6.529449462890625, -23.7877197265625, 20.696060180664062, 26.93701171875, 76.64016723632812, -8.20302963256836, -15.507080078125, 20.600341796875, 15.133651733398438, -0.6830615997314453, 13.949569702148438, 12.71807861328125, -4.759326934814453, 50.399169921875, 18.73676300048828, 8.740570068359375, 12.383270263671875, 76.44729614257812, 2.05792236328125, 41.80302429199219, 18.3372802734375, -40.65708923339844, 7.02117919921875, -10.935127258300781, 13.145744323730469, 56.907470703125, 51.559112548828125, 28.766464233398438, 19.995513916015625, 54.558616638183594, 29.68500518798828, 7.443603515625, 23.209243774414062, 0.7040081024169922, -0.088409423828125, 48.35456848144531, 1.8886680603027344, 32.93495178222656, 53.27714538574219, 49.17652893066406, -10.410202026367188, 62.89344787597656, -45.93328857421875, -79.913818359375, 12.8253173828125, 38.67762756347656, 58.78900146484375, -18.858474731445312, 14.084320068359375, -2.118408203125, 20.798736572265625, -0.9003143310546875, 63.55426025390625, -1.2568130493164062, 105.25421142578125, 13.48600959777832, 3.41070556640625, 35.59300231933594, 10.72589111328125, 11.31280517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000166.npy"}
{"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 18.40631675720215, "std": 33.43583297729492, "min": -65.20048522949219, "p10": -17.84981994628906, "median": 13.324935913085938, "p90": 64.2648910522461, "max": 102.1392822265625, "pos_frac": 0.703125, "sample": [46.24317169189453, 6.090873718261719, 14.239837646484375, 47.136871337890625, 16.832000732421875, 53.30523681640625, 48.71075439453125, 75.95468139648438, 35.2503662109375, 15.879684448242188, 60.04292297363281, -52.956634521484375, 66.32623291015625, 0.5566787719726562, 11.699356079101562, 79.4764404296875, 14.855255126953125, 72.95260620117188, -2.35986328125, 11.727447509765625, 3.5360565185546875, 31.205352783203125, -56.594573974609375, -65.20048522949219, -2.5426406860351562, 0.728851318359375, -26.846908569335938, 53.32325744628906, 31.242324829101562, 58.23384094238281, 32.448883056640625, 64.13661193847656, 18.45299530029297, 64.564208984375, 44.60563659667969, 7.265533447265625, 17.1639404296875, -33.235687255859375, 20.193206787109375, -9.005271911621094, 69.84176635742188, -4.1550445556640625, 14.800537109375, 2.3426513671875, -0.74755859375, -62.271575927734375, 0.0, 0.7861328125, -0.892333984375, 34.58134460449219, 12.4100341796875, 75.65484619140625, -32.70293426513672, 65.14743041992188, -11.076736450195312, 15.32177734375, 46.943443298339844, 10.103485107421875, 34.24481201171875, 0.322357177734375, -32.07879638671875, 3.70196533203125, 87.27322387695312, 0.20255088806152344, 27.87371826171875, 12.37811279296875, 17.311004638671875, -15.469146728515625, 37.8448486328125, 3.060068130493164, -1.6360321044921875, 1.7233657836914062, 39.88567352294922, -37.37617492675781, 49.871124267578125, 46.65156555175781, 60.59027099609375, -1.688323974609375, 27.45428466796875, 3.6319427490234375, -2.24957275390625, 48.84059143066406, 42.88031005859375, -25.743728637695312, 59.31170654296875, -6.008964538574219, 19.5064697265625, -5.529788970947266, 36.756103515625, 41.006370544433594, 37.90277099609375, 9.931816101074219, 38.756568908691406, 16.8802490234375, 39.18669128417969, 83.7877197265625, -5.356956481933594, -17.0164794921875, -19.794281005859375, -8.197677612304688, -4.52691650390625, 5.5213775634765625, -11.0234375, 70.443603515625, -8.567657470703125, 10.26861572265625, 102.1392822265625, 22.326507568359375, 46.37811279296875, -13.331100463867188, 2.8902587890625, 44.9375, -33.55792236328125, -3.3386688232421875, -3.2135467529296875, 2.769287109375, 46.376708984375, 2.764801025390625, -2.66229248046875, 51.95457458496094, 44.39414978027344, 32.571807861328125, 10.041244506835938, -2.4359130859375, -58.85333251953125, 54.769287109375, 4.037353515625, 80.58610534667969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000167.npy"}
{"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 23.761913299560547, "std": 33.5301399230957, "min": -69.72003173828125, "p10": -11.989191055297852, "median": 19.78290557861328, "p90": 66.74017944335938, "max": 98.726318359375, "pos_frac": 0.75, "sample": [29.320663452148438, 17.51873779296875, 78.52548217773438, 64.9136962890625, 43.76947021484375, 22.587915420532227, 46.06364440917969, 20.285072326660156, 49.2086181640625, -7.653800964355469, 55.71942138671875, -11.796066284179688, 6.81890869140625, 8.18939208984375, 0.5385055541992188, 2.803741455078125, 23.512802124023438, 31.952392578125, -5.832561492919922, 54.3302001953125, 17.37065887451172, 62.75286865234375, 19.446792602539062, 22.0556640625, 63.2822265625, -1.8204479217529297, 20.1190185546875, 47.87990951538086, -2.0607681274414062, 0.30358123779296875, 0.0, 14.553329467773438, 77.800537109375, 16.740249633789062, 66.60455322265625, 37.18170166015625, 18.344573974609375, 4.85107421875, 32.236785888671875, 2.4659423828125, 67.1849365234375, 17.3037109375, -1.381072998046875, 43.314605712890625, -5.9972381591796875, 97.7799072265625, 19.322891235351562, 35.17413330078125, 82.33734130859375, -14.257904052734375, 72.40771484375, -3.651702880859375, 35.95906448364258, 15.107666015625, -46.70733642578125, 36.942665100097656, 51.499237060546875, 51.247283935546875, 50.24040985107422, 7.4404296875, -25.05767822265625, -23.510009765625, 1.8924560546875, -1.2947540283203125, 46.645172119140625, -1.4374752044677734, 59.769775390625, 56.670745849609375, 98.726318359375, 26.670623779296875, 78.77313232421875, 7.922939300537109, -69.72003173828125, 44.43292236328125, 66.52078247070312, 0.6085834503173828, 3.9119949340820312, 52.34056091308594, 5.5933380126953125, -4.3853759765625, -5.855560302734375, 6.853240966796875, 39.540924072265625, 21.0069580078125, 65.92147827148438, 56.055572509765625, 67.056640625, 41.444000244140625, 81.48455810546875, 56.4097900390625, -19.678077697753906, -8.795074462890625, 61.68255615234375, -12.80621337890625, 7.6641693115234375, 44.9222412109375, -12.439815521240234, 17.947235107421875, 33.882415771484375, -58.63508605957031, 15.023345947265625, 10.355789184570312, 54.85505294799805, 32.561767578125, 28.0135498046875, 77.85308837890625, -3.271453857421875, -3.611480712890625, 4.684101104736328, -1.1554107666015625, 7.225677490234375, 0.352569580078125, -30.36883544921875, -36.981109619140625, 55.721397399902344, 23.419647216796875, 46.7628173828125, 68.43447875976562, 65.32705688476562, 46.99530029296875, -0.231292724609375, 34.7967529296875, 7.993408203125, -9.5675048828125, -37.35491943359375, 11.79119873046875, 67.44284057617188, -66.42990112304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000168.npy"}
{"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 27.492664337158203, "std": 35.23688507080078, "min": -58.06573486328125, "p10": -7.9344558715820295, "median": 18.247337341308594, "p90": 75.69701232910155, "max": 114.48541259765625, "pos_frac": 0.7734375, "sample": [-2.2518234252929688, 47.81193542480469, 89.2061767578125, 40.22991943359375, -20.968048095703125, 13.08203125, 1.920745849609375, 6.61004638671875, 1.377716064453125, 16.7518310546875, 24.540176391601562, 59.63017272949219, 57.78903579711914, 60.2100830078125, -48.04046630859375, -2.7503738403320312, 56.024383544921875, 11.08013916015625, 5.480501174926758, -2.070526123046875, -2.83514404296875, 11.973350524902344, 52.91118621826172, 15.155235290527344, 54.83392333984375, 66.90289306640625, 28.5709228515625, -15.73065185546875, 40.764984130859375, 51.4349365234375, 62.376434326171875, 67.63385009765625, 56.4322509765625, -7.491729736328125, 2.9215164184570312, 8.636024475097656, -14.442873001098633, -58.06573486328125, 30.35638427734375, -3.1107177734375, -9.968437194824219, 67.5657958984375, -13.20977783203125, 15.775238037109375, 0.0, -45.66187286376953, 9.398801803588867, 33.326812744140625, 20.353729248046875, 20.969223022460938, -0.669281005859375, 9.137176513671875, 114.48541259765625, 63.214996337890625, 42.58127975463867, 54.127655029296875, -52.087989807128906, 53.958892822265625, 0.7768726348876953, -0.3004913330078125, 77.959716796875, 50.61297607421875, 31.141693115234375, -8.967483520507812, 41.39898681640625, 76.67730712890625, 63.45806884765625, 90.2188720703125, 17.633209228515625, 70.73440551757812, 3.3015899658203125, 92.3260498046875, 7.999977111816406, 1.2860107421875, 41.026702880859375, -35.58885192871094, 15.07977294921875, 79.3883056640625, 49.42498779296875, -5.048370361328125, 51.422515869140625, 8.327766418457031, 13.975942611694336, 51.09808349609375, 24.603561401367188, 57.237396240234375, 68.72994995117188, -6.9788055419921875, 43.393280029296875, 31.816879272460938, -0.39347076416015625, 36.35015869140625, 6.163795471191406, 102.87060546875, 34.082733154296875, 10.815845489501953, 13.573516845703125, 6.126556396484375, 84.53750610351562, 42.99192810058594, 6.829538345336914, 52.530189514160156, 94.70330810546875, 17.8720703125, -32.585479736328125, -29.56048583984375, 7.1540679931640625, -1.8336944580078125, 97.302490234375, 18.622604370117188, 48.63299560546875, 0.10901641845703125, 70.912353515625, 55.033294677734375, 55.189605712890625, 0.0, 15.2010498046875, 3.1998443603515625, 20.39361572265625, 40.59722137451172, 14.353240966796875, -0.46813011169433594, 12.3028564453125, 13.685062408447266, 91.13858032226562, 89.28289794921875, 75.27688598632812, -6.288429260253906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000169.npy"}
{"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 24.13933563232422, "std": 33.03411865234375, "min": -57.948333740234375, "p10": -9.672738647460934, "median": 21.17852210998535, "p90": 67.49268188476562, "max": 117.65155029296875, "pos_frac": 0.7578125, "sample": [4.1436614990234375, 15.699249267578125, 52.35997009277344, -7.16595458984375, 49.602630615234375, 10.03277587890625, -0.8358688354492188, 106.32586669921875, 7.933021545410156, 11.16595458984375, 31.141357421875, 93.68673706054688, 44.696929931640625, 26.027381896972656, -39.8206787109375, 103.30194091796875, -25.80615234375, 32.36083984375, -0.181976318359375, 15.732294082641602, -34.837646484375, 4.1116943359375, 32.87278747558594, 44.43165588378906, 5.35247802734375, -31.49277114868164, -17.82758331298828, 3.7902908325195312, 70.97842407226562, -8.519134521484375, 29.619354248046875, 24.568763732910156, 44.630126953125, 10.375701904296875, 16.97985076904297, -2.792877197265625, 1.8084716796875, 5.041778564453125, 67.11912536621094, 58.646942138671875, 1.5297088623046875, 36.993309020996094, 3.1134796142578125, 65.70343017578125, 50.243812561035156, 28.850082397460938, 31.002471923828125, -1.96661376953125, -22.592391967773438, 68.39324951171875, 67.90142822265625, 54.965858459472656, 54.6815185546875, -48.47095489501953, 29.5057373046875, -7.576690673828125, 72.28350830078125, -11.733261108398438, 66.72962951660156, 44.12627410888672, 19.3328857421875, 77.09442138671875, 59.03089904785156, -14.474853515625, 46.050235748291016, 46.31159210205078, -1.69775390625, 12.64971923828125, 61.15647888183594, 3.3029022216796875, 33.649627685546875, 54.7022705078125, 7.662813186645508, 21.85601806640625, 48.013816833496094, 49.68719482421875, 0.0, 69.35894775390625, 20.672279357910156, 48.31690979003906, -32.337005615234375, 48.196136474609375, 3.5462646484375, 42.29681396484375, 59.330230712890625, 34.93170166015625, -29.699310302734375, 8.16253662109375, 40.21746826171875, -57.948333740234375, 27.5177001953125, 24.518646240234375, -5.626708984375, 39.890777587890625, 2.3749160766601562, -3.129230499267578, 75.1998291015625, -0.604095458984375, 28.36834716796875, 117.65155029296875, -1.8289642333984375, -8.789657592773438, 26.636207580566406, -3.455902099609375, -5.3441162109375, 12.4603271484375, 26.895004272460938, 12.203353881835938, 2.3687210083007812, 13.1927490234375, -3.6564102172851562, 4.8755645751953125, 87.133544921875, 21.358642578125, -26.397789001464844, 22.324508666992188, 9.35479736328125, 62.90460205078125, 20.998401641845703, 4.883642196655273, 2.0650691986083984, 41.34124755859375, 73.93727111816406, 62.06786346435547, 6.036640167236328, 35.8638916015625, -1.38751220703125, 67.3175048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000170.npy"}
{"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 24.29757308959961, "std": 33.903053283691406, "min": -37.70288848876953, "p10": -19.462445068359372, "median": 20.07403564453125, "p90": 73.82120361328126, "max": 95.32466125488281, "pos_frac": 0.7421875, "sample": [89.89956665039062, 17.96197509765625, 27.680862426757812, 13.39190673828125, 63.712677001953125, 68.14308166503906, 63.828514099121094, -28.7137451171875, -3.9676761627197266, 16.41900634765625, -22.077011108398438, -0.22998046875, 79.51966857910156, -29.858489990234375, 3.1806182861328125, -6.611183166503906, -14.051681518554688, 16.333343505859375, 31.01220703125, 79.82211303710938, 54.99287414550781, -0.094482421875, 20.287750244140625, 83.53030395507812, 62.908447265625, 23.898208618164062, -26.660980224609375, 11.216911315917969, -26.075164794921875, 21.9432373046875, 9.000465393066406, 13.4224853515625, -12.683746337890625, 54.45438003540039, 55.53070068359375, 76.109130859375, 55.59332275390625, 2.5517578125, 29.1134033203125, -5.7701263427734375, 14.282073974609375, -12.8199462890625, -6.704681396484375, -7.998870849609375, -15.454437255859375, 89.03643798828125, 87.53778076171875, 11.195104598999023, -35.8031005859375, 5.3087005615234375, 72.83792114257812, 27.4215087890625, 7.04925537109375, 67.47869873046875, 30.41644287109375, 26.717864990234375, 41.1177978515625, 74.39620971679688, 73.77374267578125, -18.803924560546875, 53.524505615234375, 32.29742431640625, -9.514984130859375, 58.566184997558594, 33.10672378540039, 63.59016418457031, 31.868988037109375, 10.270864486694336, 56.2210693359375, 43.24134826660156, 24.286293029785156, 78.234130859375, -5.596710205078125, 4.2572021484375, -30.33435821533203, -14.837448120117188, 78.20307922363281, 3.0787200927734375, -29.25408935546875, 13.573532104492188, 0.009523391723632812, 2.791778564453125, -13.378463745117188, -25.0863037109375, 17.712005615234375, -34.67097473144531, 51.38823318481445, 44.8397216796875, -4.490478515625, 27.873687744140625, 49.367713928222656, 91.61360168457031, -37.70288848876953, 12.958892822265625, 29.780517578125, 61.81036376953125, 56.43206787109375, 2.664379119873047, 70.36892700195312, 25.09747314453125, 20.4739990234375, 49.7113037109375, -27.118011474609375, 13.295303344726562, 24.312408447265625, 17.4940185546875, 36.18111801147461, 61.9337158203125, 22.34881591796875, 2.0781097412109375, 19.43463134765625, -20.998992919921875, 61.132110595703125, 57.57049560546875, 53.10723876953125, 16.215484619140625, 23.7562255859375, 5.412178039550781, 19.860321044921875, 8.083511352539062, 95.32466125488281, 49.33793640136719, -9.624595642089844, 1.0830726623535156, -3.5791702270507812, 73.93194580078125, 25.406387329101562, -9.886566162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000171.npy"}
{"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 16.591552734375, "std": 35.47467041015625, "min": -89.850341796875, "p10": -18.249227905273436, "median": 14.554367065429688, "p90": 63.97277374267577, "max": 87.96337890625, "pos_frac": 0.6875, "sample": [15.72552490234375, -10.8780517578125, 4.14178466796875, 21.7293701171875, 34.53883361816406, 20.83039093017578, -3.3848876953125, -11.403877258300781, 5.6557769775390625, 0.481475830078125, -3.7497406005859375, -1.1702117919921875, -64.379150390625, 6.239635467529297, 55.10179138183594, 12.455018997192383, 37.64375305175781, 73.19180297851562, 41.0279541015625, 12.511940002441406, -9.87164306640625, 8.9736328125, 24.985321044921875, 81.89266967773438, 4.1082763671875, 45.30938720703125, 37.96211242675781, -8.581085205078125, -9.437347412109375, 61.154693603515625, 4.1881103515625, 27.732696533203125, -69.15283203125, 7.76226806640625, 18.37713623046875, 48.59080505371094, -16.000457763671875, 74.11209106445312, -20.257476806640625, -0.2243804931640625, -10.673811912536621, 15.596328735351562, -0.618865966796875, -1.349557876586914, 76.93486022949219, -89.850341796875, -10.913482666015625, 35.163482666015625, 77.02435302734375, 58.150421142578125, -4.1673583984375, -27.889556884765625, 19.982330322265625, -0.8334426879882812, 32.718017578125, 16.260055541992188, 60.662841796875, 69.14265441894531, -17.3885498046875, -4.353599548339844, -4.52520751953125, 17.580810546875, 24.969558715820312, -2.846405029296875, 60.18994140625, 47.047119140625, -3.94537353515625, 4.5260009765625, 22.08380889892578, 12.196083068847656, 87.560791015625, -60.78387451171875, 1.5954437255859375, 40.331207275390625, 71.6160888671875, 58.46368408203125, 1.6524810791015625, 1.61279296875, 20.77398681640625, -71.39920043945312, 30.0430908203125, 13.51995849609375, 26.09552001953125, -59.67584228515625, 17.731155395507812, 51.449859619140625, 11.605377197265625, 66.89678955078125, 14.72674560546875, -45.99884033203125, 76.38726806640625, -5.848194122314453, 66.775390625, -37.55810546875, -38.25369644165039, 22.898300170898438, 28.952850341796875, 18.537933349609375, -21.074825286865234, 14.381988525390625, -6.2510986328125, 87.96337890625, 48.400390625, 42.58235549926758, 34.81060791015625, 27.55841064453125, -11.727737426757812, 18.992706298828125, 39.1398811340332, 8.027923583984375, 13.78961181640625, 53.140892028808594, 37.2753791809082, 62.77165222167969, 59.138580322265625, 42.67277526855469, -7.420867919921875, 7.83551025390625, 6.804250717163086, 59.07220458984375, -75.84210205078125, 67.17927551269531, 0.071990966796875, 8.703750610351562, 0.0, 38.97900390625, 36.595703125, -8.36810302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000172.npy"}
{"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 20.480083465576172, "std": 34.51216506958008, "min": -67.57861328125, "p10": -19.81518535614013, "median": 19.322168350219727, "p90": 64.400244140625, "max": 120.2396240234375, "pos_frac": 0.7109375, "sample": [-0.678619384765625, -67.57861328125, -48.005760192871094, 19.477157592773438, 28.31756591796875, -12.803619384765625, 47.897003173828125, -42.754364013671875, 47.0482177734375, 113.92742919921875, -4.930843353271484, 24.422515869140625, -18.23808479309082, 38.235595703125, 49.984130859375, 62.797698974609375, 12.47540283203125, 4.715202331542969, 20.281402587890625, 78.28887939453125, 65.29685974121094, 13.02984619140625, 7.4895477294921875, 35.7447509765625, 26.08344268798828, 15.89453125, 6.4140472412109375, 6.008148193359375, 40.827178955078125, 54.29835510253906, -5.3170166015625, 47.66009521484375, -2.2623519897460938, 6.222988128662109, 7.3701019287109375, 34.47843933105469, -35.8304443359375, 29.451904296875, 19.270008087158203, 42.021759033203125, -9.9705810546875, 30.528076171875, 41.174224853515625, 41.29125213623047, 53.50340270996094, 35.785491943359375, -4.9534759521484375, -11.316009521484375, 2.7113876342773438, 11.98016357421875, -12.390640258789062, -41.492164611816406, -36.32615661621094, 120.2396240234375, 14.875274658203125, -27.68048095703125, 7.968505859375, -23.495086669921875, 22.83666229248047, 100.37985229492188, 61.89292907714844, 64.352783203125, 2.2918357849121094, 57.870697021484375, 61.793975830078125, 67.69390869140625, -51.4625244140625, 37.980010986328125, 21.039154052734375, 23.419326782226562, 25.615406036376953, 4.28753662109375, 69.12759399414062, 48.99700164794922, 3.968353271484375, 49.68548583984375, 13.995185852050781, 71.84645080566406, -5.149810791015625, 11.86419677734375, 26.75921630859375, -27.210357666015625, 3.0274620056152344, 6.326019287109375, 57.177650451660156, 6.5206146240234375, 36.72246551513672, 62.532958984375, -4.558837890625, 5.667236328125, 20.1273193359375, -6.59490966796875, 53.9920654296875, 36.26239013671875, 68.341552734375, 28.267059326171875, 64.510986328125, 38.114044189453125, 44.404205322265625, 57.186859130859375, 71.45857238769531, 11.491607666015625, -3.6283512115478516, -12.876495361328125, -8.629659652709961, 61.59252166748047, -37.11883544921875, 56.03843688964844, 40.84349060058594, 19.37432861328125, -5.2542877197265625, 3.518951416015625, -12.418960571289062, 8.289276123046875, 0.4001274108886719, -7.654754638671875, -3.286834716796875, -38.23542022705078, 82.21041870117188, -4.858818054199219, 65.17837524414062, -10.488372802734375, 0.0, 19.416961669921875, 38.68035888671875, -38.990509033203125, -6.78350830078125, 31.816741943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000173.npy"}
{"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 26.106342315673828, "std": 33.8320426940918, "min": -56.144073486328125, "p10": -15.321395874023438, "median": 20.789627075195312, "p90": 74.211572265625, "max": 117.57748413085938, "pos_frac": 0.765625, "sample": [53.479331970214844, 1.6788368225097656, 7.636898040771484, 0.8235244750976562, 90.2587890625, 64.17416381835938, 74.86810302734375, -28.296417236328125, 7.54107666015625, -0.22570037841796875, 69.93331909179688, 54.577659606933594, 3.024017333984375, 51.122459411621094, 15.04962158203125, 30.852142333984375, 67.86923217773438, 99.89520263671875, 7.7934722900390625, 15.187149047851562, -7.69134521484375, 74.1832275390625, 1.1536407470703125, 82.5718994140625, 81.8245849609375, -4.951904296875, 46.727569580078125, 8.217620849609375, 78.75135040283203, 52.87461853027344, -4.7491455078125, 55.755096435546875, 117.57748413085938, -29.815765380859375, -26.140235900878906, -15.898422241210938, 60.5078125, 76.82855224609375, 90.3812255859375, 5.466827392578125, 17.47209930419922, 37.27201843261719, 39.754173278808594, 33.808738708496094, -2.509113311767578, 78.03787231445312, 51.858001708984375, 37.68585205078125, -10.8980712890625, 107.37081909179688, -10.19134521484375, -20.355117797851562, 59.554603576660156, 39.1644287109375, -16.954978942871094, 3.4498291015625, 12.012821197509766, -0.780426025390625, 24.94970703125, 24.091552734375, 21.947311401367188, 6.36273193359375, 74.2777099609375, 64.84892272949219, 32.98956298828125, -13.783935546875, -1.439239501953125, 40.93499755859375, 15.148910522460938, 34.90666198730469, 18.693389892578125, 49.262176513671875, 61.41267395019531, 61.48504638671875, 46.85093688964844, 30.013153076171875, 8.904083251953125, 14.887664794921875, 60.524688720703125, -16.8656005859375, -38.47718811035156, 40.99903869628906, 1.2098388671875, 41.55078125, 19.92254638671875, 2.1497955322265625, 65.49237060546875, 9.358320236206055, -15.547088623046875, 9.94687271118164, 8.434898376464844, 16.799209594726562, 36.270538330078125, 37.855499267578125, 46.63616943359375, 49.420562744140625, 6.144134521484375, 86.30191040039062, -1.7032489776611328, 22.075653076171875, 58.710540771484375, -17.383270263671875, 47.021484375, 21.656707763671875, 10.5728759765625, 13.372116088867188, -14.371543884277344, 0.680450439453125, -4.27978515625, -27.480560302734375, 47.01385498046875, -4.13482666015625, 2.8944625854492188, 34.62477493286133, 31.091094970703125, -17.76416015625, 68.78216552734375, -15.22467041015625, 7.7940673828125, -56.144073486328125, 8.747283935546875, 15.275314331054688, 52.3489990234375, 22.913848876953125, 48.97430419921875, -0.06827545166015625, 22.057464599609375, -7.8824462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000174.npy"}
{"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 20.63296127319336, "std": 36.738548278808594, "min": -95.80001831054688, "p10": -22.71699829101562, "median": 18.849876403808594, "p90": 66.18154449462888, "max": 121.83544921875, "pos_frac": 0.7734375, "sample": [54.25799560546875, 28.25592041015625, 40.73736572265625, -2.3942337036132812, 1.3720932006835938, -45.97040557861328, 24.776641845703125, 2.3562850952148438, 92.79879760742188, -54.25634765625, 18.818984985351562, 8.710670471191406, 27.169754028320312, 26.760643005371094, 49.8553581237793, 121.83544921875, 9.458763122558594, 51.28167724609375, 24.076400756835938, 74.36700439453125, -31.73822021484375, 58.39918518066406, 59.991607666015625, 47.45013427734375, 40.16761779785156, -8.9932861328125, -1.7699127197265625, -36.0404052734375, 19.354354858398438, -36.579071044921875, -5.2902069091796875, 1.1669845581054688, 21.54205322265625, 53.323760986328125, -5.625732421875, 57.957122802734375, 10.3756103515625, 0.1002197265625, 82.01461791992188, 41.21075439453125, 2.8809814453125, 10.5614013671875, 15.0648193359375, 20.58697509765625, 8.551483154296875, 70.81864929199219, 55.7432861328125, 53.90863037109375, -13.19256591796875, 85.15060424804688, 25.633010864257812, 10.8978271484375, 1.68768310546875, 10.052734375, -20.372222900390625, -28.59283447265625, 19.583648681640625, -2.1645660400390625, 0.01503753662109375, 72.85171508789062, 18.772417068481445, 71.32968139648438, 8.976264953613281, 40.47123718261719, 53.28558349609375, 8.421157836914062, 2.4888687133789062, 4.531911849975586, -1.139923095703125, 7.749603271484375, 59.402618408203125, -3.0774993896484375, 15.47833251953125, 62.69378662109375, 51.41944122314453, -34.600982666015625, 0.6121826171875, 64.1942138671875, 0.0, -95.80001831054688, 18.71796417236328, 43.43408203125, 80.68612670898438, 5.51171875, 24.9091796875, -0.6302490234375, 25.065628051757812, -16.039085388183594, -36.699462890625, 84.69454956054688, 43.6551513671875, 4.018032073974609, 48.770172119140625, 57.975860595703125, 47.6385498046875, 29.418426513671875, -21.45904541015625, 12.676603317260742, -25.6522216796875, 7.994598388671875, 51.795318603515625, 51.443572998046875, 10.880870819091797, 38.6083984375, 4.410919189453125, -77.2921142578125, 18.880767822265625, 23.163787841796875, -8.20574951171875, 24.97882080078125, 34.70779800415039, 18.674575805664062, 26.166854858398438, 32.70391082763672, 78.690673828125, 6.34295654296875, 40.521461486816406, 87.3751220703125, 56.08709716796875, 6.08392333984375, 20.515625, 25.964935302734375, 22.373809814453125, -73.43353271484375, -66.52230834960938, 0.0, 4.273887634277344, 90.97988891601562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000175.npy"}
{"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 15.656757354736328, "std": 38.56398391723633, "min": -115.0986328125, "p10": -26.73856582641601, "median": 11.008827209472656, "p90": 64.14462890624999, "max": 108.06033325195312, "pos_frac": 0.734375, "sample": [-29.37281036376953, 12.787353515625, 2.99456787109375, -25.609603881835938, 20.47064208984375, 54.68821716308594, -56.07255554199219, -57.42620849609375, 9.285903930664062, 18.804351806640625, -82.46221923828125, 4.9259033203125, 32.04434585571289, 66.99252319335938, 11.615341186523438, 41.50169372558594, 47.77833557128906, 4.7992095947265625, 1.46429443359375, -9.177810668945312, -41.62774658203125, 9.094970703125, 8.841817855834961, -73.30755615234375, 19.373275756835938, 77.15435791015625, 31.49915313720703, 99.15255737304688, 45.0313720703125, 2.376800537109375, 94.95895385742188, -33.650054931640625, -3.7776260375976562, 18.44873046875, 14.910160064697266, 2.062103271484375, -10.43359375, 24.304183959960938, 15.6671142578125, 7.70147705078125, -49.93336486816406, 12.149126052856445, 47.753631591796875, -3.933929443359375, 7.856590270996094, 61.30609893798828, 6.886833190917969, 3.4006805419921875, 35.30799865722656, 81.262451171875, -31.95226287841797, 52.224464416503906, 47.47166442871094, -23.855224609375, 12.87091064453125, 51.48516845703125, 2.7762603759765625, 46.95446014404297, 35.844970703125, 5.524166107177734, 2.776031494140625, 0.9602203369140625, -7.857452392578125, -17.642993927001953, 46.16942596435547, 61.428558349609375, 37.923675537109375, -15.68670654296875, 71.68795776367188, 3.68316650390625, -12.966854095458984, 76.01132202148438, 16.249679565429688, 10.402313232421875, 108.06033325195312, 63.945556640625, 0.8073711395263672, -15.3616943359375, 17.894393920898438, 7.9815826416015625, 32.819847106933594, 36.952239990234375, 54.581356048583984, 44.095558166503906, 45.291900634765625, 57.19196701049805, -6.468994140625, 19.229232788085938, 9.27484130859375, 84.95272827148438, 30.543701171875, -0.9423675537109375, -32.14070129394531, 33.10711669921875, -15.86163330078125, 8.369138717651367, 48.712066650390625, -71.85891723632812, -23.7393798828125, 32.477081298828125, -24.07916259765625, 53.291656494140625, 3.8676834106445312, -15.3756103515625, 6.774650573730469, -69.22998046875, 28.98760986328125, 15.915771484375, -17.902008056640625, 3.822216033935547, 64.98186492919922, -7.476432800292969, 84.35006713867188, 57.529510498046875, 86.44696044921875, 35.60272979736328, 4.23773193359375, 25.172630310058594, -25.450164794921875, -115.0986328125, 64.609130859375, 0.48480224609375, 0.6932792663574219, 34.38250732421875, 38.82122802734375, 8.335540771484375, 41.809478759765625, -3.705352783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000176.npy"}
{"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 14.967822074890137, "std": 42.468563079833984, "min": -112.17036437988281, "p10": -36.72613525390625, "median": 8.290756225585938, "p90": 71.64668579101563, "max": 107.12158203125, "pos_frac": 0.671875, "sample": [52.10215759277344, -0.16251754760742188, 3.0987625122070312, 83.79843139648438, -18.137184143066406, 71.30183410644531, 13.452716827392578, -23.844512939453125, 35.79754638671875, 7.892616271972656, 67.11666870117188, -46.03783416748047, -94.21832275390625, -13.035564422607422, 4.0962677001953125, -30.547760009765625, 44.059356689453125, 1.780914306640625, 50.848785400390625, 8.37652587890625, 49.4527587890625, 26.660507202148438, 59.73646545410156, 70.57769775390625, -25.859375, 29.504974365234375, -36.79534912109375, 85.98477172851562, 53.48345947265625, 97.93389892578125, 7.7415313720703125, 29.535747528076172, -74.503173828125, 92.58035278320312, 3.1672821044921875, 28.846786499023438, 72.87188720703125, 3.370044708251953, -9.64715576171875, 6.522216796875, 47.656280517578125, -23.276687622070312, -3.663808822631836, 2.5172500610351562, 2.90472412109375, -32.53363800048828, 4.964488983154297, 14.369037628173828, 50.66752624511719, 54.365478515625, -10.1494140625, -4.091278076171875, 93.112548828125, 26.359237670898438, -6.0310211181640625, 62.4151611328125, 24.00396728515625, -5.830291748046875, 79.07316589355469, 5.08154296875, -45.89599609375, 80.07620239257812, -0.555908203125, -36.69647216796875, 74.4066162109375, -4.6761932373046875, -48.226104736328125, -10.507049560546875, 5.492031097412109, 68.78213500976562, -16.114715576171875, 99.27264404296875, -6.70697021484375, 7.170717239379883, -5.305030822753906, 54.22483825683594, 50.872528076171875, -90.09768676757812, 22.895416259765625, 5.690826416015625, 21.3935546875, -55.780975341796875, -51.65467071533203, 26.212158203125, -74.91384887695312, 10.89453125, 36.28174591064453, 1.4246826171875, -22.5159912109375, 11.171585083007812, 23.97967529296875, 64.44677734375, 42.358055114746094, 36.42881774902344, -11.08917236328125, -16.894683837890625, 12.703445434570312, 17.1387939453125, 58.35552978515625, 3.7701797485351562, -27.168563842773438, -5.4273681640625, 14.288543701171875, 43.245819091796875, 7.98114013671875, -80.90383911132812, 8.204986572265625, -0.7408905029296875, -48.892547607421875, -112.17036437988281, 27.235572814941406, 50.16178894042969, 72.45133972167969, 8.035125732421875, 16.255718231201172, 75.1966552734375, 37.7288818359375, 40.995452880859375, 107.12158203125, 11.452621459960938, 50.068267822265625, 0.6149673461914062, -3.38043212890625, -3.4975738525390625, 18.318527221679688, 4.885833740234375, 33.628173828125, 61.561302185058594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000177.npy"}
{"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 24.86409568786621, "std": 39.86204528808594, "min": -95.49752807617188, "p10": -17.24786987304687, "median": 20.070465087890625, "p90": 82.44652404785155, "max": 148.53732299804688, "pos_frac": 0.7265625, "sample": [62.58880615234375, 17.357250213623047, 8.405914306640625, -8.01300048828125, -7.593505859375, 7.770965576171875, 42.8511962890625, 54.97815704345703, 34.42024230957031, 40.2841796875, 148.53732299804688, 63.12290954589844, 15.023895263671875, 29.515655517578125, -9.69866943359375, 86.74272155761719, 0.0, 61.18711853027344, 60.78741455078125, 36.166473388671875, 14.021240234375, 93.49163818359375, 8.275217056274414, 78.17919921875, 53.453125, -14.736763000488281, -8.876571655273438, 99.62368774414062, -24.843597412109375, 42.05474853515625, 5.277910232543945, -5.7081146240234375, -1.66552734375, 47.722259521484375, -64.89578247070312, 12.60968017578125, 31.518783569335938, 70.70921325683594, -28.83868408203125, 55.22209930419922, 78.53204345703125, 34.50212097167969, 32.65824890136719, 45.854248046875, 81.47378540039062, 59.2484130859375, 15.157562255859375, -50.44053649902344, -8.087127685546875, 93.04367065429688, 61.002532958984375, 92.01425170898438, -55.37825012207031, -95.49752807617188, -5.82281494140625, 11.670158386230469, 16.45147705078125, 36.345375061035156, -36.37840270996094, 25.57861328125, 0.5098514556884766, 102.97799682617188, 10.556709289550781, 38.534912109375, 66.221923828125, 25.21851348876953, 29.748626708984375, 32.7310791015625, 39.80145263671875, 2.7733001708984375, 96.82403564453125, 24.933868408203125, -13.575958251953125, -2.91204833984375, 12.603208541870117, 84.71624755859375, -43.1224365234375, 15.584609985351562, -3.911285400390625, 1.484130859375, 44.97625732421875, 22.31915283203125, 47.648956298828125, -21.467437744140625, -43.396209716796875, -3.9690589904785156, 89.0947265625, -1.1850738525390625, 34.60191345214844, 33.186492919921875, 16.84679412841797, 3.782064437866211, -0.532928466796875, 20.0277099609375, 15.259063720703125, 98.04006958007812, 14.957550048828125, 10.622550964355469, 1.1751708984375, 67.07196044921875, 9.505729675292969, -14.145057678222656, -27.059646606445312, 94.28466796875, 22.037025451660156, -15.42919921875, 3.8729248046875, 68.71658325195312, 69.22756958007812, -15.439483642578125, -4.030189514160156, 103.43756103515625, -38.49037170410156, 4.321533203125, 43.526611328125, 20.11322021484375, -21.724925994873047, 19.234046936035156, 48.20524597167969, 2.59423828125, 26.1510009765625, 31.02423095703125, 24.32769775390625, -4.57958984375, -5.472503662109375, 29.793914794921875, 55.729557037353516, 43.15681457519531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000178.npy"}
{"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 18.547178268432617, "std": 38.08045196533203, "min": -65.66925048828125, "p10": -33.697524261474605, "median": 14.031400680541992, "p90": 72.67248992919922, "max": 96.73245239257812, "pos_frac": 0.6953125, "sample": [45.280242919921875, 1.9685134887695312, 66.2730712890625, -9.574951171875, 14.943557739257812, 67.49842834472656, 9.658302307128906, 10.54361343383789, -59.3115234375, 1.05615234375, 14.60528564453125, -2.62255859375, 22.446807861328125, 2.9473114013671875, 73.30397033691406, -38.39329528808594, 46.45635986328125, 28.5428466796875, 6.610748291015625, -4.804283142089844, 26.32147216796875, 45.918663024902344, 1.6075057983398438, 30.458969116210938, 22.58636474609375, -51.04728698730469, -42.397247314453125, -5.71068000793457, -5.389923095703125, -55.66319274902344, 51.210845947265625, 8.142318725585938, -48.66815185546875, 30.337066650390625, -8.228546142578125, 30.9111328125, 50.582916259765625, 76.71453857421875, 22.68463134765625, -7.3099517822265625, -29.34417724609375, -52.602294921875, -8.009292602539062, 25.92437744140625, -20.26104736328125, 44.02888488769531, 6.6341094970703125, -10.920526504516602, 71.43173217773438, -41.12384033203125, -28.765884399414062, -30.442840576171875, 32.3572998046875, 17.43816375732422, 3.5430335998535156, 82.34909057617188, 72.05925750732422, 65.09662628173828, 13.8597412109375, -4.28759765625, -65.66925048828125, -11.2213134765625, 94.52944946289062, 13.82501220703125, -15.417770385742188, 37.945465087890625, 72.40185546875, -9.877426147460938, 0.003662109375, 13.54962158203125, 11.123775482177734, 45.04248046875, 46.60237121582031, 35.0399169921875, 1.125152587890625, 7.447477340698242, 80.56056213378906, 11.160049438476562, 9.38870620727539, 56.097511291503906, 11.467020034790039, 8.311538696289062, 93.84637451171875, -4.3544464111328125, 14.126777648925781, -41.1800537109375, 82.23433685302734, 78.05450439453125, 0.8638343811035156, -43.72492980957031, 51.993309020996094, 9.278656005859375, 80.6029052734375, 34.0540771484375, 32.587554931640625, 26.279754638671875, 44.54548645019531, 43.41386413574219, 16.31732177734375, 45.51737976074219, 88.27676391601562, 37.01902770996094, 68.571533203125, 18.64226531982422, -42.0206298828125, 30.305389404296875, -54.578155517578125, 3.08447265625, 32.81950378417969, 65.5501480102539, 21.88702392578125, 42.1473388671875, -31.68505096435547, 96.73245239257812, -14.99310302734375, -19.08631134033203, 45.53550720214844, 51.445831298828125, -2.0731201171875, -10.005142211914062, 80.978759765625, 57.07273864746094, -2.3728790283203125, 13.936023712158203, -11.284072875976562, -20.11199951171875, 76.30349731445312, 54.593536376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000179.npy"}
{"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 24.078582763671875, "std": 37.524322509765625, "min": -63.950775146484375, "p10": -18.718766784667967, "median": 17.248313903808594, "p90": 77.87674102783203, "max": 107.22610473632812, "pos_frac": 0.7421875, "sample": [-0.9825325012207031, -4.323024749755859, -40.74272155761719, -10.84075927734375, 2.3946456909179688, 30.7333984375, 69.71160888671875, 7.31549072265625, -21.66741943359375, 76.77281188964844, -2.337127685546875, 56.038787841796875, 100.10629272460938, -61.20738220214844, 7.78692626953125, 70.72491455078125, -5.2434844970703125, 30.975616455078125, 20.192665100097656, -63.950775146484375, 58.43670654296875, 107.22610473632812, 42.455894470214844, 36.174560546875, 23.733734130859375, 51.87885284423828, 18.01849365234375, 66.46260070800781, 98.72172546386719, -7.767940521240234, -46.02693176269531, 3.854766845703125, 78.34315490722656, 101.4500732421875, 77.67684936523438, 13.208328247070312, 15.357498168945312, 11.168991088867188, 11.844329833984375, 16.995803833007812, 69.533935546875, -7.417533874511719, 48.03997802734375, 33.05670166015625, 91.89730834960938, 17.94293212890625, -23.919189453125, -2.165130615234375, 7.56231689453125, 2.2212677001953125, -36.896278381347656, -4.765350341796875, 4.965728759765625, 30.207672119140625, 84.81396484375, 0.97613525390625, 18.872638702392578, 55.2452392578125, 12.578353881835938, 11.282051086425781, 90.05853271484375, 43.941261291503906, -6.93701171875, 95.01263427734375, 52.576698303222656, -16.568817138671875, 41.96875, 35.11761474609375, -42.46244812011719, 20.263458251953125, 32.9261474609375, 4.3268280029296875, 17.500823974609375, 85.00994873046875, 72.03941345214844, 74.07305908203125, 0.634674072265625, -0.24118423461914062, 75.21719360351562, 44.572601318359375, 100.89639282226562, 15.986618041992188, -15.282623291015625, 56.37481689453125, 58.95070266723633, 49.744850158691406, 42.05010986328125, -41.00249481201172, 1.882720947265625, 32.96186828613281, -1.3737335205078125, 16.83380126953125, 5.739452362060547, 30.682373046875, 56.00334167480469, 14.15576171875, 26.13262939453125, 31.97747802734375, 0.9638023376464844, -3.49176025390625, 5.146034240722656, 22.1358642578125, 67.90933227539062, 0.01568603515625, 4.46638298034668, -21.55770492553711, 5.740480422973633, -21.203277587890625, 78.837158203125, 6.461555480957031, -13.974578857421875, 43.6214599609375, -5.726772308349609, -27.6419677734375, 52.16828155517578, 32.028778076171875, 40.17250061035156, -20.257354736328125, -16.3558349609375, 54.52685546875, 0.7586460113525391, 36.708038330078125, 2.9781875610351562, 4.5233612060546875, -9.248703002929688, 93.30523681640625, -18.059371948242188, 30.66009521484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000180.npy"}
{"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 22.81806755065918, "std": 40.66453552246094, "min": -80.88885498046875, "p10": -31.507850646972656, "median": 21.604248046875, "p90": 74.29309692382812, "max": 131.94757080078125, "pos_frac": 0.7265625, "sample": [1.1610050201416016, 6.495025634765625, 2.7321243286132812, 36.1202392578125, 0.0, 45.24217987060547, 76.42666625976562, 34.09429931640625, 8.392791748046875, 38.302642822265625, 66.349609375, -11.853126525878906, 52.919677734375, -7.542236328125, 89.1181640625, 38.691314697265625, 0.0, 73.63337707519531, 40.00827407836914, -31.584915161132812, 51.72291564941406, 4.672088623046875, 75.126953125, 15.634796142578125, -72.09170532226562, 94.17399597167969, -38.30804443359375, 41.43034744262695, 5.610801696777344, 29.485687255859375, 63.42827224731445, 3.0245742797851562, 13.206497192382812, 65.441162109375, 26.1986083984375, 47.269737243652344, 52.15093994140625, 19.783721923828125, 21.947418212890625, 33.44677734375, 17.935928344726562, 35.47216796875, -26.626678466796875, -5.715715408325195, 11.33575439453125, 14.663185119628906, 9.486495971679688, 14.952667236328125, 95.79683685302734, -40.69757843017578, 32.642578125, -5.998085021972656, 53.80757141113281, 62.64373779296875, -21.496490478515625, -64.177734375, -3.4398574829101562, 18.550567626953125, -3.93560791015625, -10.871551513671875, 35.98675537109375, -6.0909271240234375, -24.47845458984375, 77.73648071289062, 46.35052490234375, 43.88604736328125, 11.052520751953125, 68.12205505371094, 6.597198486328125, 35.58721923828125, 21.53955078125, 30.203048706054688, 34.915565490722656, 86.02928161621094, 115.32855224609375, 131.94757080078125, -59.970916748046875, 12.672149658203125, -31.474822998046875, 69.22665405273438, -34.609466552734375, 31.21282958984375, -5.896732330322266, -68.496826171875, -11.414947509765625, 69.46821594238281, 24.8348388671875, -11.022254943847656, 21.6689453125, 20.648345947265625, -40.33052062988281, 36.568023681640625, 73.93572998046875, -38.4307861328125, 13.731986999511719, 32.640174865722656, 10.624755859375, 46.45045471191406, 23.409332275390625, 87.21365356445312, 80.7947998046875, -30.0638427734375, 46.88006591796875, 11.57391357421875, -31.400177001953125, -13.859848022460938, 11.722976684570312, 90.4442138671875, 3.69891357421875, -80.88885498046875, 61.48175048828125, 32.482208251953125, 73.03396606445312, -13.519775390625, 38.51618957519531, 83.40788269042969, -46.53489685058594, 16.31365966796875, 14.902275085449219, -12.262725830078125, 15.6153564453125, 30.803085327148438, 63.898681640625, -43.60408020019531, 65.8939437866211, 55.880615234375, 48.727638244628906, 63.012115478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000181.npy"}
{"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 28.893413543701172, "std": 34.453792572021484, "min": -67.473876953125, "p10": -10.579850769042967, "median": 29.262283325195312, "p90": 72.83034362792968, "max": 107.82412719726562, "pos_frac": 0.8125, "sample": [43.90934753417969, 64.23017883300781, 11.719635009765625, -2.357583999633789, 20.245193481445312, 19.174514770507812, 72.55880737304688, 39.40325927734375, 23.05413818359375, 6.1992340087890625, 65.832275390625, 29.0601806640625, 29.464385986328125, 53.58319091796875, 58.28521728515625, 3.87921142578125, 41.79913330078125, 54.3509521484375, 70.152099609375, 49.33818054199219, 45.731040954589844, 19.284107208251953, 49.796470642089844, 85.268798828125, -18.035003662109375, 74.8546142578125, -7.53662109375, 50.781768798828125, 97.24441528320312, 39.692596435546875, 54.51692199707031, 42.27107238769531, 37.869659423828125, 38.9193115234375, 33.488121032714844, -7.958637237548828, -11.471054077148438, 7.941650390625, 0.856658935546875, 8.716598510742188, -56.61236572265625, 25.89495849609375, -67.473876953125, 45.93260955810547, 60.487945556640625, -10.197906494140625, 83.13751220703125, -46.261444091796875, 65.24700927734375, 107.82412719726562, 32.976165771484375, 62.878021240234375, 70.04556274414062, 74.6346435546875, 50.4415283203125, -7.609466552734375, 21.21615219116211, 6.909034729003906, 7.08966064453125, 42.1046142578125, 34.862548828125, 71.01054382324219, 24.412109375, 20.162307739257812, 42.46099853515625, 23.03697967529297, 36.83244323730469, 54.59771728515625, 98.69015502929688, 27.89102554321289, -35.56217956542969, 9.336479187011719, 49.489166259765625, 14.66192626953125, 0.963775634765625, 29.4857177734375, 36.353485107421875, 49.65081024169922, 6.979423522949219, 9.955718994140625, 8.678970336914062, -4.202301025390625, 23.00067138671875, 58.24322509765625, -19.21368408203125, -1.1262359619140625, 31.809188842773438, 8.972877502441406, 13.806396484375, -5.0622406005859375, 82.91558837890625, 55.6312255859375, 82.91543579101562, 73.46392822265625, -42.87516784667969, -10.09510612487793, 82.33758544921875, 46.31568908691406, 28.27105712890625, 43.50083923339844, -23.2362060546875, -2.75262451171875, 12.1949462890625, 14.122913360595703, 2.4669189453125, 48.60396957397461, 47.328643798828125, 27.44012451171875, 80.53326416015625, 7.8858642578125, -1.686920166015625, 59.19282531738281, 71.118408203125, 2.2188663482666016, 1.7692337036132812, 6.541259765625, 23.389450073242188, 65.49386596679688, 10.09771728515625, -38.88043212890625, -20.503448486328125, 66.72396850585938, 50.884765625, 1.20269775390625, 81.4195556640625, -16.330894470214844, 31.48443603515625, -37.70381164550781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000182.npy"}
{"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 24.156522750854492, "std": 40.0928840637207, "min": -68.92253875732422, "p10": -24.466778564453122, "median": 18.92547035217285, "p90": 72.77988281249999, "max": 114.1343994140625, "pos_frac": 0.71875, "sample": [-26.01806640625, 44.776031494140625, -30.163833618164062, 57.192657470703125, -49.531768798828125, 75.01020812988281, -36.309814453125, 12.552261352539062, 63.668121337890625, 14.320724487304688, -0.127288818359375, 23.5750732421875, 43.42710876464844, 3.62530517578125, 7.792236328125, -63.326416015625, 1.6153316497802734, 69.15341186523438, 16.263153076171875, -6.279937744140625, 18.843307495117188, 11.247901916503906, 56.89942932128906, -15.736312866210938, 13.010223388671875, -23.1776123046875, 71.98880004882812, 40.903167724609375, 9.55865478515625, 2.5661392211914062, -7.8848876953125, -62.18328857421875, 16.781692504882812, 17.48858642578125, -0.4336395263671875, 59.1845703125, 15.180709838867188, 73.5579833984375, 33.145477294921875, -35.170745849609375, 96.06903076171875, -7.472557067871094, 50.26854705810547, -22.73193359375, 72.4464111328125, 24.24304962158203, 59.900634765625, -20.43218994140625, 87.46438598632812, 45.92337417602539, 9.8948974609375, 70.22573852539062, 61.873870849609375, -50.15753173828125, 0.0, -1.2190208435058594, -0.33428955078125, 8.089736938476562, -18.835487365722656, 65.40567016601562, 12.914375305175781, 74.222900390625, 0.5835666656494141, 89.67132568359375, 71.84182739257812, -36.4710693359375, 21.475265502929688, 89.0404052734375, 59.90504455566406, -17.871688842773438, 44.3612060546875, 100.73440551757812, 114.1343994140625, 51.37506103515625, -5.4725494384765625, 34.273956298828125, -60.02186584472656, 25.78460693359375, -2.41748046875, 31.133026123046875, 34.064971923828125, 35.27947998046875, 19.007633209228516, 24.08321189880371, -2.7940216064453125, 6.747314453125, 62.858184814453125, 62.859039306640625, 12.906036376953125, 12.655670166015625, -13.217788696289062, 32.53569030761719, 5.3328094482421875, 3.674142837524414, -23.80194091796875, 70.4249267578125, 76.32673645019531, 52.28118896484375, 47.0498046875, -0.46697235107421875, -68.92253875732422, 3.2298431396484375, 0.0287628173828125, 17.896499633789062, 3.4205970764160156, 59.27699279785156, 0.17833328247070312, 72.20228576660156, -5.458518981933594, 92.64004516601562, 38.24955749511719, 89.94586181640625, 58.78302001953125, 53.48356628417969, 41.17218017578125, -51.816749572753906, 21.28692626953125, 48.59625244140625, -48.375213623046875, 107.06576538085938, -7.252197265625, 56.61353302001953, 69.60861206054688, 32.313201904296875, -7.364288330078125, 33.739585876464844, 50.69450378417969, 66.188720703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000183.npy"}
{"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 36.56228256225586, "std": 44.48066329956055, "min": -84.30914306640625, "p10": -27.241339111328124, "median": 41.40620422363281, "p90": 90.25158081054686, "max": 118.65133666992188, "pos_frac": 0.7890625, "sample": [63.582618713378906, 118.65133666992188, 83.06524658203125, 5.192344665527344, 20.199310302734375, -12.917976379394531, 43.38433837890625, -7.395866394042969, 62.947265625, 45.553436279296875, 27.435134887695312, 6.714141845703125, -60.47381591796875, 27.6689453125, -36.56428527832031, -0.6992874145507812, 101.84500122070312, 78.75119018554688, 51.39630889892578, 55.596343994140625, 62.864349365234375, 88.79180908203125, -84.30914306640625, 88.21743774414062, -7.1158905029296875, 86.90093994140625, 73.28033447265625, -2.5577850341796875, -50.170631408691406, 100.19354248046875, 34.75714874267578, 95.909423828125, 57.710693359375, -45.62535095214844, 29.77874755859375, 84.0655517578125, 7.232025146484375, 87.06016540527344, 61.248870849609375, 105.86492919921875, 25.377410888671875, 98.27871704101562, -49.19117736816406, 47.01131820678711, 41.051116943359375, 107.20370483398438, -52.401580810546875, 99.0848388671875, -26.977264404296875, 45.6260986328125, 24.163982391357422, -14.196136474609375, 42.788299560546875, 6.630157470703125, -27.857513427734375, 40.664520263671875, 69.9495849609375, 42.11894226074219, -5.70635986328125, 105.9027099609375, 8.349761962890625, 5.367362976074219, 52.8558349609375, -54.91227722167969, 74.10507202148438, 60.32318115234375, 71.262939453125, 85.73878479003906, 26.280914306640625, 71.54046630859375, 9.67852783203125, 35.10646057128906, 69.19467163085938, 34.38963317871094, 41.76129150390625, 24.554298400878906, 28.17437744140625, 70.835693359375, 36.09544372558594, 82.82061767578125, 93.65771484375, 74.725830078125, 1.198944091796875, 80.08006286621094, 21.696937561035156, -46.34260559082031, 6.28082275390625, 83.54391479492188, 4.38055419921875, 12.936981201171875, 27.99560546875, 86.95574951171875, 48.60502624511719, 29.788986206054688, 69.10934448242188, -5.303474426269531, -28.889890670776367, 19.952713012695312, 1.5893821716308594, 85.1689453125, 56.964134216308594, 57.4810791015625, 49.77789306640625, 76.22244262695312, 0.3805503845214844, 2.8164424896240234, 71.22769165039062, 46.698150634765625, -13.710472106933594, 95.25958251953125, 28.982757568359375, -43.028289794921875, 52.736480712890625, 13.424041748046875, 72.87157440185547, -8.365081787109375, 95.3460693359375, 6.9840850830078125, -9.294265747070312, 67.24293518066406, -4.132110595703125, -29.20306396484375, 83.80276489257812, 17.26458740234375, 107.656494140625, 74.26046752929688, -11.267501831054688, 53.36777877807617], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000184.npy"}
{"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 23.374982833862305, "std": 41.926204681396484, "min": -72.56488037109375, "p10": -27.861217498779297, "median": 20.607074737548828, "p90": 81.83070526123046, "max": 113.76022338867188, "pos_frac": 0.671875, "sample": [26.752304077148438, -59.630279541015625, 57.9857177734375, -18.726959228515625, -33.98663330078125, 79.66998291015625, 30.55908203125, 83.99404907226562, 86.33544921875, -7.0986785888671875, 21.515228271484375, 36.904449462890625, 76.52127075195312, 57.009403228759766, 35.527442932128906, 96.00616455078125, 47.9361572265625, -16.71185302734375, 8.951896667480469, 70.81005859375, 39.61944580078125, 10.6798095703125, 55.6172981262207, 26.351455688476562, 28.788818359375, 20.1998291015625, 84.36328125, 9.855216979980469, 1.307159423828125, 86.79910278320312, -72.56488037109375, 56.14330291748047, 3.6093368530273438, -7.93170166015625, 20.72010040283203, 68.46829223632812, 27.618072509765625, 47.249366760253906, 34.737060546875, 107.9976806640625, -7.705108642578125, 40.885955810546875, -4.22979736328125, 24.601715087890625, 9.196624755859375, 83.21986389160156, -6.11712646484375, 37.31178283691406, 29.45574951171875, 64.6373291015625, -50.05024719238281, -25.147842407226562, -43.359527587890625, -18.385589599609375, 13.680252075195312, -0.03299903869628906, 12.722442626953125, 0.0, 16.056777954101562, 64.9694595336914, -40.23899841308594, -27.677001953125, 12.538162231445312, 1.6121444702148438, 34.93605041503906, 20.494049072265625, 5.353084564208984, 16.3685302734375, 37.44976806640625, -4.754098892211914, -15.925506591796875, 37.39813232421875, 54.06263732910156, 75.20281982421875, -18.408145904541016, 0.936859130859375, -0.384124755859375, 17.98760986328125, 28.980060577392578, 69.9789047241211, 113.76022338867188, -1.2934112548828125, 12.122238159179688, 0.0, -23.92657470703125, 63.50205612182617, 78.50408935546875, -22.724761962890625, 81.2353515625, -13.263763427734375, 49.533775329589844, 50.46112060546875, -20.463233947753906, -5.555538177490234, 63.52606201171875, 87.753173828125, -49.88352966308594, 69.22563171386719, 15.258598327636719, 15.7847900390625, -19.636192321777344, 40.929901123046875, 50.079681396484375, -37.163360595703125, 94.01907348632812, -56.87896728515625, -14.874420166015625, 15.69049072265625, 108.64999389648438, 112.67822265625, 62.765472412109375, 36.347625732421875, 12.79852294921875, -2.276611328125, 79.39669799804688, -65.9832763671875, 43.8416748046875, 91.47549438476562, 55.08683776855469, -31.5706787109375, 28.89299774169922, -15.973434448242188, -11.4354248046875, -21.72235107421875, -37.96005630493164, -28.291053771972656, 34.61116027832031, 27.37054443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000185.npy"}
{"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 22.744592666625977, "std": 38.76459503173828, "min": -79.7489013671875, "p10": -19.119735717773438, "median": 19.3466796875, "p90": 72.42445373535155, "max": 134.833740234375, "pos_frac": 0.7109375, "sample": [27.057815551757812, -2.7062301635742188, 11.058685302734375, 52.857940673828125, 19.781951904296875, 68.60812377929688, 29.15087890625, -2.53973388671875, 15.508041381835938, 17.246524810791016, -4.0760498046875, 8.85882568359375, 10.077957153320312, 50.45802307128906, -18.70458984375, 68.39569091796875, 57.984901428222656, 2.304250717163086, 71.63824462890625, 9.658828735351562, 9.465087890625, -9.908340454101562, -26.796630859375, 3.6253204345703125, 1.2492141723632812, 54.941070556640625, 43.48265075683594, 28.83709716796875, 6.630401611328125, 0.642059326171875, 9.46811294555664, 37.65625, 95.47564697265625, 26.300552368164062, -10.249237060546875, -47.33507537841797, 40.118011474609375, 93.21316528320312, 29.19305419921875, 31.671112060546875, 2.430938720703125, 68.8349609375, 60.24049377441406, 48.43191909790039, 58.32292556762695, 37.44952392578125, -54.37841796875, 6.642950057983398, 53.93376922607422, 26.2857666015625, 29.379638671875, 35.27093505859375, 77.98129272460938, -5.65045166015625, -17.6087646484375, -14.760498046875, 70.14260864257812, 21.656585693359375, -2.7646484375, 5.7578125, -6.317291259765625, -8.7708740234375, 84.6536865234375, -32.177032470703125, 74.25894165039062, 12.85400390625, 36.533905029296875, -4.0203857421875, 5.225738525390625, 78.51211547851562, -6.363090515136719, 20.1285400390625, 102.4580078125, 27.316192626953125, -22.461334228515625, 18.911407470703125, 60.001739501953125, -73.46435546875, -0.6300792694091797, 84.253173828125, 56.300811767578125, 7.66400146484375, 29.3154296875, 121.870361328125, 62.55987548828125, -65.30305480957031, 23.0753173828125, -23.969642639160156, 6.8270721435546875, -51.0904541015625, 1.3551483154296875, 20.165618896484375, 27.20044708251953, -6.296638488769531, 98.28326416015625, 95.21981811523438, 7.349174499511719, 14.670852661132812, -79.7489013671875, 33.916324615478516, 39.53133010864258, 70.87908935546875, 17.11516571044922, -6.21368408203125, 42.20628356933594, 53.831298828125, 26.279510498046875, -1.0009346008300781, -3.61785888671875, 49.89504623413086, 51.43009948730469, -15.595039367675781, 32.650177001953125, 4.590240478515625, 134.833740234375, 0.0, 78.76150512695312, 59.5222282409668, -20.088409423828125, 29.8216552734375, -22.2495059967041, -31.841228485107422, -1.0930633544921875, 0.31421661376953125, 36.3037109375, -13.966949462890625, 58.2987060546875, -7.456298828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000186.npy"}
{"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 22.81952476501465, "std": 45.140769958496094, "min": -84.77426147460938, "p10": -34.0394603729248, "median": 15.744606018066406, "p90": 88.48703613281249, "max": 118.08905029296875, "pos_frac": 0.6796875, "sample": [51.75927734375, 68.74111938476562, 16.7281494140625, 40.538299560546875, -55.81787109375, 9.391815185546875, 95.03286743164062, -0.07987022399902344, -36.87950134277344, 118.08905029296875, 80.11154174804688, 0.3003997802734375, -32.349456787109375, 13.385055541992188, 70.38795471191406, 114.2752685546875, 50.07794189453125, 48.72649383544922, 88.15896606445312, 12.1180419921875, 69.89462280273438, -29.166465759277344, 43.004180908203125, -4.440155029296875, -18.86224365234375, -15.250457763671875, 69.22209167480469, -69.23806762695312, -47.308074951171875, -69.3240966796875, -10.477523803710938, 79.12358093261719, 48.685791015625, -6.5536956787109375, -24.968894958496094, 41.43934631347656, -36.611083984375, -10.0274658203125, 38.20555114746094, 40.37939453125, 14.26641845703125, -8.15057373046875, 103.75341796875, 8.658248901367188, -13.99102783203125, 41.318756103515625, 89.25253295898438, 4.62200927734375, -22.4312744140625, 45.91845703125, 1.57098388671875, 27.171630859375, 27.37415313720703, 66.99066162109375, -84.77426147460938, -5.501708984375, 60.047882080078125, -37.4339599609375, 60.247745513916016, 95.5335693359375, 50.86015319824219, -10.419624328613281, 38.68939208984375, -42.15338134765625, 112.38525390625, 73.23583984375, -59.02030944824219, -55.64447021484375, 15.232269287109375, 27.238189697265625, 89.92523193359375, -32.93733596801758, -7.116962432861328, -31.44647216796875, -9.336273193359375, -10.755828857421875, 69.87335205078125, 19.633262634277344, 81.99545288085938, 15.6099853515625, 2.27960205078125, 11.805221557617188, -40.59201431274414, 0.5033340454101562, 7.1951751708984375, 1.0350723266601562, 22.626495361328125, 9.519607543945312, -62.958984375, 71.1009521484375, 55.63685607910156, 75.15048217773438, 73.13664245605469, 19.818115234375, 30.415245056152344, 93.635498046875, -0.060394287109375, 60.182830810546875, 60.79576873779297, 16.038551330566406, -20.58001708984375, 25.788482666015625, 106.259765625, 9.495819091796875, -0.21484756469726562, 94.45443725585938, 1.49468994140625, 25.146102905273438, -23.87823486328125, 91.68167114257812, 55.436614990234375, 34.57147216796875, 4.679931640625, 15.879226684570312, 10.584062576293945, 76.99066162109375, 90.7847900390625, 1.15338134765625, 8.710332870483398, 11.903182983398438, -10.130828857421875, 44.86216735839844, 20.909461975097656, -5.7310028076171875, 0.0, 79.1871337890625, -12.328826904296875, 25.812103271484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000187.npy"}
{"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 23.80295753479004, "std": 44.85831069946289, "min": -95.94265747070312, "p10": -33.62803649902344, "median": 16.92719268798828, "p90": 84.4108154296875, "max": 112.02197265625, "pos_frac": 0.7109375, "sample": [16.180490493774414, 46.0025634765625, 84.38397216796875, 0.0, 37.90937042236328, 17.327423095703125, 3.445770263671875, 88.94622802734375, 1.1200828552246094, 48.086021423339844, 49.92439270019531, 5.752716064453125, 99.4571533203125, -11.787792205810547, 27.15679931640625, -2.2507858276367188, 97.28121948242188, 58.082000732421875, 15.314186096191406, 112.02197265625, 63.32733154296875, 60.868927001953125, -19.06396484375, 2.8129501342773438, -21.931053161621094, -21.12017822265625, -33.7108154296875, 20.541000366210938, -27.447799682617188, 64.66582489013672, 77.95809936523438, -12.589523315429688, 86.88009643554688, 53.54247283935547, -2.839263916015625, 16.526962280273438, 84.47344970703125, 58.020111083984375, 67.45124816894531, 16.386993408203125, 19.62908935546875, 36.83203125, 7.735649108886719, 36.812744140625, 97.432861328125, -80.91925048828125, 43.9041748046875, 63.31536865234375, 62.472930908203125, 12.085464477539062, 9.14361572265625, 68.81967163085938, -21.855453491210938, -4.00885009765625, 43.86712646484375, 89.21975708007812, 107.17779541015625, 9.161846160888672, 59.34822082519531, 97.31549072265625, 38.83978271484375, 25.343467712402344, 53.268333435058594, 74.09292602539062, 4.645099639892578, 9.166473388671875, 9.501602172851562, 76.65522766113281, 53.04866027832031, 14.334304809570312, -2.29541015625, 85.27511596679688, 76.46044921875, -0.8845310211181641, -38.214111328125, 3.1009368896484375, 67.369140625, 93.521728515625, -1.55743408203125, 4.276580810546875, -60.540863037109375, 62.51744079589844, 18.954689025878906, 78.16970825195312, 0.6268463134765625, -8.570465087890625, 40.01615905761719, 67.468017578125, -0.57318115234375, -0.01702880859375, -95.94265747070312, 29.935760498046875, -73.04531860351562, -59.58172607421875, 38.47767639160156, -7.7249755859375, -13.308387756347656, -35.423065185546875, 29.39427947998047, 63.918052673339844, 22.569854736328125, 2.1018753051757812, -52.09523010253906, -93.21212768554688, 48.2528076171875, 12.123947143554688, 43.953208923339844, 73.40171813964844, 73.72848510742188, -11.823299407958984, -64.36962890625, 1.5593318939208984, 50.67219543457031, 43.76853942871094, 82.6822509765625, -17.991455078125, 91.33837890625, -4.847820281982422, 1.5729217529296875, -37.03509521484375, -33.592559814453125, 4.08135986328125, 0.0, 15.271026611328125, 5.645751953125, 33.9499626159668, 2.2121505737304688, -54.43621826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000188.npy"}
{"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 29.241382598876953, "std": 38.444149017333984, "min": -75.15435791015625, "p10": -16.999008178710938, "median": 26.27346420288086, "p90": 79.81834716796875, "max": 111.51873779296875, "pos_frac": 0.7890625, "sample": [82.90313720703125, 20.40264892578125, -19.110137939453125, 12.329727172851562, 11.679412841796875, 49.4727783203125, 59.40812683105469, 105.32232666015625, 49.48723602294922, 27.78778076171875, 50.31758117675781, 58.44842529296875, 100.56884765625, 36.38189697265625, 65.78155517578125, 37.72454833984375, 75.32949829101562, -2.8209571838378906, -8.932586669921875, 4.58355712890625, 49.022674560546875, 90.4464111328125, -17.330615997314453, 96.10076904296875, 40.390960693359375, 43.86772155761719, 37.209136962890625, 76.89608764648438, 10.061004638671875, -2.5689544677734375, -25.761932373046875, 4.1320953369140625, 22.499732971191406, 18.959365844726562, -2.3724288940429688, 25.91948699951172, 24.10437774658203, 4.13604736328125, 15.824203491210938, 20.206130981445312, -2.0215721130371094, 79.80047607421875, 69.65509033203125, 29.789886474609375, 60.68511962890625, 50.66841125488281, 43.85211181640625, -16.90594482421875, 77.0194091796875, -75.15435791015625, -31.6400146484375, 26.62744140625, -9.251800537109375, 82.5638427734375, 1.93133544921875, 32.26422119140625, 29.442474365234375, 1.6162872314453125, 36.59732437133789, 15.4266357421875, 79.86004638671875, 38.57362365722656, 78.27981567382812, -6.0953826904296875, 10.53369140625, 8.44155502319336, 8.201154708862305, 4.431480407714844, 7.373313903808594, 6.174171447753906, 46.70933532714844, 2.6946487426757812, -41.20326232910156, 46.1136474609375, 9.178359985351562, 2.098590850830078, 36.5491943359375, 6.660923004150391, 5.500732421875, 22.393524169921875, 1.9178314208984375, 70.23904418945312, 28.88763427734375, -17.216156005859375, 47.514747619628906, 43.40220642089844, 109.71090698242188, -34.06172180175781, -1.6842269897460938, -0.18918228149414062, 16.502197265625, 85.54971313476562, 28.518157958984375, -55.574310302734375, 2.4608306884765625, 0.0, 109.18505859375, 21.123336791992188, 18.579620361328125, 69.18544006347656, 73.9241943359375, 25.827102661132812, 76.99800872802734, 58.497467041015625, 11.2003173828125, 0.0, 29.8946533203125, 104.5679931640625, -18.15338134765625, 51.52008056640625, 76.92236328125, 17.279464721679688, 39.45245361328125, 14.460298538208008, 29.770263671875, -60.47332763671875, 34.954803466796875, 79.36968994140625, 63.96886444091797, 27.53216552734375, 42.28936767578125, 98.90069580078125, -48.223297119140625, 111.51873779296875, 73.091064453125, -27.743118286132812, -5.2340240478515625, -7.510202407836914], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000189.npy"}
{"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 27.152952194213867, "std": 43.66796875, "min": -88.48648071289062, "p10": -28.03462524414062, "median": 29.789947509765625, "p90": 80.09790802001953, "max": 126.82916259765625, "pos_frac": 0.78125, "sample": [-36.71300506591797, 14.533599853515625, 98.7490234375, 30.87896728515625, 0.0578765869140625, 38.9324836730957, -65.11961364746094, -3.7184486389160156, -21.71734619140625, 34.276397705078125, 45.82563781738281, -24.919151306152344, -19.110137939453125, 83.78973388671875, 35.179412841796875, 35.773948669433594, 9.557098388671875, -26.568527221679688, 16.603408813476562, -83.81024169921875, -33.794456481933594, -33.132408142089844, -38.90711975097656, 71.40351867675781, 76.44282531738281, 10.484039306640625, 35.884674072265625, 3.77557373046875, 49.36863708496094, 93.99490356445312, 120.14031982421875, -31.455520629882812, 12.454383850097656, 49.18965148925781, 50.60595703125, -5.616455078125, 57.662506103515625, 3.953014373779297, 54.16912078857422, 9.672409057617188, 81.4647216796875, -84.11886596679688, 4.597015380859375, 100.98948669433594, 67.49767303466797, -0.26397705078125, 61.104217529296875, 42.73474884033203, 68.57357788085938, 104.13983154296875, 16.712234497070312, 12.665111541748047, 56.95935821533203, -2.7925491333007812, 43.98774719238281, 48.023193359375, 6.19232177734375, 0.604888916015625, 36.19012451171875, -13.6578369140625, -3.71295166015625, 75.40470886230469, 36.329498291015625, 8.225250244140625, 31.62677001953125, 34.956207275390625, 28.898895263671875, -15.653045654296875, 55.240509033203125, 54.39678955078125, 54.175621032714844, 18.663410186767578, 50.702362060546875, 0.0, 71.79830932617188, 79.51213073730469, 43.11872482299805, 41.367027282714844, 114.75668334960938, 50.56428909301758, 1.2674560546875, 36.175994873046875, 30.208343505859375, 48.060089111328125, 69.55743408203125, 13.45928955078125, 13.403717041015625, 7.327239990234375, 61.42555236816406, -12.4219970703125, 18.314010620117188, 126.82916259765625, 34.18919372558594, 96.21621704101562, 1.4980087280273438, -88.48648071289062, 14.952224731445312, 2.25860595703125, 73.23542022705078, -49.9112548828125, 51.32208251953125, 21.844207763671875, 73.8802490234375, 79.44100952148438, 34.564483642578125, 55.028839111328125, 26.434288024902344, 13.005706787109375, -0.1011962890625, 29.185394287109375, 5.261268615722656, 116.41421508789062, 10.28913688659668, 29.371551513671875, 5.218780517578125, 8.025726318359375, -15.029159545898438, 54.49433517456055, -68.17568969726562, 67.09490966796875, 52.193359375, 0.7043609619140625, 11.67868423461914, 50.8299674987793, -73.41085815429688, 84.6170654296875, -36.447479248046875, 89.53349304199219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000190.npy"}
{"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 26.41290855407715, "std": 45.97270202636719, "min": -116.780517578125, "p10": -26.98627014160156, "median": 17.919654846191406, "p90": 90.11316680908203, "max": 123.198486328125, "pos_frac": 0.7578125, "sample": [98.98977661132812, 64.37103271484375, -8.117156982421875, -11.826980590820312, 26.270587921142578, 100.80191040039062, 15.380752563476562, 16.529075622558594, 2.9000473022460938, -55.5308837890625, 6.557243347167969, 55.33991241455078, -3.133596420288086, 9.132217407226562, 4.950439453125, -25.600372314453125, 60.822628021240234, -1.62872314453125, 1.293569564819336, 63.09660339355469, 40.662254333496094, 90.340087890625, 10.726951599121094, 7.695442199707031, 0.9760913848876953, 18.181983947753906, -10.873748779296875, -80.4736328125, 61.23370361328125, 52.30085754394531, 8.739799499511719, 8.233919143676758, -17.411270141601562, 88.3929443359375, -0.29766845703125, -30.22003173828125, 39.301849365234375, -7.3672332763671875, -116.780517578125, -33.632568359375, 45.753814697265625, 66.6895751953125, 107.31732177734375, 17.657325744628906, 20.936016082763672, 107.12596130371094, 15.15380859375, 18.52166748046875, 6.4029388427734375, 11.546417236328125, 2.267059326171875, 71.84365844726562, 106.932861328125, 87.17259216308594, 12.24652099609375, -6.60888671875, 60.25499725341797, 90.01591491699219, 28.65087890625, 36.83972930908203, 6.074771881103516, -4.000823974609375, 42.38958740234375, -14.036430358886719, 76.20169067382812, 7.920402526855469, 111.35482788085938, 62.650062561035156, 26.032623291015625, 3.700338363647461, 5.05389404296875, 65.40010833740234, 6.973594665527344, -34.05517578125, 8.540206909179688, -66.54779052734375, 52.56134033203125, 25.072021484375, 55.440338134765625, -6.9337158203125, 0.18719482421875, 39.275054931640625, 2.184886932373047, 95.7667236328125, -51.21208190917969, 123.198486328125, 84.21417236328125, -66.33255004882812, 81.07266235351562, 90.850341796875, 22.255096435546875, 73.30072021484375, 13.78851318359375, 3.183198928833008, -1.47259521484375, -47.9346923828125, 63.439544677734375, 38.50372314453125, 104.10824584960938, 34.45391845703125, 27.08843994140625, 11.27947998046875, 70.38673400878906, 49.657440185546875, 117.8560791015625, 81.57896423339844, 3.8966140747070312, 5.193328857421875, -47.1339111328125, 18.80633544921875, 0.9356689453125, -66.39607238769531, 57.09360885620117, 57.68180847167969, 60.59539794921875, 54.03857421875, -4.406280517578125, 11.549697875976562, -7.3195343017578125, 65.3011474609375, 41.53577423095703, 43.7841796875, 22.344505310058594, -24.2191162109375, 0.0, 76.39089965820312, -49.07582092285156, 112.738525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000191.npy"}
{"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 21.172569274902344, "std": 48.101383209228516, "min": -115.87924194335938, "p10": -37.600759887695304, "median": 17.186983108520508, "p90": 87.86587829589844, "max": 111.75033569335938, "pos_frac": 0.6796875, "sample": [7.3850555419921875, 27.9505615234375, 90.44212341308594, -84.744873046875, 81.20172119140625, 19.998291015625, -63.363555908203125, 15.781387329101562, -21.588581085205078, 93.0491943359375, 32.246978759765625, 31.26959228515625, 62.30016326904297, 89.48504638671875, 88.954345703125, -7.0503997802734375, -54.66697692871094, 55.673492431640625, -115.87924194335938, 111.75033569335938, -80.45361328125, 106.25039672851562, -14.20709228515625, -65.87591552734375, 7.569183349609375, 88.69827270507812, 31.0267333984375, 11.078628540039062, -14.248908996582031, 61.79026794433594, -92.02508544921875, 43.667686462402344, 58.98139953613281, 58.96527099609375, -0.60699462890625, 2.8504505157470703, -16.268112182617188, 64.05343627929688, 19.132766723632812, 17.75494384765625, -0.7593231201171875, -12.1845703125, -34.78578186035156, 66.23367309570312, 83.70477294921875, -28.47607421875, -5.1165771484375, -65.86370849609375, 0.0, 0.0, 2.3430862426757812, -5.029632568359375, -63.585044860839844, 34.650146484375, -67.23057556152344, 65.31373596191406, 6.103515625e-05, 49.372100830078125, -5.269355773925781, -21.01068115234375, 51.28075408935547, 43.91053009033203, -13.715362548828125, 4.123626708984375, 72.2554931640625, 88.06832885742188, 77.36181640625, 85.06295776367188, 2.6301803588867188, 71.148681640625, -43.37261962890625, 4.032501220703125, 16.765899658203125, 1.488555908203125, 33.087791442871094, 55.25470733642578, 42.03948974609375, 87.77911376953125, 89.28805541992188, 25.654312133789062, 17.60806655883789, 80.08624267578125, -19.602943420410156, -35.127105712890625, -5.17230224609375, 19.788909912109375, 6.997226715087891, -2.780853271484375, 64.97206115722656, 69.3265380859375, 1.736083984375, 21.457763671875, -74.97331237792969, 108.71307373046875, 14.204458236694336, 8.842498779296875, -0.63616943359375, -10.131317138671875, 94.605712890625, 49.253143310546875, 11.008010864257812, 26.2999267578125, 10.037109375, -23.800273895263672, 2.525432586669922, 30.43011474609375, 32.99906921386719, 11.488800048828125, 82.8944091796875, 51.20453643798828, 47.22950744628906, -13.2410888671875, 66.34522247314453, 13.178329467773438, 47.1981201171875, 14.855751037597656, 73.67831420898438, -8.283588409423828, -75.91278076171875, -0.294677734375, 4.830528259277344, 91.49200439453125, 52.747230529785156, 37.97700500488281, 20.901718139648438, 91.64596557617188, -16.671539306640625, 83.37825012207031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000192.npy"}
{"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 23.243213653564453, "std": 43.4381217956543, "min": -80.83123779296875, "p10": -27.39306182861328, "median": 19.456491470336914, "p90": 83.02535552978514, "max": 110.4737548828125, "pos_frac": 0.6796875, "sample": [78.09893035888672, -6.265205383300781, -3.5843505859375, -10.231483459472656, -9.267242431640625, 65.64984130859375, 17.42693328857422, 41.100555419921875, 48.718536376953125, 31.81060791015625, 22.407455444335938, 2.502328872680664, 18.03399658203125, 77.50581359863281, 5.952911376953125, 58.9261474609375, -24.41534423828125, 43.81539535522461, -0.205718994140625, 4.117216110229492, 72.65115356445312, -20.61724853515625, 63.25347900390625, 10.94097900390625, 28.812911987304688, 37.242950439453125, 3.202239990234375, -52.83857727050781, 60.29182434082031, -6.01922607421875, -45.74468231201172, -3.2450408935546875, 22.59893798828125, 25.5296630859375, 58.742347717285156, -20.110321044921875, 95.54803466796875, -60.354583740234375, 3.5124759674072266, 34.63105010986328, -75.0577392578125, -23.64410400390625, 0.6559829711914062, 44.945098876953125, 24.624755859375, 43.4150390625, 82.39639282226562, 33.12530517578125, 86.06890869140625, 96.59605407714844, -1.45355224609375, 10.191253662109375, 39.552650451660156, 6.541728973388672, 41.32232666015625, 3.3383255004882812, 100.35001373291016, 24.630706787109375, -11.24139404296875, 37.080810546875, -10.107025146484375, 58.68804168701172, 56.50465393066406, 7.2624969482421875, 100.8115234375, -22.315155029296875, 57.3381462097168, 41.469940185546875, 24.1197509765625, -6.74945068359375, -45.90545654296875, -27.907669067382812, -0.340911865234375, 104.17315673828125, -39.11610412597656, 60.0126953125, 68.6785888671875, 89.36087036132812, 4.74847412109375, 95.778564453125, 73.70132446289062, 31.4390869140625, 74.79605102539062, -21.86578369140625, 97.04443359375, 106.10870361328125, 9.334442138671875, 1.2929916381835938, -15.677688598632812, 20.45892333984375, 60.1136474609375, -4.9646148681640625, 110.4737548828125, 67.9910888671875, 11.79962158203125, 16.048797607421875, 36.40044403076172, 5.069892883300781, 27.59918212890625, 68.3592529296875, 54.451446533203125, 84.49293518066406, 15.695497512817383, -43.61878967285156, 43.517242431640625, -35.13245391845703, 100.45022583007812, -27.172515869140625, -68.26863098144531, -5.20660400390625, -2.5589218139648438, -5.3348388671875, -80.83123779296875, 52.262786865234375, 50.1077880859375, 18.454059600830078, 79.2718505859375, -19.62640380859375, 22.59515380859375, -3.34747314453125, 2.49188232421875, -5.773468017578125, -68.1297607421875, -0.0506591796875, 69.13067626953125, -44.560272216796875, 2.326873779296875, 63.902008056640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000193.npy"}
{"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 18.083086013793945, "std": 43.185821533203125, "min": -91.32186889648438, "p10": -28.961141967773436, "median": 9.777572631835938, "p90": 77.69088745117188, "max": 121.90557861328125, "pos_frac": 0.6328125, "sample": [114.53579711914062, 24.348175048828125, 22.34595489501953, -46.24364471435547, 25.26116943359375, 112.92752075195312, 43.43586730957031, -9.824661254882812, 3.425384521484375, -1.6300392150878906, 75.11043548583984, -23.462875366210938, -3.351633071899414, 5.5773773193359375, 16.054534912109375, 3.138751983642578, -10.828544616699219, 25.58758544921875, 90.60147094726562, -8.24224853515625, 19.384521484375, -4.808263778686523, 38.9588737487793, 6.124725341796875, 13.805389404296875, -19.413772583007812, 3.5593719482421875, 28.069305419921875, -4.863533020019531, -86.3629150390625, -11.751663208007812, 25.241554260253906, -9.320159912109375, 11.532257080078125, 17.323135375976562, 14.5076904296875, 121.90557861328125, 58.567359924316406, 12.910675048828125, 66.67807006835938, -20.681060791015625, 107.45758056640625, -8.5560302734375, -7.177604675292969, -14.699874877929688, 93.44154357910156, -32.58903121948242, 65.36640167236328, 66.96644592285156, 9.293136596679688, -10.025581359863281, 31.055908203125, 28.702072143554688, 35.5462646484375, 114.27685546875, 42.523468017578125, 43.920127868652344, 32.91998291015625, 47.14842224121094, -12.554168701171875, 17.659423828125, 12.33642578125, -4.969451904296875, 63.09104919433594, -49.86370849609375, -33.216156005859375, 8.559951782226562, -91.32186889648438, 32.883056640625, -17.8408203125, 110.24371337890625, -8.105743408203125, -30.931365966796875, 87.27005004882812, -0.593475341796875, 81.47845458984375, -2.8021240234375, -18.76323699951172, 36.06805419921875, 67.22590637207031, -16.476470947265625, 6.948486328125, 72.04708862304688, 79.26187133789062, -5.444206237792969, -63.205108642578125, 53.31749725341797, -6.981864929199219, 65.22708129882812, 77.01760864257812, 46.17425537109375, -68.16183471679688, 16.596298217773438, -42.74421310424805, 5.709745407104492, -1.164306640625, 7.571708679199219, -77.11940002441406, 109.2972412109375, 6.5911712646484375, -4.455474853515625, 53.1278076171875, 3.3472976684570312, 9.15399169921875, -46.67210388183594, 13.85885238647461, -10.5732421875, 29.69500732421875, -6.0894317626953125, -15.604034423828125, 1.193389892578125, 92.13742065429688, 60.017364501953125, 10.262008666992188, 37.46600341796875, 38.575775146484375, 61.613525390625, 2.2010326385498047, -33.06205749511719, 2.9753265380859375, 14.3935546875, -6.595672607421875, 61.742889404296875, -28.11676025390625, 3.8111572265625, -19.471908569335938, 36.51951599121094, 57.16363525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000194.npy"}
{"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 17.588884353637695, "std": 45.09249496459961, "min": -88.58514404296875, "p10": -30.408891296386717, "median": 9.157209396362305, "p90": 76.20562286376953, "max": 129.77682495117188, "pos_frac": 0.609375, "sample": [-10.312149047851562, -1.2211837768554688, -2.1701507568359375, 67.65350341796875, 76.1878662109375, 69.15394592285156, -51.441619873046875, -1.2308349609375, 59.044525146484375, 9.12575912475586, 41.01580810546875, 9.731903076171875, 3.139495849609375, -8.592996597290039, -6.9111175537109375, 20.76251220703125, -12.44439697265625, -20.06109619140625, 20.33245849609375, 29.5111083984375, -16.661270141601562, 17.91766357421875, 76.23295593261719, -13.066329956054688, -14.400146484375, -3.9358367919921875, 0.05579376220703125, -3.1044921875, -31.224044799804688, 11.882080078125, -14.52886962890625, 64.88885498046875, -25.060531616210938, -16.623016357421875, 6.2548828125, 3.2138595581054688, 91.38160705566406, 92.39712524414062, -1.519805908203125, -4.194061279296875, -22.048431396484375, -9.742889404296875, 12.268333435058594, 5.1547698974609375, 71.8121337890625, 5.7876434326171875, 76.19390869140625, -72.04763793945312, -21.4521484375, 42.451873779296875, -42.14910888671875, 41.66748046875, 33.92668151855469, 35.05015563964844, -83.805908203125, 28.10760498046875, -5.314666748046875, -73.53695678710938, 77.62471008300781, -7.28790283203125, 28.582473754882812, -88.58514404296875, 4.981658935546875, 58.504127502441406, -17.864547729492188, -79.05694580078125, -69.30702209472656, -15.3868408203125, -1.7954559326171875, 48.7490234375, 94.42134094238281, 38.69384765625, 66.29782104492188, -12.956741333007812, 32.6319580078125, 89.759033203125, 13.96710205078125, 2.994873046875, 37.70892333984375, 17.646453857421875, 45.43601989746094, 47.7491455078125, 34.1409912109375, 51.228851318359375, 5.4610137939453125, -1.8401031494140625, 1.5982284545898438, -57.28668212890625, -1.3720502853393555, 17.214767456054688, -30.059539794921875, 119.41629028320312, 70.49383544921875, -1.5746002197265625, 69.22147369384766, 51.6971435546875, 129.77682495117188, -2.000335693359375, 68.74748229980469, 98.04690551757812, -40.43787384033203, 52.446746826171875, 9.477432250976562, 70.00225830078125, 5.787811279296875, 10.075904846191406, -12.853973388671875, 65.97775268554688, 52.2811279296875, 76.40287780761719, 107.53067016601562, 8.92510986328125, -81.3001708984375, 34.7030029296875, -29.179168701171875, 7.371772766113281, 88.58343505859375, -66.25128173828125, 12.255950927734375, 66.14689636230469, 9.18865966796875, -3.758392333984375, -2.02899169921875, 52.37513732910156, 74.8106689453125, 48.18721008300781, 78.7110595703125, -13.975143432617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000195.npy"}
{"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 29.51057243347168, "std": 40.487518310546875, "min": -67.6553955078125, "p10": -14.428717041015625, "median": 18.243305206298828, "p90": 79.24295730590819, "max": 122.88998413085938, "pos_frac": 0.7421875, "sample": [-0.3530693054199219, 13.65447998046875, 18.56543731689453, 0.0397796630859375, 17.921173095703125, 9.438446044921875, -20.922698974609375, 4.281421661376953, 64.00335693359375, 0.71990966796875, 4.857185363769531, 84.22151184082031, 45.3594970703125, 46.72906494140625, 102.48211669921875, -2.551544189453125, 1.011932373046875, 3.602508544921875, 15.869583129882812, 21.080078125, 52.80760192871094, 7.242679595947266, 55.7347412109375, 78.22899627685547, -0.41497039794921875, 60.76344299316406, -12.526931762695312, -0.7098464965820312, -57.15277862548828, 65.92388916015625, 107.935546875, 85.2022476196289, 57.58587646484375, 57.0791015625, 1.87860107421875, 55.203125, 8.946563720703125, -40.0679931640625, 71.89376831054688, 51.267845153808594, -24.447952270507812, 96.65768432617188, 38.125213623046875, -10.234878540039062, 15.464397430419922, 11.062568664550781, 16.898773193359375, 98.55099487304688, 22.0274658203125, 52.832183837890625, -30.7564697265625, 0.14703750610351562, 0.6169281005859375, 5.889293670654297, -1.6399993896484375, 81.7406005859375, 76.29412841796875, 3.6320533752441406, -0.13494873046875, 51.71363830566406, 61.08636474609375, 10.170135498046875, 78.64485931396484, -2.4509735107421875, 21.383804321289062, 86.34025573730469, -36.92751693725586, 30.17462158203125, -14.312957763671875, -15.076385498046875, 15.067276000976562, 20.606842041015625, 0.9441814422607422, 63.328346252441406, 47.676795959472656, 74.81959533691406, 2.801116943359375, 21.80670166015625, 59.305633544921875, 45.390411376953125, 12.105827331542969, -48.75035095214844, 58.85545349121094, -3.1037368774414062, -13.177978515625, 68.46722412109375, 66.5814208984375, 75.96517944335938, 68.61090087890625, 1.194854736328125, 74.15020751953125, -67.6553955078125, 22.226898193359375, 108.6204833984375, 53.27423095703125, -4.972816467285156, -12.603385925292969, -14.698822021484375, 73.326904296875, 76.80543518066406, 62.327392578125, -15.947998046875, 8.700714111328125, 122.88998413085938, 92.18270874023438, 17.07569694519043, -27.475868225097656, 70.31212615966797, 78.64314270019531, 80.63851928710938, 66.94233703613281, 108.07171630859375, 74.77297973632812, 75.94171142578125, -6.707244873046875, -5.920722961425781, -6.556022644042969, 64.64682006835938, 60.693260192871094, 37.784454345703125, 1.21087646484375, -6.798364639282227, -42.741668701171875, 54.17057800292969, 12.203933715820312, -2.99884033203125, -8.19976806640625, 4.218788146972656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000196.npy"}
{"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 37.831600189208984, "std": 44.1453857421875, "min": -85.39825439453125, "p10": -13.289411926269526, "median": 41.76480484008789, "p90": 91.03369750976563, "max": 144.2955322265625, "pos_frac": 0.8125, "sample": [-69.40251159667969, 6.24896240234375, -1.26361083984375, 75.96392822265625, 65.30740356445312, -10.733154296875, 67.34394836425781, 23.439559936523438, 21.417156219482422, 88.57586669921875, 96.72314453125, 90.94773864746094, -10.468597412109375, 48.18011474609375, 47.81010437011719, -30.208057403564453, 127.82315063476562, 99.09573364257812, 59.809356689453125, 103.64859008789062, 1.2239456176757812, 54.52672576904297, 91.23426818847656, 24.154434204101562, 17.002784729003906, 32.330291748046875, 70.58746337890625, 90.43032836914062, 80.41921997070312, 82.21630859375, 71.3494873046875, 51.849769592285156, -51.359344482421875, 54.086875915527344, 28.9769287109375, 37.444725036621094, 67.68353271484375, 0.02556610107421875, 20.737361907958984, 91.8927001953125, 41.161170959472656, 56.20989990234375, -85.39825439453125, 88.59130859375, -53.00799560546875, 13.86956787109375, 99.9049072265625, -2.0373077392578125, -20.238739013671875, -11.805740356445312, 57.68218994140625, 53.73564147949219, 21.189727783203125, -16.751312255859375, 6.917930603027344, 2.1627540588378906, 70.25630187988281, 0.318878173828125, 83.107666015625, 69.83876037597656, 62.47649383544922, 144.2955322265625, 5.0028228759765625, 95.9296875, -33.339599609375, 4.807491302490234, 62.93890380859375, 76.0875473022461, 38.40919494628906, 45.756256103515625, 20.751388549804688, 40.052734375, -2.2910079956054688, 72.53604125976562, 1.3416290283203125, 80.77326965332031, 1.11944580078125, 42.368438720703125, -43.30583953857422, 4.55908203125, -4.3336181640625, 9.682968139648438, 50.7279052734375, 54.1845703125, -25.48443603515625, 70.94766235351562, 2.8382568359375, 53.35478210449219, 66.355224609375, 29.495391845703125, 34.200836181640625, -1.2080078125, 85.105712890625, 42.4566650390625, 57.694427490234375, 43.52593994140625, 24.665618896484375, -40.841217041015625, 17.715721130371094, -60.502716064453125, 12.94281005859375, 54.65940856933594, -5.538093566894531, 73.57821655273438, 78.5882568359375, 59.843902587890625, 36.49432373046875, 35.56629943847656, 89.94769287109375, 40.88409423828125, 111.38949584960938, 103.78030395507812, 13.0654296875, 5.8256072998046875, 66.7718734741211, 18.045272827148438, 96.64520263671875, 1.3736324310302734, 73.57015991210938, 16.06622314453125, 0.0, 128.70330810546875, 68.83644104003906, 59.9620361328125, 64.91214752197266, -39.7095947265625, 54.62725830078125, -2.0176010131835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000197.npy"}
{"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 28.347990036010742, "std": 50.04185104370117, "min": -107.44479370117188, "p10": -27.60458946228027, "median": 23.192413330078125, "p90": 92.03709411621094, "max": 127.61065673828125, "pos_frac": 0.7265625, "sample": [75.05328369140625, 20.201751708984375, 10.4312744140625, -15.055572509765625, 62.925079345703125, -0.30816650390625, 74.36993408203125, 92.51828002929688, 35.71038818359375, 16.51971435546875, -14.5380859375, -7.938838958740234, 42.428131103515625, 36.082977294921875, 10.20560073852539, -9.853958129882812, 101.17982482910156, 54.99456787109375, 48.2998046875, 11.5555419921875, 91.83087158203125, -87.8968505859375, -87.46597290039062, 91.22894287109375, 127.61065673828125, 114.73162841796875, -71.51806640625, 12.22967529296875, 16.463821411132812, 19.6502685546875, 84.34747314453125, 74.58009338378906, 77.90422058105469, 43.88983154296875, -0.093994140625, 50.04486083984375, -34.50982666015625, 124.3214111328125, -48.318328857421875, 118.6322021484375, -37.152923583984375, 2.74267578125, 83.60910034179688, 73.56721496582031, -12.720787048339844, 0.20348358154296875, 107.684814453125, -31.967632293701172, -26.934635162353516, 5.073162078857422, 50.09107971191406, 44.756439208984375, -3.144542694091797, 76.69801330566406, 29.240158081054688, 73.51376342773438, 100.3172607421875, 2.7320404052734375, 22.449462890625, 47.7296142578125, 14.239395141601562, 26.69708251953125, -2.5168609619140625, 89.45809936523438, 68.5015869140625, 23.93536376953125, 1.696664810180664, 101.13748168945312, 29.642120361328125, 31.216659545898438, 17.78875732421875, -77.7750244140625, -18.84844970703125, 50.150115966796875, -107.44479370117188, -18.938629150390625, 19.3193359375, 82.43344116210938, -15.140388488769531, 57.295654296875, 26.449798583984375, 80.77772521972656, 52.75267028808594, -13.58465576171875, 72.28755950927734, -0.38910675048828125, 54.13108825683594, 88.82737731933594, -72.70210266113281, 14.8992919921875, 0.0, 87.37811279296875, 11.679931640625, -17.062774658203125, -57.42852783203125, 12.064170837402344, -2.7993030548095703, 61.78917694091797, 97.4873046875, 35.6732177734375, 102.620361328125, 2.5951271057128906, 75.47758483886719, 53.94023132324219, 28.477523803710938, 55.47328186035156, 28.132171630859375, 4.625820159912109, 0.621826171875, -104.7818603515625, 103.90225219726562, 5.46282958984375, 2.4607772827148438, -1.2930679321289062, -5.857490539550781, -5.1278076171875, 108.773681640625, 28.064910888671875, 89.5626220703125, 13.79974365234375, 29.47125244140625, -29.167816162109375, 8.4951171875, -20.36084747314453, 17.05290985107422, 82.6678466796875, 75.5877685546875, 1.85498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000198.npy"}
{"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 27.912246704101562, "std": 44.790504455566406, "min": -85.75686645507812, "p10": -25.622900390624995, "median": 21.367103576660156, "p90": 87.81274185180663, "max": 123.02447509765625, "pos_frac": 0.7421875, "sample": [17.022964477539062, 42.5650634765625, 84.83085632324219, 72.84024047851562, 94.55982971191406, 67.84779357910156, 70.13021850585938, 7.348613739013672, 87.458251953125, 21.3026123046875, 116.11721801757812, 73.47344970703125, 37.592987060546875, 59.0267333984375, 108.94818115234375, -57.22052001953125, 68.14193725585938, 60.591583251953125, 7.615877151489258, -3.035205841064453, -3.8583450317382812, 2.2512893676757812, -1.8071861267089844, 3.394878387451172, 3.4300537109375, 37.90943145751953, 55.83543395996094, 1.9098587036132812, 32.221893310546875, 3.674591064453125, 46.44076156616211, 48.00883483886719, 62.39704132080078, 16.368789672851562, 29.20068359375, 56.35887145996094, -44.29065704345703, -14.045448303222656, 71.71038818359375, 14.285064697265625, 10.96319580078125, 76.24893188476562, 11.934295654296875, 4.983694076538086, 21.431594848632812, 76.59149169921875, -8.52569580078125, -6.510902404785156, -1.7831268310546875, -58.36767578125, 66.56124877929688, 109.79409790039062, 13.365081787109375, 94.64691162109375, -14.113632202148438, 27.399261474609375, 17.8975830078125, 24.152664184570312, 94.68858337402344, -37.92466735839844, -83.254638671875, 72.76882934570312, 81.42681884765625, 27.381011962890625, -11.388370513916016, -43.87888717651367, 55.498207092285156, 12.939666748046875, -17.750099182128906, 30.282806396484375, 104.07736206054688, -1.4896240234375, 4.516809463500977, 111.40362548828125, 10.951522827148438, -3.184478759765625, 47.371734619140625, 25.997207641601562, -2.267192840576172, 22.924896240234375, 100.98049926757812, 8.976936340332031, 75.53546142578125, -10.86773681640625, 72.588623046875, 92.3984375, -68.6304702758789, 65.96826171875, 78.21319580078125, 36.354583740234375, -31.250762939453125, -17.663116455078125, 53.19403839111328, 94.538330078125, -8.09625244140625, 48.25640869140625, 0.40842437744140625, 10.16583251953125, 23.777801513671875, 38.70086669921875, 24.788055419921875, -28.95897674560547, 19.982086181640625, 10.552398681640625, 65.901123046875, -85.75686645507812, -41.643775939941406, 0.8548183441162109, 12.548683166503906, 54.300262451171875, 14.862373352050781, 11.9852294921875, 76.13949584960938, -3.6879959106445312, 68.5570068359375, 17.3863525390625, 88.63988494873047, 2.7064971923828125, -5.886894226074219, 56.86802673339844, 8.787261962890625, -71.08447265625, -5.98779296875, -24.193153381347656, 123.02447509765625, 81.16656494140625, 44.13700866699219, -33.08465576171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000199.npy"}
{"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 24.0122013092041, "std": 52.707496643066406, "min": -100.8817138671875, "p10": -42.171453857421874, "median": 13.46762466430664, "p90": 92.02641906738279, "max": 156.18450927734375, "pos_frac": 0.6875, "sample": [71.6558837890625, -63.912994384765625, 116.45681762695312, 14.067192077636719, -50.065093994140625, 13.334815979003906, -37.961212158203125, 64.29354858398438, -7.008758544921875, 7.107841491699219, -3.0110015869140625, 79.55201721191406, 25.460906982421875, 3.8285903930664062, -89.9205322265625, 78.34130859375, 65.9578857421875, -13.917800903320312, 25.415115356445312, 122.12551879882812, -83.97579956054688, -41.504913330078125, 108.33908081054688, 39.328025817871094, 97.18865966796875, 69.51498413085938, -20.746734619140625, -11.980262756347656, 52.311676025390625, -5.790283203125, -36.494537353515625, -93.25875854492188, 12.406658172607422, -53.7244873046875, 13.600433349609375, 42.23846435546875, -56.839866638183594, 84.27529907226562, 66.15042114257812, -13.11651611328125, 0.0, 54.104393005371094, 90.23300170898438, 11.116050720214844, -2.882568359375, -82.49028015136719, 85.41790771484375, 7.61273193359375, 119.59890747070312, -8.168373107910156, 15.267333984375, 12.402130126953125, 88.22482299804688, 3.80364990234375, -24.202224731445312, -20.964813232421875, 17.301666259765625, -74.7981948852539, 100.92230224609375, 2.8358993530273438, 54.58445739746094, 4.54534912109375, 2.1417789459228516, 70.040771484375, 77.06961059570312, 68.05520629882812, 11.832847595214844, 96.2110595703125, -43.726715087890625, 60.513519287109375, 102.59806823730469, 52.24700927734375, -7.1192474365234375, 41.10523223876953, 45.90248107910156, 24.79046630859375, -7.679229736328125, -5.210670471191406, 7.231414794921875, 64.794677734375, -80.94802856445312, 99.4105224609375, 156.18450927734375, 0.5184326171875, 9.956573486328125, 47.19056701660156, -9.672866821289062, 67.62199401855469, -100.8817138671875, 88.225341796875, 10.787246704101562, 118.788330078125, -56.78083038330078, 88.72000122070312, 56.865386962890625, 85.06991577148438, 5.112586975097656, 103.61370849609375, -27.7137451171875, -31.176681518554688, 16.679931640625, 45.57135009765625, 73.97769165039062, 34.57398223876953, -15.431396484375, 81.54020690917969, 12.707328796386719, -29.484893798828125, 4.735626220703125, 0.56365966796875, 10.416984558105469, 69.696533203125, 61.9228515625, 116.4296875, 30.79528045654297, 38.3729248046875, -6.320335388183594, -21.06562042236328, 62.42195129394531, 39.01865005493164, 7.862518310546875, -2.1718826293945312, 59.509910583496094, 5.146614074707031, -7.57666015625, 14.504287719726562, 31.114044189453125, 2.175323486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000200.npy"}
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 28.859167098999023, "std": 43.92090606689453, "min": -81.22341918945312, "p10": -20.959474182128897, "median": 23.218196868896484, "p90": 87.07900314331054, "max": 128.06692504882812, "pos_frac": 0.7265625, "sample": [-51.1292724609375, -14.29962158203125, 57.01026153564453, 75.6474609375, 76.81536865234375, 48.21966552734375, 22.934463500976562, 20.456390380859375, 25.981536865234375, 65.4729995727539, 82.32275390625, 2.6235437393188477, -5.026771545410156, 23.679962158203125, -48.10528564453125, -37.821624755859375, 78.65965270996094, 96.89443969726562, 56.9462890625, 15.895111083984375, 13.905288696289062, 72.09742736816406, 86.78116607666016, 0.0, 66.1647720336914, 38.25791931152344, 75.25701904296875, 72.4927978515625, 2.2030563354492188, 91.49179077148438, 98.46966552734375, 53.50482177734375, 17.074356079101562, -8.152214050292969, 0.0, -60.1976318359375, 8.720626831054688, -1.1994705200195312, 9.891822814941406, 6.48095703125, 34.136253356933594, 48.82720947265625, 79.90036010742188, 75.89786529541016, -44.817901611328125, 16.456085205078125, 52.60985565185547, 62.01752471923828, -18.579025268554688, 48.369781494140625, -3.4037322998046875, 23.501930236816406, 24.9080810546875, 62.129638671875, -2.6255435943603516, 22.7314453125, 27.770423889160156, 83.82246398925781, 48.207611083984375, 78.86167907714844, -6.193992614746094, 50.185516357421875, 101.3980712890625, 12.975433349609375, 92.49078369140625, 68.76714324951172, 32.224578857421875, 0.0, 9.343887329101562, -81.22341918945312, 17.12677001953125, 10.421554565429688, 0.292755126953125, 116.1595458984375, 41.702545166015625, 9.12152099609375, 1.6630096435546875, 8.694061279296875, -10.84661865234375, 36.59307861328125, 25.464279174804688, 31.83380126953125, -14.473472595214844, -13.030776977539062, -2.10406494140625, 92.47341918945312, -7.461387634277344, 87.77395629882812, -26.51385498046875, 128.06692504882812, 57.22845458984375, 62.9364013671875, 58.687713623046875, -5.57293701171875, -30.7423095703125, -18.054779052734375, 10.185028076171875, 89.51593017578125, 6.480472564697266, 89.21084594726562, 76.08750915527344, 71.72598266601562, -59.3995361328125, 65.28530883789062, 17.850296020507812, -6.648895263671875, -32.85917663574219, 33.3719482421875, 84.62368774414062, -53.93701171875, 21.420501708984375, 25.572540283203125, 16.3807373046875, 0.123748779296875, -3.4769668579101562, 105.702880859375, 9.716949462890625, 53.672325134277344, 52.57720947265625, 86.4979248046875, 8.285816192626953, -57.795440673828125, 10.97430419921875, 96.76913452148438, -6.7725677490234375, 65.49485778808594, -13.696296691894531, -63.5179443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000201.npy"}
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 28.030349731445312, "std": 50.37781524658203, "min": -106.9072265625, "p10": -22.943344116210938, "median": 15.671173095703125, "p90": 99.79293060302734, "max": 114.89273071289062, "pos_frac": 0.703125, "sample": [106.35394287109375, 6.526100158691406, 11.7025146484375, 15.646331787109375, 91.80006408691406, -0.11205291748046875, -13.117218017578125, 103.84776306152344, -50.93499755859375, -14.23077392578125, -6.625244140625, 99.81915283203125, 22.51568603515625, 65.171630859375, 69.05200958251953, -2.08734130859375, -21.445098876953125, 71.04158782958984, 5.455558776855469, 111.11007690429688, 12.945789337158203, 77.35325622558594, -8.672027587890625, 16.643844604492188, -23.07568359375, 9.1524658203125, 105.78271484375, 0.9218978881835938, 55.87748718261719, 12.90972900390625, -43.26647186279297, 28.966888427734375, -18.902587890625, 102.27520751953125, 50.14495849609375, 75.99992370605469, 65.6546630859375, 108.70022583007812, 11.130516052246094, -68.17546081542969, 2.0709457397460938, 107.181640625, 63.893653869628906, 79.44561767578125, -1.2486114501953125, 55.64404296875, 46.0576171875, 30.473480224609375, 82.58123779296875, -5.7519683837890625, -78.40766906738281, 65.39390563964844, 86.14643859863281, 1.7423095703125, -77.17852783203125, 57.76737976074219, 52.72222900390625, -11.158447265625, 27.263153076171875, -9.443351745605469, 26.444580078125, 68.30143737792969, 101.02006530761719, -26.673004150390625, -9.108154296875, -17.486053466796875, 107.94528198242188, 0.2768402099609375, 14.361083984375, 7.3017425537109375, 2.3514022827148438, 83.36993408203125, 112.46600341796875, 1.9459075927734375, 65.02362060546875, 95.77604675292969, 70.64694213867188, 38.96600341796875, 91.32585144042969, 79.72027587890625, 0.0, 114.89273071289062, 0.7678680419921875, 15.696014404296875, 17.50726318359375, -0.4354400634765625, 0.5325775146484375, -16.57940673828125, 5.946540832519531, -50.953399658203125, 58.4786376953125, 79.1453857421875, 3.6344146728515625, 18.350448608398438, 97.06460571289062, 59.264976501464844, -20.379531860351562, -22.886627197265625, -20.67742919921875, -28.6019287109375, -106.9072265625, 96.18533325195312, 36.812889099121094, 83.449951171875, 36.015350341796875, -1.2060546875, -13.6630859375, 16.4542236328125, 28.644729614257812, 0.1510639190673828, 61.10198974609375, 76.10563659667969, -56.82825469970703, 11.28564453125, 71.94610595703125, -102.03117370605469, 13.9854736328125, 71.27467346191406, -61.399078369140625, -13.21746826171875, 5.893465042114258, 86.72525024414062, 1.4055976867675781, -2.5997180938720703, 6.3800506591796875, 99.78169250488281, -13.027877807617188, 111.36978149414062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000202.npy"}
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 21.55194664001465, "std": 38.65181350708008, "min": -94.3304443359375, "p10": -23.98894805908203, "median": 12.806145668029785, "p90": 76.20313110351563, "max": 121.31248474121094, "pos_frac": 0.7109375, "sample": [2.1589412689208984, 9.997322082519531, 8.19024658203125, 69.66588592529297, 54.129058837890625, 32.80015563964844, 10.555105209350586, 13.636581420898438, 46.66314697265625, 69.78898620605469, 72.21795654296875, 30.84161376953125, 4.609897613525391, 40.060028076171875, 37.36268615722656, 38.78515625, -13.980972290039062, -52.673927307128906, -8.247627258300781, 8.72113037109375, 18.214492797851562, 0.0, -41.72503662109375, 57.44432067871094, -25.674392700195312, 10.905181884765625, -23.34112548828125, 48.49462890625, 2.11932373046875, -3.3608856201171875, 26.78021240234375, 42.97206115722656, 30.144287109375, 78.8634033203125, 53.85078430175781, -7.196987152099609, 29.605682373046875, 109.34036254882812, 18.029937744140625, -12.253280639648438, 39.90379333496094, 85.52247619628906, 26.71630859375, 64.3880844116211, 2.3388710021972656, -5.554718017578125, 10.985984802246094, 60.91841125488281, -4.83184814453125, 61.528053283691406, 0.18270111083984375, 23.92919158935547, 27.373626708984375, 40.09160614013672, 121.31248474121094, -20.315948486328125, 49.815765380859375, -1.5882091522216797, 80.83197021484375, 117.04660034179688, 5.503761291503906, 47.478118896484375, 55.962677001953125, -2.842193603515625, -57.647552490234375, 83.99549102783203, -46.829498291015625, 76.06494140625, 10.237133026123047, -0.506195068359375, 32.833282470703125, 3.5531082153320312, -94.3304443359375, 68.1728515625, -6.0400390625, -2.8143482208251953, -26.309600830078125, 1.43218994140625, -28.091278076171875, 2.027496337890625, 32.340301513671875, 11.660781860351562, -48.3206787109375, 48.43363952636719, 79.56736755371094, 18.031692504882812, 57.60003662109375, 22.28955078125, 93.80047607421875, 76.52557373046875, 12.528297424316406, -6.0189208984375, -10.09075927734375, 58.888427734375, 7.919950485229492, -18.874710083007812, 38.07666015625, -15.000923156738281, 75.43000030517578, 10.640335083007812, 71.67845153808594, 2.4693756103515625, -3.88787841796875, 12.512481689453125, 13.083993911743164, -53.2061767578125, -4.605194091796875, -31.242828369140625, 14.808441162109375, 101.9141845703125, 0.0, 13.498138427734375, 23.701629638671875, 29.547943115234375, 0.3766670227050781, -3.578277587890625, 28.916046142578125, 82.19317626953125, 10.720848083496094, -2.55303955078125, 34.922576904296875, -35.77294921875, 8.235748291015625, 84.53521728515625, 12.3599853515625, -25.500534057617188, 11.808980941772461, 25.345434188842773], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000203.npy"}
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 31.898340225219727, "std": 44.2192497253418, "min": -82.30755615234375, "p10": -17.448995971679686, "median": 24.112075805664062, "p90": 89.9466323852539, "max": 140.6654052734375, "pos_frac": 0.7734375, "sample": [109.59573364257812, 70.85948181152344, 110.86572265625, 7.9158477783203125, 14.140647888183594, 90.02294921875, 5.957096099853516, 56.755943298339844, 4.75547981262207, 102.66885375976562, 12.4251708984375, 9.288116455078125, 71.30615234375, -17.2396240234375, -37.81573486328125, -82.30755615234375, 15.963836669921875, -30.53973388671875, 8.121772766113281, 23.450286865234375, 85.37033081054688, 5.958213806152344, 43.76544189453125, 55.73280334472656, 24.77386474609375, 3.84552001953125, 11.176673889160156, -47.031494140625, 12.957382202148438, 26.654541015625, 32.231903076171875, 140.6654052734375, 3.1591339111328125, -30.235183715820312, 40.06072998046875, -13.668212890625, 8.127067565917969, 119.3236083984375, -2.67010498046875, 79.51589965820312, 100.90516662597656, 13.201889038085938, 47.003173828125, -15.238750457763672, 71.57803344726562, 74.21873474121094, 13.171173095703125, 113.4468994140625, 85.99758911132812, 105.22833251953125, 74.5130844116211, 59.4652099609375, -3.20709228515625, 63.985984802246094, 32.40153503417969, -2.914470672607422, 72.95858764648438, 29.64252471923828, -2.3326416015625, -29.399261474609375, 32.476470947265625, 72.5206298828125, 29.640472412109375, -17.937530517578125, -8.698318481445312, 106.30769348144531, 53.507720947265625, 72.8043212890625, -73.333984375, 81.777587890625, -8.228113174438477, 76.6055908203125, -0.8896331787109375, -12.6929931640625, 5.94781494140625, 4.632610321044922, 30.681316375732422, 83.53731536865234, -2.9767227172851562, 19.099365234375, 7.284858703613281, -1.50396728515625, -36.59254455566406, 11.474853515625, 66.810302734375, -6.6015625, 67.71292114257812, 15.712600708007812, 76.7867431640625, 67.62820434570312, 88.57565307617188, 15.0888671875, -37.780426025390625, 90.74371337890625, 9.961723327636719, 2.057830810546875, 81.90750122070312, -37.908447265625, 19.017013549804688, 86.8836669921875, 9.766334533691406, -1.32666015625, 8.81561279296875, 45.90312194824219, 66.0363540649414, 103.2105712890625, 25.487594604492188, 33.13427734375, 63.40362548828125, 40.30560302734375, 58.06439208984375, 1.0193328857421875, 50.754150390625, 76.02536010742188, 12.46514892578125, 27.57421875, 54.03865051269531, 3.333953857421875, -7.7054443359375, 27.22119140625, -36.065086364746094, 11.222152709960938, 89.91392517089844, 51.95158386230469, -37.13037109375, 115.18222045898438, 5.784523010253906, 2.030517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000204.npy"}
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 30.460939407348633, "std": 46.24976348876953, "min": -78.090576171875, "p10": -24.085264587402342, "median": 28.694347381591797, "p90": 89.22168884277343, "max": 133.86785888671875, "pos_frac": 0.78125, "sample": [40.246246337890625, 0.0, -3.5422515869140625, 3.1561203002929688, 8.99471664428711, 78.94546508789062, -25.503509521484375, -23.119720458984375, -24.929214477539062, 0.9087657928466797, 133.86785888671875, 66.70938873291016, 5.665443420410156, 11.943977355957031, -13.8057861328125, 6.790309906005859, 110.85699462890625, 101.7607421875, 75.1220703125, 11.991851806640625, 10.007476806640625, 21.80768585205078, -17.52178955078125, 93.83804321289062, 56.04595947265625, 13.6737060546875, 68.99411010742188, 16.369796752929688, 102.79638671875, 38.620880126953125, 50.04107666015625, 52.72551345825195, 2.3991851806640625, 7.727691650390625, 23.97314453125, 6.6729583740234375, 64.90685272216797, 11.51361083984375, 52.45526885986328, 8.291664123535156, 29.1614990234375, 0.279052734375, -46.92071533203125, 79.4146728515625, 14.196929931640625, 49.48095703125, 0.0, 30.52703857421875, -65.4787368774414, 31.060333251953125, 2.5654144287109375, 22.328941345214844, 87.95321655273438, 93.0965576171875, -58.363525390625, 30.082443237304688, 81.18690490722656, 28.698455810546875, 3.01275634765625, 56.6185302734375, -73.91200256347656, -1.97808837890625, 7.301727294921875, 41.87663269042969, -23.72357177734375, 5.1119384765625, 92.18145751953125, -3.329498291015625, -64.11917114257812, 125.90435791015625, 78.4600830078125, 74.89022827148438, -55.150482177734375, 44.251861572265625, -2.5052108764648438, 73.19657897949219, 5.7589569091796875, 69.01936340332031, 68.89120483398438, -5.87530517578125, 31.219741821289062, 111.26374816894531, 64.66000366210938, 83.63800048828125, 38.57855224609375, 8.714286804199219, -69.69610595703125, 3.07623291015625, 70.40018463134766, 70.66511535644531, 112.8731689453125, 72.32156372070312, 114.25445556640625, -22.937957763671875, 32.3363037109375, 97.17051696777344, 82.01083374023438, 9.644577026367188, 15.41191291809082, 2.7252349853515625, 111.269775390625, 69.36065673828125, -5.50653076171875, -42.627166748046875, 44.881134033203125, 56.21435546875, 19.828857421875, 69.88818359375, 63.861297607421875, 54.427520751953125, 28.69023895263672, 10.58917236328125, -78.090576171875, 7.567119598388672, 40.8116455078125, -29.713623046875, -67.23912048339844, 87.28160095214844, -8.984077453613281, 66.79338073730469, 56.411865234375, 57.66343688964844, 1.0373687744140625, 52.368499755859375, 9.552200317382812, 82.59259033203125, 0.0, 35.18994903564453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000205.npy"}
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 28.661041259765625, "std": 43.67868423461914, "min": -114.20599365234375, "p10": -24.38556671142578, "median": 21.681991577148438, "p90": 88.59205474853516, "max": 111.98150634765625, "pos_frac": 0.7578125, "sample": [94.49761962890625, 23.472843170166016, -67.24594116210938, 75.45217895507812, 110.09674072265625, -0.9964141845703125, -18.411468505859375, 36.751373291015625, 15.39044189453125, 86.8967514038086, 25.04596710205078, 88.93559265136719, 88.32040405273438, -18.619258880615234, -28.2615966796875, 27.0291748046875, 47.13623046875, 7.265605926513672, -4.430320739746094, 90.53901672363281, 0.5855445861816406, 77.578369140625, -4.16297721862793, -27.396438598632812, -4.586578369140625, 7.0080413818359375, -37.90863037109375, 42.787506103515625, 59.46656036376953, 7.344810485839844, 3.025146484375, -10.18682861328125, -46.69416809082031, 72.99125671386719, 73.08721923828125, -12.97188949584961, 109.20870971679688, 88.44482421875, 83.23095703125, 53.477256774902344, 15.111846923828125, 29.71501922607422, -5.1036224365234375, 27.90606689453125, 46.259002685546875, 30.4976806640625, -25.107879638671875, 88.02639770507812, 59.103668212890625, 79.46875, 41.792816162109375, 62.10374450683594, 65.62645721435547, 41.19940185546875, 2.6628246307373047, 41.16795349121094, 54.25657653808594, 106.03948974609375, -114.20599365234375, 22.442474365234375, 47.62371826171875, -31.411705017089844, -24.076004028320312, 73.22360229492188, 28.907913208007812, 13.5107421875, 19.1861572265625, 13.276397705078125, -4.835628509521484, 49.21978759765625, 12.8944091796875, 74.42524719238281, 43.4376220703125, -65.84344482421875, 93.771728515625, 3.187469482421875, -11.523468017578125, 39.40101623535156, 9.84442138671875, 2.006206512451172, -17.608978271484375, -31.099212646484375, 96.9798583984375, 69.4395751953125, 71.76332092285156, 13.23480224609375, 49.66693115234375, -34.710060119628906, 9.987041473388672, 0.7125511169433594, 64.45684814453125, 72.73696899414062, 64.74383544921875, 7.622341156005859, -17.162261962890625, 46.040283203125, 19.60321044921875, -12.025398254394531, 3.55694580078125, -3.647258758544922, 72.635986328125, 20.9215087890625, 101.42636108398438, 88.973388671875, 6.1722412109375, 76.44515991210938, -37.07281494140625, -9.493095397949219, -37.400115966796875, 18.19671630859375, 2.403472900390625, 111.98150634765625, 74.55104064941406, 4.245941162109375, 1.8664016723632812, 95.57672119140625, 64.24334716796875, 6.394981384277344, 89.21983337402344, 36.7738037109375, 6.990478515625, 9.59429931640625, 1.168365478515625, 73.64358520507812, 44.993896484375, 8.264266967773438, -16.5184326171875, 9.740692138671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000206.npy"}
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 24.486730575561523, "std": 53.78083801269531, "min": -108.065185546875, "p10": -32.03695220947265, "median": 11.7154541015625, "p90": 96.19310607910155, "max": 147.3148193359375, "pos_frac": 0.703125, "sample": [74.870361328125, 49.38883972167969, 14.83306884765625, -9.791175842285156, 17.8021240234375, 7.314277648925781, 26.739776611328125, 121.49850463867188, 47.84016418457031, -4.482366561889648, 28.33506202697754, 128.3768310546875, -61.8839111328125, 4.509178161621094, -5.106201171875, 86.68148040771484, 1.755950927734375, 42.83183288574219, 134.63601684570312, -13.34417724609375, 82.20613098144531, -13.488357543945312, -101.00967407226562, -17.25824737548828, 71.5677490234375, 32.28434753417969, -10.893539428710938, 12.479217529296875, -71.5242919921875, 65.69062042236328, 79.21182250976562, 15.620040893554688, 9.637725830078125, -30.5570068359375, 72.45883178710938, 77.32150268554688, 16.268798828125, -2.4251747131347656, 21.274799346923828, 101.81100463867188, -12.144195556640625, 8.274642944335938, 53.337554931640625, 8.8336181640625, 10.0985107421875, -23.262603759765625, 3.513336181640625, 3.7829437255859375, -89.22860717773438, 8.060760498046875, -108.065185546875, -17.4644775390625, -47.61773681640625, 30.3468017578125, 86.69952392578125, 0.0, 10.274078369140625, -30.582489013671875, 54.0859375, -5.831901550292969, 112.47637939453125, -11.556442260742188, -57.31360626220703, 28.496368408203125, 15.006866455078125, -92.81561279296875, 98.77610778808594, -68.94029998779297, 7.5893707275390625, -2.7135162353515625, 48.2069091796875, -17.94732666015625, 26.970458984375, 87.47726440429688, 102.7427978515625, 93.04824829101562, 66.56607818603516, -21.238174438476562, 2.3842620849609375, 95.11540222167969, 0.2257061004638672, -69.12399291992188, 6.49456787109375, 34.05661392211914, 10.061086654663086, 147.3148193359375, 13.7476806640625, 78.64473724365234, -28.824249267578125, 1.219146728515625, 9.44305419921875, 8.598495483398438, -30.61358642578125, 105.78660583496094, 20.60406494140625, 70.73580932617188, 51.010986328125, 2.722259521484375, 54.26874542236328, -84.12088012695312, 105.01708984375, 89.30174255371094, -15.2158203125, 82.02119445800781, 6.6800537109375, 115.82705688476562, 9.700469970703125, 5.076751708984375, 9.680292129516602, 81.38717651367188, 10.221343994140625, 57.819610595703125, -0.9112319946289062, 98.70774841308594, 92.01361083984375, -7.76617431640625, 84.029541015625, 85.45001220703125, 113.14471435546875, -18.89208984375, 10.951690673828125, -71.41748046875, 81.81283569335938, 86.01457214355469, -35.35813903808594, 23.561553955078125, 56.61004638671875, 17.6578369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000207.npy"}
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 27.744281768798828, "std": 47.2112922668457, "min": -117.2005615234375, "p10": -14.365550613403318, "median": 18.016963958740234, "p90": 88.61479187011717, "max": 144.70587158203125, "pos_frac": 0.734375, "sample": [-3.1470947265625, -8.793594360351562, -28.9246826171875, 19.613800048828125, -91.72705078125, 5.530029296875, -6.012979507446289, 25.890594482421875, 5.4903106689453125, 82.72161865234375, -11.255054473876953, 7.76068115234375, 59.052734375, 48.90507507324219, 47.423397064208984, 123.4935302734375, 17.313369750976562, 12.427223205566406, 18.51776123046875, 135.47201538085938, -89.7879638671875, 82.93644714355469, -2.167083740234375, 20.32806396484375, 14.8507080078125, 72.1905517578125, 115.20370483398438, 73.28700256347656, 15.645240783691406, -12.575576782226562, 144.70587158203125, 62.83971405029297, 58.64752197265625, 59.89910888671875, -13.909065246582031, 106.59658813476562, -117.2005615234375, 15.625656127929688, 17.632476806640625, 4.4365997314453125, 30.004549026489258, -101.33157348632812, 79.57369995117188, -15.430683135986328, 84.65008544921875, 31.808349609375, 56.54052734375, 107.09197998046875, 0.0, 130.06060791015625, -33.49357604980469, 2.1443490982055664, 22.902999877929688, -29.793167114257812, 3.004180908203125, 128.76129150390625, 10.488265991210938, 28.252120971679688, 17.880218505859375, 94.17044067382812, 41.298152923583984, 37.933624267578125, -10.094688415527344, 51.5936279296875, -5.3473358154296875, 5.02490234375, 4.926177978515625, -0.7576904296875, 25.192848205566406, 59.18357849121094, 129.39990234375, 0.0, 25.542633056640625, 58.92510986328125, 12.393436431884766, 87.4117431640625, -7.896629333496094, 87.36199951171875, 19.509521484375, 65.46772003173828, -3.354248046875, 45.3812255859375, 79.33223724365234, 95.28359985351562, -7.0867919921875, 56.89833068847656, -9.307952880859375, -1.45574951171875, 0.4066925048828125, 37.859527587890625, -12.83551025390625, 24.880859375, 52.503204345703125, 12.991851806640625, 75.68521118164062, 33.16839599609375, 75.4761734008789, 24.66058349609375, 98.23553466796875, -16.362586975097656, 3.2399749755859375, 2.851806640625, 6.489044189453125, 14.567615509033203, 17.23065948486328, 10.360115051269531, -2.2347183227539062, 77.35321044921875, 18.153709411621094, 26.968307495117188, 23.081405639648438, -26.86273193359375, 68.79460906982422, 14.163848876953125, 6.19000244140625, -51.564910888671875, 15.723579406738281, 0.9424247741699219, -27.553756713867188, 52.71112060546875, 61.642547607421875, 17.215606689453125, -11.296550750732422, 91.42190551757812, 48.72125244140625, -38.94232177734375, -2.79046630859375, 49.0401611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000208.npy"}
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 23.462268829345703, "std": 48.79105758666992, "min": -105.1490478515625, "p10": -32.877685546875, "median": 14.776702880859375, "p90": 87.88175048828126, "max": 125.44158935546875, "pos_frac": 0.625, "sample": [118.60105895996094, -8.90411376953125, 76.69035339355469, 76.29016876220703, -8.386260986328125, -4.118560791015625, 28.005157470703125, -31.55224609375, -38.039154052734375, 20.69427490234375, 84.88650512695312, -12.1004638671875, -5.176239013671875, 7.009492874145508, 113.392822265625, 89.68358612060547, 30.245254516601562, 5.440254211425781, 8.842864990234375, 94.73617553710938, 94.81048583984375, -21.595245361328125, 9.10101318359375, -68.53196716308594, 22.448944091796875, 7.8483734130859375, 12.604053497314453, 8.6976318359375, -11.357086181640625, 75.64875793457031, 80.4912109375, -51.802490234375, -0.9698486328125, -58.119140625, -43.04962158203125, -57.71295166015625, 118.41036987304688, 5.804534912109375, 0.0, 11.434600830078125, -32.21636962890625, -48.68310546875, 62.22401428222656, -32.7755126953125, 14.336273193359375, 56.27142333984375, 37.00979995727539, 4.878284454345703, -19.88818359375, 102.75444030761719, 40.592315673828125, 33.74079895019531, 61.4921875, -32.397308349609375, 39.8690185546875, 0.650054931640625, -20.26617431640625, 14.180313110351562, 71.50090026855469, -14.974008560180664, 22.095550537109375, -23.503189086914062, -6.184234619140625, 85.58392333984375, 88.025390625, 27.570755004882812, -13.661712646484375, -20.407470703125, 102.93569946289062, -14.07525634765625, 20.9112548828125, 96.5308837890625, 15.217132568359375, 34.16074752807617, -1.4913330078125, 0.0, 58.7440185546875, 77.05079650878906, -22.927215576171875, 79.9403076171875, -16.462615966796875, -5.3568878173828125, 58.62861633300781, -0.69818115234375, 34.401824951171875, 87.47238159179688, 78.70918273925781, -62.17704772949219, 15.231842041015625, 8.8480224609375, 28.904991149902344, -64.29066467285156, -19.57977294921875, 65.74922180175781, -2.056673049926758, -2.01666259765625, 30.76727294921875, 50.052589416503906, 125.44158935546875, -105.1490478515625, -9.857559204101562, 5.232124328613281, -45.531951904296875, -8.282058715820312, -2.8514404296875, -78.25736999511719, -9.148818969726562, 33.9354248046875, 75.0357666015625, 78.05902099609375, 40.71308898925781, 81.00736999511719, 74.25526428222656, 24.482803344726562, 61.04121398925781, 49.65936279296875, 35.92274475097656, 83.58370971679688, 3.607177734375, 118.9586181640625, 34.32476806640625, -33.1160888671875, 59.920196533203125, 41.635223388671875, 99.426025390625, 83.06304931640625, -13.105308532714844, 87.8201904296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000209.npy"}
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 30.136491775512695, "std": 49.42402267456055, "min": -104.83062744140625, "p10": -29.153161621093748, "median": 27.276641845703125, "p90": 89.0184814453125, "max": 173.3775634765625, "pos_frac": 0.703125, "sample": [-7.4139556884765625, -4.007240295410156, 44.50787353515625, -5.0601806640625, 77.26504516601562, 28.928421020507812, 107.86187744140625, 112.68179321289062, 88.91619873046875, 73.03608703613281, 88.5718994140625, 66.7205810546875, 2.906585693359375, 67.8853759765625, 71.55487060546875, -1.495941162109375, 69.70780944824219, -5.294670104980469, 63.68273162841797, 6.4403533935546875, 32.63261413574219, -0.1299285888671875, -59.9862060546875, 34.0257568359375, 65.32380676269531, 52.20671081542969, 5.72991943359375, 173.3775634765625, -54.14166259765625, -6.075164794921875, 58.63372802734375, 27.463226318359375, 17.59710693359375, -0.300537109375, 77.77593994140625, -15.217926025390625, 5.44401741027832, 55.9527587890625, 0.8919677734375, 30.06488037109375, 1.4351806640625, 79.37980651855469, 18.56975746154785, -7.249977111816406, 12.528022766113281, 65.3851318359375, 29.428558349609375, -6.887359619140625, 75.87380981445312, 13.263151168823242, 27.090057373046875, -1.722625732421875, 58.034027099609375, 65.82672119140625, 39.83092498779297, 65.43941497802734, 9.89447021484375, -43.84593200683594, 101.61126708984375, -7.232513427734375, 56.57965087890625, 61.916748046875, 83.87094116210938, 70.63212585449219, 30.9456787109375, 47.6004638671875, -86.70880126953125, -8.907440185546875, 7.39697265625, 83.93621826171875, -73.61297607421875, -29.688201904296875, 88.1309814453125, 84.69314575195312, 95.32485961914062, 91.70578002929688, 0.5551185607910156, 8.966064453125, -10.58990478515625, -104.83062744140625, 73.31481170654297, 9.537338256835938, -1.790212631225586, 94.38531494140625, 54.376747131347656, 84.20635986328125, 124.9603271484375, 61.768768310546875, -3.22210693359375, -36.33039855957031, -22.641464233398438, 13.698623657226562, 3.027219772338867, -30.092727661132812, 25.159536361694336, 70.9705810546875, -30.747467041015625, 116.41627502441406, -16.64794921875, 54.57830810546875, 59.74017333984375, 124.5247802734375, 7.922605514526367, -7.11151123046875, 101.92825317382812, 24.332611083984375, 96.63495635986328, -35.08168029785156, 9.233856201171875, -18.271392822265625, 56.32969665527344, 27.9573974609375, -8.152496337890625, 18.203216552734375, 18.725143432617188, -96.36140441894531, 15.034706115722656, 89.25714111328125, 31.79583740234375, -73.26365661621094, 73.1715087890625, 0.290283203125, 74.00949096679688, 58.009033203125, -28.923858642578125, -7.706451416015625, -6.638763427734375, 63.72096252441406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000210.npy"}
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 39.02486801147461, "std": 44.98902130126953, "min": -74.30116271972656, "p10": -10.554984283447265, "median": 37.79566192626953, "p90": 96.73341369628906, "max": 142.17495727539062, "pos_frac": 0.8125, "sample": [68.73636627197266, 63.71272277832031, 112.7135009765625, 36.815643310546875, 77.18367767333984, 96.3363037109375, 13.162410736083984, 6.283592224121094, -13.405746459960938, 63.30078125, 89.12432861328125, -71.44644165039062, 40.963287353515625, 44.7025146484375, -11.343803405761719, 87.35873413085938, 1.723876953125, -0.048187255859375, -4.516571044921875, -17.70556640625, 10.188591003417969, 72.93952941894531, 19.744834899902344, 47.023681640625, 77.34088134765625, 26.710296630859375, 8.7392578125, -57.004730224609375, 73.88584899902344, -10.2169189453125, 0.09550857543945312, 76.5362777709961, 0.526580810546875, 70.62987518310547, -6.301582336425781, 65.4385986328125, 26.332067489624023, -54.18775177001953, 52.98029327392578, 77.84747314453125, 4.48974609375, 26.002288818359375, 8.979095458984375, 0.060028076171875, -3.193359375, 126.2779541015625, 94.52984619140625, 55.82666015625, 24.4942626953125, 9.855396270751953, 67.82492065429688, -19.718048095703125, 3.95947265625, 3.353363037109375, 142.17495727539062, 74.33853149414062, 96.30563354492188, -19.32628631591797, 21.158187866210938, 63.28466796875, -7.415374755859375, -25.064666748046875, 13.886138916015625, 131.51904296875, 6.5637359619140625, 11.054824829101562, -4.1060791015625, 62.97303771972656, 97.66000366210938, 16.050357818603516, -25.56341552734375, -1.87744140625, -2.23834228515625, 68.53288269042969, 1.273874282836914, 12.93199348449707, 77.97311401367188, 65.25515747070312, 75.3352279663086, 31.891326904296875, 57.024169921875, 60.16229248046875, -0.1032867431640625, 84.287109375, 67.5777587890625, 127.55610656738281, 25.624847412109375, 2.7464141845703125, 70.852294921875, 8.751981735229492, -14.6387939453125, 120.94198608398438, 104.34918212890625, -34.514617919921875, 102.58258056640625, 23.061988830566406, 91.8094482421875, 50.442047119140625, 12.730499267578125, 9.186614990234375, 76.10726928710938, 112.72198486328125, 43.09906005859375, 7.593528747558594, -74.30116271972656, 4.663673400878906, 45.40667724609375, 43.788726806640625, 105.330078125, 46.42176818847656, 95.91067504882812, 11.878849029541016, 87.84751892089844, 39.750885009765625, 14.364969253540039, 34.78369140625, 109.08920288085938, 98.1832275390625, 68.01962280273438, 73.18861389160156, 22.114547729492188, 39.440948486328125, -0.8149356842041016, 87.52142333984375, 53.44110870361328, 38.77568054199219, 1.7874908447265625, 88.42451477050781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000211.npy"}
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 28.91183090209961, "std": 49.70034408569336, "min": -126.7755126953125, "p10": -34.17593994140625, "median": 25.22315216064453, "p90": 88.79950942993162, "max": 129.36587524414062, "pos_frac": 0.7265625, "sample": [4.145603179931641, -3.208221435546875, 32.334808349609375, 61.48932647705078, -16.625869750976562, -44.29460144042969, 2.2061767578125, 16.37322998046875, 3.6717453002929688, -15.160415649414062, 103.09925079345703, -1.818145751953125, 121.90206909179688, 2.9731369018554688, 36.257080078125, -52.547264099121094, 21.04823112487793, 77.19833374023438, 66.23292541503906, 31.684814453125, 85.91084289550781, 28.0809326171875, -35.63848876953125, 19.185302734375, 82.50677490234375, 17.97271728515625, -32.538726806640625, -71.80574798583984, 44.88728332519531, 62.63804626464844, -17.940448760986328, -63.26313781738281, 4.7899322509765625, 40.95428466796875, 17.863990783691406, 60.1632080078125, 23.737640380859375, 95.30010986328125, 63.12440490722656, -5.6873016357421875, 79.75662231445312, -92.3204345703125, -3.685302734375, -2.713808059692383, -8.549896240234375, 80.22232055664062, 80.295654296875, 76.52224731445312, -4.13348388671875, -53.716636657714844, 68.53290557861328, 30.469207763671875, -37.389434814453125, 106.0120849609375, 45.855987548828125, -6.108123779296875, -11.706802368164062, 71.50030517578125, 22.5130615234375, -3.1548423767089844, -7.0403289794921875, 6.586004257202148, 77.35566711425781, 31.92742919921875, -8.514490127563477, 75.6882553100586, -11.475833892822266, 56.37303161621094, 80.22601318359375, 94.35787963867188, 49.905975341796875, 81.02957153320312, -66.435791015625, 33.11505126953125, 82.80500793457031, 17.050979614257812, 57.35234832763672, 73.80609130859375, -0.7690811157226562, 103.489013671875, 39.46417236328125, 107.524658203125, 25.008392333984375, 60.2950439453125, -115.68917846679688, 22.579620361328125, -4.74072265625, -50.382659912109375, 25.437911987304688, 7.694297790527344, 80.98751831054688, 41.9595947265625, 12.42230224609375, 6.717670440673828, 121.40863037109375, 62.913848876953125, 21.70330810546875, 96.13915252685547, 63.305938720703125, 51.89915466308594, 51.99855041503906, 129.36587524414062, 52.1788330078125, 0.6693382263183594, 21.666168212890625, -126.7755126953125, 84.83792114257812, 113.9954833984375, -33.54913330078125, -13.515609741210938, 7.8946380615234375, -40.143646240234375, 34.71466064453125, 8.151937484741211, 73.18072509765625, 1.3367462158203125, 40.19761657714844, 17.074630737304688, 64.45999145507812, 10.227714538574219, 96.85723876953125, 11.983474731445312, 57.58904266357422, 86.41735076904297, 8.303955078125, -3.6767578125, 109.660400390625, 26.725982666015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000212.npy"}
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 27.497570037841797, "std": 49.83884048461914, "min": -82.31961059570312, "p10": -30.682968139648438, "median": 12.947036743164062, "p90": 101.38451385498047, "max": 142.8701171875, "pos_frac": 0.6796875, "sample": [107.57553100585938, 79.15583801269531, -40.98999786376953, 86.8768310546875, -34.800537109375, 68.37848663330078, 77.82421875, 70.865966796875, -4.10748291015625, 30.64678955078125, 5.5838623046875, 24.06982421875, 21.67840576171875, -24.611053466796875, 15.845285415649414, 130.2496337890625, -3.518524169921875, 19.076080322265625, 24.75384521484375, 63.423583984375, 46.2515869140625, -8.927032470703125, 142.29641723632812, -6.31268310546875, 102.23435974121094, 7.4984130859375, -8.288970947265625, 14.34783935546875, 21.979507446289062, 15.922683715820312, 7.57086181640625, -29.057281494140625, -16.901275634765625, -0.0029144287109375, 93.26823425292969, 33.62623596191406, 80.52406311035156, -2.0135955810546875, 14.312423706054688, -20.857147216796875, 89.90798950195312, 46.68644714355469, 11.581649780273438, 71.31838989257812, -18.07489013671875, 79.4781494140625, 25.6785888671875, 6.367462158203125, -44.498870849609375, 47.4774169921875, 7.2998809814453125, 101.02029418945312, 100.44769287109375, 4.5878448486328125, 127.65679931640625, -41.31438446044922, -6.214813232421875, 70.01483154296875, -24.237396240234375, 102.36140441894531, 17.450225830078125, 17.443115234375, 31.29936981201172, -67.29623413085938, 7.1915283203125, 31.1339111328125, 110.85333251953125, -40.57562255859375, -17.074172973632812, -8.873992919921875, 75.46708679199219, -16.04254150390625, 31.145721435546875, -82.31961059570312, 7.1326904296875, 7.720405578613281, 54.72615051269531, 94.7711181640625, 8.952095031738281, 24.73773193359375, -18.339431762695312, 8.107841491699219, -2.3285388946533203, -9.2237548828125, -1.0747909545898438, 109.63995361328125, -62.527496337890625, -6.849151611328125, 36.778900146484375, 123.85476684570312, 71.21044921875, 100.69393920898438, 5.461151123046875, 8.62408447265625, 47.56903076171875, -5.479972839355469, 106.43191528320312, -30.64208984375, -32.4969482421875, 142.8701171875, 0.7950096130371094, 7.831047058105469, -9.36163330078125, 93.08956146240234, -34.694000244140625, 112.513916015625, 88.4141845703125, -8.55401611328125, 50.34419250488281, 0.0, 9.112483978271484, 32.31866455078125, 7.027435302734375, 96.87873840332031, 10.805389404296875, -70.33677673339844, 86.705078125, 39.85248947143555, -2.8164291381835938, 20.3653564453125, 2.892498016357422, 82.2059326171875, -30.778350830078125, 5.8005828857421875, 112.99020385742188, 0.25467681884765625, -31.500152587890625, 4.417869567871094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000213.npy"}
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 24.63863182067871, "std": 48.18707275390625, "min": -95.95193481445312, "p10": -24.787927246093748, "median": 17.17022705078125, "p90": 88.1263801574707, "max": 158.75088500976562, "pos_frac": 0.71875, "sample": [56.6142578125, 69.41950988769531, -25.027259826660156, 7.867431640625, 15.720947265625, 93.69662475585938, 21.922561645507812, -13.443954467773438, -54.008056640625, -19.528549194335938, 11.44439697265625, 35.485107421875, -20.386260986328125, 1.291961669921875, 54.17425537109375, 68.77606201171875, -95.95193481445312, 80.3035659790039, 17.365921020507812, 32.8343505859375, -5.419197082519531, -19.103759765625, 24.794532775878906, 68.4093017578125, 37.479248046875, 17.04510498046875, 68.28064727783203, -24.42901611328125, 7.0640869140625, 3.6071395874023438, 6.129852294921875, 6.280242919921875, 14.665666580200195, 4.6719512939453125, 32.3489990234375, 22.701629638671875, 158.75088500976562, 74.71046447753906, 102.812744140625, -76.79351806640625, 32.14524841308594, -36.509124755859375, 4.57392692565918, 67.26608276367188, 3.439239501953125, 4.591041564941406, -27.0357666015625, 88.1827392578125, 31.112945556640625, -6.650796890258789, 68.52397918701172, 22.23870849609375, -46.427268981933594, 20.541534423828125, 2.281261444091797, 83.34632873535156, 63.244659423828125, 53.757781982421875, -15.2203369140625, 79.98373413085938, 71.69410705566406, 114.68841552734375, 63.86469268798828, 14.308609008789062, 19.79571533203125, -45.82366180419922, -16.777023315429688, -24.68535614013672, -0.95965576171875, 0.0, -16.560821533203125, 8.879791259765625, -13.283966064453125, 3.502471923828125, 69.40200805664062, 2.1174774169921875, 50.06597900390625, -59.116580963134766, 6.331573486328125, 15.925987243652344, -18.823974609375, -71.3592529296875, 89.69293212890625, 0.595489501953125, 77.34584045410156, -21.624542236328125, 88.10222625732422, 42.062034606933594, 23.395263671875, 17.389739990234375, -80.95553588867188, 18.9466552734375, 28.139495849609375, 97.9332275390625, 90.07974243164062, 9.985321044921875, -16.39422607421875, 59.02339172363281, 6.50860595703125, 60.42060852050781, 86.84547424316406, 116.8153076171875, 8.95050048828125, 69.55284118652344, -13.13787841796875, 98.92874145507812, 77.60614013671875, 29.2576904296875, -22.034576416015625, 76.85362243652344, 142.5400390625, 17.29534912109375, -21.003997802734375, -41.49542236328125, 1.903839111328125, 90.93843078613281, 14.21638298034668, -67.54139709472656, -5.04693603515625, -0.9801025390625, 2.9201812744140625, 33.36456298828125, 20.890289306640625, 55.120819091796875, -4.370658874511719, 151.3597412109375, 51.79579162597656, 62.4031982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000214.npy"}
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 26.114883422851562, "std": 53.263519287109375, "min": -107.46868896484375, "p10": -43.493325805664064, "median": 21.181312561035156, "p90": 91.77504806518553, "max": 159.53631591796875, "pos_frac": 0.671875, "sample": [32.068809509277344, 13.756561279296875, 63.63739013671875, 83.77420043945312, 118.58773803710938, 76.849609375, 82.25128173828125, 40.626243591308594, 67.8172607421875, -11.226242065429688, 106.70472717285156, 75.70489501953125, 102.50445556640625, -28.294876098632812, 9.465118408203125, 29.31634521484375, -13.78094482421875, 136.02951049804688, 138.60992431640625, 76.44808959960938, 82.17898559570312, 0.054931640625, 0.0, -26.9678955078125, -4.8549041748046875, -34.68385314941406, 41.388885498046875, 72.76031494140625, 132.49087524414062, -10.414863586425781, 85.42973327636719, 0.2152557373046875, -17.991851806640625, -47.00743103027344, 2.7804412841796875, 10.262771606445312, 87.4180679321289, -3.810791015625, 20.222122192382812, 128.79806518554688, -21.220703125, 27.6126708984375, -28.6279296875, -32.791412353515625, -0.2413501739501953, -9.787139892578125, -75.5743408203125, 82.53711700439453, 77.29678344726562, 46.922821044921875, 44.63606262207031, 159.53631591796875, 16.141265869140625, 22.1405029296875, -55.883514404296875, 0.8349723815917969, 5.7545623779296875, -43.467315673828125, 0.0, 82.15902709960938, -1.1593856811523438, -1.2173023223876953, 50.915435791015625, -48.268341064453125, 55.07940673828125, 54.86663818359375, -71.04048156738281, 6.1949005126953125, 94.32734680175781, 25.9927978515625, 27.943771362304688, 8.517602920532227, -11.566307067871094, 79.50262451171875, -59.90673828125, 74.2357177734375, 6.18365478515625, -107.46868896484375, 60.33018493652344, 47.4002685546875, 11.311355590820312, 46.92858123779297, 19.495361328125, -21.415740966796875, -75.956298828125, -4.30561637878418, 78.971923828125, 17.22637939453125, 95.49163818359375, -36.80267333984375, -13.544105529785156, 31.703094482421875, 24.86278533935547, -3.7576904296875, 9.264450073242188, 13.592971801757812, 2.2136154174804688, 31.510177612304688, 116.045654296875, 150.00970458984375, -32.91552734375, 27.38439178466797, -75.17227172851562, 0.4314594268798828, -43.55401611328125, -45.37115478515625, 61.23980712890625, 90.68120574951172, -61.37754821777344, 89.62405395507812, 24.0723876953125, -7.887866973876953, 24.70339584350586, -16.95263671875, 13.030105590820312, 34.384765625, 24.586822509765625, 87.12126159667969, -50.8251953125, 29.90118408203125, 58.41825866699219, -4.313446044921875, 17.7646484375, 96.52728271484375, 23.994598388671875, 59.85624694824219, 28.968994140625, 87.57765197753906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000215.npy"}
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 29.555112838745117, "std": 57.099876403808594, "min": -146.11117553710938, "p10": -43.32147064208984, "median": 25.45672607421875, "p90": 102.74732513427735, "max": 143.016845703125, "pos_frac": 0.7265625, "sample": [36.935302734375, 48.73799133300781, 67.85678100585938, 33.89844512939453, 16.8604736328125, -45.07801818847656, -146.11117553710938, 49.34466552734375, 71.32623291015625, -8.322975158691406, 79.9872817993164, 87.80630493164062, 79.659423828125, -17.198776245117188, 8.320907592773438, 52.06840515136719, 143.016845703125, 9.311363220214844, 111.59228515625, 119.22573852539062, 20.335693359375, -86.68475341796875, 66.74783325195312, -2.10174560546875, -9.724956512451172, 20.6641845703125, -97.1181640625, 12.8671875, 61.032073974609375, 82.5491943359375, -49.98243713378906, -64.28257751464844, -110.11978149414062, 0.07611846923828125, 70.76434326171875, 16.638391494750977, 85.54824829101562, -46.76165771484375, 113.542724609375, 79.3670425415039, -24.971553802490234, 90.15951538085938, 123.77569580078125, 82.5481185913086, 85.85865783691406, 61.154388427734375, 41.16119384765625, 56.12457275390625, 5.99407958984375, -1.3204345703125, -50.38861083984375, -6.397560119628906, 135.03826904296875, 43.40910339355469, 9.1495361328125, 27.71636962890625, 13.025360107421875, 107.73751831054688, -13.14056396484375, 5.101993560791016, 0.0, 62.54731750488281, -19.0614013671875, 78.01028442382812, 103.09637451171875, 93.1300048828125, 97.5625, -14.720001220703125, -6.3369903564453125, 11.19171142578125, 45.68724060058594, 38.38275146484375, -117.90084838867188, 16.02960205078125, 116.80531311035156, 102.59773254394531, 10.728118896484375, 107.61334228515625, 59.101806640625, 10.625030517578125, 2.98773193359375, 16.970367431640625, 95.05072021484375, 26.641448974609375, 111.90020751953125, 77.06349182128906, -35.73338317871094, -8.75726318359375, 107.03909301757812, 91.58526611328125, -42.56866455078125, -66.50173950195312, 2.22064208984375, 51.91278076171875, -5.7220306396484375, -18.11962890625, 15.893310546875, -18.770294189453125, 54.63479995727539, 24.272003173828125, 123.7901611328125, 12.419334411621094, 0.0, 64.98114776611328, 82.06443786621094, -78.24310302734375, 2.762237548828125, 64.46163940429688, 69.50081634521484, 59.17327117919922, 0.09112739562988281, 62.48138427734375, 87.6175537109375, -10.14593505859375, 45.12651824951172, -9.251907348632812, 32.37867736816406, 36.770050048828125, 5.0458984375, 53.575469970703125, 2.17596435546875, -107.6878662109375, 76.53111267089844, 85.78080749511719, 4.84796142578125, 2.51171875, -22.4404296875, 5.317626953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000216.npy"}
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 40.0655632019043, "std": 54.32268524169922, "min": -86.4425048828125, "p10": -20.924554443359373, "median": 33.57879638671875, "p90": 108.88633575439452, "max": 216.1087646484375, "pos_frac": 0.7578125, "sample": [10.655677795410156, -2.7472915649414062, -4.36566162109375, 82.57696533203125, 108.15715026855469, -79.10841369628906, 3.212860107421875, 99.01025390625, 6.512298583984375, 46.89697265625, 101.74540710449219, -2.22369384765625, -1.373565673828125, 30.494964599609375, -28.7943115234375, 102.02330780029297, 75.91133117675781, 85.56358337402344, 72.56129455566406, 33.491912841796875, 36.85359191894531, 77.23056030273438, 6.866428375244141, 81.73863220214844, 166.2939453125, 69.56973266601562, 87.62832641601562, 89.4832534790039, 90.71286010742188, 100.02796173095703, 38.24285888671875, 52.675201416015625, -13.686813354492188, 125.12408447265625, 54.351287841796875, 27.492950439453125, 216.1087646484375, 83.64959716796875, 91.51763916015625, 49.50933837890625, -86.4425048828125, 94.17721557617188, 37.45220947265625, 150.28668212890625, -82.28099060058594, 41.73383331298828, 12.207183837890625, -23.48285675048828, 41.23626708984375, 23.32049560546875, 3.8663978576660156, 1.3220672607421875, 11.3682861328125, 40.02803039550781, 56.44725036621094, -48.46185302734375, -3.9343490600585938, 89.32026672363281, 60.839698791503906, -20.416656494140625, -1.7291183471679688, 42.26116943359375, -8.376266479492188, 119.53652954101562, 5.2383270263671875, 60.28459930419922, 12.715005874633789, 99.07937622070312, 75.34493255615234, -3.054412841796875, 74.35140991210938, 82.75257110595703, 33.665679931640625, -3.3792381286621094, 102.217041015625, -18.116806030273438, -28.0081787109375, -22.705894470214844, 59.26177978515625, 157.64663696289062, 10.790184020996094, 114.109619140625, 32.140541076660156, -8.726631164550781, -86.376953125, 125.66505432128906, 111.696044921875, 110.5877685546875, 17.062255859375, -58.1822509765625, 17.3721923828125, 90.89346313476562, -13.69390869140625, 81.88040161132812, 0.67315673828125, 60.08111572265625, 1.9644012451171875, 14.771942138671875, 11.184440612792969, 78.50747680664062, -3.1602020263671875, 0.0, 17.040679931640625, 33.46998596191406, -29.146881103515625, 128.37234497070312, 82.09416198730469, -7.261810302734375, 34.37469482421875, 121.61882019042969, 64.04849243164062, 30.302871704101562, 23.63818359375, 134.21466064453125, 83.754638671875, 10.541854858398438, 5.391044616699219, 27.922119140625, -28.9749755859375, 8.026763916015625, 57.21905517578125, 45.833595275878906, 51.6365966796875, 23.160980224609375, 17.451873779296875, -22.109649658203125, -6.2147216796875, 3.5838050842285156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000217.npy"}
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 32.49669647216797, "std": 48.41775131225586, "min": -92.26166534423828, "p10": -22.59193725585937, "median": 23.15442657470703, "p90": 92.21218414306641, "max": 153.5760498046875, "pos_frac": 0.765625, "sample": [45.227020263671875, 15.269309997558594, 2.746185302734375, 97.26406860351562, -11.024673461914062, 6.4261016845703125, -6.903900146484375, 111.42529296875, 29.977741241455078, 6.140178680419922, -66.79460144042969, 13.411224365234375, 49.810546875, 62.150482177734375, 46.45538330078125, -82.35638427734375, 8.42041015625, 80.70320129394531, 73.67113494873047, 74.96945190429688, -14.731372833251953, 37.19024658203125, 85.48257446289062, 24.549835205078125, 36.3206787109375, 24.126358032226562, 13.26519775390625, -12.71636962890625, -20.4656982421875, 66.73150634765625, 71.14368438720703, 123.4244384765625, 6.115936279296875, 90.67658233642578, -24.6124267578125, 14.73614501953125, 79.45210266113281, -27.531494140625, 79.59489440917969, 66.68063354492188, 144.38327026367188, 55.65232849121094, 21.117095947265625, 1.840169906616211, -19.828079223632812, 94.08866882324219, 19.20772933959961, -63.51116943359375, -40.75579833984375, 89.73037719726562, 37.34559631347656, -27.4278564453125, 80.79800415039062, 31.63531494140625, 95.8292236328125, -2.2108612060546875, 92.96694946289062, 17.546142578125, 88.30792236328125, -21.72601318359375, 17.4815673828125, -8.651611328125, -30.12358856201172, 10.545852661132812, 0.0, 68.9912109375, 16.87841796875, 1.70111083984375, -71.53010559082031, 85.52178955078125, 14.060073852539062, 47.859405517578125, -62.58365249633789, 92.10777282714844, 32.05706787109375, 90.87420654296875, -2.6092453002929688, 13.20526123046875, 27.14898681640625, 12.110466003417969, 153.5760498046875, 92.455810546875, 113.3818359375, 6.74212646484375, -5.841064453125, 20.43121337890625, -5.8622589111328125, 2.3073654174804688, 112.44900512695312, 31.674560546875, 18.68859100341797, 66.19603729248047, 18.22705078125, -57.576690673828125, 80.38919830322266, 71.29320526123047, 94.06672668457031, 90.68548583984375, -41.5389404296875, 45.3665771484375, 5.507781982421875, 81.47467041015625, -7.687015533447266, 16.5018310546875, 51.951171875, -9.995208740234375, -92.26166534423828, 61.75215148925781, 18.11993408203125, 69.08248901367188, 10.587005615234375, 15.2457275390625, 52.585693359375, 78.87750244140625, -1.740325927734375, 22.1824951171875, 117.74163818359375, 3.36627197265625, 10.339408874511719, -14.57513427734375, 91.95254516601562, 65.2015151977539, 68.233154296875, 27.26123046875, 46.477935791015625, 84.0543212890625, 44.085182189941406, 9.6868896484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000218.npy"}
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 32.032752990722656, "std": 52.32183837890625, "min": -119.68707275390625, "p10": -19.734898376464837, "median": 26.16541862487793, "p90": 102.9668029785156, "max": 134.12716674804688, "pos_frac": 0.7734375, "sample": [-119.68707275390625, 18.38262939453125, 49.4525146484375, 10.661239624023438, 94.52403259277344, 50.65020751953125, -14.706161499023438, 67.68475341796875, 87.85336303710938, 30.45880126953125, 64.89059448242188, -1.099172592163086, 40.256103515625, -55.90802001953125, 0.7287368774414062, 71.11051177978516, 60.96124267578125, -12.630767822265625, 16.970108032226562, 77.19949340820312, 33.29844665527344, 8.664112091064453, 45.76373291015625, 120.53546142578125, -5.295356750488281, 26.120697021484375, 82.77255249023438, 4.99005126953125, 73.90621948242188, -24.266067504882812, -10.368389129638672, -4.169797897338867, -99.17422485351562, 42.72942352294922, 15.04327392578125, 94.55839538574219, -4.45166015625, 26.210140228271484, 21.2088623046875, 90.91650390625, 107.60528564453125, 132.45794677734375, 7.415966033935547, 53.597412109375, 9.964668273925781, -81.9197998046875, 8.3529052734375, 72.69287109375, 11.384490966796875, 122.72280883789062, 0.0, 26.669647216796875, 23.255386352539062, 27.706268310546875, -49.055450439453125, 107.87628173828125, -25.791290283203125, 100.9788818359375, 114.02725219726562, 16.311981201171875, 49.79002380371094, 41.82470703125, 33.262359619140625, 4.80078125, 1.5577268600463867, -10.8497314453125, 7.597686767578125, 6.305208206176758, 2.2869796752929688, 72.37440490722656, -12.317352294921875, 23.72015380859375, 13.523887634277344, 116.13333129882812, 20.71990966796875, -1.9376983642578125, -84.906005859375, 12.689605712890625, 55.48883056640625, 99.5859375, 1.635324478149414, 97.07589721679688, 24.2984619140625, 27.16583251953125, -5.738260269165039, -38.786712646484375, 115.47903442382812, 6.408653259277344, 115.06117248535156, -4.920257568359375, 115.5531005859375, 70.42031860351562, 45.2265625, 74.6109619140625, 78.51541900634766, 86.4747314453125, 41.28310012817383, 52.011260986328125, 26.40264892578125, 90.11512756347656, -17.79296875, 89.0235595703125, 81.09576416015625, -55.46275329589844, 98.04803466796875, 63.50482177734375, 59.219146728515625, 1.4993743896484375, 25.621856689453125, 25.795166015625, 40.30003356933594, -2.0123291015625, 42.165496826171875, -114.47613525390625, 132.89730834960938, 108.2520751953125, 40.10752868652344, -45.84539794921875, 13.466476440429688, 14.416488647460938, 51.325439453125, 19.3331298828125, 40.7520751953125, 12.698036193847656, 134.12716674804688, -10.101663589477539, 2.9767074584960938, -83.68441772460938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000219.npy"}
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 34.95800018310547, "std": 53.96216583251953, "min": -129.1527099609375, "p10": -29.185453796386717, "median": 39.498741149902344, "p90": 108.35106658935547, "max": 140.97744750976562, "pos_frac": 0.71875, "sample": [109.97817993164062, 116.87808227539062, 71.31147766113281, 57.16844940185547, 23.735998153686523, 72.7647705078125, 4.23065185546875, 85.669189453125, 25.984375, 71.9466552734375, -8.991275787353516, -0.61572265625, 78.09586334228516, -59.69158935546875, 10.827346801757812, 32.682373046875, 35.656829833984375, 106.10556030273438, -28.3770751953125, 9.412094116210938, 59.58949279785156, -9.9102783203125, 25.264129638671875, 2.126985549926758, 0.5490188598632812, 140.97744750976562, -4.939178466796875, 97.77781677246094, -72.83981323242188, 19.7618408203125, 56.630767822265625, 4.3722686767578125, -9.0068359375, -31.071670532226562, 31.368408203125, 8.871673583984375, -23.8411865234375, 54.404998779296875, 1.6679115295410156, 70.11083984375, 50.57770538330078, 31.944580078125, 94.29824829101562, 47.51507568359375, 114.35078430175781, 56.739891052246094, -6.659671783447266, -129.1527099609375, 70.73004150390625, 44.29339599609375, 56.50274658203125, 81.67320251464844, -10.83343505859375, 115.24960327148438, -26.46674346923828, 16.5025634765625, -12.296554565429688, -24.088539123535156, 7.85858154296875, 51.709327697753906, -25.652618408203125, 64.77716827392578, 43.34065246582031, 58.83399963378906, 11.850303649902344, -13.897605895996094, -9.318328857421875, 54.60089111328125, 2.57904052734375, 87.84820556640625, 105.42903137207031, 115.37010192871094, -7.9801483154296875, 26.786972045898438, 78.78634643554688, 72.81818389892578, -22.209991455078125, 63.75779724121094, 77.09135437011719, -9.639862060546875, 12.251220703125, 15.61578369140625, 83.57687377929688, 58.05230712890625, -5.915153503417969, 117.59017181396484, -86.74127197265625, 6.62451171875, 129.62295532226562, 130.08865356445312, -37.49957275390625, 35.07763671875, -87.45175170898438, 0.30692291259765625, 64.77667236328125, 120.80084228515625, 110.3363037109375, 17.903839111328125, 77.10775756835938, -27.928115844726562, 84.3170166015625, 61.7635498046875, -38.520263671875, -34.910797119140625, -1.5243301391601562, -19.504302978515625, 80.23638916015625, -34.12615966796875, 116.41580200195312, 107.65373229980469, 73.19895935058594, 46.002899169921875, 58.4095458984375, -39.481414794921875, 127.44189453125, -32.02886962890625, 98.49337768554688, 73.28083801269531, 11.6614990234375, 85.50436401367188, -93.72703552246094, 52.1024169921875, 69.20398712158203, 56.87043762207031, 87.4261474609375, -12.095794677734375, 70.07791137695312, 44.02946472167969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000220.npy"}
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 27.19757080078125, "std": 52.1388053894043, "min": -117.31094360351562, "p10": -34.95707397460937, "median": 19.51758575439453, "p90": 94.95101165771484, "max": 130.75576782226562, "pos_frac": 0.7109375, "sample": [2.799468994140625, 65.54141235351562, -15.857666015625, 19.764892578125, 44.56017303466797, 21.312164306640625, 19.270278930664062, 54.027496337890625, -13.2603759765625, -6.298480987548828, 66.20040130615234, 52.365264892578125, 2.2796630859375, 58.6474609375, -2.1543731689453125, 81.7430419921875, 13.869781494140625, -67.10223388671875, 8.36431884765625, -14.097297668457031, 72.61248779296875, 36.744415283203125, 130.75576782226562, -22.400314331054688, 31.16790771484375, 90.92525482177734, 98.63848876953125, 70.85787963867188, 88.28379821777344, -1.014617919921875, 12.87225341796875, 12.459182739257812, 117.01904296875, -117.31094360351562, -41.638397216796875, 15.244873046875, 42.03923034667969, 78.6905517578125, -0.4940948486328125, 111.8033447265625, -46.765541076660156, 9.938629150390625, -16.687820434570312, 34.42529296875, -70.59934997558594, -4.82586669921875, 90.78854370117188, -68.88116455078125, 80.99107360839844, 33.8193359375, 0.0021152496337890625, 79.64598083496094, 21.9619140625, -19.21624755859375, -5.630842208862305, 63.868560791015625, 60.2393798828125, -63.61708068847656, 2.64410400390625, 53.44274139404297, 8.17596435546875, 7.3233489990234375, 62.361114501953125, 12.131780624389648, 41.63751220703125, -0.042606353759765625, 76.67573547363281, 94.44136047363281, -2.7711715698242188, 78.31570434570312, 71.1248550415039, -15.09820556640625, 65.63800048828125, 49.75898742675781, 105.22737121582031, 64.59576416015625, 108.55978393554688, 25.9893798828125, -54.503684997558594, -82.739013671875, 96.14019775390625, 84.38134765625, -17.365234375, 8.564743041992188, -7.33221435546875, 92.91688537597656, 13.070098876953125, -9.0906982421875, 27.697433471679688, 72.84494018554688, 116.43801879882812, 101.75234985351562, 0.6768455505371094, 7.46522331237793, -39.122894287109375, 6.765926361083984, -21.18328857421875, 118.04776000976562, -103.41400146484375, 49.060211181640625, 5.7985382080078125, -77.3779296875, -4.49322509765625, 115.68911743164062, 16.67657470703125, 30.5467529296875, 15.296188354492188, 96.22442626953125, 92.9364013671875, 10.401512145996094, 55.148651123046875, 48.110443115234375, -33.171722412109375, -18.02199935913086, 5.913322448730469, 6.663043975830078, 87.84814453125, 28.54241943359375, 81.588134765625, -5.933868408203125, 50.582672119140625, 82.27110290527344, -27.1285400390625, 96.54562377929688, 29.276214599609375, 7.650115966796875, -97.20281982421875, 11.018890380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000221.npy"}
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 35.12825393676758, "std": 49.47154235839844, "min": -158.1397705078125, "p10": -19.37053070068359, "median": 22.329788208007812, "p90": 105.19799804687499, "max": 132.4549560546875, "pos_frac": 0.765625, "sample": [62.897300720214844, 15.51800537109375, 88.57057189941406, 76.33975219726562, -25.22198486328125, -26.179641723632812, 75.63458251953125, 45.09352493286133, 16.672576904296875, 9.724960327148438, -31.89581298828125, 21.155029296875, 86.44406127929688, 42.4222412109375, 23.754852294921875, 23.61821746826172, 77.78182220458984, 120.54360961914062, 62.9698486328125, 42.65876770019531, 7.992706298828125, 1.42633056640625, -26.775436401367188, -39.483551025390625, 60.528656005859375, 124.5157470703125, -11.46600341796875, -1.841796875, -72.14454650878906, 7.800235748291016, 11.079345703125, 42.15336608886719, 130.406494140625, 5.710662841796875, 19.00275421142578, 10.781566619873047, 18.7823486328125, 64.05828857421875, 77.89878845214844, 36.1348876953125, 2.3938446044921875, -28.672515869140625, 73.62896728515625, 3.050750732421875, 90.11259460449219, 8.134517669677734, -34.81920623779297, 50.72453308105469, -158.1397705078125, 41.243408203125, 99.54942321777344, 71.36698913574219, 0.18820953369140625, 104.2242431640625, -24.483699798583984, 132.4549560546875, 22.99333953857422, 103.9178466796875, 2.248065948486328, 7.11126708984375, 25.553924560546875, 24.960174560546875, 22.807373046875, 109.52337646484375, -8.723358154296875, 17.67529296875, 86.86425018310547, 130.01919555664062, 57.00372314453125, 67.24432373046875, 97.32435607910156, 85.82240295410156, 2.171783447265625, 88.8984375, -21.040023803710938, 13.87371826171875, 19.117156982421875, -1.8208427429199219, 109.50927734375, -7.43389892578125, 7.9272308349609375, 21.682538986206055, -9.527933120727539, 57.13517761230469, -18.697341918945312, -1.3962554931640625, 107.4700927734375, -19.956466674804688, -7.086585998535156, 61.19792175292969, 88.45696258544922, 84.79331970214844, 13.367729187011719, 35.551544189453125, 5.978483200073242, -14.919219970703125, 4.6193695068359375, 76.88032531738281, 26.97479248046875, 110.43331909179688, 119.5264892578125, -8.312576293945312, -1.355072021484375, 40.29060363769531, 111.90422058105469, -5.4829559326171875, 94.7811279296875, 35.33001708984375, 15.522064208984375, 96.10841369628906, 21.852203369140625, 45.245330810546875, -31.79595947265625, 19.542327880859375, -0.834991455078125, -10.490936279296875, -15.14190673828125, 99.02084350585938, -19.119415283203125, 78.02777099609375, 55.1610107421875, 9.72210693359375, 118.45260620117188, 84.08277893066406, 20.097511291503906, 0.8347644805908203, 13.176254272460938, 117.74525451660156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000222.npy"}
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 39.13160705566406, "std": 55.10234069824219, "min": -98.68626403808594, "p10": -31.850978851318356, "median": 38.40703201293945, "p90": 107.50683288574217, "max": 149.85748291015625, "pos_frac": 0.7734375, "sample": [10.120521545410156, -19.3858642578125, 5.106292724609375, 84.01535034179688, 10.583267211914062, 11.553741455078125, 1.4776611328125, -10.604366302490234, 21.229564666748047, 101.49467468261719, 26.821502685546875, 100.74009704589844, 65.08367919921875, 8.6575927734375, 34.160736083984375, 143.87139892578125, 105.13581848144531, -98.68626403808594, 84.41551208496094, -40.33294677734375, 29.727752685546875, 125.57333374023438, -3.8936309814453125, 62.8564453125, 41.50205993652344, -35.825775146484375, 17.574127197265625, 12.68218994140625, 115.04461669921875, -16.119049072265625, 41.555389404296875, 67.8875732421875, 100.76347351074219, 100.02557373046875, 89.35150146484375, 94.83988952636719, 142.7999267578125, -95.48513793945312, 64.05374908447266, -79.45033264160156, 15.006011962890625, 64.49713134765625, 9.4005126953125, 73.61344909667969, 105.87762451171875, 87.45205688476562, 114.6505126953125, 16.558197021484375, 120.79660034179688, 25.994598388671875, 1.259063720703125, 25.13494873046875, 125.53060913085938, 132.24911499023438, 83.09696960449219, 56.592254638671875, 4.632354736328125, 111.73602294921875, 21.007606506347656, 48.43890380859375, 41.518218994140625, 20.46417999267578, 85.5272216796875, 149.85748291015625, 40.00977325439453, 44.54838562011719, 24.7821044921875, 61.127685546875, 80.40960693359375, -7.1793365478515625, -13.615142822265625, 60.31610107421875, 5.622344970703125, 97.85157775878906, -45.650718688964844, 27.015705108642578, -33.64116668701172, -31.083755493164062, 94.947265625, -14.587661743164062, 38.19200897216797, -10.538299560546875, 0.28446197509765625, 23.521846771240234, 93.74526977539062, 65.63125610351562, -66.44204711914062, -15.469161987304688, 73.4053955078125, -15.165046691894531, -82.26139068603516, 38.62205505371094, 60.908843994140625, 86.5400390625, 88.30523681640625, 125.75274658203125, 78.24969482421875, 81.15815734863281, 65.80763244628906, 16.591476440429688, -1.9124279022216797, 65.75749206542969, 23.54638671875, 73.44544219970703, 54.93145751953125, -13.067413330078125, 102.549560546875, 137.82266235351562, 0.5380020141601562, 14.300079345703125, 111.30831909179688, -63.972442626953125, -8.08740234375, 8.745460510253906, 99.200927734375, 59.7850341796875, -66.27965545654297, -2.443817138671875, 71.18423461914062, 5.37213134765625, -30.28249740600586, 92.08460998535156, 56.90788269042969, 44.4747314453125, 33.40782928466797, 17.99872589111328, -33.846099853515625, -44.1497802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000223.npy"}
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 39.025794982910156, "std": 51.501102447509766, "min": -102.61734008789062, "p10": -22.344960784912107, "median": 34.55650329589844, "p90": 108.34341125488281, "max": 146.92510986328125, "pos_frac": 0.796875, "sample": [125.79754638671875, 146.5533447265625, 60.267547607421875, 58.618568420410156, 2.303478240966797, 2.9872970581054688, -48.157745361328125, 43.35923767089844, 85.03160095214844, 19.799728393554688, 6.45263671875, 77.1645278930664, -2.57403564453125, 75.84022521972656, 40.30805969238281, 33.483001708984375, 10.8267822265625, 83.0584716796875, 18.82695770263672, -60.85337829589844, 77.31524658203125, -1.757659912109375, 117.92598724365234, 119.92904663085938, 22.408416748046875, 70.95701599121094, 74.86337280273438, 22.826828002929688, 17.42291259765625, 69.029296875, 123.60394287109375, 87.23561096191406, -102.61734008789062, 72.18330383300781, 37.23390197753906, 21.024795532226562, 35.804443359375, 78.15815734863281, 70.02439880371094, 11.484893798828125, 11.6710205078125, -18.652008056640625, 86.71128845214844, 65.79761505126953, 54.929718017578125, 27.911178588867188, 56.733673095703125, -35.87602233886719, 107.92160034179688, 0.11356544494628906, 73.21397399902344, 54.99517822265625, 0.0, 111.04147338867188, 87.906494140625, 36.39966583251953, 11.850250244140625, -63.82234191894531, 4.2733917236328125, 31.1602783203125, 61.401397705078125, 23.05035400390625, 1.7319049835205078, 91.91091918945312, -2.9193115234375, 110.637451171875, -20.52801513671875, -31.880218505859375, 55.701385498046875, -35.35160827636719, 6.013763427734375, 130.63873291015625, 25.649063110351562, 109.32763671875, 16.84063720703125, -9.85968017578125, -12.517196655273438, 67.130126953125, 12.571060180664062, 113.46746826171875, -21.867355346679688, -14.1021728515625, -90.4014892578125, 9.595355987548828, 12.362174987792969, 95.24465942382812, 46.5084228515625, -28.1646728515625, 102.3759765625, -85.13211822509766, 92.0982666015625, 83.7457275390625, 45.72343444824219, 98.60202026367188, 32.09967041015625, 82.91082763671875, 101.20278930664062, 60.64945983886719, 35.6300048828125, 6.479911804199219, -35.73399353027344, 28.77276611328125, 8.447540283203125, 17.120025634765625, 21.207595825195312, 43.5333251953125, 146.92510986328125, -23.459373474121094, 111.72164916992188, 74.53717041015625, 18.681365966796875, -45.87779235839844, 11.360763549804688, 2.69964599609375, -3.10308837890625, 30.756759643554688, 93.94314575195312, 9.061111450195312, 61.454315185546875, -18.7626953125, 7.067066192626953, 101.62692260742188, 74.73587036132812, 77.91387939453125, 98.63858032226562, 77.9866943359375, -2.2612648010253906, 119.30142211914062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000224.npy"}
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 26.338563919067383, "std": 60.51341247558594, "min": -106.94903564453125, "p10": -56.34955596923828, "median": 21.585857391357422, "p90": 115.59788513183594, "max": 144.8580322265625, "pos_frac": 0.640625, "sample": [119.283203125, -6.9879150390625, 35.825927734375, 115.0831298828125, 103.55482482910156, 6.0088653564453125, -76.71784973144531, -71.243408203125, 74.35150146484375, 111.74267578125, 27.348236083984375, 73.756103515625, 10.678329467773438, -11.870040893554688, 121.3438720703125, 29.034912109375, 135.51693725585938, 82.81689453125, 3.2240943908691406, 26.448837280273438, -10.884246826171875, -12.05706787109375, 21.51103973388672, -37.188720703125, 102.3724365234375, 28.245513916015625, 116.79898071289062, -9.180648803710938, 0.9117431640625, -56.09368896484375, -8.36993408203125, -89.4913330078125, 85.26295471191406, 85.83251953125, 12.330062866210938, 8.346046447753906, 84.902099609375, -0.6885490417480469, 0.79827880859375, -48.45796203613281, 56.619781494140625, 122.75167846679688, -42.2935791015625, -69.03558349609375, -91.82418823242188, -68.2588882446289, 50.6307373046875, 29.565338134765625, -6.670551300048828, 43.1968994140625, -61.231597900390625, 20.6568603515625, 74.08181762695312, 90.590576171875, -11.28369140625, 16.55316162109375, -34.25506591796875, 36.4427490234375, 105.3763427734375, 82.64741516113281, 88.09091186523438, 121.51646423339844, 128.6888427734375, -30.496978759765625, 29.371826171875, -12.1749267578125, -63.08392333984375, 66.85955810546875, 51.22914123535156, 136.72860717773438, 1.463409423828125, 112.24566650390625, 30.49127197265625, 70.5520248413086, -5.775604248046875, 53.139984130859375, 98.4092025756836, 127.3292236328125, -22.982421875, 87.16279602050781, -44.146636962890625, -58.324493408203125, -48.644981384277344, 28.697763442993164, -88.85322570800781, -10.220367431640625, 6.980558395385742, -0.3975257873535156, 42.12413024902344, -106.94903564453125, -4.250640869140625, -31.11688232421875, -16.16021728515625, -16.265869140625, 28.532623291015625, -31.4075927734375, 0.1805267333984375, 43.895729064941406, 144.8580322265625, 65.74214172363281, 5.571533203125, 73.71871185302734, 121.009765625, 108.79632568359375, -28.06781005859375, 14.299850463867188, -68.27527618408203, 27.694320678710938, 12.36102294921875, 135.315185546875, -16.799270629882812, 25.694091796875, 75.55853271484375, 135.9530487060547, 10.54541015625, 67.87757110595703, 56.446434020996094, -22.61083984375, 36.15830612182617, -56.94657897949219, 21.660675048828125, 2.637187957763672, -9.261634826660156, -5.62652587890625, -24.733428955078125, 26.598648071289062, 36.795623779296875, 107.5672607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000225.npy"}
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 27.449565887451172, "std": 54.06696319580078, "min": -101.38796997070312, "p10": -44.043359375, "median": 14.992507934570312, "p90": 103.06221466064453, "max": 134.81906127929688, "pos_frac": 0.734375, "sample": [18.25799560546875, 134.81906127929688, -14.6385498046875, 67.85386657714844, 96.946533203125, 112.33291625976562, 7.935516357421875, -1.4880828857421875, 66.80368041992188, 95.1602783203125, 111.32424926757812, 53.065677642822266, 7.61016845703125, 112.27616882324219, -5.48284912109375, 13.887706756591797, 35.51237487792969, -70.31829833984375, 92.90560913085938, 104.36190795898438, 10.155158996582031, 1.4833602905273438, 92.54331970214844, 61.73584747314453, 3.1246566772460938, 22.486572265625, -10.105745315551758, -52.13946533203125, 83.70977020263672, 65.50726318359375, 1.8558063507080078, 93.99383544921875, 60.64752197265625, 111.27642822265625, 1.8177947998046875, 15.12261962890625, 95.0620346069336, -43.883544921875, 1.635833740234375, 77.74610900878906, -84.774658203125, 57.508941650390625, -43.10917663574219, 0.10430717468261719, -17.830902099609375, 45.772064208984375, 13.842239379882812, -9.686294555664062, -101.38796997070312, 67.6092529296875, 36.228973388671875, -52.18267822265625, 95.83282470703125, 1.8676490783691406, 9.496376037597656, -74.61941528320312, 2.5705184936523438, 24.09832763671875, 2.225759506225586, 0.529022216796875, -7.899017333984375, 19.314102172851562, 11.621109008789062, -93.65618896484375, 24.540191650390625, -6.356544494628906, 12.074676513671875, 102.61918640136719, 2.163785934448242, 10.291259765625, -27.299209594726562, 7.232385635375977, 79.9517822265625, -24.041580200195312, 51.91957092285156, 2.0980224609375, 14.250946044921875, 82.05242919921875, 6.143951416015625, 115.45364379882812, -58.69818115234375, 98.91231536865234, 120.57196044921875, -84.06433868408203, 112.05325317382812, 1.977630615234375, 15.538467407226562, 2.4059066772460938, 33.7332763671875, 83.0238037109375, 65.850341796875, 34.861083984375, 92.02801513671875, 69.00418090820312, 104.095947265625, 14.862396240234375, -19.284027099609375, -0.923004150390625, 83.90775299072266, 36.95213317871094, 116.89579010009766, 16.49102783203125, 111.80224609375, 1.463531494140625, 6.82794189453125, 22.135604858398438, 101.54216003417969, 37.18963623046875, 71.46049499511719, -24.276519775390625, 49.538909912109375, -44.416259765625, -1.4860687255859375, 71.2767333984375, 32.39995574951172, -4.721778869628906, 82.63333129882812, -18.568103790283203, -25.67333984375, 119.302734375, 71.11737060546875, -63.80155944824219, -51.27287292480469, 22.267498016357422, -50.101837158203125, -36.10369873046875, 14.302947998046875, -10.9613037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000226.npy"}
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 39.268699645996094, "std": 56.08241271972656, "min": -124.28253173828125, "p10": -26.06492462158203, "median": 33.417503356933594, "p90": 111.68383026123047, "max": 153.47158813476562, "pos_frac": 0.7734375, "sample": [-82.89154052734375, 41.505828857421875, 20.595672607421875, 29.47064208984375, 74.40115356445312, 54.07170867919922, 121.57940673828125, 29.10999298095703, 7.4134674072265625, 33.18156433105469, 111.0406494140625, 75.22340393066406, 13.588088989257812, -5.357818603515625, 24.470413208007812, 45.559532165527344, 0.0, 117.3028564453125, -110.48095703125, -72.26643371582031, 74.599853515625, -62.08380126953125, 88.7291259765625, 105.64097595214844, 141.77850341796875, 153.47158813476562, 110.325927734375, -90.56315612792969, 97.41658782958984, -16.6046142578125, -7.360603332519531, 94.18777465820312, 13.4456787109375, 16.71649169921875, 33.6534423828125, -11.50244140625, 68.57678985595703, 74.7154769897461, 54.683990478515625, 121.810546875, 126.57257080078125, 95.86474609375, 14.6832275390625, 74.26080322265625, 38.120391845703125, -10.725341796875, 30.15899658203125, 59.384063720703125, -12.587936401367188, 31.64190673828125, 109.73236083984375, 27.06926727294922, -10.81451416015625, -26.514312744140625, 43.0535888671875, 139.03372192382812, 6.86199951171875, 78.135498046875, 113.18458557128906, 28.140827178955078, 115.74795532226562, -43.40325927734375, -20.4803466796875, 85.9615249633789, -124.28253173828125, 24.80181884765625, 30.278213500976562, 130.98870849609375, -47.517181396484375, 39.4739990234375, 77.26620483398438, 38.3170166015625, 0.7867546081542969, 83.122314453125, 110.90951538085938, 58.14288330078125, 59.3153076171875, 81.25772094726562, -41.822967529296875, 83.18344116210938, 118.86788940429688, 9.271697998046875, 28.713058471679688, 109.150390625, 47.620330810546875, 31.047317504882812, 51.3057861328125, 32.998809814453125, 31.336700439453125, 61.150146484375, -32.650665283203125, 78.1735610961914, 15.94232177734375, -63.591636657714844, 93.312744140625, 43.65081787109375, 88.80282592773438, 18.956832885742188, -12.393058776855469, 99.72814178466797, -1.3504638671875, 124.40673828125, 1.5747966766357422, 114.37579345703125, -56.86359405517578, 96.93072509765625, -3.4991073608398438, 0.392242431640625, 3.467132568359375, 1.6392669677734375, 76.66949462890625, 18.873428344726562, 96.04754638671875, 97.4178466796875, 12.883544921875, 31.121734619140625, 36.05015563964844, 46.349945068359375, 96.29074096679688, -7.423187255859375, -18.39593505859375, -23.276199340820312, 26.319549560546875, 18.847938537597656, -25.872329711914062, 5.462575912475586, 48.07720184326172, 72.02243041992188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000227.npy"}
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 40.310821533203125, "std": 54.926918029785156, "min": -75.98016357421875, "p10": -23.124162292480467, "median": 30.576366424560547, "p90": 109.14007568359374, "max": 140.2755126953125, "pos_frac": 0.7265625, "sample": [-65.90811157226562, -16.42639923095703, 71.9171142578125, 98.3548583984375, 68.57565307617188, -4.976917266845703, 128.8922119140625, 31.465240478515625, 74.0853271484375, 104.90725708007812, 5.87432861328125, -15.195777893066406, -12.141525268554688, 114.40866088867188, 55.689971923828125, 67.12069702148438, -7.277923583984375, 87.89647674560547, -30.133193969726562, 16.665008544921875, 93.26537322998047, 22.334041595458984, -75.98016357421875, -14.84271240234375, 14.75274658203125, 91.48934936523438, 23.483154296875, 2.3468475341796875, -17.307220458984375, 60.345550537109375, 14.018280029296875, 23.505035400390625, -34.713623046875, 103.015869140625, 38.587982177734375, 98.53848266601562, 68.15182495117188, 78.52326965332031, 8.0572509765625, -8.561233520507812, 100.32037353515625, -7.642364501953125, -15.98956298828125, -23.8460693359375, 80.11965942382812, 10.69171142578125, -8.374176025390625, 15.7911376953125, 43.197601318359375, 91.80502319335938, 39.02880859375, 125.0792236328125, 64.74259948730469, 20.95612335205078, 103.52859497070312, 36.967926025390625, 86.98284912109375, 82.80941772460938, 101.7342529296875, 120.99945068359375, -0.9272003173828125, -69.06779479980469, -27.18230438232422, -0.1256999969482422, 87.222900390625, 11.075698852539062, 0.0, 28.369232177734375, 42.375457763671875, 18.58539581298828, 12.368194580078125, 139.418701171875, 17.43822479248047, -6.6942596435546875, -37.11444091796875, 96.85653686523438, -17.7939453125, 105.30938720703125, -14.814346313476562, 112.15280151367188, -14.420074462890625, -62.667823791503906, 18.951904296875, 66.53363037109375, 106.26225280761719, 117.49984741210938, 9.870162963867188, 2.752410888671875, 123.20594787597656, 78.1728515625, -22.436538696289062, 104.5216064453125, 17.20782470703125, 91.97772216796875, 77.17726135253906, 53.06810760498047, -64.73727416992188, 4.669948577880859, 17.935314178466797, 106.19999694824219, 107.84890747070312, -71.44337463378906, -22.814773559570312, 58.06819152832031, 80.34291076660156, 60.4864501953125, 103.20077514648438, -54.29864501953125, -16.990203857421875, -5.1588134765625, 14.858100891113281, 116.20733642578125, 4.6564178466796875, 87.84634399414062, 102.72634887695312, 121.10195922851562, 130.55426025390625, 70.3291244506836, 11.320013046264648, -43.858062744140625, 81.24575805664062, 84.79388427734375, 128.76852416992188, 60.49220275878906, 140.2755126953125, 7.061492919921875, 11.573486328125, 29.68749237060547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000228.npy"}
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 50.18384552001953, "std": 57.52732849121094, "min": -93.64686584472656, "p10": -20.062901306152344, "median": 52.10560989379883, "p90": 118.88795928955076, "max": 170.6185302734375, "pos_frac": 0.8046875, "sample": [81.7020263671875, -27.82501220703125, 94.5003662109375, 78.53759765625, -9.553192138671875, -91.22039794921875, 76.83031463623047, 95.867919921875, 39.68450927734375, -62.3543701171875, 152.916748046875, 122.64591979980469, 84.57244873046875, -0.6887626647949219, 133.370849609375, 102.10455322265625, 19.71478271484375, 2.68194580078125, 18.48737335205078, 13.8187255859375, 41.06230163574219, 22.516311645507812, 103.4951171875, -9.239959716796875, 26.274253845214844, -66.96153259277344, 9.698318481445312, 9.690101623535156, 112.8961181640625, 79.47401428222656, 45.260498046875, 89.09130859375, 123.8583984375, 82.17303466796875, 33.522216796875, 130.23638916015625, -1.8271484375, 19.1900634765625, 147.442626953125, -93.64686584472656, 9.45074462890625, -58.3743896484375, 61.86286926269531, 99.52024841308594, 68.50785827636719, 139.12075805664062, 170.6185302734375, 109.02803039550781, 85.54344177246094, 97.24354553222656, 63.87535095214844, -1.0142402648925781, 62.18733215332031, 6.0960693359375, 89.68392181396484, 108.11891174316406, 100.30648803710938, 13.293121337890625, 11.560409545898438, -20.146011352539062, 32.049224853515625, -81.41213989257812, 104.69720458984375, 100.05108642578125, -11.253044128417969, -7.6515655517578125, 103.595703125, 15.912132263183594, 79.44287872314453, -3.7640380859375, 3.11431884765625, 59.40483856201172, -0.4348735809326172, 6.161865234375, 101.68624877929688, 60.23370361328125, 34.437522888183594, 95.05756378173828, 115.30960083007812, 44.7138671875, 112.92790222167969, 64.09886169433594, 39.13922119140625, 99.61247253417969, 42.142642974853516, 14.44256591796875, -37.10992431640625, 98.71902465820312, 133.28375244140625, 48.42658233642578, 141.62660217285156, 100.35781860351562, 13.223308563232422, 75.67108154296875, 14.649627685546875, 107.81474304199219, 46.35310363769531, 117.26461791992188, 43.170135498046875, 29.468307495117188, 87.88571166992188, 6.897529602050781, 44.735321044921875, 125.1524658203125, -0.4668121337890625, 12.591522216796875, 35.665679931640625, -20.02728271484375, 10.889095306396484, 117.27740478515625, 65.1453857421875, -8.978368759155273, 55.784637451171875, 48.33203125, 84.14035034179688, 68.25802612304688, 99.59793090820312, -44.708465576171875, 88.80284118652344, -75.53887939453125, 132.10580444335938, 70.59954833984375, 75.6627197265625, 30.169036865234375, 65.01885986328125, -65.73960876464844, 140.84658813476562, -29.684226989746094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000229.npy"}
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 38.205726623535156, "std": 58.00857162475586, "min": -172.18557739257812, "p10": -20.020603942871094, "median": 29.19525146484375, "p90": 113.3275634765625, "max": 156.344482421875, "pos_frac": 0.75, "sample": [117.8299560546875, 74.24850463867188, 30.64764404296875, 94.29398345947266, 154.40237426757812, 73.30419921875, 120.12626647949219, -6.320556640625, 113.94540405273438, 35.59046936035156, 17.538406372070312, 28.706085205078125, 12.835464477539062, -73.42008209228516, -172.18557739257812, 95.82254028320312, 96.5615234375, -2.391490936279297, -3.779449462890625, -57.7791748046875, 4.499267578125, 12.391605377197266, 120.29562377929688, -20.215972900390625, 0.0, 104.041259765625, -1.829071044921875, 18.018707275390625, 66.05154418945312, 96.07254028320312, 87.55438995361328, 103.08377075195312, 7.8984375, 39.96812438964844, -0.67022705078125, -77.48452758789062, 83.35113525390625, 92.09727478027344, 71.4765625, -22.610794067382812, 0.05489158630371094, -51.345916748046875, 9.722885131835938, 29.684417724609375, 20.73736572265625, 106.07818603515625, 75.15779113769531, 35.660003662109375, 94.23355102539062, -19.936874389648438, -11.199142456054688, 55.70133972167969, -34.89361572265625, 69.53729248046875, 100.68756103515625, -14.43743896484375, 40.91893005371094, 16.8123779296875, 102.946044921875, 27.741424560546875, -70.93791961669922, 86.41854858398438, 128.51995849609375, 82.33232879638672, 16.982322692871094, 4.667327880859375, -18.850631713867188, 93.7388916015625, -40.078765869140625, 98.2762451171875, 46.59539794921875, 5.8656005859375, -14.787139892578125, -13.578399658203125, 58.508056640625, 18.912261962890625, 9.722412109375, 41.12833786010742, 132.11404418945312, -8.619644165039062, -2.5242671966552734, 66.58787536621094, 27.818939208984375, -25.856945037841797, 20.04052734375, -125.26850891113281, -2.7240676879882812, -22.16710662841797, 135.15521240234375, 78.78192138671875, 143.0955810546875, 43.50779342651367, 11.815673828125, 8.251739501953125, 113.06277465820312, 112.41317749023438, 1.8439216613769531, 30.436859130859375, 95.18767547607422, 80.56681823730469, 109.74830627441406, 2.156808853149414, 80.92466735839844, 4.419183731079102, -13.541282653808594, 89.50215148925781, -2.04168701171875, 127.0496826171875, 77.36103820800781, 6.389900207519531, 79.49191284179688, 120.15055847167969, 77.23223876953125, 2.5591869354248047, 57.7811279296875, 32.418663024902344, 0.9512557983398438, 132.3735809326172, 156.344482421875, 0.4394989013671875, 0.01251220703125, 50.8701171875, 1.8666610717773438, 9.871734619140625, 26.918487548828125, 40.15650939941406, -1.3932647705078125, -14.46356201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000230.npy"}
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 30.512165069580078, "std": 52.85794448852539, "min": -106.09226989746094, "p10": -35.887414550781244, "median": 25.310312271118164, "p90": 97.02281799316405, "max": 145.76690673828125, "pos_frac": 0.765625, "sample": [80.63877868652344, 18.961883544921875, -37.760162353515625, -56.895416259765625, 86.38701629638672, 77.127685546875, 48.05755615234375, -4.2307586669921875, 79.11627197265625, -106.09226989746094, 103.24803161621094, -35.084808349609375, 60.226959228515625, 76.4677734375, -7.7314300537109375, 102.07891845703125, 10.054885864257812, 9.117000579833984, 39.15900421142578, 0.10040283203125, -23.396209716796875, -69.83628845214844, -13.09735107421875, 49.078338623046875, 9.249847412109375, 34.80553436279297, 33.50806427001953, 94.92369842529297, 2.81182861328125, -51.445220947265625, 26.48651123046875, 89.79884338378906, 96.73429870605469, 2.6954078674316406, 49.96282958984375, -11.876937866210938, 124.0955810546875, 45.5570068359375, 15.304641723632812, 41.82940673828125, -0.6281280517578125, 36.7039794921875, 0.48089599609375, 0.9663772583007812, 108.0592041015625, 96.25762939453125, -6.817138671875, 50.90965270996094, 23.08148193359375, 33.221832275390625, 89.54638671875, -51.43377685546875, 13.418975830078125, 15.377105712890625, 4.48443603515625, 13.468505859375, 93.61297607421875, -21.760345458984375, 22.963058471679688, 144.51800537109375, -68.6402587890625, 5.405887603759766, -1.6100311279296875, 45.20466613769531, 2.6627197265625, 89.0684814453125, 40.9464111328125, 24.0531005859375, 20.1497802734375, 51.90879821777344, 28.835372924804688, 58.37127685546875, 134.04522705078125, -27.112548828125, -92.40516662597656, -2.30084228515625, 48.85655975341797, -9.546539306640625, 88.07745361328125, 36.785614013671875, 16.5433349609375, 18.060516357421875, 36.94075012207031, 66.52053833007812, 6.7522125244140625, 19.747161865234375, 9.73760986328125, 105.83984375, 57.555145263671875, 7.8604278564453125, 60.242645263671875, 73.15361022949219, 77.24871826171875, 31.902297973632812, -69.27799987792969, 78.6768798828125, 24.134113311767578, 72.119873046875, 145.76690673828125, 0.6917877197265625, -77.66787719726562, 22.26220703125, 37.76765441894531, 136.72059631347656, 83.16059875488281, -14.25860595703125, -0.48867034912109375, 39.201416015625, 10.966812133789062, 57.50165557861328, 0.3604583740234375, 44.74664306640625, 33.717041015625, -88.4849853515625, 75.65866088867188, 135.6142578125, 11.729034423828125, 108.50509643554688, 14.292343139648438, 57.29240417480469, -18.038185119628906, 1.62591552734375, 97.69602966308594, -45.251220703125, 83.14993286132812, 0.0, -87.25450134277344, 115.49176025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000231.npy"}
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 35.229644775390625, "std": 56.74714279174805, "min": -92.65237426757812, "p10": -38.00269546508789, "median": 25.530906677246094, "p90": 107.63683319091797, "max": 170.13458251953125, "pos_frac": 0.6796875, "sample": [-41.19342041015625, 104.58407592773438, -3.3609237670898438, 115.5137939453125, 31.198028564453125, 72.67219543457031, 93.14242553710938, 48.104461669921875, -4.39703369140625, -6.0781097412109375, 84.30955505371094, 98.67950439453125, 14.629669189453125, -24.952056884765625, -43.594146728515625, 32.05615234375, 6.283683776855469, 5.679023742675781, 20.20562744140625, 21.459609985351562, 27.5677490234375, 10.493743896484375, -91.55996704101562, 74.68408203125, 99.69052124023438, 93.38226318359375, 23.060386657714844, 31.540725708007812, -1.1437873840332031, 135.7535400390625, 28.715789794921875, -8.056182861328125, -41.60675048828125, 71.49136352539062, 82.06655883789062, 97.08901977539062, -63.101043701171875, 40.014923095703125, 18.17169189453125, 104.4249267578125, 73.12333679199219, -37.41442108154297, -1.0312252044677734, 22.96587371826172, 11.458084106445312, 1.4693450927734375, 81.62174987792969, 14.02764892578125, 109.29713439941406, -0.29944610595703125, 28.565093994140625, -25.9451904296875, 94.36077117919922, 21.317108154296875, 4.6709136962890625, 87.4847412109375, 20.0859375, 98.32550811767578, -0.871795654296875, 85.9067611694336, 19.35388946533203, 136.6505126953125, 107.90711975097656, -24.59478759765625, 92.43278503417969, 97.48808288574219, -4.2069091796875, 32.134765625, -7.5130615234375, -10.834068298339844, 10.015983581542969, 111.04962158203125, 11.118316650390625, -41.51287841796875, 128.56884765625, 125.28836059570312, -2.2349090576171875, -39.375335693359375, 48.05230712890625, 63.27032470703125, 0.5831756591796875, 10.889251708984375, -17.14886474609375, 87.50882720947266, 105.2042236328125, -18.638263702392578, -6.555229187011719, 76.08370971679688, -5.0438232421875, -31.9522705078125, 90.75605773925781, 70.93106079101562, 0.0, 118.79513549804688, 67.38550567626953, 5.91546630859375, 55.76068115234375, 27.492935180664062, 102.73675537109375, 62.40557861328125, 91.66228485107422, 28.009567260742188, 132.86367797851562, -87.65487670898438, 88.74354553222656, 170.13458251953125, 11.71826171875, -21.639495849609375, 98.15542602539062, 70.41105651855469, -4.705657958984375, 117.62139892578125, -83.62255096435547, -12.955062866210938, -18.69116973876953, -52.20037841796875, 37.82067108154297, -40.22097396850586, 51.36737060546875, -56.79425048828125, 107.52099609375, -2.28326416015625, -92.65237426757812, 130.94677734375, 63.94120788574219, 79.03465270996094, 23.568878173828125, -23.614013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000232.npy"}
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 37.11216735839844, "std": 55.63750076293945, "min": -94.40464782714844, "p10": -30.19591064453125, "median": 26.607168197631836, "p90": 113.87651824951172, "max": 164.6378173828125, "pos_frac": 0.78125, "sample": [-48.1097412109375, 96.22710418701172, -0.3153076171875, -1.12158203125, 12.260398864746094, 108.89028930664062, 45.22633361816406, 39.80694580078125, 16.711589813232422, 14.119987487792969, 20.94012451171875, 26.148296356201172, 119.60826110839844, 55.027679443359375, 20.468490600585938, 0.102264404296875, 29.664093017578125, 113.498291015625, 30.381866455078125, 154.99649047851562, 100.37677001953125, 92.71627044677734, 106.392578125, 97.03572082519531, 67.81387329101562, 107.944580078125, 98.5604248046875, 9.1353759765625, 94.867431640625, -24.801666259765625, 27.787353515625, -6.872200012207031, 89.178466796875, 4.622211456298828, 114.75904846191406, 48.392730712890625, 24.31097412109375, -78.1694564819336, 77.6407470703125, 128.8084716796875, -52.96022033691406, -4.625150680541992, 3.1310997009277344, 25.8277587890625, -29.300537109375, 145.75534057617188, 63.058258056640625, 17.69525909423828, 23.6800537109375, 22.086509704589844, 118.72915649414062, 134.7733154296875, 16.92864990234375, -43.42344665527344, 37.95947265625, -9.83502197265625, 149.68878173828125, 33.01318359375, 101.47637939453125, 8.7164306640625, 54.05572509765625, 97.11639404296875, -47.05609130859375, 38.564762115478516, 10.076904296875, -29.954193115234375, -30.759918212890625, 62.305545806884766, -70.95526123046875, -78.165283203125, 68.59307861328125, 108.523193359375, 125.67987060546875, 64.32096862792969, 146.07684326171875, 53.651580810546875, 19.651870727539062, 15.169326782226562, 15.29644775390625, 3.617645263671875, 2.8656005859375, 6.2414398193359375, 13.644805908203125, 57.43756103515625, -40.05891418457031, 9.230379104614258, 164.6378173828125, 54.487518310546875, -16.081146240234375, 45.255828857421875, 20.11963653564453, -4.9467010498046875, -48.8162841796875, 134.6998291015625, 23.398536682128906, -26.2412109375, 1.5533161163330078, 88.87814331054688, -3.13623046875, 101.95069885253906, -14.956268310546875, 5.744873046875, -10.111907958984375, -94.40464782714844, 83.02301025390625, 100.30050659179688, 32.026153564453125, 80.5342025756836, 32.6986083984375, -78.5256118774414, 0.944488525390625, 92.7083740234375, 53.616302490234375, -5.274322509765625, 36.16123962402344, 27.0660400390625, -42.7860107421875, 116.57388305664062, 57.31053924560547, 40.45355224609375, 38.01964569091797, 4.473335266113281, 2.8323974609375, 68.28457641601562, 53.56854248046875, 17.730941772460938, 23.51348114013672, 20.494598388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000233.npy"}
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 38.551780700683594, "std": 65.4608154296875, "min": -116.96255493164062, "p10": -46.683200073242176, "median": 34.26715087890625, "p90": 119.66107940673828, "max": 188.0291748046875, "pos_frac": 0.765625, "sample": [-81.71565246582031, -103.9647216796875, -62.603607177734375, 0.46691131591796875, 47.94024658203125, 120.43391418457031, 119.32986450195312, 48.210906982421875, -30.112152099609375, 139.01235961914062, 107.21311950683594, -105.99957275390625, 101.0274658203125, 88.53889465332031, -24.75958251953125, 98.83944702148438, 35.581214904785156, 54.64447021484375, 79.76869201660156, 125.60820007324219, 87.87983703613281, 61.678680419921875, 69.06466674804688, 134.0716552734375, 13.848541259765625, 5.2473297119140625, 95.802734375, 87.41316223144531, 93.05166625976562, 34.745208740234375, 22.755691528320312, 54.63134765625, 105.71298217773438, 135.15438842773438, 91.02713012695312, 24.776527404785156, 32.99383544921875, 11.46307373046875, 1.8720703125, 56.3326416015625, 61.8148193359375, 18.209686279296875, 6.3377685546875, -55.119537353515625, 23.671142578125, -19.655044555664062, 89.47549438476562, 87.99771118164062, -3.387683868408203, 111.2658920288086, 27.130939483642578, 4.58392333984375, 135.3651123046875, 67.06011962890625, -14.056312561035156, 158.56707763671875, 93.05143737792969, 79.79180145263672, 99.10116577148438, 39.120025634765625, 10.87795639038086, 7.550294876098633, 126.9295654296875, 11.061492919921875, 7.803781509399414, -3.12554931640625, -34.40888977050781, 0.0, 25.2197265625, -13.7034912109375, 15.456573486328125, 28.0179443359375, -97.80284118652344, 2.2627487182617188, 103.34535217285156, 144.70327758789062, 104.469970703125, 28.7994384765625, -85.157470703125, 54.548492431640625, -1.28802490234375, 91.08003234863281, 115.20797729492188, -17.590972900390625, 43.88810729980469, 33.789093017578125, 86.12930297851562, 44.22235107421875, -116.96255493164062, 137.1544189453125, -76.1151123046875, 150.8380126953125, 19.735015869140625, 22.936431884765625, 91.28810119628906, 1.885406494140625, 93.18128967285156, 36.38507080078125, 1.53973388671875, 10.70068359375, 91.88442993164062, 0.0, 23.009971618652344, 56.53050231933594, 188.0291748046875, 136.2489013671875, -107.60447692871094, -43.067626953125, 117.0235595703125, 71.87872314453125, 81.01947784423828, 5.181640625, -60.107879638671875, 19.62310791015625, 62.879638671875, 108.82354736328125, -31.61553955078125, -32.385284423828125, -38.657928466796875, 1.31390380859375, 48.631805419921875, 8.34698486328125, -69.842041015625, 10.508857727050781, -93.863525390625, 105.7943115234375, -31.255401611328125, 114.13702392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000234.npy"}
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 36.05876922607422, "std": 62.565914154052734, "min": -127.27828979492188, "p10": -33.541762542724605, "median": 31.281654357910156, "p90": 111.1055923461914, "max": 154.95977783203125, "pos_frac": 0.7109375, "sample": [21.79840087890625, 6.912269592285156, -7.19281005859375, 0.7026901245117188, 9.647098541259766, -120.20632934570312, -61.585899353027344, -6.854072570800781, 97.77267456054688, 99.78945922851562, 11.328514099121094, 146.6409912109375, 80.27081298828125, 51.081512451171875, 79.14497375488281, 30.04941177368164, 92.56346893310547, 111.00961303710938, -1.7409477233886719, -7.21661376953125, 83.88592529296875, 24.34979248046875, 39.413787841796875, 94.63345336914062, 136.53382873535156, 134.8275146484375, 76.98309326171875, -18.344955444335938, 92.17335510253906, -6.800374984741211, 115.7305908203125, 1.720123291015625, 77.22286987304688, 103.1871566772461, 44.386566162109375, 97.6352767944336, 131.31719970703125, 102.06246948242188, 7.7569580078125, 146.2698974609375, -13.349639892578125, 63.789337158203125, -33.136451721191406, -39.020721435546875, 53.859832763671875, 133.954833984375, -111.18197631835938, -25.423721313476562, -93.20562744140625, 62.662315368652344, 56.461334228515625, -16.641510009765625, -0.7564163208007812, -21.32818603515625, 35.29638671875, 92.0738525390625, -26.918609619140625, 95.19841003417969, -9.0048828125, -68.2221908569336, 10.405216217041016, 8.306365966796875, 107.76168823242188, -26.15911865234375, 144.19964599609375, 72.50939178466797, -10.23687744140625, -93.53897094726562, 7.842094421386719, 41.55165100097656, -20.7137451171875, 49.27630615234375, -67.26655578613281, 91.28096008300781, 7.091026306152344, 10.25726318359375, 7.66064453125, 118.24629211425781, 6.677490234375, 53.728240966796875, 59.592529296875, 22.4013671875, -127.27828979492188, 31.192398071289062, 38.621124267578125, -15.466941833496094, 80.55268096923828, 102.09210205078125, 66.76812744140625, 3.1624908447265625, 146.68191528320312, 104.33248901367188, 72.51445007324219, 154.95977783203125, 111.32954406738281, 83.11614990234375, 40.35612487792969, 19.5325927734375, 102.1185531616211, -8.968904495239258, -0.9755630493164062, 105.8233642578125, -73.23672485351562, 63.36126708984375, -3.171142578125, 86.77173614501953, -5.468376159667969, 56.8077392578125, 88.53485107421875, -5.430267333984375, -1.931396484375, 31.37091064453125, 3.4217681884765625, -34.48748779296875, 23.03069305419922, 14.878864288330078, 102.16383361816406, -122.90142822265625, -51.01537322998047, 5.203125, 84.77363586425781, 83.2663803100586, 25.501800537109375, 112.00776672363281, 84.24176025390625, 54.09581756591797, 28.089019775390625, 0.3405303955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000235.npy"}
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 41.193450927734375, "std": 57.963417053222656, "min": -123.45724487304688, "p10": -30.49826354980468, "median": 33.45738220214844, "p90": 110.97137451171875, "max": 159.3360595703125, "pos_frac": 0.7734375, "sample": [59.08094787597656, 22.62890625, -72.32846069335938, 63.938201904296875, 75.08909606933594, 92.19186401367188, 9.474319458007812, 104.6005859375, -13.472793579101562, -39.93083190917969, -83.613525390625, 25.967140197753906, -47.57392120361328, 63.09971618652344, 76.80426025390625, 104.75515747070312, -0.6786212921142578, 93.37393188476562, 74.90589904785156, 22.47967529296875, 26.510162353515625, 86.88060760498047, 40.206146240234375, -55.379676818847656, 52.92085647583008, 93.32530212402344, 91.03416442871094, 6.152839660644531, 69.36737060546875, 106.29385375976562, 22.953048706054688, 34.42120361328125, 76.13684844970703, 19.282371520996094, 14.690315246582031, 18.791831970214844, 11.5146484375, 133.61630249023438, 102.62026977539062, 2.5253448486328125, 9.637466430664062, 104.7644271850586, 108.10153198242188, 33.400970458984375, 143.0765380859375, 1.5049190521240234, -47.59696960449219, 4.1881103515625, 25.676925659179688, 0.0, -5.230255126953125, 29.481353759765625, 94.430908203125, 99.21694946289062, 10.764001846313477, -27.933258056640625, -102.68618774414062, 103.26458740234375, 11.1292724609375, -40.7703857421875, 82.48822021484375, 117.22280883789062, 53.03912353515625, 59.87517547607422, 3.86151123046875, -36.4832763671875, 89.8154296875, -123.45724487304688, 72.70797729492188, 6.136077880859375, -17.346885681152344, 159.3360595703125, 62.296897888183594, 142.3988037109375, 1.62908935546875, -5.9627685546875, -51.433929443359375, -4.752532958984375, 111.01570129394531, 73.9757080078125, 46.60725402832031, 65.93402099609375, -46.24700927734375, 86.51742553710938, 102.39358520507812, 33.5137939453125, 116.95477294921875, 139.751708984375, -6.933807373046875, 80.52035522460938, 58.814697265625, 109.41336059570312, 73.4566650390625, 21.801986694335938, 5.95904541015625, 110.95237731933594, 104.38204956054688, 74.47135925292969, 92.66207122802734, 118.85400390625, 103.81637573242188, 33.20855712890625, -10.523284912109375, 6.3263397216796875, 0.0, 45.4371337890625, 8.987152099609375, 28.588058471679688, -101.40240478515625, 102.56904602050781, 4.0999908447265625, -8.800346374511719, -19.70867919921875, -20.553298950195312, 46.300140380859375, 105.7306900024414, 123.68666076660156, 11.4779052734375, 8.963905334472656, -22.578414916992188, 117.26454162597656, 92.912353515625, 116.86207580566406, 19.87432861328125, -1.99566650390625, 116.21954345703125, 8.30209732055664, 32.478851318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000236.npy"}
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 34.96562194824219, "std": 59.21298599243164, "min": -130.1951904296875, "p10": -29.08511123657226, "median": 25.048126220703125, "p90": 115.17004470825195, "max": 194.24746704101562, "pos_frac": 0.734375, "sample": [38.47125244140625, 13.64190673828125, 55.45967483520508, 119.98446655273438, -10.31658935546875, 30.995285034179688, 76.96964263916016, 88.17926788330078, 42.47747802734375, 39.28010559082031, -3.130615234375, 54.09629821777344, 140.8741455078125, -55.29541778564453, -39.91278839111328, 4.180858612060547, 46.1669921875, 10.003013610839844, 194.24746704101562, 89.82266998291016, 80.37232971191406, 31.458450317382812, 114.14459228515625, 68.11006927490234, 11.593620300292969, 3.0701904296875, 77.6400146484375, 0.0, 78.66934204101562, 86.74722290039062, 0.6794891357421875, 14.270950317382812, -33.371185302734375, 22.561180114746094, -16.162933349609375, 38.45941162109375, 1.7242889404296875, 0.241424560546875, -130.1951904296875, -27.24822235107422, -3.0298995971679688, -1.6221923828125, 1.4739227294921875, 96.20169067382812, 8.098358154296875, 29.45501708984375, 113.82223510742188, 30.728271484375, 39.5098876953125, 3.4418563842773438, 0.4552764892578125, 147.81265258789062, 66.17697143554688, 55.521934509277344, 13.700469970703125, 85.10877990722656, 109.86004638671875, 62.2371826171875, 70.80645751953125, 22.734634399414062, 124.78741455078125, 36.86475372314453, 24.832916259765625, 17.3519287109375, 61.693511962890625, -7.078582763671875, 124.10787963867188, 108.78834533691406, -56.330596923828125, -65.61468505859375, 0.6678962707519531, 134.56973266601562, -87.53756713867188, 11.28546142578125, 108.49168395996094, -3.335926055908203, -72.83248901367188, -4.8782501220703125, 147.52964782714844, -20.4268798828125, 16.712860107421875, 25.6571044921875, 122.16683959960938, 29.009796142578125, 119.300537109375, -69.1566162109375, -3.8481216430664062, 6.53143310546875, 117.5627670288086, 25.263336181640625, 10.7733154296875, 143.71084594726562, -8.333641052246094, -0.594451904296875, -37.3560791015625, 133.060302734375, -51.98626708984375, 1.915740966796875, 110.45101165771484, 88.45219421386719, 16.58428955078125, 9.335418701171875, 98.69039916992188, -1.7578506469726562, 2.3793716430664062, 6.194890975952148, 99.02290344238281, -23.731597900390625, -15.1466064453125, -12.689239501953125, 3.027252197265625, 113.55073547363281, -53.90380859375, -1.4052200317382812, 63.95610809326172, 38.8267822265625, -126.58474731445312, 65.93266296386719, 90.62335205078125, 108.79458618164062, 100.083740234375, -17.8048095703125, 32.7574462890625, 86.24192810058594, -2.685056686401367, 32.4573974609375, 19.49853515625, 69.66796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000237.npy"}
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 39.01617431640625, "std": 62.743995666503906, "min": -96.2444076538086, "p10": -33.67469329833984, "median": 27.845252990722656, "p90": 122.14327163696288, "max": 170.06484985351562, "pos_frac": 0.6953125, "sample": [-52.82563781738281, 2.4797439575195312, 5.9510040283203125, 13.390724182128906, -57.66961669921875, 146.882080078125, 102.99281311035156, 44.838043212890625, 99.31622314453125, -50.66265869140625, -2.91094970703125, 84.57179260253906, -3.899169921875, 17.06378173828125, -16.366073608398438, -72.01644897460938, 130.62744140625, 107.22833251953125, -2.4380645751953125, 51.496910095214844, 107.38885498046875, 75.00283813476562, 39.5618896484375, 115.72411346435547, 150.5076446533203, 29.466934204101562, -12.96221923828125, 2.9222946166992188, 14.755279541015625, -88.95582580566406, -90.09201049804688, -96.2444076538086, 133.84521484375, -85.4012451171875, -8.851104736328125, 132.13070678710938, 32.4466552734375, 97.35226440429688, 3.8419570922851562, 90.69427490234375, 146.90380859375, 12.7950439453125, 54.442535400390625, 160.43017578125, 81.08370971679688, 115.34765625, 65.62564086914062, 61.067840576171875, -12.4276123046875, 34.87548828125, 118.86112976074219, 155.7279052734375, 83.33334350585938, -9.526344299316406, 18.4256591796875, 13.79107666015625, -16.512832641601562, 5.668806076049805, -9.363067626953125, 80.82713317871094, -7.41571044921875, 3.248931884765625, -24.873260498046875, -94.92242431640625, 100.14134216308594, 65.35235595703125, 92.86903381347656, 109.66034698486328, 86.6668701171875, 112.38443756103516, -49.42681884765625, -34.8675537109375, 57.90824890136719, -5.560211181640625, -1.4754180908203125, 74.11453247070312, 128.41986083984375, -16.60467529296875, 88.87612915039062, 102.2127685546875, 84.28639221191406, 126.18925476074219, -45.0921745300293, -10.514545440673828, 4.1484375, 25.284454345703125, 95.76155090332031, -3.470449447631836, 73.5336685180664, 35.6182861328125, 0.3870506286621094, -13.9512939453125, -33.16346740722656, 32.967498779296875, 170.06484985351562, -9.4500732421875, 34.639556884765625, 75.45003509521484, 16.94115447998047, 7.917819976806641, 25.219406127929688, 129.20358276367188, 120.4092788696289, 32.587158203125, 61.33187484741211, 5.4798431396484375, -14.529006958007812, 110.47134399414062, 24.401123046875, -48.438140869140625, 26.22357177734375, 112.83816528320312, 14.76375961303711, 13.806549072265625, 108.31259155273438, -12.841476440429688, 0.51324462890625, 108.992431640625, -9.59185791015625, -9.802703857421875, 112.67630004882812, 140.65255737304688, 36.105133056640625, -29.270660400390625, 38.806732177734375, 7.32293701171875, 86.08767700195312, -4.4796142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000238.npy"}
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 32.755279541015625, "std": 63.66354751586914, "min": -158.4063720703125, "p10": -40.69602355957031, "median": 18.500411987304688, "p90": 127.00680618286133, "max": 169.02276611328125, "pos_frac": 0.7109375, "sample": [-66.52055358886719, -51.18522644042969, 139.84869384765625, 68.1251220703125, -14.670684814453125, 1.091278076171875, -149.791015625, 78.45008087158203, 134.12481689453125, 1.1809310913085938, 86.3839111328125, -158.4063720703125, -99.71615600585938, -5.155975341796875, 92.88826751708984, 84.4129638671875, 102.96636962890625, 127.22911834716797, 53.52581787109375, -11.3966064453125, -1.0578460693359375, 109.08868408203125, 2.6121463775634766, 24.066574096679688, 2.8181686401367188, 83.32180786132812, 0.05201530456542969, 98.8143539428711, 18.920196533203125, -14.342071533203125, -44.91680908203125, 97.57559204101562, 69.11042022705078, -2.8152008056640625, 147.07867431640625, -14.967926025390625, 26.042572021484375, 100.69525146484375, 169.02276611328125, 115.80963134765625, 88.27544403076172, -5.3070068359375, 25.286788940429688, 0.79071044921875, 149.384765625, -11.462013244628906, -14.1688232421875, 73.74554443359375, -79.47639465332031, 99.83877563476562, -73.12373352050781, 61.08290481567383, 17.450294494628906, 64.18496704101562, 39.96710968017578, -17.6080322265625, 48.1497802734375, 9.169151306152344, 5.038726806640625, -38.887115478515625, 18.537750244140625, 119.93455505371094, 7.279187202453613, 95.48162841796875, 28.98841094970703, 30.843017578125, 101.49120330810547, 0.76025390625, 6.9268035888671875, 141.4874267578125, 1.1440677642822266, 11.91827392578125, 12.9554443359375, 39.104644775390625, 104.43455505371094, -8.695068359375, 89.3067626953125, 18.513671875, -71.93832397460938, 10.116165161132812, -0.37103271484375, 14.855690002441406, 144.13858032226562, -19.724884033203125, -3.27178955078125, 135.24853515625, -51.207733154296875, 30.777427673339844, 10.297409057617188, -47.69793701171875, 114.44233703613281, -4.219699859619141, -10.691871643066406, 109.65853118896484, 3.1902999877929688, 66.23506164550781, 7.975593566894531, 3.236572265625, 141.77850341796875, 40.648681640625, 0.25072479248046875, 140.65921020507812, 15.93194580078125, -29.35723876953125, 52.59027099609375, -2.617717742919922, -27.804046630859375, 29.673370361328125, 39.684356689453125, 131.41357421875, 6.8827056884765625, 45.83599853515625, 53.666412353515625, 145.85009765625, 126.91152954101562, 0.0, 21.58355712890625, 0.5424680709838867, 110.90304565429688, -5.02728271484375, -45.439666748046875, 83.40300750732422, 18.487152099609375, 24.305206298828125, 34.34844970703125, -81.72715759277344, -2.444438934326172, 17.635665893554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000239.npy"}
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 42.2647705078125, "std": 58.10175323486328, "min": -113.33140563964844, "p10": -22.228525733947748, "median": 37.9495849609375, "p90": 116.08681640625, "max": 162.41586303710938, "pos_frac": 0.7265625, "sample": [8.003896713256836, 116.42532348632812, 68.64419555664062, 111.26956176757812, -18.300003051757812, -4.3536376953125, 70.61572265625, 38.854248046875, 0.0, 138.89657592773438, 35.42694091796875, -27.17852783203125, -85.17420959472656, 10.850738525390625, -20.866283416748047, 100.40386199951172, 66.60577392578125, 75.74146270751953, 95.37139892578125, 67.36077117919922, 16.800140380859375, -0.78594970703125, 82.72528076171875, 106.975341796875, 16.196868896484375, 105.79203796386719, 85.66444396972656, -43.35270690917969, 119.65078735351562, 115.94174194335938, 40.423736572265625, 15.5517578125, 135.59942626953125, 94.23190307617188, 5.350341796875, -13.7091064453125, 126.78176879882812, 78.71493530273438, 96.80006408691406, 14.082145690917969, 23.53179931640625, 48.8194580078125, 121.05534362792969, -9.296142578125, 81.490234375, 48.94580078125, 93.15631103515625, 42.55656433105469, 67.54339599609375, 58.948638916015625, 112.70755004882812, 69.27454376220703, 0.0, 72.45386505126953, -113.33140563964844, 24.50244140625, -4.7762451171875, -6.8852081298828125, -7.74139404296875, -7.757598876953125, 88.73417663574219, 100.39862060546875, -6.169525146484375, -12.0550537109375, 117.73126220703125, 105.39462280273438, 46.575469970703125, 88.93903350830078, 114.20191955566406, 54.61602783203125, 140.06146240234375, -31.09991455078125, 0.0, 17.073806762695312, -17.5003662109375, -12.747894287109375, 146.56610107421875, -1.7159271240234375, 19.006881713867188, 29.8504638671875, -10.235145568847656, 37.808868408203125, 96.25570678710938, 27.31884765625, 86.08204650878906, 162.41586303710938, -34.682861328125, 159.373291015625, -104.679443359375, 115.2261962890625, 6.405200958251953, -11.112060546875, 13.67901611328125, 101.85380554199219, 100.82554626464844, 44.96482849121094, 43.18743896484375, 85.96910095214844, 15.884061813354492, -25.40709114074707, -19.061477661132812, 7.451723098754883, 1.1733531951904297, 10.497833251953125, 104.21247863769531, 39.07789611816406, 83.87741088867188, 115.3131103515625, 38.090301513671875, 10.9991455078125, 117.05106353759766, 18.126068115234375, 29.9075927734375, -85.53551483154297, 29.523391723632812, 3.5348052978515625, -49.7479248046875, 78.97538757324219, -32.6263427734375, 98.27777099609375, 0.0, 9.971717834472656, 69.46107482910156, 19.4052734375, 22.0689697265625, -69.05237579345703, -38.94677734375, 135.63943481445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000240.npy"}
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 32.96603775024414, "std": 62.19008255004883, "min": -114.26561737060547, "p10": -40.75841369628906, "median": 17.41010284423828, "p90": 115.40965728759764, "max": 173.07028198242188, "pos_frac": 0.703125, "sample": [0.902923583984375, 118.16839599609375, 86.90884399414062, 61.578338623046875, 34.528465270996094, -30.7611083984375, 23.564453125, 0.8774890899658203, 73.26506042480469, 97.533447265625, 108.82681274414062, 0.0, -12.048583984375, 118.236083984375, 93.44158935546875, 149.7020263671875, 7.53656005859375, -2.3105316162109375, 14.36236572265625, 124.35101318359375, 73.26296997070312, 106.12857818603516, -15.486907958984375, -12.01336669921875, 96.23970794677734, 17.37640380859375, -40.64508056640625, 27.498016357421875, 102.36383056640625, -8.9720458984375, 124.63587188720703, -37.4044189453125, 106.29457092285156, 50.94989013671875, 11.020477294921875, 0.843536376953125, -8.3133544921875, -9.957962036132812, 5.119777679443359, 108.6541748046875, 10.187889099121094, 18.777877807617188, 97.0664291381836, 147.509521484375, 12.966598510742188, 89.539794921875, 58.95562744140625, -112.133056640625, 13.498764038085938, 24.8204345703125, 21.230224609375, 2.1458358764648438, 2.479747772216797, 92.86463928222656, -22.97052001953125, -9.651901245117188, 112.71682739257812, 8.176300048828125, 23.166473388671875, -66.91166687011719, 49.115142822265625, 98.61485290527344, 45.80706787109375, 97.74444580078125, -5.3587799072265625, 19.523773193359375, -12.881759643554688, 31.700027465820312, 2.5791244506835938, 99.40631103515625, -72.41535186767578, 7.8018646240234375, 131.6270294189453, 77.40985107421875, -22.845458984375, -51.7513427734375, 126.4140625, 79.78324890136719, 2.6682891845703125, -0.287994384765625, 1.900390625, 0.18955612182617188, 146.75845336914062, 32.361114501953125, 32.10267639160156, 31.839828491210938, -51.8853759765625, 103.82416534423828, -8.782424926757812, 114.22734069824219, 6.547119140625, -74.46492004394531, -2.9683151245117188, -0.4371185302734375, 105.96792602539062, 88.08290100097656, 3.3208389282226562, 17.443801879882812, 125.16607666015625, -41.022857666015625, 85.92729187011719, -55.58770751953125, 21.63153076171875, 4.402679443359375, 110.173095703125, 111.21441650390625, -78.99264526367188, 6.937767028808594, 131.44097900390625, 70.62736511230469, -23.2740478515625, -114.26561737060547, -29.9661865234375, -7.5497894287109375, 89.7337875366211, -0.6458797454833984, 139.42788696289062, 54.01091003417969, -71.71598815917969, 75.983642578125, 173.07028198242188, 4.445343017578125, 15.723236083984375, -111.16323852539062, 10.59161376953125, -21.809295654296875, 55.672760009765625, -51.91131591796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000241.npy"}
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 39.71366882324219, "std": 56.49221420288086, "min": -123.35182189941406, "p10": -20.74032592773437, "median": 42.75908279418945, "p90": 116.49526672363281, "max": 146.229248046875, "pos_frac": 0.7578125, "sample": [98.60017395019531, 13.318199157714844, 79.439208984375, 11.10672378540039, 16.868682861328125, 58.13865661621094, 34.45729064941406, 77.96983337402344, -27.484954833984375, 11.493507385253906, 137.156982421875, 92.89849853515625, 54.58202362060547, -14.6314697265625, -96.1019515991211, 51.883216857910156, 55.15074157714844, 104.51013946533203, -123.35182189941406, 49.76700973510742, 12.848669052124023, 135.59190368652344, 128.10888671875, 51.90785217285156, 55.64935302734375, 142.45602416992188, -0.8714447021484375, 131.6031494140625, -0.665435791015625, 3.6253414154052734, 80.42095947265625, 100.79339599609375, -1.8253173828125, 126.55349731445312, 40.5025634765625, 144.32901000976562, 123.12284851074219, 2.4717140197753906, -87.5387954711914, 1.4921798706054688, 94.78152465820312, -20.050384521484375, -2.04931640625, 58.641571044921875, -22.350189208984375, -6.5768890380859375, 30.50323486328125, 85.13919067382812, 78.3199462890625, 2.7946243286132812, -7.768646240234375, -3.437358856201172, -2.6244163513183594, 45.41081237792969, -7.526451110839844, -3.4532318115234375, 104.55592346191406, 12.250900268554688, -12.67919921875, 98.08868408203125, 46.087249755859375, 0.2600727081298828, -28.90740966796875, 122.49740600585938, 8.3946533203125, -1.5593948364257812, 146.229248046875, 65.05972290039062, 69.06379699707031, 22.355972290039062, 30.413543701171875, 57.198333740234375, -24.844146728515625, 13.210090637207031, 89.38671875, 23.996017456054688, 84.7701187133789, 19.083240509033203, 16.46142578125, -69.67448425292969, 57.71538543701172, 86.56654357910156, 22.5933837890625, 46.12908935546875, -72.94619750976562, 14.864639282226562, 5.56407356262207, 50.16735076904297, 43.18865966796875, 90.26567077636719, 145.76638793945312, 6.144134521484375, 62.14604187011719, 28.972591400146484, -3.2753753662109375, 17.83270263671875, 72.907470703125, 102.46710205078125, 75.45439147949219, 15.978179931640625, 1.4907646179199219, 46.27264404296875, -30.47756576538086, -17.921035766601562, 99.55325317382812, 55.36389923095703, 57.803680419921875, -113.90963745117188, 73.277099609375, 54.88359069824219, 39.208221435546875, 9.51861572265625, 116.47555541992188, 130.3059539794922, 78.52328491210938, 84.22088623046875, 116.541259765625, 42.329505920410156, -69.43533325195312, 89.4144287109375, -11.98236083984375, 59.38214874267578, 12.941970825195312, -48.761260986328125, 68.7001953125, 98.68379211425781, -17.519195556640625, 102.1630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000242.npy"}
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 45.022064208984375, "std": 55.994441986083984, "min": -114.63984680175781, "p10": -15.192016601562496, "median": 33.30680847167969, "p90": 119.22822570800781, "max": 164.352294921875, "pos_frac": 0.7578125, "sample": [-49.52874755859375, -23.569305419921875, 29.210037231445312, 76.70553588867188, 111.81076049804688, 74.72528076171875, 19.9820556640625, 131.3739471435547, -68.95651245117188, 95.21794128417969, 24.51397705078125, 143.88677978515625, 26.421594619750977, -25.559844970703125, 55.595672607421875, -12.477249145507812, -28.79034423828125, 74.89712524414062, 133.8050537109375, -79.10841369628906, -0.37532806396484375, 16.63018798828125, -14.085540771484375, 0.3024139404296875, 67.0010757446289, 117.82830810546875, 164.352294921875, 95.80972290039062, -4.9394683837890625, 49.625030517578125, 9.828023910522461, 94.14356994628906, 10.700752258300781, 25.649658203125, 107.31134033203125, 12.102508544921875, 122.03549194335938, 15.027984619140625, 77.031005859375, 103.26498413085938, -13.687744140625, 101.16015625, 60.325927734375, 70.81097412109375, 72.38189697265625, 105.86785888671875, 127.3385009765625, -47.01519775390625, 54.78857421875, 67.66444396972656, 58.76593017578125, -114.63984680175781, 2.4009742736816406, 39.5675048828125, 6.6966552734375, 3.97027587890625, 23.510818481445312, 15.949371337890625, 15.234527587890625, -1.87103271484375, 144.53900146484375, 61.153167724609375, 92.52120208740234, -12.1470947265625, -9.206878662109375, 89.98287200927734, 36.32135009765625, 102.67831420898438, 90.06703186035156, -5.017822265625, 31.969085693359375, -28.171417236328125, -14.30889892578125, 117.35157775878906, 88.45864868164062, 23.511489868164062, -0.286407470703125, 64.61233520507812, 4.541923522949219, 22.240623474121094, -10.141170501708984, 98.3109359741211, -26.690635681152344, 76.97659301757812, 9.426055908203125, 0.0, 3.290252685546875, 65.14460754394531, 99.7005386352539, 108.10238647460938, 43.415077209472656, 80.73419189453125, 86.22467041015625, -5.481056213378906, 117.82318115234375, 119.54769897460938, 126.38519287109375, 11.212417602539062, 3.719207763671875, 34.64453125, 22.93243408203125, -5.0953369140625, -8.35845947265625, 27.840087890625, -17.25262451171875, 119.09130859375, 137.7786865234375, 95.63467407226562, 3.1816177368164062, 29.16168212890625, 11.58871078491211, 12.037940979003906, 111.41943359375, -13.43994140625, -22.3836669921875, 153.67819213867188, 20.357379913330078, 115.66278076171875, -11.81396484375, 27.354400634765625, 66.20024108886719, 94.42819213867188, 149.20834350585938, 46.92951965332031, 136.5249786376953, 56.36181640625, 53.38397216796875, -21.362640380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000243.npy"}
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 45.58747100830078, "std": 61.83949279785156, "min": -135.96133422851562, "p10": -20.05463867187499, "median": 40.002227783203125, "p90": 125.60113525390625, "max": 209.96646118164062, "pos_frac": 0.75, "sample": [-135.96133422851562, 52.190948486328125, 46.423614501953125, 1.0618438720703125, 30.026351928710938, 15.324161529541016, 115.5113525390625, -30.632644653320312, 144.6702880859375, 23.432228088378906, 103.9931640625, 56.19561767578125, 0.7711639404296875, -6.282562255859375, 10.434288024902344, 19.317153930664062, 89.42449951171875, 89.71905517578125, 91.41413879394531, 97.6868896484375, 18.535491943359375, 104.37328338623047, -1.8557510375976562, -8.386184692382812, 110.90214538574219, -3.31298828125, -34.82991027832031, -63.186317443847656, 113.04692840576172, -26.0191650390625, 7.4160919189453125, -17.05231475830078, -4.859710693359375, -4.539253234863281, -102.25784301757812, 9.606391906738281, 146.20123291015625, -7.746337890625, 88.65652465820312, 111.052490234375, 3.18707275390625, 52.3963623046875, 132.82369995117188, 79.34484100341797, 134.8212890625, 64.4593505859375, 0.8914794921875, 73.28691101074219, 37.94831848144531, 79.227783203125, 55.69281005859375, 101.31292724609375, 0.0, 95.6038818359375, -17.4984130859375, -46.890533447265625, -54.287109375, 97.79051208496094, -2.041351318359375, -7.3266143798828125, 125.353759765625, -33.66502380371094, 4.871307373046875, 84.89458465576172, 5.187095642089844, 126.1783447265625, 118.0702896118164, 134.73455810546875, 209.96646118164062, 15.300811767578125, 53.80949401855469, 70.63505554199219, -39.94390869140625, 71.72042846679688, 35.49971008300781, 5.9540557861328125, 132.20947265625, -9.233734130859375, 134.7962646484375, 86.87701416015625, 0.24462890625, 63.479400634765625, 11.479904174804688, -69.22052001953125, 20.933685302734375, 83.29486083984375, -4.495697021484375, 94.56947326660156, 100.76570129394531, 5.380279541015625, 4.23309326171875, -2.2896728515625, 41.47955322265625, 70.70068359375, 143.07382202148438, 127.80975341796875, 115.3919677734375, 3.8456954956054688, 154.380859375, 83.86509704589844, 0.883087158203125, 90.71536254882812, 38.52490234375, 97.82329559326172, -12.668930053710938, 94.20114135742188, 24.997421264648438, 17.1297607421875, 123.5374984741211, 69.38290405273438, -26.524139404296875, 67.6864013671875, 3.3087234497070312, 102.0484619140625, 69.9981689453125, 148.90869140625, -2.315305709838867, 84.91983795166016, 88.48153686523438, -10.271133422851562, 12.539794921875, 15.752090454101562, 104.38024139404297, -112.72811889648438, 29.879913330078125, 83.28689575195312, 119.60791015625, -1.637481689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000244.npy"}
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 46.042259216308594, "std": 62.433719635009766, "min": -132.34483337402344, "p10": -24.32706604003906, "median": 40.226463317871094, "p90": 118.12777252197264, "max": 168.86328125, "pos_frac": 0.78125, "sample": [18.817138671875, -13.804534912109375, 35.27848815917969, 57.14818572998047, 2.0030517578125, 26.64361572265625, 90.71514892578125, 7.10107421875, 11.584320068359375, 96.1748046875, 36.822723388671875, -25.314666748046875, 21.37652587890625, 138.40777587890625, 80.80621337890625, 15.3970947265625, 154.31155395507812, 145.72943115234375, 13.741561889648438, 168.86328125, -132.34483337402344, -27.531005859375, -64.07901000976562, -28.33465576171875, 30.96282958984375, 84.71603393554688, 0.0, 36.6466064453125, 78.8949966430664, -8.350555419921875, 94.22264099121094, 98.32106018066406, 89.67959594726562, 112.60482788085938, 4.39337158203125, 147.64324951171875, 3.61810302734375, 38.88179016113281, 95.47840118408203, 68.04458618164062, 133.0477294921875, 13.02484130859375, 72.55259704589844, 98.46572875976562, 9.5294189453125, 97.66265869140625, 29.10491943359375, 45.382598876953125, 52.77436447143555, -1.6709442138671875, -23.90380859375, 51.620361328125, 8.234466552734375, 116.19912719726562, 17.967315673828125, -7.005462646484375, 89.13479614257812, 62.55029296875, 41.571136474609375, 122.62794494628906, 148.336669921875, 6.484800338745117, -2.5772857666015625, 73.20230102539062, 49.68254089355469, 101.21820831298828, 19.27410125732422, -40.48814392089844, 56.2847900390625, -89.00387573242188, 52.52952575683594, 100.18868255615234, 87.61344146728516, 90.83737182617188, 100.69463348388672, -15.353607177734375, -65.57797241210938, 33.26526641845703, 163.01065063476562, 115.16708374023438, 7.390138626098633, 108.86173248291016, -82.29513549804688, 7.732597351074219, 10.0076904296875, -54.73780822753906, 4.388946533203125, 115.40994262695312, 102.22457885742188, 108.37200927734375, 84.88084411621094, 20.349517822265625, -3.1539840698242188, 146.76712036132812, 61.68359375, 162.42312622070312, 17.892364501953125, 29.86865234375, 56.71947479248047, 90.1802978515625, -93.88478088378906, 115.73519897460938, 81.65301513671875, -20.305999755859375, -6.712882995605469, 82.81671142578125, 31.50439453125, 106.86273193359375, 80.89897155761719, 107.38053894042969, 10.77154541015625, -14.334030151367188, 38.387481689453125, 41.900665283203125, 100.15487670898438, 20.85675048828125, -3.4614410400390625, -21.923011779785156, 140.55078125, 0.4631996154785156, 110.74136352539062, 18.62744140625, 75.753173828125, 164.4864501953125, -125.19863891601562, -36.60845184326172, 0.0, 72.39756774902344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000245.npy"}
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 29.66974639892578, "std": 65.29949951171875, "min": -127.91751098632812, "p10": -47.52252502441406, "median": 15.327301025390625, "p90": 126.39483337402343, "max": 162.534912109375, "pos_frac": 0.6484375, "sample": [77.77701568603516, 95.48367309570312, 87.11225891113281, -2.86517333984375, 74.3115234375, -5.522560119628906, 162.534912109375, 11.425445556640625, 19.913898468017578, 49.60740661621094, -28.480754852294922, 86.48853302001953, 24.149169921875, 18.462188720703125, -17.261024475097656, -54.513641357421875, -14.899627685546875, 138.2590789794922, 138.31753540039062, -58.615028381347656, 118.45013427734375, 150.20742797851562, 20.64776611328125, 101.36736297607422, 40.994537353515625, -4.683135986328125, 9.939926147460938, -59.829444885253906, 24.857345581054688, -7.234588623046875, 40.47772216796875, -33.138427734375, 123.29638671875, 21.387313842773438, 14.50933837890625, -6.3355865478515625, -114.17919921875, 122.21200561523438, 3.66845703125, 0.9338626861572266, 15.64678955078125, -108.07769775390625, -22.6104736328125, 125.3189697265625, -47.147735595703125, -20.907470703125, -70.70625305175781, -21.194107055664062, 2.877685546875, 81.64114379882812, 27.889984130859375, 144.579833984375, -55.2225341796875, 20.96221923828125, 130.39895629882812, -16.878297805786133, 32.39898681640625, -23.611526489257812, -127.91751098632812, 35.87289810180664, 140.60205078125, 73.69769287109375, 91.62413024902344, 59.69647216796875, -34.7442626953125, 72.90201568603516, 141.02005004882812, 80.86233520507812, -16.11810302734375, 142.27285766601562, 128.9193115234375, 29.27752685546875, -43.07463073730469, 0.77044677734375, 22.28973388671875, 1.17205810546875, 110.53375244140625, 20.396148681640625, -26.98455810546875, -70.21568298339844, -48.39703369140625, -6.752349853515625, 100.95465087890625, 80.71792602539062, -26.092178344726562, -8.7972412109375, -97.05758666992188, -71.066162109375, 15.0078125, 8.275749206542969, 56.116485595703125, 125.72354125976562, 86.3743896484375, -65.68443298339844, 80.9049301147461, 23.939971923828125, 1.82586669921875, -9.878021240234375, 0.0, 116.91278076171875, 129.6393585205078, 108.091796875, -15.332511901855469, -44.41407775878906, 100.22244262695312, 1.8342132568359375, 2.63543701171875, 87.48086547851562, 12.00472640991211, 73.2918701171875, 40.49958801269531, 6.149406433105469, 118.26437377929688, 73.57803344726562, 14.124847412109375, 127.961181640625, -12.209781646728516, -31.970245361328125, -22.164154052734375, -19.656402587890625, 130.87109375, -17.507232666015625, 0.5961227416992188, 45.35069274902344, 8.240867614746094, 9.871719360351562, -6.211822509765625, 116.00865173339844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000246.npy"}
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 41.00924301147461, "std": 59.17033004760742, "min": -134.04153442382812, "p10": -23.03504638671875, "median": 30.448447227478027, "p90": 122.70651702880859, "max": 177.0555419921875, "pos_frac": 0.7421875, "sample": [115.61003112792969, -31.248291015625, 85.16265869140625, 78.2658462524414, -16.934249877929688, 62.05352783203125, 121.36538696289062, 124.291748046875, 101.24603271484375, 63.18574523925781, 0.0, 96.09185028076172, 122.33961486816406, 17.959197998046875, 6.3754119873046875, 39.068328857421875, -25.17986297607422, 9.006011962890625, -37.039642333984375, 94.57536315917969, -2.3861083984375, 123.5626220703125, -23.628662109375, -85.8892822265625, 67.84613037109375, 95.07363891601562, 45.03741455078125, -5.829803466796875, 99.53286743164062, 123.952392578125, 67.83255004882812, 112.44585418701172, 14.547042846679688, 150.9625244140625, 72.1863784790039, -25.82904052734375, -2.351226806640625, 140.71035766601562, -6.4283447265625, 9.31463623046875, -11.286163330078125, 7.978736877441406, 27.4088134765625, -12.902297973632812, 10.555755615234375, 69.5240478515625, -1.992767333984375, 38.653076171875, 3.1814804077148438, -68.6123046875, 6.741920471191406, 86.93136596679688, 67.87265014648438, 67.59417724609375, -31.19635009765625, 113.12342071533203, 156.06576538085938, 17.462127685546875, 2.5509185791015625, -13.799728393554688, -17.69219207763672, -12.843925476074219, -1.42926025390625, 134.51385498046875, -4.155517578125, 1.1442108154296875, 79.63456726074219, -93.24151611328125, 5.8543853759765625, -22.7806396484375, 13.057601928710938, 165.21676635742188, -134.04153442382812, 1.6971397399902344, 108.89501953125, 20.009536743164062, 20.17749786376953, 7.563140869140625, 32.496185302734375, 92.56346130371094, 39.789886474609375, 87.02867126464844, 177.0555419921875, 12.865264892578125, -27.795745849609375, -7.318992614746094, 2.1154251098632812, 40.98902893066406, 97.89157104492188, 32.447662353515625, 28.5286865234375, 92.345947265625, 75.81642150878906, 26.46368408203125, 16.53302001953125, 100.391845703125, 127.70309448242188, 110.56343078613281, 142.83673095703125, -17.068252563476562, 31.85004425048828, 16.91992950439453, 59.87712097167969, 44.5655517578125, 93.14596557617188, 93.43617248535156, 30.393310546875, -30.450469970703125, 134.0330810546875, 13.339874267578125, -98.25244140625, 97.6365966796875, 64.31388092041016, -21.771484375, 30.503583908081055, 51.39900207519531, 79.60809326171875, -11.730083465576172, 22.16961669921875, -4.507080078125, 85.48355102539062, 39.07073974609375, 129.23802185058594, 4.554622650146484, 104.85354614257812, 5.161102294921875, 23.9915771484375, 71.81658935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000247.npy"}
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 32.12859344482422, "std": 62.98286437988281, "min": -152.77862548828125, "p10": -40.86435470581053, "median": 23.36026382446289, "p90": 120.8006591796875, "max": 180.9361572265625, "pos_frac": 0.75, "sample": [12.120094299316406, 119.00616455078125, -20.139678955078125, 12.600242614746094, 144.7130126953125, -19.98419189453125, 23.099273681640625, 3.704803466796875, 74.079833984375, 12.432327270507812, 26.34271240234375, 44.14082336425781, 80.3721923828125, 37.7946891784668, 1.569244384765625, 33.45941162109375, 137.91131591796875, 24.522552490234375, -1.7458953857421875, 94.0443115234375, -7.703216552734375, 21.978469848632812, 86.84175872802734, 48.12664031982422, 132.12115478515625, 37.459747314453125, 76.29608154296875, 65.32318115234375, 0.829559326171875, 99.95718383789062, 20.437881469726562, -22.616127014160156, 141.2464599609375, -15.084747314453125, 9.555198669433594, 31.576339721679688, 126.49720764160156, -152.77862548828125, 12.658706665039062, 1.747701644897461, -35.8953857421875, 4.004386901855469, 23.621253967285156, -28.4383544921875, -9.451114654541016, 38.8818359375, -3.654146194458008, -113.24853515625, 37.92970275878906, 57.248504638671875, 15.891860961914062, 146.9423065185547, 34.455535888671875, 22.478553771972656, 160.6373291015625, -57.427032470703125, 125.84246826171875, 0.64923095703125, -23.30548095703125, 132.77566528320312, 44.853546142578125, 100.4807357788086, -25.53424072265625, -79.17903137207031, 17.806564331054688, 14.550506591796875, 180.9361572265625, 120.81689453125, 80.2091064453125, 91.099609375, 89.01537322998047, -51.74534606933594, 4.11646842956543, 77.76126098632812, 2.55755615234375, -100.84844970703125, 0.0, 29.44488525390625, 107.00634765625, 120.793701171875, 110.43421936035156, 75.77017211914062, 82.86209106445312, 81.06707763671875, 6.910572052001953, 42.344757080078125, -27.36700439453125, -95.17376708984375, -8.87103271484375, 10.055061340332031, -61.977081298828125, 12.76470947265625, 121.97079467773438, 99.65036010742188, 116.21761322021484, -89.40032958984375, 35.003875732421875, 11.455924987792969, 7.897804260253906, 106.09101867675781, 0.9072113037109375, -51.945281982421875, 38.23204040527344, -17.927947998046875, 57.82720947265625, 33.300628662109375, 97.93722534179688, 7.753654479980469, 9.654327392578125, -97.66946411132812, 32.223297119140625, 22.59820556640625, 48.521034240722656, -1.3103179931640625, 150.53683471679688, 73.4935302734375, 74.05860900878906, 50.14469909667969, 64.21723937988281, -81.10289001464844, -36.201072692871094, -4.299102783203125, 18.917007446289062, 23.9647216796875, 97.47843933105469, -72.37890625, 10.650245666503906, 16.57757568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000248.npy"}
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 39.27798080444336, "std": 63.542823791503906, "min": -135.46810913085938, "p10": -29.82530822753906, "median": 29.1807861328125, "p90": 118.82371063232422, "max": 169.05303955078125, "pos_frac": 0.6953125, "sample": [-14.617919921875, 4.418857574462891, -62.48881530761719, -87.09378051757812, 59.35774230957031, 4.237449645996094, -15.604736328125, -3.7421112060546875, -5.20635986328125, 27.3333740234375, -1.4876556396484375, -60.145530700683594, 97.51011657714844, 145.74200439453125, -11.752273559570312, 92.94241333007812, 44.84735107421875, 112.67816162109375, 23.63397216796875, 116.169677734375, 14.176376342773438, -135.46810913085938, 113.31856536865234, 18.777679443359375, -31.293731689453125, 149.42962646484375, 77.458740234375, -15.810150146484375, 35.74468994140625, 70.75469970703125, 19.05541229248047, -9.806365966796875, 38.57110595703125, -14.368339538574219, -8.062652587890625, 30.331695556640625, 128.95889282226562, 54.292633056640625, 119.67221069335938, -90.72862243652344, 118.46006774902344, 105.8719482421875, -10.973068237304688, 114.67311096191406, 155.820556640625, 52.731231689453125, -95.99211120605469, 101.75982666015625, 58.2359619140625, 9.317962646484375, -12.4888916015625, 18.724693298339844, 169.05303955078125, 83.57816314697266, 138.25689697265625, -2.38671875, 14.608245849609375, 75.88033294677734, -29.19598388671875, 94.77218627929688, 37.2857666015625, -1.848388671875, -0.866302490234375, 4.389923095703125, 102.828369140625, 10.267898559570312, 94.61215209960938, 3.3436813354492188, 81.79313659667969, 94.237060546875, -22.489036560058594, -66.18852996826172, 28.029876708984375, 93.98492431640625, 56.62513732910156, 86.30194091796875, 168.8775634765625, 139.4487762451172, 109.9493179321289, 106.24349212646484, 147.10638427734375, 84.66356658935547, 20.100685119628906, -68.03225708007812, 93.81918334960938, 139.8321075439453, -5.840555191040039, -26.097381591796875, -17.067325592041016, 95.78445434570312, 25.879180908203125, -50.09222412109375, 78.15457153320312, 96.81913757324219, 27.607818603515625, 98.22479248046875, 49.25250244140625, 26.036026000976562, 9.887176513671875, -41.346466064453125, 123.2967529296875, 1.86773681640625, 32.20842742919922, 82.30496215820312, 54.20945739746094, 6.13677978515625, -5.880867004394531, -93.54327392578125, -38.6318359375, -12.584541320800781, -28.251129150390625, 13.008371353149414, 53.430511474609375, 1.70123291015625, 14.83367919921875, -4.166168212890625, 84.3470458984375, 38.345733642578125, 101.82130432128906, 5.39984130859375, 109.85531616210938, 98.84848022460938, 103.16964721679688, -27.228485107421875, -3.4827804565429688, 36.4229736328125, 40.259910583496094, 135.92054748535156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000249.npy"}
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 39.83466339111328, "std": 67.02472686767578, "min": -119.00189208984375, "p10": -32.07691650390625, "median": 36.4092960357666, "p90": 123.62122192382812, "max": 186.13743591308594, "pos_frac": 0.7265625, "sample": [39.825225830078125, 14.849395751953125, -61.33302307128906, 120.95166015625, 97.76010131835938, 170.22280883789062, 157.43446350097656, 91.03164672851562, 41.33856201171875, -94.98164367675781, 93.34162902832031, 22.280351638793945, 130.03863525390625, 34.250396728515625, 19.70709228515625, 136.13711547851562, 60.077972412109375, -10.042388916015625, 123.44036865234375, 105.5972900390625, 118.44606018066406, 48.07220458984375, 61.34446716308594, 105.0394287109375, -29.303955078125, 1.61297607421875, -119.00189208984375, 22.013336181640625, -7.5947265625, -0.4739990234375, 120.95904541015625, -97.77163696289062, 52.480865478515625, 165.01824951171875, -10.287872314453125, 7.268310546875, 61.37127685546875, 0.9815883636474609, 134.49603271484375, -111.27078247070312, -2.1958541870117188, 63.66008758544922, 124.043212890625, 29.36376190185547, 95.36013793945312, 79.10118103027344, -31.38397216796875, -90.34555053710938, -1.4245681762695312, 73.83427429199219, 95.81465148925781, 10.853889465332031, 44.9403076171875, 9.8057861328125, 35.819091796875, 13.2081298828125, -11.004714965820312, -33.69378662109375, 69.68738555908203, 12.792167663574219, 105.87939453125, 139.245361328125, 77.890625, -0.7011566162109375, 26.464309692382812, 42.589599609375, 155.84432983398438, 106.41593933105469, -3.680206298828125, 40.05513000488281, 88.80448150634766, 99.89436340332031, -5.397308349609375, 123.27317810058594, 34.33050537109375, -25.533905029296875, 83.6768798828125, 114.10685729980469, 136.08990478515625, 36.9995002746582, 26.459674835205078, -26.136993408203125, 39.150634765625, -27.64892578125, 114.77996826171875, 6.512821197509766, -2.7341766357421875, -112.28616333007812, 97.72555541992188, 144.74456787109375, -4.722747802734375, 186.13743591308594, -73.82583618164062, -59.770111083984375, -107.61215209960938, -0.3678741455078125, 7.10015869140625, -41.332183837890625, -94.7923583984375, 103.52323913574219, 78.81562805175781, 100.66622161865234, 64.7816162109375, 86.10855102539062, 95.91162872314453, 44.85095977783203, -29.50482177734375, 44.031494140625, 16.37622833251953, -4.948751449584961, 4.813262939453125, 81.12472534179688, 155.5410919189453, -11.77996826171875, 42.063934326171875, 89.06126403808594, 104.65263366699219, 7.607879638671875, 7.669189453125, 1.6105117797851562, -27.627525329589844, 94.37698364257812, 45.27783203125, 2.177520751953125, 6.599853515625, 24.256500244140625, 16.179000854492188, 3.428813934326172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000250.npy"}
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 38.977455139160156, "std": 63.77779006958008, "min": -134.44277954101562, "p10": -31.65152359008789, "median": 24.541702270507812, "p90": 126.3145751953125, "max": 244.267578125, "pos_frac": 0.7265625, "sample": [83.13812255859375, 80.64312744140625, 24.633819580078125, -14.486846923828125, -52.066802978515625, -45.597991943359375, -76.53753662109375, 49.64092254638672, 8.8006591796875, 19.87371826171875, 7.959775924682617, 61.70294189453125, 107.9599609375, 81.05925750732422, 5.853157043457031, -4.450717926025391, 32.313446044921875, 1.9316253662109375, 5.800210952758789, 146.61712646484375, -1.0612945556640625, 4.803802490234375, -31.684234619140625, 117.25006103515625, 93.89558410644531, 130.63963317871094, -59.004150390625, 107.48599243164062, 136.12005615234375, 22.632537841796875, 107.7218017578125, 6.289478302001953, -31.019309997558594, 43.1287841796875, 34.4425048828125, -59.63679885864258, 61.4542236328125, -1.9286785125732422, -40.325592041015625, -21.654281616210938, 93.46694946289062, 38.6336669921875, 107.84774780273438, 130.21322631835938, 129.88478088378906, -12.009902954101562, 0.190582275390625, -49.77061462402344, 3.442777633666992, 5.3102569580078125, -31.63750457763672, 18.51629638671875, 244.267578125, 33.55804443359375, 80.67078399658203, 99.82782745361328, 21.0113525390625, 117.51399230957031, 6.033287048339844, 73.66754913330078, 83.30752563476562, -134.44277954101562, 29.566184997558594, 93.52322387695312, 20.458694458007812, -18.650772094726562, 6.94964599609375, -4.695930480957031, 24.4495849609375, -0.34871482849121094, 108.16375732421875, 90.29034423828125, 7.63702392578125, 4.521947860717773, 52.110565185546875, -5.652374267578125, 1.1978073120117188, 28.853485107421875, -4.858573913574219, 84.63368225097656, 1.851654052734375, 68.13597106933594, 2.7495927810668945, -9.540756225585938, 73.98312377929688, 102.94324493408203, 126.53033447265625, -89.37193298339844, -3.343658447265625, 3.4107818603515625, 66.02068328857422, 139.991943359375, 110.75881958007812, 59.759490966796875, 151.41592407226562, -8.47232437133789, 197.230224609375, 126.22210693359375, 90.83597564697266, -6.197658538818359, 74.2659912109375, 34.08935546875, 75.16290283203125, 43.66741943359375, 97.00289916992188, 32.37603759765625, 119.00059509277344, 4.027351379394531, 78.03237915039062, -55.57989501953125, -0.247344970703125, 34.06494140625, 12.828689575195312, -26.176902770996094, 5.4095611572265625, -86.5845947265625, -26.107421875, -18.362060546875, 2.3362579345703125, 149.14080810546875, 112.12142944335938, -42.44322204589844, 133.18861389160156, 138.88327026367188, -11.972076416015625, 4.578723907470703, 85.53150177001953, 93.97422790527344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000251.npy"}
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 43.28020095825195, "std": 70.6903305053711, "min": -148.59231567382812, "p10": -35.05940704345703, "median": 35.94558525085449, "p90": 129.95043487548827, "max": 212.0577392578125, "pos_frac": 0.703125, "sample": [171.51193237304688, 138.96316528320312, 135.05615234375, 63.99516296386719, -85.7572021484375, 83.435546875, -78.9648208618164, 10.085746765136719, 4.0715484619140625, -20.513153076171875, -5.803337097167969, -34.79820251464844, 16.70465087890625, 33.491668701171875, 212.0577392578125, 32.69940185546875, 126.37431335449219, 63.650245666503906, 119.87699890136719, 107.93716430664062, 7.773895263671875, 55.04737854003906, -18.399688720703125, 131.2768096923828, -148.59231567382812, 8.489501953125, 5.535919189453125, 78.62329864501953, -33.822845458984375, 54.72938537597656, 1.56427001953125, 126.46240234375, 118.6436767578125, 100.43313598632812, 86.37744903564453, 24.2613525390625, -51.06842041015625, -13.910402297973633, 74.36712646484375, 201.3331298828125, 99.84344482421875, 0.16538619995117188, 90.59768676757812, 29.271820068359375, 40.334136962890625, 13.01171875, 26.128921508789062, -9.154205322265625, -0.11370849609375, 15.1923828125, 117.40357971191406, 6.97344970703125, 133.29556274414062, 91.58038330078125, -20.770751953125, -4.28265380859375, 104.37788391113281, -47.240509033203125, 34.26702880859375, 45.11016845703125, 91.55558776855469, -19.257232666015625, 34.028778076171875, 141.95428466796875, 68.00288391113281, 65.58192443847656, 123.7510986328125, 38.48583984375, 73.13690185546875, 124.74075317382812, 3.13458251953125, 108.59932708740234, 142.01397705078125, -8.020286560058594, -70.03016662597656, 107.57305908203125, -100.13845825195312, 5.563316345214844, 92.36688232421875, 90.11558532714844, 88.18380737304688, 3.20550537109375, 49.950927734375, 102.34020233154297, -7.963348388671875, -1.0714569091796875, 0.0, 98.1058578491211, -108.75054931640625, 116.6886978149414, -0.4036407470703125, 107.2587890625, -12.648681640625, -19.9110107421875, 32.86322021484375, 143.3145751953125, -2.45855712890625, 129.38198852539062, 45.9180908203125, 120.46513366699219, -3.63153076171875, 99.43912506103516, 35.110347747802734, -5.237640380859375, 136.78802490234375, -1.88812255859375, 103.36065673828125, -121.8052978515625, 21.194580078125, 8.634506225585938, -25.225860595703125, -21.625732421875, 39.610107421875, -112.0438232421875, -109.1265869140625, -1.1907958984375, 103.39828491210938, 145.39682006835938, -53.0008544921875, 110.80007934570312, 36.78082275390625, 146.06051635742188, 89.92660522460938, -35.66888427734375, 56.34747314453125, 103.52572631835938, 14.373249053955078, 112.74458312988281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000252.npy"}
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 33.13809585571289, "std": 65.56414031982422, "min": -139.35537719726562, "p10": -38.7713134765625, "median": 18.182937622070312, "p90": 126.3173828125, "max": 160.101806640625, "pos_frac": 0.7421875, "sample": [153.10372924804688, -6.41583251953125, 8.829483032226562, 69.2014389038086, 18.56939697265625, 23.1812744140625, 2.2260684967041016, 75.99577331542969, 160.101806640625, 24.169189453125, 39.3699951171875, 126.10850524902344, 7.857421875, -66.69448852539062, 73.1905517578125, -105.24371337890625, 79.24727630615234, -72.4177017211914, 110.17367553710938, -30.006317138671875, 0.91339111328125, 5.3936920166015625, 38.12141418457031, 13.57325553894043, -1.554443359375, -32.40667724609375, 33.2406005859375, 133.99639892578125, 104.9772720336914, 95.25007629394531, -5.135507583618164, -11.94925308227539, -106.75411987304688, 29.54461669921875, 159.91534423828125, 8.229583740234375, -9.442367553710938, 128.35174560546875, -139.35537719726562, -22.3516845703125, 120.87008666992188, 49.852783203125, 133.3482666015625, 32.21778869628906, 19.2403564453125, -39.1466064453125, 111.52214050292969, 5.588226318359375, -10.267547607421875, 17.88116455078125, 125.7197265625, 46.933563232421875, 123.92666625976562, 10.647964477539062, -31.428863525390625, 112.251953125, 0.34676361083984375, 10.302116394042969, -34.2813720703125, 86.97645568847656, 19.1966552734375, -31.6270751953125, 5.4568328857421875, 108.05823516845703, 18.484710693359375, 120.81016540527344, -44.586395263671875, -21.85181427001953, 78.9615478515625, -30.8287353515625, 22.487281799316406, -12.923431396484375, -112.45178985595703, 0.2279052734375, -117.81640625, 53.72068786621094, -77.29485321044922, 4.123649597167969, -4.7626495361328125, -30.511260986328125, 79.64733123779297, 1.1375389099121094, 59.079345703125, 102.18008422851562, 35.27909851074219, -3.741241455078125, 28.126731872558594, 11.7337646484375, -50.0718994140625, 144.98556518554688, -66.52342224121094, 7.331962585449219, 105.16220092773438, -0.367095947265625, 6.5099639892578125, 113.95738220214844, -59.204612731933594, 150.65869140625, 7.685203552246094, 78.27247619628906, 89.49150848388672, 13.690841674804688, -38.6104736328125, 2.435455322265625, 38.71534729003906, 126.80476379394531, 14.321746826171875, 134.42520141601562, 14.477584838867188, 3.042644500732422, 22.1043701171875, 132.47119140625, 79.53160858154297, 3.622650146484375, 97.71627807617188, 13.509635925292969, 10.97918701171875, 15.086685180664062, 34.10498046875, 157.1658935546875, 7.8506927490234375, 43.400115966796875, 134.82257080078125, 99.47419738769531, 12.13568115234375, 94.52239990234375, 74.10973358154297, 105.95404052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000253.npy"}
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 42.141700744628906, "std": 60.24184799194336, "min": -144.66827392578125, "p10": -27.293464660644528, "median": 28.770785331726074, "p90": 121.87696533203125, "max": 161.2619171142578, "pos_frac": 0.7734375, "sample": [149.24127197265625, 1.4871063232421875, -76.77809143066406, 29.25139617919922, -9.945510864257812, -3.5542240142822266, 17.954742431640625, 14.010208129882812, -1.0503120422363281, 20.556777954101562, 5.677665710449219, -15.860870361328125, -14.0560302734375, -2.93182373046875, 7.315185546875, 130.92959594726562, 6.1947021484375, 111.33700561523438, 47.61279296875, -31.696563720703125, -34.45062255859375, 133.61920166015625, 15.878982543945312, 10.54327392578125, -18.422698974609375, 15.048294067382812, 104.61209106445312, -67.06019592285156, 103.6956787109375, 80.65475463867188, 104.4747314453125, 141.14236450195312, -91.41998291015625, 84.01849365234375, 1.7545299530029297, 22.773773193359375, -26.506912231445312, 71.55290222167969, 8.448074340820312, 13.098419189453125, 65.84855651855469, 56.664215087890625, 90.82110595703125, 45.508392333984375, -54.43183135986328, 16.20435333251953, 100.24139404296875, 47.93304443359375, 68.70654296875, 104.08219909667969, 73.49331665039062, 5.207868576049805, 28.365270614624023, 117.28536987304688, 86.4031982421875, 29.176300048828125, 50.95383071899414, -61.808319091796875, 24.46649169921875, 77.45687866210938, 2.5456008911132812, 23.3162841796875, 138.83349609375, 6.7379302978515625, 57.56834411621094, 119.95880889892578, 47.073394775390625, 120.46954345703125, 122.5338134765625, 33.35423278808594, 8.991279602050781, 38.73187255859375, 131.93206787109375, 6.931434631347656, 46.67413330078125, -23.5321044921875, -5.28961181640625, 31.40631103515625, 1.0690383911132812, 2.9891510009765625, -2.30230712890625, -10.442169189453125, 136.1226806640625, 121.20068359375, 87.60210418701172, 143.76104736328125, 70.95428466796875, 106.31974792480469, 15.9666748046875, 115.44735717773438, 26.597396850585938, 120.23301696777344, 120.44464111328125, 96.81396484375, 111.19298553466797, 22.911376953125, 0.12961387634277344, 110.41571807861328, -5.3261566162109375, -11.084457397460938, -29.128753662109375, 51.445526123046875, 47.39244079589844, -2.498537063598633, 0.0, -29.947250366210938, 128.70928955078125, 88.60748291015625, 35.8870849609375, 24.4267578125, 11.968284606933594, 24.20977020263672, 31.88341522216797, 121.595458984375, 87.570068359375, 0.3148956298828125, -144.66827392578125, -62.328834533691406, 150.39459228515625, 22.640079498291016, -43.98095703125, -37.4742431640625, 5.214237213134766, 161.2619171142578, 152.34715270996094, 102.19630432128906, 85.83401489257812, 89.28428649902344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000254.npy"}
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 46.12751770019531, "std": 70.96184539794922, "min": -123.747314453125, "p10": -31.68120956420897, "median": 29.697174072265625, "p90": 136.81844635009764, "max": 186.41708374023438, "pos_frac": 0.765625, "sample": [23.851566314697266, -80.913330078125, 145.8507080078125, 47.22550964355469, 47.35844421386719, -95.437255859375, 4.1710205078125, 128.4696807861328, -11.7567138671875, 110.35260009765625, -7.981597900390625, 9.225692749023438, 11.895713806152344, 11.5625, 156.40829467773438, 130.941650390625, 13.051094055175781, 15.7332763671875, 147.7587890625, 9.988006591796875, 118.33370971679688, -91.34454345703125, 0.629241943359375, 112.18824768066406, 43.8148193359375, 134.15957641601562, 48.538055419921875, 109.97821044921875, -23.3458251953125, 29.464080810546875, 83.92889404296875, 11.773956298828125, 20.820266723632812, 12.115966796875, 68.11429595947266, 34.06378173828125, 128.22531127929688, 66.63255310058594, -75.47470092773438, -79.72000122070312, 186.41708374023438, 156.94085693359375, 137.4697723388672, 29.63787841796875, -86.41044616699219, -25.99407958984375, -123.747314453125, 140.56298828125, 98.60929107666016, 92.61836242675781, 125.32373046875, 128.59237670898438, 99.74832153320312, 98.04214477539062, 153.0773468017578, -28.15808868408203, 53.5670166015625, -96.3182373046875, -58.1156005859375, 114.56153869628906, 24.1435546875, 124.06747436523438, 113.96969604492188, 105.9227294921875, 30.76263427734375, 101.99659729003906, 7.218650817871094, -104.39810943603516, 131.28367614746094, 18.299942016601562, 151.55599975585938, 29.7564697265625, 48.32468795776367, -9.406623840332031, 3.528167724609375, 19.84994888305664, 5.749721527099609, 28.391586303710938, 12.428924560546875, 102.24597930908203, -6.3614501953125, 50.080963134765625, -119.02925109863281, 149.9794921875, 122.8343505859375, 101.79878234863281, -4.1180419921875, 27.935821533203125, 17.301376342773438, 16.06488800048828, 136.539306640625, 114.88751220703125, 81.8680648803711, 131.62014770507812, -15.64129638671875, 155.94454956054688, 118.25970458984375, 3.00494384765625, 4.097686767578125, -14.720855712890625, -1.426137924194336, -39.901824951171875, 82.39152526855469, 5.434661865234375, 138.02880859375, 76.41302490234375, 8.11212158203125, 110.90664672851562, -3.5571060180664062, 93.131591796875, 3.0682373046875, 9.618362426757812, -12.785980224609375, 124.87759399414062, 21.986770629882812, 34.17716979980469, -7.7489471435546875, -12.847991943359375, 112.47341918945312, 62.120452880859375, 15.213584899902344, 113.44612884521484, -5.7786407470703125, 90.58197021484375, -54.149169921875, 161.71337890625, 21.785812377929688, -28.077850341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000255.npy"}
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 46.28375244140625, "std": 74.0203857421875, "min": -137.90554809570312, "p10": -48.771479797363284, "median": 41.82659912109375, "p90": 135.5734069824219, "max": 170.63922119140625, "pos_frac": 0.7265625, "sample": [-1.2473602294921875, 117.9549560546875, 135.20858764648438, -39.68597412109375, 18.726730346679688, 97.655029296875, 123.94989013671875, 99.38519287109375, 116.685791015625, -2.91387939453125, 108.93583679199219, -40.59925842285156, -35.53472900390625, -108.53958892822266, 96.55094909667969, 127.06159973144531, 133.96719360351562, -19.264923095703125, 90.11451721191406, 125.763427734375, 2.257781982421875, 116.64945983886719, -34.09330749511719, 170.63922119140625, 16.3251953125, 131.83082580566406, 106.86477661132812, -64.335205078125, 145.65328979492188, 17.0504150390625, 143.4638671875, 2.247455596923828, 62.15972900390625, 37.423004150390625, 40.734771728515625, 12.543243408203125, -48.84698486328125, -124.53143310546875, 96.68441772460938, 62.6617431640625, 145.20248413085938, 27.140106201171875, 87.23159790039062, 136.2553253173828, 58.29170227050781, 54.7679443359375, -74.54402923583984, -1.9904937744140625, -30.346588134765625, 160.2435302734375, 6.512386322021484, -68.17325592041016, 138.26611328125, 141.88833618164062, -2.8640899658203125, 2.640523910522461, -137.41688537597656, -33.2086181640625, 117.39231872558594, 83.16519165039062, 101.87849426269531, 18.874465942382812, 113.22323608398438, 142.9991455078125, -3.580876350402832, 94.86149597167969, 50.95855712890625, -103.85543060302734, 108.48046875, -4.803741455078125, -39.25128173828125, -99.85336303710938, 123.7081298828125, 42.918426513671875, 33.167022705078125, 117.7164306640625, -20.678741455078125, 121.09173583984375, 26.69305419921875, 145.07052612304688, 153.79489135742188, 93.53511047363281, 90.63770294189453, 0.93402099609375, 2.089447021484375, 64.62484741210938, -101.90492248535156, 11.160179138183594, 34.41685485839844, 135.39471435546875, -70.1636962890625, 128.96804809570312, 103.45767974853516, 137.9119873046875, 15.072998046875, 105.90392303466797, 71.06980895996094, -14.064178466796875, 20.72118377685547, 29.66680908203125, 39.27752685546875, 60.271873474121094, 134.25262451171875, 86.06146240234375, 35.77728271484375, 51.50927734375, 20.419403076171875, 126.57119750976562, -51.823936462402344, -137.90554809570312, 7.911224365234375, 134.75222778320312, 135.9903564453125, 15.296775817871094, -13.075332641601562, -0.101043701171875, 133.63497924804688, -32.38079833984375, -18.07269287109375, -48.73912048339844, 104.49479675292969, -5.4656982421875, 0.0760040283203125, 63.76222229003906, 21.7357177734375, 40.17507553100586, 72.822021484375, 116.23739624023438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000256.npy"}
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 42.685611724853516, "std": 66.19627380371094, "min": -112.75910949707031, "p10": -25.972665405273435, "median": 28.300003051757812, "p90": 131.34043731689454, "max": 189.56568908691406, "pos_frac": 0.671875, "sample": [80.81246948242188, 41.0704345703125, 189.56568908691406, -22.461349487304688, -2.4806480407714844, 188.310791015625, 23.156707763671875, 131.77818298339844, 78.2950439453125, 114.13365936279297, 78.3373794555664, 147.89053344726562, 116.71981811523438, 73.79736328125, 4.239847183227539, 27.49182891845703, 27.402053833007812, -40.78642272949219, -26.57868194580078, 90.49020385742188, 131.1043701171875, -37.15021514892578, 91.85082244873047, 166.37307739257812, -18.214111328125, -64.09161376953125, 22.12249755859375, 22.719009399414062, -6.06097412109375, -23.94264030456543, -25.71294403076172, -31.59588623046875, 119.6417236328125, 23.94091796875, 86.48739624023438, 126.37274169921875, 182.66326904296875, 1.229217529296875, 140.67938232421875, 100.66617584228516, -23.8125, 8.350494384765625, -22.493080139160156, 110.32412719726562, 134.51638793945312, -20.205596923828125, -2.9208145141601562, 43.7091064453125, 25.282913208007812, -111.69232177734375, 89.93316650390625, 164.30145263671875, 45.859336853027344, 32.21631622314453, 123.20779418945312, 27.571746826171875, 82.92967224121094, -20.566848754882812, 79.76686096191406, 35.581787109375, -8.758026123046875, 134.4722900390625, 111.64486694335938, 17.995418548583984, -64.88773345947266, -18.746383666992188, 76.45095825195312, 3.9085254669189453, 21.164642333984375, 80.735595703125, 1.581878662109375, 107.8646240234375, 80.50383758544922, 35.540740966796875, 26.387828826904297, 12.825668334960938, -112.75910949707031, -1.9887619018554688, 99.74946594238281, 129.5, 5.2761993408203125, 2.5068912506103516, 63.08984375, 29.02825927734375, 33.06193161010742, -14.957443237304688, -0.90045166015625, 131.15283203125, 65.961181640625, -12.443695068359375, -18.448822021484375, 95.94891357421875, -33.626495361328125, 130.74407958984375, 55.31818771362305, 99.03662109375, 148.28659057617188, 52.230865478515625, -14.318817138671875, 7.283973693847656, 53.209503173828125, 12.418106079101562, 115.66241455078125, 32.8394775390625, -3.6230316162109375, -14.3973388671875, -8.708770751953125, -96.52145385742188, -29.63751220703125, 132.8491973876953, 23.937652587890625, 122.438720703125, -20.458358764648438, 171.07891845703125, -52.27142333984375, -7.975128173828125, 68.60516357421875, -25.141998291015625, 84.18206787109375, 35.01519775390625, 98.30799865722656, -19.460548400878906, -10.179450988769531, 62.54954528808594, -23.54003143310547, -12.63916015625, -29.828964233398438, 115.5035400390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000257.npy"}
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 47.62141418457031, "std": 61.66843795776367, "min": -152.06402587890625, "p10": -23.269418334960935, "median": 36.81494903564453, "p90": 125.36917724609374, "max": 161.69094848632812, "pos_frac": 0.7578125, "sample": [110.83090209960938, 13.824541091918945, 17.363021850585938, 5.263145446777344, 84.24623107910156, -23.097015380859375, 108.9720458984375, -23.0489501953125, 73.23703002929688, 114.20654296875, -44.201202392578125, 36.383758544921875, 57.327423095703125, 20.676025390625, 123.99664306640625, 83.36214447021484, -24.594024658203125, 29.0855712890625, 123.12554931640625, 87.11978149414062, 133.54611206054688, 87.06185913085938, -40.26933288574219, 11.3896484375, 158.11279296875, -35.97889709472656, -152.06402587890625, 120.08499145507812, 131.98638916015625, 48.27032470703125, 113.88677978515625, 111.72509765625, -36.752105712890625, 59.313629150390625, 86.8558349609375, 34.489166259765625, 3.1950950622558594, -27.288238525390625, 62.38563537597656, 112.0264892578125, 107.84451293945312, 98.43995666503906, 6.276588439941406, 38.686767578125, 13.784103393554688, 21.899810791015625, 23.4818115234375, 19.553466796875, 19.9136962890625, 14.408233642578125, 72.95499420166016, 6.6697845458984375, 115.80813598632812, 22.78204345703125, -132.0565185546875, 73.7526626586914, 39.746368408203125, -6.28607177734375, -50.7042236328125, 4.99456787109375, -17.050552368164062, 30.13043975830078, 34.343414306640625, 34.6474609375, 124.912841796875, 2.0623550415039062, 156.13180541992188, 120.3802490234375, -6.754486083984375, 104.59921264648438, 68.63165283203125, 66.611328125, -4.072978973388672, -4.905426025390625, -0.5421104431152344, 75.32232666015625, 137.27264404296875, -53.76835632324219, 84.42583465576172, 25.9215087890625, -8.140775680541992, -23.67169189453125, 100.93997955322266, 110.68064880371094, -9.839447021484375, -1.0079574584960938, 95.67420959472656, 37.005950927734375, 126.4339599609375, -67.73486328125, 48.61099624633789, 15.557632446289062, 31.410125732421875, 12.832721710205078, -5.333404541015625, 55.198333740234375, 103.97991943359375, 114.92469787597656, 8.46743392944336, -14.8568115234375, 20.505935668945312, -3.6997833251953125, 101.29704284667969, -6.9433746337890625, 161.69094848632812, -2.75897216796875, 113.49160766601562, 137.55685424804688, 128.68972778320312, 108.26936340332031, 45.012481689453125, -15.20001220703125, 157.68356323242188, 91.137939453125, 147.81097412109375, 130.62127685546875, 75.71157836914062, 3.450531005859375, 77.44223022460938, 99.30119323730469, -13.63458251953125, 87.64482116699219, 89.44796752929688, 149.59426879882812, 15.777435302734375, 36.62394714355469, 22.94793701171875, -41.37013244628906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000258.npy"}
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 41.94192886352539, "std": 66.40558624267578, "min": -118.81887817382812, "p10": -38.80557098388672, "median": 38.895240783691406, "p90": 131.87129669189454, "max": 169.32427978515625, "pos_frac": 0.734375, "sample": [102.14035034179688, -38.008941650390625, 131.73318481445312, 10.724349975585938, -29.845046997070312, 13.6405029296875, 14.28167724609375, 146.1389617919922, 122.81097412109375, -9.91082763671875, 62.06890869140625, 16.84100341796875, 42.063446044921875, 107.26223754882812, 16.117698669433594, 71.76322937011719, 105.55513000488281, -118.81887817382812, 86.95083618164062, 119.26091003417969, 96.70437622070312, 155.02197265625, -65.9764404296875, 26.239566802978516, 141.12783813476562, 24.303482055664062, 84.46773529052734, 90.15156555175781, -77.64910888671875, -7.1975250244140625, 11.043434143066406, 8.094917297363281, -22.360992431640625, 95.79998779296875, 78.22564697265625, 42.976104736328125, 73.3017578125, 12.838516235351562, 17.894012451171875, 98.3062744140625, -40.47613525390625, 22.76226806640625, 136.07489013671875, 14.703619003295898, 3.029052734375, -70.80020141601562, 87.92435455322266, 138.87265014648438, -2.1678543090820312, 99.1466064453125, 94.93855285644531, 74.9854736328125, 127.56982421875, 5.41424560546875, 20.29485321044922, 18.150489807128906, -2.1761016845703125, 142.77935791015625, 57.038909912109375, 17.810508728027344, 169.32427978515625, 67.52617645263672, 18.7464599609375, 1.1701774597167969, -4.6343841552734375, -105.9333724975586, 51.5902099609375, -38.08961486816406, -88.56134033203125, 138.40118408203125, 41.339019775390625, -9.7779541015625, -7.6214447021484375, 76.42849731445312, 132.1935577392578, -44.874603271484375, 54.33380126953125, 0.0, -9.870361328125, 123.09857177734375, 109.23728942871094, 117.32186889648438, -23.725204467773438, 16.891357421875, 52.4375, 12.215744018554688, 134.50164794921875, -91.85196685791016, -0.9670028686523438, 51.590179443359375, -118.55426025390625, -19.5081787109375, 134.8938446044922, 137.04177856445312, 25.54840087890625, 35.9398193359375, 85.04071807861328, -22.926300048828125, 113.5766830444336, -18.76080322265625, 38.54585266113281, 61.17372131347656, -22.054641723632812, 116.19155883789062, 9.7689208984375, 62.863037109375, 116.71900939941406, -101.07171630859375, 58.041168212890625, 124.302734375, 39.24462890625, 120.17532348632812, 11.815826416015625, -42.5064697265625, 10.373291969299316, 43.032379150390625, 141.28677368164062, 6.121124267578125, 51.09089660644531, -21.2298583984375, 98.50128173828125, -51.693878173828125, -23.63482666015625, 112.22604370117188, 4.096160888671875, 99.51400756835938, 110.00599670410156, 98.97859191894531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000259.npy"}
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 19.933805465698242, "std": 66.57269287109375, "min": -157.064697265625, "p10": -57.45498352050781, "median": 9.871917724609375, "p90": 112.49362182617186, "max": 160.18429565429688, "pos_frac": 0.640625, "sample": [131.2220458984375, -10.202873229980469, 97.10066223144531, 8.387527465820312, -127.0565185546875, 109.48089599609375, 46.588165283203125, 133.0184326171875, 15.833847045898438, 77.75201416015625, -19.354782104492188, 22.949737548828125, 0.0, -21.555389404296875, 19.895065307617188, 4.18950080871582, 2.347198486328125, 160.18429565429688, 39.822662353515625, 55.828975677490234, 108.71394348144531, -157.064697265625, -0.27440643310546875, 9.74774169921875, 154.33773803710938, -57.27642822265625, 15.058361053466797, 96.23127746582031, 30.315261840820312, 14.146095275878906, 33.182769775390625, -52.73301696777344, 15.1578369140625, 3.8633785247802734, 4.837175369262695, 150.67556762695312, -44.750335693359375, -83.114013671875, -10.077499389648438, 5.325218200683594, 1.762176513671875, 16.52166748046875, 33.36175537109375, -50.992431640625, -29.408477783203125, 102.20060729980469, -89.8688735961914, -23.195293426513672, -57.871612548828125, 12.794918060302734, 2.90826416015625, -59.03741455078125, 122.70953369140625, 103.90240478515625, 8.531379699707031, 10.820465087890625, 121.43698120117188, -121.87960815429688, 110.70294189453125, 93.71278381347656, -30.254302978515625, -3.7204742431640625, 123.41789245605469, -17.700592041015625, 7.327152252197266, -96.48594665527344, 67.41342163085938, 120.02322387695312, -5.11663818359375, 133.86166381835938, 10.341552734375, 89.79067993164062, -43.33097839355469, 9.99609375, 8.126022338867188, -23.231292724609375, -15.680557250976562, 40.75569152832031, 16.52874755859375, 17.034149169921875, -98.31915283203125, -45.36213684082031, 145.61508178710938, 1.7248687744140625, 0.100860595703125, 31.2889404296875, 26.69232177734375, -18.290695190429688, -3.3591690063476562, -111.89326477050781, -28.71563720703125, 4.167938232421875, -109.47029113769531, 140.29470825195312, 66.29910278320312, 8.9183349609375, -8.533515930175781, 55.153282165527344, -27.690399169921875, 116.671875, 78.22274780273438, 74.05801391601562, 20.16015625, 75.11407470703125, 48.10052490234375, 5.71435546875, -5.5022430419921875, 61.87244415283203, -63.06340789794922, 87.81817626953125, -8.595703125, 105.64189147949219, 107.11737060546875, 15.415374755859375, -49.456573486328125, 5.6404876708984375, -5.35980224609375, -120.48007202148438, -11.866729736328125, 35.40715026855469, 97.24298095703125, -46.1435546875, 18.6123046875, 11.60107421875, 98.0101318359375, -5.516471862792969, 99.88715362548828, -24.358917236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000260.npy"}
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 36.66789245605469, "std": 70.74263763427734, "min": -128.52828979492188, "p10": -53.5313133239746, "median": 32.134300231933594, "p90": 126.24106292724609, "max": 197.563720703125, "pos_frac": 0.7109375, "sample": [-17.057144165039062, -125.3468017578125, 41.93678283691406, -32.07373046875, 90.38043212890625, 30.310699462890625, 100.50445556640625, -20.27227783203125, 31.533401489257812, 119.27217102050781, 133.46588134765625, 0.0, -24.601823806762695, -25.15057373046875, 4.296285629272461, 52.20098876953125, -17.94342041015625, 44.215850830078125, 124.40480041503906, 35.4659423828125, 78.11563110351562, -94.63267517089844, 114.87268829345703, -109.955078125, 87.97454833984375, -21.30475616455078, -50.34124755859375, 116.848876953125, -85.95564270019531, -98.22268676757812, 88.15670776367188, 119.27822875976562, 127.81719970703125, 85.94740295410156, 134.2225799560547, 62.918792724609375, -16.590511322021484, 15.787242889404297, -49.359466552734375, 86.11897277832031, 14.508689880371094, 12.30029296875, 135.23052978515625, -25.86688232421875, -1.501617431640625, 24.428085327148438, -0.7434978485107422, 119.05994415283203, -93.88546752929688, 101.05958557128906, 32.7867431640625, -24.996185302734375, 45.434967041015625, 141.96853637695312, 122.9788818359375, 53.19903564453125, 73.66995239257812, -42.42207336425781, 68.57865905761719, 120.57658386230469, 122.190673828125, 125.65287780761719, 66.20228576660156, 15.37896728515625, 83.19686889648438, -0.7684135437011719, 67.90217590332031, -83.94979858398438, 26.918533325195312, -0.5200424194335938, 116.4527587890625, 197.563720703125, 1.59173583984375, 156.6888427734375, 40.61415100097656, -10.849395751953125, 9.65972900390625, 2.633495330810547, 7.589141845703125, 13.89678955078125, 32.735198974609375, -6.5053558349609375, 123.90324401855469, -67.0076904296875, 154.57449340820312, 93.1051025390625, 44.051422119140625, 79.13160705566406, 13.360420227050781, -100.54647827148438, 13.729682922363281, 12.24169921875, -99.5977783203125, 6.3262939453125, -69.9051513671875, 137.02865600585938, 98.16802978515625, 117.49375915527344, 117.51130676269531, 39.267333984375, -22.467666625976562, 75.91795349121094, 0.5789909362792969, 112.61856079101562, 15.292984008789062, 48.524658203125, -59.9332275390625, 155.51254272460938, 131.73931884765625, -50.787635803222656, 127.61349487304688, -6.654411315917969, 33.490760803222656, -128.52828979492188, 45.985076904296875, 10.485359191894531, 9.0780029296875, 23.40606689453125, 83.77236938476562, 14.869674682617188, 80.83100891113281, 158.74838256835938, 12.480728149414062, -37.88348388671875, 0.14572715759277344, 48.744720458984375, 2.9188995361328125, 96.27540588378906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000261.npy"}
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 45.45140838623047, "std": 65.21623992919922, "min": -122.73136901855469, "p10": -30.97913055419922, "median": 37.37303924560547, "p90": 132.99302978515624, "max": 167.420166015625, "pos_frac": 0.7421875, "sample": [86.2330322265625, 17.62530517578125, 86.19549560546875, 74.34393310546875, 104.855712890625, 4.366432189941406, 160.093505859375, 126.58123779296875, 27.291107177734375, 147.66830444335938, 167.420166015625, -18.33294677734375, 32.13141632080078, 22.949356079101562, 10.766944885253906, 4.225244522094727, 103.17452239990234, 34.066802978515625, 12.269645690917969, 91.7220687866211, 129.89370727539062, -5.5349884033203125, -74.61042785644531, 128.90420532226562, 40.779541015625, 142.80731201171875, 24.683135986328125, 35.554107666015625, 110.601318359375, 152.38128662109375, 43.8359375, -122.73136901855469, 31.287429809570312, -19.422119140625, -0.9678497314453125, -101.02543640136719, -46.409637451171875, 63.5797119140625, 3.7974853515625, 125.65216064453125, 29.121200561523438, -6.186855316162109, -7.647064208984375, -42.206787109375, 34.72138977050781, 56.081031799316406, 57.96797180175781, -20.912994384765625, 74.75833129882812, -19.783203125, 76.95535278320312, 166.37838745117188, 67.80276489257812, 51.055259704589844, -34.76446533203125, -0.399810791015625, 84.16668701171875, 45.48895263671875, -24.42236328125, 15.94580078125, 68.64920043945312, -15.777557373046875, -5.6035308837890625, 76.115234375, -1.5491485595703125, 13.266357421875, 141.38580322265625, 8.4019775390625, 138.13226318359375, 42.54939270019531, -90.81854248046875, -2.9687271118164062, 134.9959716796875, 0.183685302734375, 72.72244262695312, -79.4947509765625, -73.57584381103516, 125.75418090820312, 2.04656982421875, -37.390625, -104.88832092285156, -16.703773498535156, 0.0, 129.0228271484375, 156.66195678710938, 40.904815673828125, 16.05316162109375, 97.94767761230469, 39.769309997558594, -0.3736572265625, 112.08275604248047, 115.6595458984375, 114.71443176269531, 20.87841796875, 134.8461456298828, 80.54742431640625, -6.894622802734375, 81.55673217773438, 57.46527099609375, 132.19883728027344, 5.546756744384766, 17.203125, 114.83203125, 122.21743774414062, -30.852005004882812, 15.112678527832031, 26.635009765625, 29.97125244140625, 134.88275146484375, 62.367584228515625, -1.7274436950683594, 39.19197082519531, 150.17498779296875, 93.82212829589844, 119.95974731445312, 21.62640380859375, 130.7522430419922, 73.83537292480469, 126.87327575683594, 117.54716491699219, 21.89971923828125, -53.27831268310547, 39.449066162109375, 78.89511108398438, 12.05596923828125, 94.43952941894531, 4.32720947265625, -31.2757568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000262.npy"}
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 37.774131774902344, "std": 70.00778198242188, "min": -153.00119018554688, "p10": -38.239368438720696, "median": 16.338448524475098, "p90": 140.128369140625, "max": 198.244140625, "pos_frac": 0.6796875, "sample": [139.94125366210938, 70.48516845703125, 139.43637084960938, 22.225341796875, -7.174934387207031, 91.30964660644531, -61.544647216796875, 8.376007080078125, 65.0739974975586, 59.1728515625, -36.39214324951172, 79.84719848632812, 27.83648681640625, 26.14422607421875, -43.54620361328125, 5.7096405029296875, 5.958396911621094, -89.3719482421875, 56.3350830078125, -6.57940673828125, 59.29412841796875, -42.549560546875, -20.031784057617188, 92.12660217285156, -9.9483642578125, 127.04290771484375, 55.45606994628906, 55.495826721191406, -13.4310302734375, -34.187835693359375, -17.562538146972656, 58.924041748046875, 145.155029296875, 143.500244140625, 147.60321044921875, -15.879631042480469, -9.976593017578125, -14.30426025390625, -67.37393188476562, 128.1720428466797, 7.748992919921875, -5.22015380859375, -7.264068603515625, 150.80523681640625, 58.135833740234375, 15.682748794555664, 102.81924438476562, -11.615097045898438, 124.93679809570312, 14.487586975097656, 97.1173095703125, -20.971588134765625, -18.721511840820312, 97.20686340332031, 140.56497192382812, 88.75106811523438, -104.87564086914062, 8.31768798828125, 115.59072875976562, 121.91671752929688, 0.0, 38.76361083984375, -153.00119018554688, 26.18621253967285, 63.77130126953125, 7.056243896484375, 1.620635986328125, 1.4943389892578125, 13.9974365234375, 21.175918579101562, 147.4355926513672, 3.6287994384765625, -59.13471984863281, -15.817792892456055, 11.835342407226562, 9.57611083984375, 93.1551513671875, 116.10284423828125, 72.43560791015625, 6.7977294921875, -5.5209197998046875, 7.97998046875, -50.71832275390625, 6.9381256103515625, 14.244064331054688, 74.2830810546875, -99.04412841796875, 147.47909545898438, 133.59988403320312, 16.99414825439453, -21.122802734375, 148.38446044921875, 11.687423706054688, -124.889404296875, -4.0763397216796875, -8.356964111328125, -92.5133056640625, 140.72296142578125, 19.088775634765625, -44.006439208984375, 102.55615234375, 5.79266357421875, 0.0, 9.675498962402344, 98.25186157226562, 77.0009765625, 198.244140625, 33.750213623046875, 179.2967529296875, -9.495307922363281, 101.48681640625, 0.8574600219726562, 134.1724090576172, 0.0, 143.63531494140625, 108.8765640258789, 19.424041748046875, 23.63287353515625, 113.63533020019531, 2.3602523803710938, 128.13134765625, 121.58595275878906, 57.87104797363281, -3.5025634765625, 178.21365356445312, -17.41495132446289, 74.57110595703125, -21.933502197265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000263.npy"}
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 48.66482162475586, "std": 63.974571228027344, "min": -129.25741577148438, "p10": -16.60979766845703, "median": 32.806785583496094, "p90": 139.14859313964843, "max": 200.1654052734375, "pos_frac": 0.75, "sample": [-6.162931442260742, -10.317108154296875, 161.23275756835938, 65.27679443359375, 148.7025146484375, 115.36127471923828, 97.83995056152344, 99.59320068359375, 1.1289844512939453, 12.116943359375, 2.2223472595214844, 130.84097290039062, 49.59417724609375, 31.532241821289062, -20.26666259765625, 110.1078109741211, 90.52870178222656, 119.79905700683594, -32.50048828125, 1.9284515380859375, 117.38323974609375, 95.69113159179688, -15.26165771484375, 69.3413314819336, -45.92823028564453, 170.072509765625, 129.607177734375, -0.5034370422363281, 64.74176025390625, 71.44134521484375, -28.358413696289062, 109.34353637695312, 133.52499389648438, 123.42680358886719, 151.76895141601562, 0.0, -9.8973388671875, 157.42999267578125, 38.56776428222656, 25.143882751464844, -30.72650146484375, 142.00234985351562, -16.096054077148438, 64.14202880859375, 89.02873992919922, 117.84228515625, -34.38340759277344, 161.91409301757812, -61.18865966796875, 35.574859619140625, 3.56231689453125, 18.51129150390625, 116.36456298828125, 37.75114440917969, 114.49124145507812, -96.20199584960938, 128.3969268798828, 102.3580322265625, 57.246986389160156, 18.69182586669922, 53.9888916015625, 7.367435455322266, 13.858978271484375, 56.35718536376953, 53.06535339355469, -24.0367431640625, -6.14892578125, 22.646438598632812, 19.29150390625, 125.5118408203125, -0.669097900390625, 20.485382080078125, -129.25741577148438, -15.485710144042969, 24.55794906616211, -2.0633468627929688, 16.526195526123047, 10.255348205566406, 138.56069946289062, 141.01419067382812, 77.17266845703125, 2.8402862548828125, -0.7209548950195312, 14.931777954101562, -1.83837890625, 200.1654052734375, -5.324165344238281, 0.0, 96.51470947265625, 33.37431335449219, -44.816864013671875, 82.85447692871094, 116.5197525024414, 139.47195434570312, 19.43670654296875, 105.18769836425781, -3.958263397216797, 40.75579833984375, 98.06430053710938, 20.717239379882812, 139.010009765625, 75.55438232421875, 12.668632507324219, -17.80853271484375, 3.679931640625, 9.902191162109375, 32.2392578125, 8.22027587890625, -14.999359130859375, 106.56055450439453, 1.79693603515625, 47.97607421875, 46.15777587890625, 181.04518127441406, 69.07659912109375, 152.27960205078125, -5.458038330078125, 21.881126403808594, 145.55499267578125, -7.342842102050781, -61.959068298339844, 12.267852783203125, 92.03634643554688, 6.6895294189453125, 103.80657958984375, 2.2079524993896484, 39.287567138671875, 14.214599609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000264.npy"}
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 44.611412048339844, "std": 71.26216125488281, "min": -120.712646484375, "p10": -41.766395568847656, "median": 36.426177978515625, "p90": 139.73466796875, "max": 183.59771728515625, "pos_frac": 0.71875, "sample": [2.221881866455078, 157.29031372070312, 124.14093017578125, -94.3863525390625, 46.290863037109375, 6.082977294921875, 124.46550750732422, 83.76179504394531, 87.28872680664062, -83.52261352539062, 50.64436340332031, 40.518829345703125, 97.71456146240234, -16.04632568359375, 143.0072021484375, 51.18241882324219, 34.456329345703125, 134.20440673828125, 100.2962646484375, 95.42613220214844, 183.59771728515625, 152.1256103515625, 152.8543701171875, 13.873123168945312, 167.46627807617188, 110.10601806640625, 131.29666137695312, 143.8166961669922, 138.93356323242188, -6.0550079345703125, -36.22924041748047, 119.61703491210938, 158.25747680664062, 7.6270904541015625, 124.71076965332031, 141.60391235351562, 83.99703979492188, 119.4273681640625, 38.396026611328125, 135.82614135742188, -46.80841064453125, -0.137115478515625, -38.709716796875, 61.747955322265625, -10.193031311035156, -13.563919067382812, -19.454345703125, -40.57470703125, -3.5511741638183594, 103.73506164550781, 2.02203369140625, -17.931686401367188, -36.50343322753906, 2.348968505859375, -41.37013244628906, 135.05224609375, 115.97088623046875, 13.133560180664062, 92.46418762207031, -120.712646484375, 18.78396987915039, 113.070556640625, 4.880615234375, 105.63780212402344, 22.157470703125, -51.3267822265625, 137.10581970214844, 23.18817138671875, 50.81608581542969, -27.15375518798828, -43.78009033203125, 128.8521728515625, 149.7978057861328, 24.893707275390625, 0.0, -5.3492584228515625, -82.44598388671875, -104.55218505859375, 134.58969116210938, 122.44996643066406, 39.285888671875, 4.149322509765625, 69.4292984008789, 1.651153564453125, 113.19051361083984, 46.362823486328125, 0.0, -4.408149719238281, -10.370574951171875, 92.98834228515625, 73.9228515625, -27.537109375, 0.178314208984375, 1.1022872924804688, 47.72882080078125, 4.783943176269531, 57.9686279296875, 175.4854736328125, 14.11444091796875, -102.77099609375, 16.882301330566406, 59.44256591796875, 27.7900390625, 50.17526626586914, 33.511993408203125, 30.437896728515625, -106.7823486328125, 153.58477783203125, 119.808349609375, -43.1533203125, -5.287055969238281, 64.35107421875, 6.02557373046875, -11.545509338378906, 72.46312713623047, -42.691009521484375, 0.4677314758300781, 143.51055908203125, 90.80885314941406, 96.12202453613281, 99.59719848632812, -12.532623291015625, 94.20836639404297, 29.484771728515625, 22.4162654876709, -82.09588623046875, 28.363449096679688, 52.773681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000265.npy"}
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 48.459449768066406, "std": 66.00502014160156, "min": -153.27078247070312, "p10": -12.925703430175782, "median": 38.27596664428711, "p90": 130.5476287841797, "max": 178.36697387695312, "pos_frac": 0.8203125, "sample": [43.4088134765625, 105.69843292236328, 178.36697387695312, 19.678741455078125, -13.091400146484375, -12.854690551757812, 105.78208923339844, 0.361297607421875, -22.391082763671875, 12.966560363769531, 37.550537109375, -0.830535888671875, -4.4537506103515625, 55.7135009765625, 86.23664855957031, -9.392486572265625, -10.239410400390625, 13.00054931640625, 45.35508728027344, -93.08614349365234, 123.17713928222656, 16.54151153564453, 3.1436614990234375, 30.644012451171875, 134.66293334960938, 99.99542236328125, -1.0263671875, 64.50611877441406, 144.0421142578125, 130.93106079101562, 2.677642822265625, 159.28973388671875, -14.197525024414062, 46.78765869140625, 7.016567230224609, 113.11724853515625, -44.529541015625, 10.318851470947266, 82.49490356445312, 101.83889770507812, 3.060546875, 24.3773193359375, 14.901412963867188, 1.887725830078125, 122.13231658935547, 110.22128295898438, 92.48704528808594, 125.1346206665039, 39.00139617919922, 30.281494140625, 14.979339599609375, 109.29386901855469, -119.17514038085938, -105.16349792480469, 0.299102783203125, -42.22026062011719, 5.784313201904297, 42.12693786621094, 46.1832275390625, 80.68179321289062, 128.36825561523438, 96.51268005371094, 15.908164978027344, 17.596588134765625, 21.286222457885742, 150.86630249023438, 98.87471008300781, 124.33357238769531, 11.420654296875, 31.358551025390625, 36.159820556640625, 143.84161376953125, 69.259521484375, 46.65350341796875, 124.4843978881836, 169.9368896484375, 41.11395263671875, -3.8135986328125, -79.25152587890625, -103.90482330322266, -153.27078247070312, 137.43943786621094, 15.65521240234375, 125.09391784667969, -15.358169555664062, 53.46600341796875, 41.897979736328125, 20.07568359375, 130.38330078125, 30.372711181640625, -81.59336853027344, 5.8746337890625, -0.4909553527832031, 28.497283935546875, 119.85174560546875, 65.80984497070312, 109.61985778808594, 89.5567626953125, 42.91552734375, 6.901214599609375, 22.00048828125, 49.11383056640625, 28.45263671875, 149.65643310546875, 20.42925453186035, 22.829116821289062, -11.802017211914062, 0.7805557250976562, 47.94044494628906, -5.5697021484375, 118.927490234375, 145.8160400390625, 150.90602111816406, 13.41412353515625, 125.35650634765625, 9.8367919921875, 124.8380126953125, 93.28237915039062, 12.877212524414062, 40.511077880859375, 174.87762451171875, 15.729866027832031, 125.36537170410156, 9.659984588623047, 122.1229248046875, 100.94465637207031, 107.01425170898438, 128.30836486816406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000266.npy"}
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 40.872371673583984, "std": 74.23519897460938, "min": -151.48312377929688, "p10": -45.1136245727539, "median": 32.47819137573242, "p90": 133.39461212158204, "max": 228.197998046875, "pos_frac": 0.703125, "sample": [3.678171157836914, 122.07254028320312, 110.88542175292969, 96.15945434570312, 169.8505859375, 1.540679931640625, 8.73318099975586, 32.02191162109375, -58.9561767578125, -3.673187255859375, -0.1888580322265625, 27.62109375, -36.88616943359375, 132.51614379882812, 32.934471130371094, -27.803924560546875, 17.82263946533203, 149.830078125, 149.65989685058594, 35.859771728515625, -73.94320678710938, 160.34515380859375, 118.29373168945312, 7.133995056152344, 63.81499099731445, -21.182403564453125, 64.94682312011719, 107.15232849121094, 96.56829833984375, -127.98431396484375, 114.00076293945312, 23.52093505859375, 28.31201171875, -19.415130615234375, 13.294570922851562, 98.14590454101562, -15.793193817138672, 149.830078125, 127.80377197265625, 83.83699798583984, -90.39077758789062, 148.53903198242188, 72.95093536376953, 107.55300903320312, 13.083473205566406, 188.09146118164062, 8.55419921875, -25.787628173828125, 77.48965454101562, 133.85284423828125, -109.81639099121094, -86.93008422851562, 111.7071533203125, 113.1030044555664, 10.79083251953125, 181.06427001953125, 112.91400909423828, 98.96978759765625, 145.86148071289062, 20.00834083557129, 151.42608642578125, 71.17011260986328, -108.6671142578125, 11.6241455078125, 57.718536376953125, 2.4365234375, 73.53155517578125, -71.0447998046875, 49.97161865234375, 110.70401000976562, 12.707122802734375, 85.60888671875, 77.29820251464844, -151.48312377929688, 54.685394287109375, 91.63845825195312, -35.78916931152344, 78.502197265625, 4.917381286621094, -0.3174552917480469, -49.62507629394531, 100.1455078125, -9.855331420898438, 47.88275146484375, -21.847244262695312, 87.56777954101562, -51.590667724609375, 53.772369384765625, 17.0225830078125, 96.46792602539062, -26.993736267089844, -27.028579711914062, 62.07716369628906, -5.389373779296875, 2.9734954833984375, 109.54998016357422, 90.01119995117188, 77.64555358886719, 57.37190246582031, -13.583648681640625, 125.69583129882812, 40.809356689453125, -28.237335205078125, 39.077667236328125, -12.375717163085938, 28.574310302734375, 27.582443237304688, 115.19544982910156, -6.559783935546875, 228.197998046875, -23.395423889160156, -40.90374755859375, 104.95211029052734, 30.22186279296875, 98.06709289550781, 2.173961639404297, 133.9996337890625, 109.49737548828125, -6.376953125, -43.180145263671875, 133.19822692871094, -142.53456115722656, -4.5927734375, 4.184562683105469, 68.67652130126953, -81.31967163085938, -36.07859802246094, 9.9324951171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000267.npy"}
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 43.368831634521484, "std": 65.44271087646484, "min": -115.34080505371094, "p10": -23.074153137207023, "median": 31.48253631591797, "p90": 133.08974609375, "max": 181.98907470703125, "pos_frac": 0.734375, "sample": [31.45989990234375, 13.577827453613281, 92.2793960571289, 117.645751953125, 103.74569702148438, 6.576751708984375, -2.6738853454589844, -67.42267608642578, -20.952743530273438, -66.6031494140625, 2.4048919677734375, 167.79910278320312, 20.236572265625, 133.36309814453125, 1.691436767578125, 93.36184692382812, 181.98907470703125, -5.203779220581055, -1.24200439453125, 108.77975463867188, 117.23807525634766, 128.77561950683594, 54.867034912109375, -115.34080505371094, 58.52813720703125, 31.505172729492188, -4.195734024047852, 45.76568603515625, 50.9503173828125, 35.375953674316406, 9.036088943481445, 105.62385559082031, 112.56509399414062, 97.0172119140625, -19.640060424804688, 1.5859375, -114.47099304199219, -75.06442260742188, -28.02410888671875, 178.74981689453125, 116.32473754882812, 45.44446563720703, 27.1220703125, 134.41107177734375, -2.878753662109375, 13.935226440429688, 66.55429077148438, -106.10504150390625, -46.53009033203125, -12.78558349609375, 32.58259582519531, 25.59454345703125, 53.60357666015625, 132.68423461914062, 5.369411468505859, 41.499114990234375, 161.85836791992188, -3.6394119262695312, 40.397216796875, 4.5845947265625, 172.93978881835938, 20.498374938964844, 124.97516632080078, -7.354400634765625, -28.60491943359375, 66.7108154296875, 135.97900390625, 49.014923095703125, 6.265625, -8.764892578125, 103.5509033203125, 20.3548583984375, 118.09957885742188, 121.05807495117188, 17.507293701171875, -9.298774719238281, 49.38543701171875, 147.07513427734375, 132.97259521484375, 71.96173095703125, 79.55230712890625, 36.79742431640625, 68.55535888671875, -7.226036071777344, 75.8966064453125, 101.89862060546875, -33.50926208496094, 54.555938720703125, 62.02107238769531, 6.986053466796875, 42.43963623046875, 118.31692504882812, 158.54110717773438, 83.90036010742188, -0.9937744140625, 131.29623413085938, 18.280853271484375, 5.90570068359375, 37.3018798828125, 44.1407470703125, -8.522979736328125, 10.018508911132812, 56.3973388671875, -2.0807647705078125, 5.96009635925293, 77.27545166015625, 12.39471435546875, -34.711090087890625, 0.43988037109375, 40.5369873046875, 171.69906616210938, 11.316474914550781, 107.0462646484375, 129.5718231201172, -0.16754150390625, 150.23565673828125, -82.69296264648438, -7.0129852294921875, 175.28021240234375, 14.327713012695312, -0.39079856872558594, -39.378662109375, 19.844139099121094, 28.440170288085938, 8.881103515625, 12.650901794433594, -6.819976806640625, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000268.npy"}
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 42.236236572265625, "std": 72.64724731445312, "min": -190.77389526367188, "p10": -29.487522888183587, "median": 32.977325439453125, "p90": 134.12164154052735, "max": 179.54461669921875, "pos_frac": 0.71875, "sample": [-111.08631896972656, 62.584800720214844, 115.38848876953125, -58.4952392578125, 68.851806640625, -1.356536865234375, 11.489433288574219, 125.50079345703125, 12.5089111328125, -27.00799560546875, 104.24940490722656, -6.018333435058594, -8.919685363769531, 91.5746078491211, -34.2059326171875, 119.67153930664062, 5.80950927734375, -3.830474853515625, 127.31314849853516, 104.39952087402344, 179.54461669921875, 164.370849609375, 45.562774658203125, 147.1313934326172, 163.71505737304688, -22.032821655273438, 35.471160888671875, 127.74650573730469, -12.33343505859375, 76.09136962890625, 3.7793102264404297, 147.5658721923828, 115.0975341796875, -93.64459228515625, 108.89108276367188, 0.0, 36.352874755859375, -68.1378173828125, 92.23863220214844, 110.41058349609375, -42.67017364501953, 18.560317993164062, 141.22732543945312, 80.82969665527344, 21.367027282714844, 91.85372161865234, 147.6068115234375, 0.0, 95.59812927246094, 43.199859619140625, 60.22554016113281, -20.90869140625, 133.05177307128906, -128.40713500976562, -80.18954467773438, 130.83201599121094, 104.3270034790039, -15.994956970214844, -117.67556762695312, -0.40464019775390625, 126.073974609375, 28.683395385742188, 60.73480224609375, 6.816925048828125, 109.53334045410156, -22.38616943359375, 12.73321533203125, 89.91641235351562, 125.09831237792969, 11.551170349121094, 54.915771484375, 118.68399047851562, 0.0, 138.9608612060547, 15.211814880371094, 20.99542236328125, -135.81524658203125, 2.480579376220703, 46.31640625, 4.9866485595703125, -6.75531005859375, -12.4315185546875, 138.98016357421875, 23.173255920410156, 83.06120300292969, 31.10296630859375, 31.801025390625, 57.29364013671875, 74.02186584472656, 135.32244873046875, 58.00736999511719, 30.546966552734375, 15.313812255859375, -20.216064453125, 125.3505859375, 67.02520751953125, -16.844161987304688, 111.31820678710938, 7.37811279296875, 97.80436706542969, 36.619903564453125, 34.15362548828125, 1.8675727844238281, -19.911956787109375, 4.514898300170898, -115.35757446289062, 91.40560150146484, 139.94256591796875, 17.722091674804688, 99.13545989990234, -24.400177001953125, 13.258003234863281, 94.46566772460938, 2.47607421875, 107.25169372558594, -58.40341567993164, -27.465347290039062, -5.291297912597656, 18.986190795898438, 1.640777587890625, 105.00796508789062, 137.56497192382812, 77.421630859375, -190.77389526367188, -16.48223876953125, 24.183876037597656, 133.6070098876953, 157.6820068359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000269.npy"}
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 49.7177848815918, "std": 78.10649108886719, "min": -181.77041625976562, "p10": -37.053469848632815, "median": 49.087961196899414, "p90": 140.68714599609376, "max": 221.45904541015625, "pos_frac": 0.734375, "sample": [-152.6295166015625, 48.51167297363281, 104.294921875, 113.85616302490234, -28.316722869873047, 141.867431640625, -9.849967956542969, 30.3829345703125, 79.43051147460938, -57.614837646484375, 129.43115234375, 31.986114501953125, 19.4261474609375, 44.16291809082031, 24.528915405273438, 118.44317626953125, -15.571624755859375, 143.76100158691406, 86.42768096923828, -119.08070373535156, -37.480010986328125, 65.43302154541016, 4.9385528564453125, 39.18359375, 12.666849136352539, -3.58154296875, 121.76841735839844, 116.86925506591797, -19.866943359375, 0.9347248077392578, 40.349334716796875, 131.69886779785156, 115.29348754882812, 3.402984619140625, 8.474449157714844, 10.834465026855469, -51.6373291015625, 34.47816467285156, -153.90008544921875, 98.57302856445312, -64.9920654296875, 140.55377197265625, 30.466217041015625, 13.844589233398438, 106.09381103515625, 13.485763549804688, -4.893669128417969, 138.46612548828125, 86.50210571289062, -124.5946044921875, -120.63934326171875, 134.32693481445312, 139.4827880859375, 137.3932647705078, 59.49406433105469, 111.97048950195312, 152.55630493164062, -36.87066650390625, 131.00665283203125, 5.775146484375, 31.36895751953125, -19.8094482421875, 11.74774169921875, 37.23321533203125, 133.843505859375, 112.01309204101562, 146.16897583007812, 4.240688323974609, 131.617919921875, 60.591888427734375, 97.55703735351562, -5.5562744140625, -140.95989990234375, 123.23395538330078, -54.844879150390625, 140.99835205078125, 67.92790222167969, 49.664249420166016, 110.37821960449219, -17.51983642578125, -12.20147705078125, 10.7135009765625, 87.21723937988281, 126.10153198242188, 20.21932601928711, -6.618812561035156, -0.073638916015625, -7.956451416015625, 0.740936279296875, -4.874412536621094, -35.84381103515625, -47.36277770996094, 105.07330322265625, 169.13995361328125, -8.303009033203125, 81.0706558227539, 56.693206787109375, 143.943115234375, 59.7269287109375, -6.426483154296875, 124.13613891601562, 163.224365234375, 42.945404052734375, 94.98058319091797, 63.09641647338867, -181.77041625976562, 28.19818115234375, -7.79595947265625, 155.11074829101562, 7.12286376953125, 88.20849609375, 113.627685546875, 77.80368041992188, 125.951416015625, 121.71456909179688, 150.4659423828125, 141.8880615234375, 54.26101303100586, 9.588958740234375, 156.05645751953125, -11.8145751953125, 221.45904541015625, 125.58316040039062, 130.95697021484375, 101.79451751708984, 108.0704345703125, 115.6456298828125, -28.815673828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000270.npy"}
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 46.53667449951172, "std": 66.95491027832031, "min": -100.3685531616211, "p10": -18.573547363281246, "median": 31.398681640625, "p90": 128.58844299316405, "max": 215.2921142578125, "pos_frac": 0.7578125, "sample": [110.85326385498047, -100.3685531616211, -8.119659423828125, 31.925048828125, 17.070709228515625, 0.453094482421875, 10.186225891113281, -11.463844299316406, 36.625732421875, -6.010986328125, 9.591461181640625, 16.818130493164062, 126.30327606201172, 1.4716796875, 13.669570922851562, 57.906158447265625, 93.37274169921875, 35.78978729248047, 30.52734375, 48.4549560546875, 18.545146942138672, 10.578414916992188, 122.70681762695312, 108.33547973632812, 2.9236984252929688, 4.343446731567383, -33.6282958984375, 114.01591491699219, 28.71860122680664, -10.520538330078125, 112.20356750488281, 143.50311279296875, -2.526947021484375, 44.35107421875, 1.45428466796875, 28.72662353515625, 35.99444580078125, 54.02549743652344, 98.80105590820312, -81.82244873046875, 120.75115966796875, 39.3189697265625, 1.9793701171875, 124.36752319335938, 110.05674743652344, 26.654205322265625, 47.45477294921875, -0.9000473022460938, 113.89717102050781, 10.51177978515625, -4.589996337890625, 61.634185791015625, 139.273681640625, 66.54383850097656, 215.2921142578125, 159.27508544921875, 1.0201740264892578, 141.40670776367188, 28.42181396484375, 117.88551330566406, 104.39739990234375, 9.49664306640625, -5.691688537597656, 192.18646240234375, 111.95968627929688, -59.408294677734375, -21.215850830078125, 129.66632080078125, -8.13995361328125, 35.848968505859375, 126.20368957519531, 173.36428833007812, -27.299163818359375, 94.68583679199219, -14.80023193359375, 22.971893310546875, -86.71292877197266, 30.872314453125, 41.06951904296875, 96.73490142822266, -24.695465087890625, -98.68411254882812, 141.19952392578125, -6.9695281982421875, 166.84173583984375, 93.84573364257812, -6.187530517578125, -35.723236083984375, 62.67671585083008, 35.3935546875, 7.810264587402344, 140.2490234375, 114.30206298828125, -14.163299560546875, 3.173583984375, 109.30326843261719, 8.104557037353516, 105.98479461669922, 126.6766357421875, 38.52793884277344, 9.524520874023438, 24.596817016601562, -4.671142578125, 105.4595947265625, 158.572265625, -7.3254547119140625, -99.53997802734375, 11.686065673828125, 16.806076049804688, 101.18212890625, 41.5787353515625, 110.29708862304688, 102.69243621826172, -17.441131591796875, 36.382843017578125, 16.4920654296875, 121.91888427734375, 126.66900634765625, -2.47503662109375, -8.955474853515625, 121.02629089355469, -96.73858642578125, 128.12649536132812, 155.98074340820312, 19.163291931152344, 1.4107818603515625, 98.439453125, -64.06051635742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000271.npy"}
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 43.434303283691406, "std": 69.29296112060547, "min": -142.9742431640625, "p10": -24.418345642089843, "median": 30.103897094726562, "p90": 138.23759155273436, "max": 193.34637451171875, "pos_frac": 0.7578125, "sample": [97.07791137695312, 0.0, -4.6395263671875, 19.890274047851562, -76.07943725585938, 62.805908203125, 168.0484619140625, 68.194091796875, 122.19418334960938, 4.8520050048828125, -23.916549682617188, 23.532428741455078, 50.16888427734375, 122.05706787109375, 66.64654541015625, 154.419189453125, 157.25067138671875, 107.54769897460938, 43.86309814453125, 22.88146209716797, 117.64149475097656, 9.699600219726562, -10.2728271484375, -2.011138916015625, 119.48033142089844, -27.945541381835938, 4.5590362548828125, 167.92474365234375, 1.1089744567871094, 15.46514892578125, 117.8251953125, 51.39606475830078, 72.12289428710938, 2.5773277282714844, 136.78012084960938, 99.24127197265625, 161.65286254882812, 137.29498291015625, 13.268463134765625, -2.0864105224609375, 89.23448181152344, -46.45880126953125, -57.20086669921875, 82.33392333984375, -5.994415283203125, 29.848602294921875, 121.20310974121094, -47.685333251953125, -1.0466537475585938, 1.5529499053955078, 141.55197143554688, 114.59078979492188, -15.253866195678711, 20.733360290527344, 124.37104797363281, 157.72222900390625, 37.021453857421875, 46.4344482421875, 103.42813110351562, 0.7331275939941406, 3.0731334686279297, 9.679473876953125, 8.322799682617188, 16.404754638671875, 99.57313537597656, 60.08216857910156, 193.34637451171875, 75.65386962890625, 149.7587890625, 10.25750732421875, 128.8397216796875, -21.6026611328125, 49.643035888671875, 12.595664978027344, -132.0887451171875, -104.92259216308594, -9.045745849609375, 57.216766357421875, 5.843597412109375, 126.61248779296875, 102.13339233398438, 52.0882568359375, -8.94024658203125, 164.26051330566406, 2.0528564453125, -142.9742431640625, 20.526914596557617, 13.776012420654297, 31.626617431640625, 4.5669708251953125, 36.0555419921875, 90.78959655761719, 6.4578704833984375, -90.96292114257812, 90.1114501953125, 64.8236083984375, 50.30316162109375, 114.76268768310547, -14.918243408203125, 3.7008056640625, 0.0, -25.589202880859375, 11.475845336914062, -109.53076171875, -20.36834716796875, 5.16595458984375, 17.839874267578125, -0.3337249755859375, -2.3397369384765625, 30.35919189453125, 99.02937316894531, 191.17230224609375, -48.93353271484375, 64.17301177978516, 130.39776611328125, -95.41607666015625, 13.036567687988281, 3.592945098876953, 7.60191535949707, -11.47613525390625, 148.70904541015625, 65.22399139404297, 30.670440673828125, 47.3868408203125, 70.95553588867188, 140.43701171875, 113.63661193847656, 83.59352111816406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000272.npy"}
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 47.872642517089844, "std": 74.1922378540039, "min": -151.772705078125, "p10": -40.14247283935547, "median": 44.36719512939453, "p90": 142.7968292236328, "max": 207.8701171875, "pos_frac": 0.6953125, "sample": [23.674407958984375, 166.7060546875, 154.91748046875, 10.08912467956543, 162.7269287109375, -20.661346435546875, 185.1240234375, 22.99609375, 61.7449951171875, 8.015068054199219, -33.095916748046875, 48.88426971435547, -0.2824592590332031, 97.69914245605469, 207.8701171875, -151.772705078125, 73.11468505859375, 180.15371704101562, 13.96685791015625, 54.653106689453125, -40.03083801269531, 87.20178985595703, 126.36651611328125, 24.68943977355957, 147.76132202148438, 27.02471923828125, -119.61160278320312, 34.809295654296875, 115.70635986328125, -25.744491577148438, 18.520828247070312, -40.4029541015625, -26.098304748535156, 124.40286254882812, -22.189743041992188, -0.806915283203125, 8.04132080078125, -3.093994140625, -5.766265869140625, 130.95355224609375, -6.07151985168457, -8.135986328125, 1.0965728759765625, 47.74822998046875, 57.51849365234375, 25.17889404296875, 104.95572662353516, -0.1586151123046875, -73.20132446289062, 4.8673858642578125, 99.87545776367188, -23.90618896484375, 149.99642944335938, -20.698944091796875, 90.5157470703125, -6.33770751953125, 1.16204833984375, 57.62689208984375, 74.13313293457031, 82.58519744873047, -55.56134033203125, 113.98741912841797, 7.409965515136719, 107.70782470703125, -21.42327880859375, 151.50762939453125, 100.98963928222656, 0.27954864501953125, 104.16757202148438, 5.10577392578125, 122.76620483398438, 2.8340606689453125, -49.879791259765625, -13.057952880859375, 62.47015380859375, 138.88848876953125, 93.82916259765625, 131.41201782226562, 117.6754150390625, -77.30877685546875, 103.61384582519531, 117.75408172607422, 70.63739013671875, -50.52793502807617, 59.57171630859375, 115.12877655029297, 29.68756103515625, 43.29766082763672, 145.48583984375, 63.72032165527344, 114.03091430664062, -43.940162658691406, -2.231781005859375, 120.58041381835938, -144.47830200195312, 34.25238037109375, -100.2730712890625, 122.84657287597656, 139.0257568359375, -8.666839599609375, 156.08953857421875, 92.63468933105469, 144.58282470703125, 130.0521697998047, 133.99313354492188, -5.807430267333984, -10.157470703125, 2.8841552734375, 125.22067260742188, 108.75153350830078, 19.2120361328125, 44.79034423828125, -1.687469482421875, 44.048858642578125, 44.68553161621094, 66.59002685546875, -67.632568359375, 142.03140258789062, 26.007598876953125, -5.411773681640625, -11.115833282470703, 136.12832641601562, 47.960113525390625, 120.09105682373047, 154.70077514648438, -61.407867431640625, 128.6983642578125, -38.55555725097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000273.npy"}
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 51.271785736083984, "std": 72.46210479736328, "min": -146.73773193359375, "p10": -20.94597473144531, "median": 42.32715606689453, "p90": 134.4045684814453, "max": 219.38409423828125, "pos_frac": 0.7890625, "sample": [122.56063842773438, 124.9820556640625, -55.8665771484375, 120.68096923828125, -1.1697998046875, 70.7652587890625, 64.51248168945312, 35.25401306152344, 106.1059341430664, 11.813695907592773, 19.75860595703125, 113.09455108642578, 170.3134002685547, 28.35932159423828, 25.276718139648438, 107.1255874633789, 182.9342041015625, 153.40174865722656, -61.42822265625, -9.196613311767578, 19.610626220703125, 158.68423461914062, 126.22439575195312, 42.46661376953125, 47.362579345703125, -3.3974609375, -112.79763793945312, -22.4295654296875, 14.113004684448242, 95.24334716796875, 15.81939697265625, 35.50531005859375, 110.55430603027344, 67.38037109375, 122.19560241699219, 59.148193359375, 44.00775146484375, 114.12806701660156, 7.8100128173828125, 150.91738891601562, 152.05911254882812, -12.331731796264648, -20.310150146484375, 9.621467590332031, -32.983612060546875, 128.54861450195312, 14.209335327148438, 13.322578430175781, 28.882232666015625, 157.62210083007812, -9.043014526367188, 7.265380859375, 118.68190002441406, 71.21157836914062, -128.55996704101562, 29.779159545898438, -63.68919372558594, 28.58270263671875, 1.32855224609375, 8.748809814453125, 19.287752151489258, 4.459623336791992, -4.259010314941406, 107.12129211425781, -92.25033569335938, -1.6990776062011719, 132.10935974121094, -1.069284439086914, 57.862579345703125, 120.5814208984375, 13.0584716796875, -146.73773193359375, 133.3653564453125, 42.18769836425781, 63.39166259765625, 53.48223876953125, 48.17559814453125, 44.62339782714844, 219.38409423828125, 7.819114685058594, 188.49801635742188, 107.23814392089844, 22.29931640625, 132.68661499023438, -2.9909210205078125, 61.37689208984375, 118.85873413085938, 133.8863525390625, 74.64413452148438, 133.35211181640625, 128.76828002929688, 31.7884521484375, -75.51515197753906, 91.08248901367188, 116.29205322265625, 111.77808380126953, 23.201370239257812, 140.41830444335938, 159.66452026367188, 8.206497192382812, -93.23602294921875, 16.864166259765625, 6.264678955078125, -16.215225219726562, 25.229202270507812, 117.80313110351562, 11.453804016113281, 133.04466247558594, 81.21951293945312, 55.09814453125, 135.61373901367188, 105.65882110595703, -30.8319091796875, 57.5665283203125, -3.96221923828125, -20.124664306640625, 1.272186279296875, 4.037384033203125, 28.791656494140625, 65.61251831054688, 13.615758895874023, -143.151123046875, 121.2186279296875, -0.7947864532470703, 26.70180892944336, 108.87945556640625, 121.142578125, 182.8175048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000274.npy"}
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 57.661197662353516, "std": 70.03778076171875, "min": -122.22819519042969, "p10": -12.764611816406248, "median": 41.48224639892578, "p90": 143.0098419189453, "max": 179.8038330078125, "pos_frac": 0.7890625, "sample": [178.429443359375, -10.08013916015625, -2.982952117919922, -102.0264892578125, 138.3477783203125, 100.04133605957031, 136.07473754882812, 10.30279541015625, 13.53021240234375, -107.9683837890625, 15.504623413085938, 154.2789306640625, -36.2659912109375, 143.23104858398438, 0.0, 9.447250366210938, 16.607925415039062, 103.46676635742188, 3.217357635498047, 4.057708740234375, 0.0, 0.0, 120.88934326171875, 122.90499877929688, 104.25858306884766, 4.09149169921875, 151.0025634765625, 109.24519348144531, 20.30426025390625, 24.344192504882812, 58.789947509765625, 7.042079925537109, 124.25971221923828, 136.62103271484375, 161.60585021972656, 162.10903930664062, 0.5698509216308594, -0.79931640625, -122.22819519042969, 44.09777069091797, 115.242431640625, 30.882659912109375, 86.68245697021484, 26.887039184570312, 14.913711547851562, -4.8764495849609375, -23.53076171875, 131.78030395507812, 141.2669677734375, 17.78033447265625, 141.37820434570312, -99.6053466796875, 2.269990921020508, 59.30787658691406, 101.5419921875, 26.41888427734375, 101.19969177246094, 35.3153076171875, 137.38592529296875, 0.0, 23.14111328125, 38.866722106933594, 24.668182373046875, 142.9150390625, -6.1558685302734375, 166.05923461914062, 166.29678344726562, 61.46881103515625, 48.85359191894531, 0.0, 10.760139465332031, 148.70791625976562, 21.215789794921875, 114.50885009765625, -8.30694580078125, 119.76519775390625, 138.68011474609375, 179.8038330078125, 54.47007751464844, 7.370700836181641, 118.79267883300781, 129.60736083984375, 90.39056396484375, 94.05130004882812, 84.10652923583984, 2.5320777893066406, 23.061431884765625, 117.15277099609375, 38.08873748779297, 34.22088623046875, -14.3880615234375, 94.37079620361328, 9.414901733398438, 159.555908203125, -0.502044677734375, 130.6378173828125, -15.183547973632812, 53.54869079589844, 171.49298095703125, 10.459884643554688, 73.33253479003906, 31.5516357421875, 138.39114379882812, -5.564399719238281, 135.48011779785156, 124.28057098388672, 12.46197509765625, -48.3624267578125, 1.32373046875, -20.05999755859375, 136.59384155273438, -97.42185974121094, 115.43267822265625, 1.60894775390625, 66.7442626953125, 6.894464492797852, 18.551849365234375, 129.95751953125, 172.15789794921875, 119.99663543701172, -45.616180419921875, 129.42123413085938, 56.06919860839844, -19.901535034179688, 103.23356628417969, -12.06884765625, 126.34262084960938, 106.73760986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000275.npy"}
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 55.65225601196289, "std": 73.7024154663086, "min": -115.84225463867188, "p10": -26.926992988586427, "median": 36.637969970703125, "p90": 145.43585815429685, "max": 232.75469970703125, "pos_frac": 0.75, "sample": [232.75469970703125, 17.99707794189453, 118.1218032836914, -5.224800109863281, 19.78729248046875, -0.9510498046875, 38.09423828125, 58.883331298828125, 102.67190551757812, 81.01175689697266, 107.52323913574219, 8.74810791015625, -14.949111938476562, 2.29351806640625, -4.33489990234375, 80.28038787841797, 117.32553100585938, 32.124786376953125, -58.79541015625, 125.9100341796875, 147.739013671875, 81.10664367675781, 8.00088119506836, 148.04721069335938, 88.99923706054688, -24.537445068359375, -26.979705810546875, -87.7120361328125, -8.17803955078125, 126.79989624023438, 98.30392456054688, -3.3205718994140625, 59.45147705078125, 121.30585479736328, 41.708587646484375, 20.051910400390625, 31.48236083984375, -1.199300765991211, 113.73536682128906, 83.14942932128906, 9.10369873046875, 24.571563720703125, 129.56216430664062, 59.238311767578125, 9.357845306396484, 165.48590087890625, 109.28488159179688, 144.01385498046875, 14.8197021484375, -10.0626220703125, 8.599632263183594, 134.27499389648438, 112.90435791015625, 35.18170166015625, 140.71530151367188, 11.564529418945312, 45.22218322753906, 28.80859375, 138.46395874023438, -43.239410400390625, 0.0, -94.39456176757812, 53.1463623046875, -104.76551818847656, 9.869552612304688, -45.829193115234375, 111.86361694335938, 173.44989013671875, -1.6561737060546875, -16.68548583984375, 65.87660217285156, -20.263916015625, 107.06015014648438, 136.33282470703125, 116.30548858642578, 142.9036865234375, 137.42581176757812, -25.76751708984375, 34.537864685058594, 118.60383605957031, 140.1022186279297, 117.97177124023438, 117.18574523925781, 34.76678466796875, 12.200355529785156, 93.42698669433594, -52.010528564453125, 117.5438003540039, -2.2783584594726562, 48.16705322265625, -3.3118362426757812, 192.650390625, 17.5989990234375, 21.536376953125, 11.543472290039062, 10.237655639648438, -0.9469757080078125, 156.58258056640625, 14.941970825195312, 137.17724609375, 137.63430786132812, 34.40779113769531, 149.99484252929688, 2.5230712890625, -1.55816650390625, 105.70071411132812, -26.904401779174805, -40.1512451171875, 29.72589111328125, 152.59197998046875, 5.7000885009765625, 123.35733032226562, 17.793983459472656, 163.9576416015625, 134.6660919189453, -102.70315551757812, 172.97222900390625, 19.002166748046875, 117.90829467773438, -52.51947021484375, 160.54916381835938, 8.823699951171875, 122.68193054199219, -115.84225463867188, -76.26632690429688, 144.44879150390625, 174.84463500976562, 131.95172119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000276.npy"}
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 37.44212341308594, "std": 78.69465637207031, "min": -172.57330322265625, "p10": -42.47250747680664, "median": 30.860275268554688, "p90": 134.98449401855467, "max": 220.21319580078125, "pos_frac": 0.6875, "sample": [4.734649658203125, 5.29681396484375, -6.852195739746094, 121.25445556640625, 31.45513916015625, -24.138763427734375, -121.8485107421875, -7.59912109375, 118.9521484375, 7.52008056640625, 20.84454345703125, 97.43998718261719, 128.66177368164062, 94.98944091796875, -6.9958648681640625, 143.91748046875, 95.19102478027344, -163.3704833984375, 82.6982421875, -0.55377197265625, 30.627120971679688, 10.433652877807617, 105.19832611083984, 50.68963623046875, 104.1016845703125, 148.34371948242188, -0.613184928894043, -2.8724517822265625, 136.73049926757812, 46.53498077392578, 89.4657211303711, -156.269287109375, -2.993488311767578, -33.01506805419922, -54.511322021484375, 7.4010009765625, 117.26945495605469, 6.639430999755859, 95.98624420166016, 89.40301513671875, -172.57330322265625, 2.8123321533203125, -18.208213806152344, -28.082366943359375, 10.18658447265625, 91.23193359375, 43.979949951171875, 105.4290771484375, 3.4162750244140625, 40.457672119140625, 70.51617431640625, -12.82719898223877, 1.4449729919433594, 220.21319580078125, 44.071624755859375, 0.0, 71.86874389648438, 101.6159896850586, 116.3697509765625, 29.26702880859375, 161.98416137695312, 109.17279052734375, -123.90199279785156, -4.507183074951172, 1.49969482421875, 5.563018798828125, 97.71478271484375, -105.40609741210938, 192.7056884765625, 187.24246215820312, -48.02156448364258, 19.703277587890625, 165.32940673828125, 55.858367919921875, 7.845306396484375, -163.4750213623047, 23.030517578125, -41.08164978027344, -40.87876892089844, 43.190547943115234, 20.04229736328125, 110.71729278564453, 53.787506103515625, -21.3642578125, 82.6185302734375, 113.74299621582031, 109.53780364990234, 106.62211608886719, -97.61738586425781, 153.61822509765625, 2.8220062255859375, -1.49688720703125, -15.94964599609375, 83.3608627319336, 68.32489013671875, -42.348876953125, 56.649932861328125, 108.28369140625, 38.70368957519531, 124.58566284179688, 4.0320892333984375, 13.933311462402344, 122.88778686523438, 143.06387329101562, 35.05458068847656, 126.21279907226562, 43.86181640625, 140.08074951171875, -24.872100830078125, 0.0, 139.54922485351562, 174.2979736328125, -36.268524169921875, 65.92256164550781, -23.26080322265625, -10.156204223632812, -126.93963623046875, 134.2362060546875, 15.311874389648438, 31.093429565429688, 106.1248779296875, -107.93695068359375, -0.8524723052978516, -2.5262985229492188, 6.291965484619141, 96.768310546875, 39.89757537841797, -42.76097869873047], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000277.npy"}
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 34.40441131591797, "std": 80.69239044189453, "min": -183.14352416992188, "p10": -78.58966674804687, "median": 22.868427276611328, "p90": 133.0704833984375, "max": 215.22244262695312, "pos_frac": 0.7109375, "sample": [130.58563232421875, 104.60116577148438, -46.068603515625, 46.56974792480469, 19.375717163085938, 30.439544677734375, 65.206298828125, 17.450469970703125, 1.9680747985839844, 140.45423889160156, 42.593719482421875, 123.82523345947266, 100.48318481445312, 27.869277954101562, 109.50944519042969, 111.173828125, -119.26882934570312, -18.551605224609375, 25.168746948242188, -5.446014404296875, 7.741203308105469, 78.69873046875, 20.952728271484375, -26.195205688476562, 8.875946044921875, 185.30001831054688, 157.21566772460938, 65.33926391601562, 11.12359619140625, -37.722900390625, -177.94247436523438, 192.8282470703125, -105.69640350341797, 4.560394287109375, 113.88507080078125, 110.19075012207031, 84.32441711425781, -73.77572631835938, 160.526123046875, 23.764442443847656, 142.3062744140625, 161.98944091796875, 90.44332885742188, 12.019775390625, -24.003021240234375, 61.18133544921875, -1.4676055908203125, 157.57009887695312, 63.038734436035156, 86.09464263916016, 8.816770553588867, -0.34466552734375, 2.3341140747070312, -97.63389587402344, -18.569931030273438, -183.14352416992188, 2.376161575317383, 164.78424072265625, -76.38751220703125, -104.88987731933594, 55.76591491699219, 114.78915405273438, 27.950103759765625, 8.627685546875, -25.21209716796875, 134.96600341796875, -97.89691162109375, 125.60826110839844, 124.58870697021484, 35.45442199707031, 166.87945556640625, 215.22244262695312, -28.19974708557129, -42.66148376464844, 75.25732421875, -109.04463195800781, 107.5775375366211, 67.24730682373047, 62.925445556640625, 50.3367919921875, -139.69728088378906, 10.737556457519531, 3.059215545654297, 52.8408203125, 110.9391860961914, -20.763214111328125, 37.36370849609375, 64.69646453857422, 14.777099609375, 112.07109069824219, -0.57537841796875, -7.22705078125, 121.60084533691406, 181.8682861328125, -27.361724853515625, 132.01280212402344, -94.05018615722656, 17.813385009765625, 89.72894287109375, 1.8192596435546875, -11.170242309570312, 7.513160705566406, 21.972412109375, -102.01041412353516, -16.790313720703125, 27.521499633789062, 113.257080078125, -17.298133850097656, 18.64422607421875, 13.004467010498047, 117.83235168457031, 3.837512969970703, 42.42236328125, 95.92657470703125, 0.2895851135253906, -28.839500427246094, 16.262338638305664, -83.72802734375, 98.18284606933594, -35.44200134277344, 4.571250915527344, 38.83796691894531, 14.090576171875, 132.25811767578125, -126.67498779296875, 79.29603576660156, -36.13880920410156, 125.85313415527344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000278.npy"}
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 54.27723693847656, "std": 72.90532684326172, "min": -128.60101318359375, "p10": -21.63639678955078, "median": 39.36711883544922, "p90": 151.30076599121094, "max": 227.0712890625, "pos_frac": 0.7890625, "sample": [43.4864501953125, 119.14688110351562, 40.02421569824219, -21.372825622558594, 17.626693725585938, -12.37994384765625, 87.04521179199219, 64.0079345703125, 6.1217498779296875, 132.60971069335938, 90.15850830078125, 167.76095581054688, 36.634521484375, -60.012786865234375, -3.67828369140625, 35.527008056640625, 148.72726440429688, 112.6158218383789, 37.5120849609375, 118.11398315429688, 18.259872436523438, 25.312557220458984, -6.1904296875, 35.95855712890625, 45.60546875, 28.49566650390625, 21.012229919433594, 7.931915283203125, -122.03511047363281, 7.469779968261719, 100.4971923828125, 170.04791259765625, 131.1929931640625, 141.510009765625, 70.08845520019531, -47.576171875, 6.380409240722656, 109.81546020507812, -27.769485473632812, 11.827987670898438, 113.95365142822266, 132.38319396972656, 126.30201721191406, 155.15196228027344, 12.464752197265625, 151.75958251953125, -19.917617797851562, 11.470367431640625, -49.61534118652344, 90.15638732910156, 22.934417724609375, 30.6241455078125, 10.950698852539062, 40.1127815246582, 49.275115966796875, -6.7143402099609375, 227.0712890625, -128.60101318359375, 119.14483642578125, 169.57455444335938, -14.410697937011719, -27.82220458984375, 16.638153076171875, 23.7490234375, -20.931900024414062, -10.376113891601562, -23.341201782226562, 131.39212036132812, -1.479034423828125, 192.40576171875, -45.909820556640625, 196.77197265625, 75.81947326660156, 97.28721618652344, 10.071239471435547, -121.66090393066406, 10.827392578125, 137.6224365234375, 175.71829223632812, -110.16476440429688, 6.90106201171875, 23.108749389648438, 73.93258666992188, 92.348388671875, 122.903564453125, 98.09536743164062, 51.01141357421875, 11.12009048461914, 145.04649353027344, 19.459442138671875, 94.41438293457031, -20.439712524414062, 116.56571960449219, 49.83905029296875, 22.28087615966797, 17.924942016601562, 146.89154052734375, 2.1040802001953125, 49.95013427734375, 135.44578552246094, 89.25836181640625, 151.22042846679688, -11.527145385742188, 135.60385131835938, 165.067626953125, 146.88247680664062, 16.5577392578125, 9.84478759765625, 14.76593017578125, 151.48822021484375, -4.963165283203125, 38.62969970703125, 47.8223876953125, 178.2318115234375, 119.57423400878906, 54.3157958984375, 18.1435546875, 42.8194580078125, -91.21064758300781, -22.25139617919922, 163.30941772460938, -7.845672607421875, 91.7479248046875, 2.572601318359375, 134.28375244140625, 90.400390625, 38.71002197265625, 86.93234252929688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000279.npy"}
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 47.71937561035156, "std": 76.69596099853516, "min": -160.77938842773438, "p10": -35.054876708984374, "median": 28.120929718017578, "p90": 150.40852966308594, "max": 188.19207763671875, "pos_frac": 0.75, "sample": [119.22293090820312, 121.68794250488281, 1.3127899169921875, 133.3634033203125, -4.706415176391602, -160.77938842773438, 145.76220703125, 125.30242919921875, 164.1080322265625, 6.5645294189453125, -20.791213989257812, 34.326454162597656, 17.656883239746094, 93.86566162109375, 17.4849853515625, -26.992141723632812, 159.8480224609375, 88.01470947265625, 159.62237548828125, -6.23052978515625, 3.575103759765625, 149.69332885742188, -69.31317138671875, 153.2593536376953, -26.988189697265625, 115.92539978027344, 26.94598388671875, 20.45806884765625, -8.88165283203125, 122.99172973632812, 0.0, 23.156524658203125, 28.447998046875, -36.2950439453125, 49.19451904296875, 188.19207763671875, 2.8720779418945312, 141.7121124267578, 10.446308135986328, -0.15679931640625, -8.214080810546875, 103.45539855957031, 9.409332275390625, 23.159515380859375, 92.69918060302734, 42.02410888671875, 115.66351318359375, -67.04342651367188, 154.0699462890625, 19.38232421875, 11.773345947265625, 9.754341125488281, 7.09063720703125, 5.517459869384766, 113.54634094238281, 154.36459350585938, 70.28289794921875, -26.474929809570312, 4.4742431640625, 125.7679443359375, 147.84832763671875, 29.34564208984375, 60.608768463134766, -93.54528045654297, -8.489173889160156, 17.136444091796875, -148.27792358398438, 143.51080322265625, 7.386756896972656, 100.62358093261719, 132.46261596679688, 13.993949890136719, 152.07733154296875, 155.8834228515625, 154.47265625, -24.015869140625, 105.71270751953125, 92.25070190429688, 25.053680419921875, 51.8565673828125, 114.58518981933594, 149.67697143554688, -65.30337524414062, 119.27154541015625, 141.6702880859375, 1.8351974487304688, 16.615203857421875, 60.369384765625, -42.569122314453125, -0.792022705078125, 147.0576171875, 48.915283203125, 92.33574676513672, -89.93794250488281, 50.32927322387695, 167.64334106445312, -9.540283203125, 107.75186157226562, 147.07803344726562, 29.32696533203125, 138.65843200683594, 8.71343994140625, 126.07330322265625, -21.345382690429688, 18.048171997070312, 23.277740478515625, -0.693267822265625, -5.251708984375, 161.25662231445312, 7.660621643066406, 43.293243408203125, 27.793861389160156, 5.496337890625, 104.85755920410156, -72.11959838867188, 110.12934875488281, 128.74844360351562, 182.2279052734375, -77.03414916992188, -143.80357360839844, -10.53143310546875, 11.149166107177734, -92.89079284667969, 38.96043014526367, 40.22198486328125, 2.4949569702148438, 60.41522216796875, -34.52337646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000280.npy"}
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 49.09894943237305, "std": 71.12525177001953, "min": -128.99609375, "p10": -18.296440124511715, "median": 30.384666442871094, "p90": 148.3703399658203, "max": 258.7254638671875, "pos_frac": 0.7265625, "sample": [115.24786376953125, 59.47135925292969, 1.987823486328125, 4.426261901855469, 78.94564819335938, 9.32568359375, 50.48187255859375, 69.0598373413086, 17.088653564453125, 158.77810668945312, 144.3812255859375, -107.41818237304688, -117.36553955078125, 161.41256713867188, 17.21706771850586, 28.070953369140625, 114.98543548583984, 110.69247436523438, 3.3534393310546875, 8.441734313964844, 153.4157257080078, 11.517539978027344, -8.628890991210938, -6.51495361328125, 128.5103759765625, -11.868209838867188, -47.9730224609375, 154.079833984375, 19.420806884765625, 23.845535278320312, 0.0, 50.38592529296875, -3.181793212890625, 56.39825439453125, 43.63625717163086, 139.6743927001953, 37.754425048828125, -9.330230712890625, 151.5532684326172, 100.96978759765625, -128.99609375, 12.898712158203125, 130.96551513671875, -12.874557495117188, 0.0, 29.56787109375, 35.7498779296875, 128.43185424804688, -26.415283203125, 35.78965759277344, 100.96893310546875, -27.999664306640625, 37.41887283325195, 39.81071472167969, 1.240753173828125, 179.70111083984375, -9.25994873046875, 30.875457763671875, -1.7559051513671875, 71.88851928710938, 14.3956298828125, 51.943267822265625, -19.93572998046875, -1.5538806915283203, 106.4896240234375, 70.35665893554688, 25.58043670654297, -9.322708129882812, -4.53936767578125, 6.978515625, 18.151458740234375, 143.9557647705078, 2.154052734375, 146.15789794921875, 132.93109130859375, 55.77777099609375, 143.75531005859375, 1.97418212890625, 34.019439697265625, 107.43051147460938, 37.58446502685547, 102.04685974121094, 258.7254638671875, 151.5242919921875, 139.21017456054688, -35.102935791015625, -0.9451751708984375, 110.0589599609375, 29.893875122070312, 3.5745468139648438, 181.02890014648438, 32.246307373046875, 108.98158264160156, -3.184162139892578, 20.43408203125, -43.98675537109375, -8.446868896484375, 163.964599609375, 128.38323974609375, 4.1446533203125, -25.88934326171875, 55.96514892578125, 0.0, 159.8115234375, -6.2267303466796875, -20.66180419921875, 106.46737670898438, -113.7696533203125, 169.40365600585938, 156.12286376953125, 124.82250213623047, 147.01864624023438, 131.39431762695312, -2.739501953125, 146.42919921875, 14.86865234375, -17.593887329101562, 73.81484985351562, 7.451297760009766, 23.391448974609375, 26.862884521484375, -65.82870483398438, -6.850578308105469, 140.55746459960938, 14.04339599609375, 72.22358703613281, 31.75128173828125, -5.2662200927734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000281.npy"}
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 43.33411407470703, "std": 81.05497741699219, "min": -167.35653686523438, "p10": -42.213714599609375, "median": 27.875947952270508, "p90": 149.19047241210936, "max": 212.2679443359375, "pos_frac": 0.71875, "sample": [142.5304412841797, -125.93559265136719, 4.58917236328125, 212.2679443359375, 89.75021362304688, 11.324211120605469, -167.35653686523438, 148.27294921875, 104.67364501953125, -4.154716491699219, 11.493476867675781, 93.422119140625, 22.141937255859375, -44.4266357421875, 12.038330078125, 10.092727661132812, 154.90728759765625, -39.23699951171875, 129.80441284179688, -7.356864929199219, -32.38104248046875, 60.162109375, -98.93052673339844, 42.5416259765625, 6.928680419921875, 210.54745483398438, 135.3654022216797, 54.39306640625, -2.50091552734375, 14.98223876953125, -41.5625, 21.996261596679688, 122.6920166015625, 57.63240051269531, -5.036865234375, 56.086151123046875, 104.93640899658203, 32.771575927734375, 45.57562255859375, 62.91693115234375, 49.83189392089844, 99.93988037109375, 8.014617919921875, -37.10499572753906, 128.23379516601562, -31.453109741210938, -122.73684692382812, 39.74165344238281, 156.98818969726562, 6.379083633422852, 25.28729248046875, 10.65948486328125, -26.16094970703125, 11.853370666503906, -36.317626953125, 185.57345581054688, 53.08909606933594, 1.162689208984375, 15.71475601196289, -81.99800109863281, 3.5868377685546875, 4.288818359375, -23.7265625, 132.9447021484375, 139.28445434570312, 108.64431762695312, 104.15602111816406, -10.239532470703125, -5.640594482421875, 138.21902465820312, -110.3489990234375, -6.225982666015625, 133.97314453125, 73.79071044921875, 15.504150390625, 171.4249267578125, 128.35382080078125, -71.54550170898438, 84.1964111328125, -27.451446533203125, -22.428573608398438, 10.073657989501953, -4.2685546875, 171.26795959472656, 61.228240966796875, -64.12953186035156, 19.742721557617188, 16.491928100585938, 171.02490234375, 147.9915771484375, 177.17782592773438, 35.328369140625, 13.480133056640625, 88.41915893554688, -126.87530517578125, 44.70466613769531, 125.63406372070312, 151.33135986328125, 8.307292938232422, -7.41180419921875, 141.01119995117188, 207.94970703125, 121.34896850585938, 4.818553924560547, 120.33815002441406, -1.1136322021484375, -151.48068237304688, 4.066650390625, 36.74285888671875, 30.464603424072266, 138.08499145507812, 30.740707397460938, 56.536712646484375, 107.96399688720703, 7.899814605712891, 141.982177734375, 156.32620239257812, 20.990936279296875, 151.60365295410156, 43.410125732421875, 63.998199462890625, -120.29212951660156, -4.894748687744141, -43.73321533203125, 101.70260620117188, -7.3895263671875, -12.066116333007812, 134.8216552734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000282.npy"}
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 48.91754913330078, "std": 77.228271484375, "min": -126.94564819335938, "p10": -61.769219970703126, "median": 35.36249542236328, "p90": 142.74484100341797, "max": 205.41519165039062, "pos_frac": 0.75, "sample": [-36.663482666015625, 25.017684936523438, 146.10675048828125, -26.057281494140625, 12.761423110961914, -55.041656494140625, 135.1944580078125, -84.99899291992188, 51.00457763671875, 9.480377197265625, -8.756927490234375, 135.96234130859375, 99.64079284667969, 96.81786346435547, -1.9224853515625, -75.47195434570312, -20.812713623046875, 112.72151947021484, 1.4590301513671875, -37.501312255859375, 201.36328125, 160.511962890625, 33.7952880859375, -71.94635009765625, 145.234619140625, 125.26388549804688, 109.07479858398438, -2.335693359375, 83.20350646972656, 16.529830932617188, 87.59288024902344, 71.99215698242188, 22.705352783203125, 128.6464080810547, 52.56414031982422, 109.17023468017578, 10.984115600585938, 0.8473148345947266, 74.55474853515625, 100.90277099609375, -28.0299072265625, 121.30400085449219, -38.409210205078125, -88.87710571289062, 115.78324890136719, 117.974609375, 35.256134033203125, 15.8597412109375, -102.81576538085938, 190.99472045898438, 108.22586059570312, 138.84591674804688, 120.70466613769531, 89.35820770263672, 162.71434020996094, 6.047187805175781, 1.698465347290039, 30.067306518554688, 32.31903076171875, 103.08305358886719, 96.99945068359375, 128.4716796875, 148.93011474609375, 153.86224365234375, 141.66506958007812, -6.49609375, 138.1717529296875, -0.76263427734375, 44.36320495605469, 39.86869812011719, -77.74756622314453, 111.20976257324219, -101.58328247070312, 158.78927612304688, 0.62921142578125, 67.79278564453125, 142.3171844482422, 114.17110443115234, 0.0, 6.808197021484375, -69.919189453125, 19.460693359375, 110.36491394042969, -116.46412658691406, 134.49755859375, 111.4886474609375, 87.941650390625, 29.742202758789062, -61.33831787109375, 33.049102783203125, 1.5804443359375, 35.46885681152344, 12.20098876953125, -17.19512939453125, 45.221832275390625, 127.70550537109375, 121.72158813476562, 65.26593017578125, -126.94564819335938, 120.5126953125, 0.31795692443847656, 29.0333251953125, 205.41519165039062, -6.38232421875, 30.65240478515625, 23.034500122070312, 120.42391967773438, -90.42739868164062, 19.449432373046875, 2.7672882080078125, 158.54705810546875, 19.82244873046875, -0.3181419372558594, 2.683755874633789, 0.0, 107.880126953125, 127.33661651611328, 202.59854125976562, 13.669685363769531, 5.079500198364258, 65.61585998535156, 119.1880874633789, 68.75376892089844, 41.53654479980469, -62.774658203125, 143.74270629882812, -99.77569580078125, -29.944419860839844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000283.npy"}
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 47.6872673034668, "std": 87.41604614257812, "min": -158.17633056640625, "p10": -72.46289062499999, "median": 38.39372253417969, "p90": 156.95265197753906, "max": 271.84320068359375, "pos_frac": 0.75, "sample": [-70.05780029296875, 97.37091064453125, -99.14659881591797, -158.17633056640625, 124.40113830566406, 36.844635009765625, 151.88302612304688, 148.90313720703125, -91.72883605957031, 39.51246643066406, 56.68248748779297, 156.8690185546875, 96.12993621826172, 43.46897888183594, 27.523345947265625, 111.57066345214844, 139.93130493164062, 72.313232421875, 162.09515380859375, 7.502174377441406, 134.14549255371094, 234.5439453125, -19.347259521484375, 147.21929931640625, -7.34528923034668, 14.670120239257812, 11.96710205078125, -94.72238159179688, 15.4896240234375, -44.18743896484375, -82.840576171875, 41.040435791015625, 69.56515502929688, -23.54314422607422, 6.4322662353515625, 38.71697998046875, 89.80902099609375, 100.88168334960938, 74.89735412597656, 10.76580810546875, -125.88214874267578, 12.115264892578125, 76.97508239746094, 37.05615234375, -9.630531311035156, -16.635406494140625, -46.405181884765625, 47.2191162109375, 12.82769775390625, 195.81155395507812, 144.6681671142578, -78.07476806640625, 48.22564697265625, 20.89306640625, 174.40093994140625, 98.98568725585938, 12.86676025390625, -42.905372619628906, 154.7135009765625, 71.60324096679688, 120.90481567382812, 167.41246032714844, 33.23594665527344, -7.158119201660156, 144.3025360107422, 53.96795654296875, 180.39251708984375, 29.452392578125, 98.08396911621094, 11.223176956176758, 114.24258422851562, 28.290939331054688, -22.282594680786133, 90.0780029296875, 0.7540664672851562, 271.84320068359375, 10.664031982421875, 19.755477905273438, 12.468887329101562, 1.4312591552734375, 178.2920684814453, 140.22860717773438, 113.99967956542969, 158.85028076171875, -66.97286987304688, -140.04437255859375, 4.662624359130859, -20.865966796875, 55.722991943359375, -60.462547302246094, 115.4207763671875, 38.070465087890625, 108.67684936523438, 126.57365417480469, 0.9409904479980469, -2.5060577392578125, 40.16363525390625, -122.84239196777344, 1.9287109375, 121.61807250976562, 157.14779663085938, 37.72088623046875, 115.3558349609375, 11.09869384765625, 0.6324634552001953, 141.46975708007812, -18.517425537109375, 25.788177490234375, 24.70391845703125, 85.47392272949219, 140.41342163085938, 203.093505859375, 107.6905517578125, -22.507904052734375, -145.74563598632812, -114.32867431640625, 108.37879180908203, -94.12692260742188, 1.2515945434570312, -32.13720703125, 183.35028076171875, 57.639434814453125, 113.73138427734375, 201.41653442382812, 119.99810791015625, -2.5019187927246094, -120.189453125, 106.27224731445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000284.npy"}
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 36.329620361328125, "std": 75.710693359375, "min": -142.61099243164062, "p10": -46.871603393554686, "median": 26.22628402709961, "p90": 137.59915466308593, "max": 184.89849853515625, "pos_frac": 0.703125, "sample": [15.539215087890625, 107.44479370117188, 111.94659423828125, 30.32415771484375, 133.1785888671875, 138.33624267578125, 122.42913818359375, -122.90664672851562, 132.36587524414062, 23.63275146484375, 33.212799072265625, -17.53656005859375, 9.702812194824219, -100.1422119140625, -19.30731201171875, 33.49290466308594, 16.822036743164062, -38.27215576171875, -16.64093017578125, -13.894500732421875, -7.27142333984375, 127.44134521484375, -107.28760528564453, 16.259674072265625, 28.81981658935547, 62.20574951171875, 169.7171630859375, 45.64520263671875, 60.76019287109375, 37.666473388671875, 184.89849853515625, -98.27203369140625, 9.371170043945312, 37.445587158203125, 139.127197265625, 137.84165954589844, 127.506103515625, -40.023712158203125, -29.124374389648438, 16.95550537109375, -46.757904052734375, -31.25726318359375, -4.8811798095703125, -3.668243408203125, 4.125566482543945, 112.27496337890625, 69.89593505859375, 131.72842407226562, -0.19370269775390625, 10.26385498046875, 137.49522399902344, 1.3208560943603516, 9.906929016113281, -88.41962432861328, 156.84921264648438, 1.9580402374267578, -8.524009704589844, 0.8791179656982422, 126.62513732910156, 16.668922424316406, 70.6998291015625, 98.83489990234375, 135.18228149414062, 40.03509521484375, 48.77496337890625, 17.334686279296875, -115.39016723632812, 11.687431335449219, -114.56953430175781, 44.99566650390625, -140.59161376953125, 143.33038330078125, 33.55769348144531, -0.9869537353515625, 105.20524597167969, -20.317581176757812, 57.165252685546875, 20.796459197998047, 33.68798828125, 16.427093505859375, 106.48121643066406, 121.5096435546875, 10.459625244140625, 3.5495758056640625, -13.869430541992188, 5.3824462890625, 37.179901123046875, 133.3284912109375, 156.7266845703125, 115.49688720703125, 112.26962280273438, 128.77024841308594, -3.3692398071289062, -30.274078369140625, -25.108489990234375, 43.69598388671875, 39.9442138671875, 22.09869384765625, -84.00953674316406, 19.7388916015625, 100.53428649902344, -48.805511474609375, 109.90803527832031, -37.566925048828125, -42.888671875, -29.492691040039062, 18.22747802734375, -15.280628204345703, 149.88241577148438, 61.36529541015625, -142.61099243164062, 31.747398376464844, 30.40496826171875, 2.460784912109375, -11.85516357421875, -131.53289794921875, 132.09609985351562, 4.697944641113281, 127.88412475585938, 64.50665283203125, 39.39813232421875, 79.46670532226562, 106.35757446289062, 141.79415893554688, -47.13690185546875, 153.49771118164062, 167.15142822265625, 154.42037963867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000285.npy"}
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 37.94257354736328, "std": 79.4830093383789, "min": -143.29766845703125, "p10": -56.85056495666504, "median": 29.177440643310547, "p90": 141.11922302246091, "max": 199.67718505859375, "pos_frac": 0.671875, "sample": [-8.20953369140625, -24.16534423828125, 91.10763549804688, 17.723587036132812, 12.056800842285156, 28.837356567382812, 11.23333740234375, 9.553253173828125, 9.09588623046875, 93.83364868164062, 64.40524291992188, 18.7706298828125, -2.3115692138671875, 65.09584045410156, -122.04934692382812, 159.78407287597656, 11.518783569335938, 159.62457275390625, 116.80167388916016, 152.3985595703125, -32.634368896484375, 149.66213989257812, 39.965576171875, 28.775222778320312, 19.19652557373047, 26.588531494140625, -19.002273559570312, -6.826416015625, 98.43954467773438, -20.173973083496094, -143.29766845703125, 14.232284545898438, 135.17098999023438, 130.0247039794922, -28.85894775390625, 23.84531593322754, -9.36407470703125, 50.0517578125, 9.651962280273438, 103.90232849121094, -7.5545654296875, 138.4798583984375, 118.57943725585938, -99.45943450927734, 91.3603515625, -45.51312255859375, 40.5654296875, 199.67718505859375, -12.7818603515625, -14.158843994140625, 137.3577880859375, -138.90261840820312, 137.87722778320312, 102.45611572265625, 38.40045166015625, 121.908935546875, 4.98223876953125, 101.864990234375, -26.816551208496094, 162.9189453125, -11.333732604980469, 0.0, -4.787841796875, -124.41812133789062, 46.28434371948242, 55.07948303222656, -5.652507781982422, -29.948959350585938, -47.1685791015625, 71.6659927368164, -56.2523193359375, 123.4930419921875, 123.52294921875, 41.300811767578125, -58.2464714050293, 34.942108154296875, 134.41281127929688, 76.25563049316406, -101.95829772949219, 8.689422607421875, 57.75201416015625, 29.51752471923828, -0.6771011352539062, 199.25970458984375, 134.45913696289062, 36.14222717285156, 30.7122802734375, -93.15277099609375, 114.19833374023438, -111.47918701171875, 77.67906951904297, 159.83987426757812, 34.4461669921875, 132.32977294921875, 147.27774047851562, 41.2181396484375, 23.75628662109375, 105.51730346679688, 104.18891906738281, 101.71681213378906, 168.7327880859375, 35.44354248046875, 149.84909057617188, -16.009868621826172, 78.11346435546875, -48.1497802734375, 162.41921997070312, 35.03569793701172, -37.577117919921875, -17.510318756103516, -102.63873291015625, 138.281494140625, -6.839454650878906, -114.50617218017578, -104.33235931396484, 189.56655883789062, -95.89593505859375, -37.07525634765625, 20.512283325195312, 12.468994140625, 47.87800598144531, 125.24545288085938, 0.39972877502441406, 15.888198852539062, 69.4251708984375, -20.73248291015625, 116.71492004394531, 5.66412353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000286.npy"}
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 44.66077423095703, "std": 76.29500579833984, "min": -172.6531982421875, "p10": -42.83561706542967, "median": 42.61367416381836, "p90": 135.79679412841796, "max": 174.53994750976562, "pos_frac": 0.71875, "sample": [121.64874267578125, -30.700828552246094, -58.628662109375, -11.260299682617188, -25.110031127929688, -35.58360290527344, 135.41204833984375, 130.66650390625, 101.1880111694336, 166.58062744140625, 121.08592224121094, 18.915130615234375, 7.947492599487305, 130.48812866210938, 6.84649658203125, 17.3521728515625, -29.099166870117188, -80.100830078125, -23.68665313720703, 13.706413269042969, -24.9134521484375, 174.53994750976562, 129.51873779296875, -11.21337890625, 102.20428466796875, -24.137969970703125, -13.434162139892578, 53.884185791015625, 58.29937744140625, 42.733856201171875, 8.006744384765625, 143.06768798828125, 90.61843872070312, 82.12081909179688, -7.268035888671875, 4.321502685546875, 96.98399353027344, 120.0667724609375, -1.3516826629638672, 122.41505432128906, 129.1487274169922, 123.486083984375, 19.113739013671875, 104.30755615234375, 142.96434020996094, 7.295991897583008, -1.5740814208984375, 132.05459594726562, 61.20355224609375, 163.98641967773438, 41.78825378417969, 135.38894653320312, 129.02215576171875, 56.8187255859375, 26.058746337890625, 46.26336669921875, 35.38202667236328, -22.185455322265625, 116.43109130859375, -66.783447265625, 121.54341125488281, 107.52825927734375, 47.116424560546875, 25.534862518310547, 3.8867034912109375, 30.079925537109375, 102.89981079101562, 130.00396728515625, 113.75454711914062, 58.579833984375, -134.7818603515625, 1.448751449584961, 149.89044189453125, 147.595703125, 53.031158447265625, 144.6119842529297, 130.214111328125, 105.50239562988281, 18.7188720703125, 104.21624755859375, 171.9666290283203, -85.035400390625, -104.03776550292969, -14.153785705566406, -39.18389892578125, 64.83638000488281, -31.068832397460938, 97.65283203125, -55.92926025390625, 12.388397216796875, 111.70677185058594, 159.5089111328125, -172.6531982421875, -12.496246337890625, 42.493492126464844, -10.67083740234375, 98.61546325683594, 4.547651290893555, -128.15084838867188, 44.31890869140625, 5.68994140625, 16.700927734375, 7.54205322265625, 57.822021484375, -31.121002197265625, -30.69287109375, 120.00489044189453, -134.04965209960938, 159.72378540039062, -15.294792175292969, 127.86300659179688, -56.31575012207031, 115.94903564453125, 61.17784118652344, 150.74497985839844, 8.514434814453125, 17.75151824951172, 76.36978149414062, 59.29351806640625, 9.759212493896484, 33.843360900878906, 9.00497817993164, -113.25845336914062, -51.356292724609375, 66.92849731445312, 136.6945343017578, -19.73614501953125, 106.69255065917969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000287.npy"}
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 42.091400146484375, "std": 78.14470672607422, "min": -140.50128173828125, "p10": -42.13714141845701, "median": 25.72252655029297, "p90": 139.06411437988282, "max": 211.7069091796875, "pos_frac": 0.7265625, "sample": [47.67303466796875, 4.121803283691406, -24.271636962890625, -30.431549072265625, 99.08627319335938, 3.7174072265625, -17.007522583007812, -21.542404174804688, 25.120529174804688, 105.94515991210938, 19.57867431640625, 167.2587890625, -4.0267333984375, 94.27191162109375, 45.69964599609375, -10.829833984375, 8.952171325683594, 49.75874328613281, 27.5198974609375, 0.0, 17.66705322265625, -9.6392822265625, 57.0113525390625, 9.539409637451172, 56.21388244628906, 2.19525146484375, -29.70037841796875, 2.83477783203125, -100.97177124023438, 19.4698486328125, 171.21572875976562, 126.3443603515625, 48.18968200683594, 91.9638671875, 5.723579406738281, 0.0, -9.371429443359375, 40.56463623046875, 13.40338134765625, 18.59783935546875, -6.1638946533203125, 121.38645935058594, -101.26394653320312, -110.81942749023438, 16.699533462524414, 121.8458251953125, 166.9815673828125, 158.93020629882812, -140.50128173828125, 0.3539257049560547, 138.95046997070312, -7.3507843017578125, 138.03131103515625, 0.29693603515625, 78.07012939453125, -129.62652587890625, 4.5185089111328125, 45.37591552734375, 128.00367736816406, -102.79458618164062, 211.7069091796875, -84.24778747558594, -57.867462158203125, 94.78770446777344, 181.71173095703125, 18.27105712890625, -4.6238250732421875, -52.2366943359375, 132.38037109375, 5.949089050292969, 134.36016845703125, 30.060638427734375, -3.247406005859375, 67.16763305664062, 14.514068603515625, 11.666656494140625, 21.39447021484375, 106.71885681152344, -30.469970703125, 66.03936767578125, 101.62434387207031, 37.35807800292969, -5.3560333251953125, -108.19769287109375, -26.23682403564453, -113.515380859375, 15.090709686279297, 159.51654052734375, 137.56484985351562, 25.153610229492188, 43.38922119140625, 3.058135986328125, 111.7821044921875, -7.319086074829102, 98.5684814453125, 121.66481018066406, 102.43767547607422, -70.96290588378906, 144.9388427734375, 26.29144287109375, 46.860382080078125, 88.50798034667969, -108.77395629882812, 206.50839233398438, 8.892250061035156, 35.453369140625, 15.70134162902832, 97.48236083984375, 40.697723388671875, -15.433074951171875, -37.80876159667969, 108.43441772460938, 39.125457763671875, 139.32928466796875, 105.53961181640625, 128.418212890625, 28.14532470703125, 12.2322998046875, 16.920135498046875, 128.1077423095703, 38.10561752319336, 210.88705444335938, 130.15159606933594, 173.93240356445312, 138.93692016601562, -27.636016845703125, 208.8455810546875, 126.47840881347656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000288.npy"}
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 30.22136116027832, "std": 78.81938171386719, "min": -159.34042358398438, "p10": -69.3524848937988, "median": 14.971923828125, "p90": 134.3367889404297, "max": 177.37774658203125, "pos_frac": 0.6484375, "sample": [-128.60177612304688, -22.131195068359375, 15.26953125, -82.7795639038086, -5.163970947265625, 1.536773681640625, 13.776458740234375, -26.008544921875, 62.8895263671875, -96.65350341796875, 130.36322021484375, -5.762908935546875, 1.1495170593261719, 45.308837890625, 1.8845367431640625, 87.1009521484375, -31.450363159179688, 150.77362060546875, 15.44012451171875, -62.87031555175781, 172.743408203125, 151.28285217285156, 82.86553955078125, 11.942222595214844, -5.53582763671875, -22.121185302734375, -11.502838134765625, 1.76947021484375, 10.03997802734375, 89.29847717285156, -8.186416625976562, 123.3643798828125, -2.052276611328125, 25.01251220703125, 1.4129486083984375, 0.0, 8.829055786132812, 95.02627563476562, -29.073516845703125, 15.982666015625, 11.212776184082031, -31.0362548828125, -123.24017333984375, -131.72515869140625, 79.71514892578125, -31.065032958984375, 56.30712890625, 119.9140625, -0.9366455078125, -3.463531494140625, -9.546432495117188, 130.42379760742188, -139.63897705078125, -134.59048461914062, 14.67431640625, 23.388870239257812, -7.271341323852539, -5.7215728759765625, 26.250097274780273, 168.15829467773438, -100.19189453125, 127.51324462890625, 97.56129455566406, 87.45896911621094, 42.6231689453125, 119.68033599853516, -87.35618591308594, -117.25697326660156, -59.88996124267578, 16.241775512695312, 136.82785034179688, -26.76885986328125, -0.285125732421875, 116.28480529785156, 17.548133850097656, 20.540771484375, 8.254875183105469, 45.24298095703125, 116.09042358398438, 6.51019287109375, 73.90982055664062, -21.24786376953125, -159.34042358398438, 144.54434204101562, -14.727508544921875, 110.92276763916016, -10.7728271484375, 134.08604431152344, -43.77687072753906, 141.15185546875, 39.1739501953125, 162.6378173828125, 126.35136413574219, 10.74127197265625, 36.10163879394531, 116.58273315429688, 92.75537109375, 7.968414306640625, 99.8451156616211, 94.05415344238281, 139.44403076171875, 54.94532012939453, 163.87005615234375, 108.94279479980469, -21.622894287109375, 126.41873168945312, 15.296905517578125, 6.14801025390625, 102.34796142578125, 177.37774658203125, -121.08642578125, 125.29085540771484, 133.52630615234375, -36.95892333984375, 134.92185974121094, -63.5980224609375, -123.37246704101562, 63.771461486816406, -5.953765869140625, 133.49850463867188, 27.714263916015625, 27.897918701171875, 157.5560302734375, 1.6763458251953125, 14.18731689453125, -6.138069152832031, 8.579620361328125, 29.08416748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000289.npy"}
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 38.025882720947266, "std": 74.1874008178711, "min": -152.79367065429688, "p10": -36.359216308593744, "median": 21.001976013183594, "p90": 141.69223022460938, "max": 204.5487060546875, "pos_frac": 0.671875, "sample": [148.0723419189453, 15.00457763671875, -7.4486083984375, 119.26637268066406, 54.2191162109375, -152.79367065429688, 2.1405029296875, 120.39459228515625, -15.1309814453125, 24.60479736328125, 169.80096435546875, 136.01422119140625, 48.94488525390625, -14.103462219238281, 128.524658203125, 119.13529968261719, 38.52708435058594, 33.846527099609375, 142.651611328125, 123.57723999023438, 129.95819091796875, -6.931297302246094, 3.442962646484375, -1.3032073974609375, 185.8057861328125, 12.935577392578125, -95.05914306640625, -122.97210693359375, -22.29681396484375, 1.39080810546875, 109.1318359375, 1.5719337463378906, 2.54522705078125, 52.361328125, 150.59243774414062, 159.38238525390625, 146.57569885253906, 30.87428855895996, -2.67626953125, -68.15774536132812, 112.14375305175781, 46.38273620605469, -1.99700927734375, 129.6761474609375, 47.547576904296875, -0.1023101806640625, 0.44049072265625, -27.636810302734375, 48.49530029296875, 56.42254638671875, -16.009246826171875, -7.38629150390625, 23.946990966796875, 95.78890228271484, 17.5887451171875, -30.223358154296875, 48.60459899902344, 112.8056640625, 13.058135986328125, 19.793853759765625, 12.011688232421875, 13.896160125732422, -12.769691467285156, 114.70343017578125, 28.17108154296875, 25.983551025390625, 16.36102294921875, 76.2730712890625, 21.069625854492188, 106.52374267578125, -32.470611572265625, 25.862167358398438, 3.44854736328125, -6.873863220214844, -21.888107299804688, 165.0152587890625, -6.825277328491211, -48.1279296875, -3.6395263671875, 100.28509521484375, -2.404998779296875, -13.258636474609375, -35.512847900390625, 134.73446655273438, 120.48458862304688, 131.35708618164062, 131.0723876953125, -15.68450927734375, -144.782958984375, 7.2537384033203125, 70.12179565429688, -41.31288146972656, 52.158050537109375, 6.24810791015625, -91.84652709960938, 5.2045745849609375, -14.638458251953125, -13.106109619140625, -92.51704406738281, 112.6201171875, 7.405727386474609, -97.91472625732422, 42.757232666015625, 204.5487060546875, 158.77685546875, 49.253326416015625, 135.60394287109375, 141.28106689453125, 143.0777587890625, 9.880592346191406, -98.95692443847656, 20.934326171875, 113.4559326171875, 47.8282470703125, 53.33148193359375, -81.99563598632812, -13.400917053222656, 7.599151611328125, -38.334075927734375, 119.27265167236328, -15.3411865234375, -1.1338882446289062, 26.19500732421875, 118.70207214355469, -30.048324584960938, 78.23233032226562, 148.383544921875, 142.95916748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000290.npy"}
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 57.1813850402832, "std": 78.76441192626953, "min": -175.14529418945312, "p10": -39.26625671386718, "median": 48.94026184082031, "p90": 155.5992462158203, "max": 201.55816650390625, "pos_frac": 0.7421875, "sample": [158.90728759765625, 0.24465179443359375, 184.4649658203125, 13.965850830078125, 178.45138549804688, 51.840118408203125, 153.64620971679688, 131.6807098388672, 75.40617370605469, -91.7432861328125, -5.282318115234375, 110.22744750976562, 29.304122924804688, -6.98529052734375, 154.593017578125, 12.171051025390625, 169.02093505859375, -14.8751220703125, 119.49557495117188, 32.63658905029297, 106.72640991210938, 87.69927978515625, 127.43106079101562, 23.924560546875, -11.894668579101562, 138.01144409179688, 44.30720520019531, 8.060653686523438, 108.750732421875, -27.716644287109375, -139.3813934326172, -11.16415023803711, -49.66508483886719, 10.13302230834961, -175.14529418945312, 27.98992919921875, 114.3931884765625, 98.07293701171875, 25.590576171875, 6.38641357421875, 152.31463623046875, -23.5174560546875, -100.07381439208984, 82.73051452636719, 148.61024475097656, 36.4332275390625, 13.176773071289062, 100.98738098144531, 75.86322021484375, 46.0404052734375, 161.32208251953125, 117.82459259033203, 101.68914794921875, -19.8707275390625, 201.55816650390625, 116.86555480957031, 86.96063232421875, -37.179412841796875, -52.47557067871094, 122.68927001953125, 159.7846221923828, 11.186431884765625, -10.553375244140625, -6.08453369140625, 139.48681640625, -60.744964599609375, 5.7037200927734375, 90.50021362304688, -72.6180419921875, 151.55105590820312, -44.13555908203125, 16.4539794921875, 76.03013610839844, 191.8988037109375, 22.780487060546875, 33.798583984375, 113.43065643310547, 129.3101806640625, 148.68463134765625, -7.35260009765625, 82.06890869140625, 0.8543796539306641, -8.61590576171875, -18.888671875, 27.073165893554688, 159.825439453125, 91.90190124511719, -68.59628295898438, 0.0, -9.795883178710938, -50.99266052246094, 133.6976318359375, 20.928436279296875, -78.15670013427734, 189.4041748046875, 119.79829406738281, 136.18099975585938, 41.67510986328125, 157.94711303710938, -31.12750244140625, 74.12484741210938, 137.6140594482422, 31.94000244140625, 131.88082885742188, 1.319061279296875, 68.2275390625, 17.61578369140625, 36.04217529296875, 184.19952392578125, 60.53778076171875, -87.83831787109375, 133.21054077148438, -7.847206115722656, 86.2260513305664, 117.73077392578125, 72.98443603515625, 127.6949462890625, -1.0817375183105469, 164.1666259765625, 140.48284912109375, 86.987548828125, 21.577346801757812, 4.55426025390625, -10.790245056152344, 146.44451904296875, 147.56353759765625, 117.02659606933594, 30.700576782226562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000291.npy"}
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 43.44450759887695, "std": 75.72868347167969, "min": -192.77438354492188, "p10": -35.98470916748047, "median": 28.115156173706055, "p90": 144.0520462036133, "max": 218.73486328125, "pos_frac": 0.75, "sample": [11.977783203125, 2.99505615234375, 116.24211883544922, -63.59674072265625, 84.10968017578125, 118.18329620361328, 133.9273681640625, 195.96295166015625, 21.692977905273438, 149.08554077148438, 203.86529541015625, 104.82241821289062, -29.66619873046875, -9.647781372070312, 56.867950439453125, 27.32110595703125, -35.43116760253906, -10.311918258666992, 17.778564453125, -4.177978515625, 30.110275268554688, 44.2369384765625, 38.34210205078125, 68.630859375, 134.8011474609375, 53.97247314453125, 143.8749237060547, 27.816619873046875, 26.16179656982422, -5.771171569824219, 139.05361938476562, 113.71224975585938, 1.7825431823730469, 40.749298095703125, 122.65960693359375, 25.109004974365234, 26.0849609375, -69.10873413085938, -192.77438354492188, 10.248092651367188, 72.59208679199219, 11.629417419433594, 52.58404541015625, 49.9254150390625, -5.9996337890625, 7.753517150878906, 32.269126892089844, -100.375, 2.2783126831054688, 149.29010009765625, -20.09625244140625, 159.01513671875, -73.57737731933594, -12.600555419921875, -4.08551025390625, 19.632537841796875, -0.2789306640625, -65.3944091796875, 21.71624755859375, 15.657562255859375, 117.35044860839844, 13.84320068359375, 129.95042419433594, -46.12812805175781, -12.241912841796875, 14.446037292480469, 14.497695922851562, 81.29925537109375, 218.73486328125, 57.148681640625, 5.899658203125, 58.875030517578125, 63.456024169921875, 36.98388671875, 115.1446762084961, 175.9410400390625, 8.620899200439453, 146.3787078857422, 28.077503204345703, -15.083513259887695, -58.251861572265625, -187.9390869140625, 175.550048828125, 81.20704650878906, 137.21023559570312, 95.37509155273438, 126.18695068359375, 65.62155151367188, 7.512365341186523, 144.46533203125, -37.27630615234375, 28.224395751953125, 7.244274139404297, 17.460693359375, 88.81248474121094, -22.841094970703125, -10.40969467163086, -25.70623779296875, 17.79281234741211, 153.425537109375, 99.68115234375, 28.152809143066406, 11.981613159179688, 10.07219123840332, 32.811744689941406, 80.60693359375, 103.19580078125, -1.1785888671875, 4.7230377197265625, 5.450714111328125, -20.93817138671875, 54.784820556640625, 43.13981628417969, -0.085723876953125, 132.70599365234375, 126.40167236328125, 184.6756591796875, -113.66558837890625, 94.1070556640625, 83.10624694824219, 151.91143798828125, -140.6162109375, -59.163124084472656, 116.07647705078125, 63.50624084472656, 133.36805725097656, 111.096435546875, 24.50091552734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000292.npy"}
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 46.78425979614258, "std": 77.50768280029297, "min": -149.54437255859375, "p10": -27.82194671630859, "median": 25.98883056640625, "p90": 153.0638275146484, "max": 203.40106201171875, "pos_frac": 0.7421875, "sample": [147.585693359375, -10.521286010742188, 132.88075256347656, 163.77499389648438, 25.133758544921875, -64.7755126953125, 16.614349365234375, 109.9486083984375, 3.1996612548828125, 18.52661895751953, -5.3519134521484375, 139.8819580078125, 40.73565673828125, 21.902923583984375, 19.640199661254883, 14.212860107421875, -78.49469757080078, -0.55914306640625, -80.10539245605469, 8.590179443359375, 0.7666854858398438, 1.610107421875, 160.39874267578125, 123.62652587890625, 18.438873291015625, 0.0, 160.2197265625, 40.46763610839844, 158.4476318359375, 16.238853454589844, 123.02468872070312, -29.98944091796875, -12.240554809570312, 6.459686279296875, 15.12506103515625, -24.154525756835938, 8.781005859375, 117.28291320800781, -1.925079345703125, 65.6849365234375, 146.69314575195312, -1.929779052734375, 156.8653564453125, 139.21829223632812, 26.463226318359375, -149.54437255859375, 151.43460083007812, 143.09144592285156, 185.0185546875, 26.82663345336914, 69.20509338378906, 42.42315673828125, 123.81201171875, 101.73333740234375, -0.29736328125, 129.83009338378906, 25.592193603515625, -10.33209228515625, 4.9170379638671875, 20.67291259765625, -2.2553558349609375, 150.93064880371094, 52.485870361328125, -13.976926803588867, -131.50942993164062, 18.096519470214844, 14.562103271484375, -44.47509765625, 112.13644409179688, -22.600013732910156, -141.70681762695312, -35.714385986328125, 110.51789855957031, 203.40106201171875, -125.09344482421875, 11.0989990234375, -9.6600341796875, 121.442138671875, -16.392120361328125, 41.4737548828125, 21.46856689453125, 19.1566162109375, 47.156982421875, 16.59284210205078, 150.48602294921875, 31.408935546875, 48.881317138671875, 134.08358764648438, 119.15505981445312, -26.893020629882812, -15.8204345703125, 162.795166015625, 9.463409423828125, 128.66368103027344, 48.241119384765625, 92.2685546875, 85.33041381835938, -107.19711303710938, 42.74310302734375, 11.306221008300781, 95.89787292480469, 180.50924682617188, 197.76791381835938, 178.36383056640625, 68.99029541015625, 129.8036346435547, 23.250526428222656, 29.6324462890625, -47.49037170410156, 0.014942169189453125, -2.5707244873046875, 20.531158447265625, -1.13623046875, 78.430908203125, 200.74850463867188, 193.63809204101562, 13.9840087890625, 31.305282592773438, 120.12495422363281, -110.77090454101562, 110.12701416015625, -9.916084289550781, 9.010574340820312, 104.55442810058594, 33.675384521484375, 58.281585693359375, 40.411468505859375, 26.385467529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000293.npy"}
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 41.42200469970703, "std": 77.4525146484375, "min": -135.31690979003906, "p10": -39.51404113769531, "median": 20.992488861083984, "p90": 145.57760620117188, "max": 212.6673583984375, "pos_frac": 0.734375, "sample": [143.782958984375, 117.860595703125, 131.25009155273438, 0.7423248291015625, -10.652801513671875, 7.549674987792969, 0.6710338592529297, 2.793060302734375, 17.04120635986328, 141.91000366210938, 47.6602783203125, 159.2132568359375, 132.64175415039062, -8.359420776367188, -3.87298583984375, 7.841545104980469, 134.9463653564453, 170.96189880371094, 144.8365478515625, 25.05267333984375, 110.119873046875, 132.71533203125, 101.09623718261719, -78.12816619873047, 9.658149719238281, 145.4281005859375, 11.89520263671875, -27.284378051757812, 3.4229736328125, 13.199028015136719, 169.69692993164062, 25.97265625, 12.604568481445312, 99.42266845703125, -12.839218139648438, 19.201751708984375, -31.869415283203125, -125.44512939453125, 52.19993591308594, 23.725128173828125, 81.28163146972656, 164.2982177734375, -40.3441162109375, -116.0284194946289, 149.30880737304688, 111.4845962524414, 109.8118667602539, -3.4431838989257812, -4.730583190917969, 38.74357604980469, 145.92645263671875, -4.503795623779297, 156.26626586914062, 140.23855590820312, 26.061573028564453, 149.0273895263672, 148.86000061035156, -25.818267822265625, 20.802459716796875, 14.131156921386719, 106.882568359375, 26.5533447265625, 29.0657958984375, 130.0963897705078, -126.19009399414062, 115.31029510498047, 110.17832946777344, 50.551177978515625, 4.70233154296875, 7.95440673828125, 14.706756591796875, 166.77513122558594, 94.8011474609375, 0.0, 140.92193603515625, 212.6673583984375, -93.67268371582031, -9.768234252929688, -12.862762451171875, 53.61126708984375, -4.952644348144531, 147.49374389648438, 6.915225982666016, -132.05528259277344, -15.29278564453125, 61.58721923828125, -39.158294677734375, 4.740684509277344, 113.61662292480469, -45.50596618652344, 25.076934814453125, 1.3476486206054688, 21.182518005371094, -135.31690979003906, 17.319190979003906, 101.64485168457031, 11.854949951171875, 2.1003189086914062, 36.555328369140625, -15.780792236328125, 13.024253845214844, 40.629547119140625, 13.717376708984375, 14.291643142700195, 60.36430358886719, 106.73031616210938, -71.00079345703125, -130.72674560546875, 35.661376953125, -0.1240386962890625, -37.312225341796875, 126.13015747070312, 104.37237548828125, 20.6429443359375, 96.47874450683594, 92.09930419921875, 4.57763671875, -27.261985778808594, 101.54071044921875, -14.8267822265625, 13.3502197265625, -75.38375854492188, 75.62288665771484, 2.309417724609375, 151.63229370117188, -135.0294189453125, 114.57632446289062, 140.23703002929688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000294.npy"}
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 46.14934539794922, "std": 81.39994812011719, "min": -136.35195922851562, "p10": -41.799149322509756, "median": 25.321937561035156, "p90": 147.51366119384764, "max": 230.86962890625, "pos_frac": 0.7265625, "sample": [164.08474731445312, 17.817413330078125, -33.60816955566406, -117.62832641601562, -8.812019348144531, 74.47601318359375, -47.532615661621094, 139.05551147460938, -119.71694946289062, -14.689193725585938, -15.25225830078125, 26.26610565185547, 142.9070281982422, 20.2130126953125, -13.53717041015625, -19.1492919921875, -125.20697021484375, 116.05728149414062, -114.48641967773438, 28.982555389404297, 94.25205993652344, 185.27117919921875, -11.81378173828125, -136.35195922851562, -24.654251098632812, 114.66655731201172, 28.871566772460938, -83.94221496582031, 112.79104614257812, -2.3049774169921875, 5.5421142578125, 24.730056762695312, 54.7547607421875, 119.46231079101562, 136.27566528320312, 9.028228759765625, 96.72442626953125, 10.131011962890625, 65.637451171875, 95.42361450195312, 189.3118438720703, -10.572101593017578, 62.9169921875, 104.93923950195312, -2.498016357421875, 91.93011474609375, 4.785301208496094, 174.46954345703125, 141.0587158203125, 123.24493408203125, -125.3360595703125, -92.47209167480469, 139.35035705566406, 230.86962890625, 138.92250061035156, 192.4700927734375, -21.467056274414062, 104.84735107421875, 55.73065185546875, 13.515369415283203, -14.521392822265625, 2.3599853515625, 61.37541198730469, 2.4417495727539062, 121.4544677734375, 81.33513641357422, 19.26691436767578, 74.28732299804688, 117.38075256347656, 20.710189819335938, 6.1742095947265625, 8.5068359375, 1.4671955108642578, 145.8759765625, -30.121307373046875, 115.99992370605469, 144.00299072265625, 126.25099182128906, 15.243942260742188, 100.48822021484375, 9.864028930664062, 36.73500061035156, 46.37665557861328, 18.784812927246094, 6.988677978515625, 25.913818359375, 0.0, 54.73300552368164, 11.011093139648438, 16.460235595703125, 146.80355834960938, 175.183837890625, 58.46250915527344, 46.4427490234375, 147.19793701171875, 0.3652992248535156, -129.69033813476562, 163.7701416015625, 21.546905517578125, 146.36940002441406, 96.73410034179688, -29.17157745361328, 9.008930206298828, 24.292724609375, 28.67364501953125, -111.76416015625, 134.85763549804688, 15.083999633789062, -1.527618408203125, 102.3040771484375, -26.210567474365234, 138.72813415527344, 198.37109375, 149.447021484375, -0.079132080078125, 173.12948608398438, 1.3990974426269531, -11.101020812988281, 22.794960021972656, -48.193115234375, -39.341949462890625, 6.5162353515625, 148.25035095214844, -13.02935791015625, 123.61886596679688, 158.71681213378906, 78.82962036132812, -57.2730712890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000295.npy"}
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 61.36614227294922, "std": 81.29344940185547, "min": -154.87823486328125, "p10": -34.73428344726561, "median": 52.60082244873047, "p90": 159.02751922607422, "max": 190.4530029296875, "pos_frac": 0.765625, "sample": [117.16841888427734, 11.771522521972656, 22.549072265625, 123.43856811523438, 123.76048278808594, -12.969108581542969, -0.9144134521484375, -1.6685810089111328, 98.97309875488281, 6.840755462646484, 67.49566650390625, -94.353759765625, 47.41560363769531, 39.0484619140625, 7.096954345703125, 132.18927001953125, 141.844970703125, 120.62261962890625, 124.3475570678711, -69.982177734375, 136.20835876464844, 103.07037353515625, -11.3931884765625, 134.67666625976562, 62.788818359375, 156.4530029296875, -10.299945831298828, 16.10354995727539, 149.05422973632812, -47.092926025390625, 15.235122680664062, 10.713119506835938, 182.55450439453125, 6.922210693359375, -133.74349975585938, 171.39163208007812, 50.95086669921875, 127.80621337890625, 145.73736572265625, 41.59137725830078, -30.437850952148438, 23.124359130859375, 170.01788330078125, 135.71185302734375, -11.422355651855469, -31.3897705078125, 36.830596923828125, -79.41586303710938, 186.67828369140625, 14.971145629882812, 166.58041381835938, 88.67398071289062, -20.06243896484375, 153.86944580078125, 162.20123291015625, 186.06399536132812, 12.558349609375, 169.904296875, 66.31344604492188, 129.877685546875, 4.8704833984375, -67.7667465209961, 127.9149169921875, -74.44268798828125, 52.025634765625, 105.46000671386719, 31.98919677734375, 150.06829833984375, -129.72959899902344, 76.588134765625, 139.90206909179688, 53.17601013183594, -42.53814697265625, 133.4580078125, 106.82225036621094, 44.985809326171875, 174.264404296875, 188.28610229492188, -3.021404266357422, 116.47500610351562, 49.75494384765625, 136.92449951171875, 21.080856323242188, -154.87823486328125, 157.2899169921875, 116.6827392578125, 0.46820068359375, -92.41773986816406, 159.01083374023438, -8.116119384765625, 122.73649597167969, 8.6424560546875, 108.73849487304688, -44.12799072265625, 41.68145751953125, 0.0, 115.21017456054688, 131.18862915039062, 130.26556396484375, 82.6986083984375, 19.87493896484375, 142.6502685546875, 77.77992248535156, 134.17218017578125, 11.273513793945312, 43.03924560546875, 15.68548583984375, 11.938262939453125, 45.189422607421875, 132.26333618164062, 173.88327026367188, 98.87948608398438, 4.561733245849609, -13.166046142578125, -8.158119201660156, 137.72210693359375, 135.80032348632812, 2.2639923095703125, -12.45843505859375, 159.0664520263672, -1.5704193115234375, 104.694091796875, 22.50927734375, 157.1744384765625, 101.18515014648438, 190.4530029296875, -138.61773681640625, -8.896438598632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000296.npy"}
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 50.3233642578125, "std": 77.19222259521484, "min": -150.43417358398438, "p10": -25.884660339355467, "median": 30.877685546875, "p90": 153.60957031249998, "max": 186.6107177734375, "pos_frac": 0.65625, "sample": [179.64804077148438, 67.95636749267578, 134.72140502929688, 154.80731201171875, -9.6729736328125, -2.9546279907226562, 138.5069580078125, 52.75624084472656, 153.09625244140625, -25.191574096679688, -2.51409912109375, -1.58209228515625, -30.43567657470703, 5.1634063720703125, 164.84298706054688, -89.57232666015625, -9.105734825134277, 34.277099609375, 134.9868621826172, 102.84468078613281, 11.595237731933594, 34.471099853515625, -1.7775192260742188, 48.3939208984375, -57.116607666015625, -67.33319091796875, -22.66619873046875, -150.43417358398438, 16.66973876953125, -19.460269927978516, -58.12947082519531, 117.24630737304688, 30.035751342773438, 76.71218872070312, -9.572860717773438, -1.278594970703125, 147.45928955078125, -0.515289306640625, 185.10568237304688, 22.552642822265625, -12.831779479980469, 11.364875793457031, 93.53108215332031, -27.501861572265625, 3.355438232421875, -17.07904052734375, -1.66619873046875, 18.6668701171875, -118.84841918945312, 140.5155029296875, 61.96623229980469, -22.707916259765625, -0.816009521484375, -34.13311767578125, 0.97100830078125, 186.6107177734375, 138.921142578125, 109.55613708496094, -3.3533058166503906, -23.721343994140625, 108.24786376953125, 137.4083251953125, -139.51882934570312, 25.46771240234375, 169.70216369628906, 40.31646728515625, 15.746612548828125, -71.85812377929688, 164.494140625, -23.76519775390625, 31.196136474609375, 111.73797607421875, 163.97189331054688, 156.80760192871094, 178.0628662109375, -62.38800048828125, 36.59466552734375, 4.654991149902344, 27.8944091796875, -7.227771759033203, -4.08465576171875, 131.7021026611328, 65.270263671875, -8.281791687011719, 1.914133071899414, 30.559234619140625, 39.664642333984375, -11.06982421875, 148.83694458007812, 8.583740234375, 128.80108642578125, -0.649932861328125, 79.18817138671875, 125.16697692871094, 119.397705078125, 101.76063537597656, 8.00341796875, 149.79164123535156, -4.076080322265625, 168.3701171875, 112.94696044921875, 140.3057098388672, 98.80615234375, 142.1092529296875, -8.670829772949219, 110.35797119140625, 58.5755615234375, 41.562255859375, 126.5449447631836, 137.88043212890625, 172.8992919921875, 128.02255249023438, 5.064910888671875, 26.56561279296875, -1.172607421875, -24.150421142578125, -12.188995361328125, 87.416015625, 129.09722900390625, 23.6407470703125, 32.80937957763672, 137.00985717773438, 170.47317504882812, -56.532012939453125, -4.852630615234375, 87.056396484375, 123.19574737548828, 152.88739013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000297.npy"}
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 56.43511199951172, "std": 85.09032440185547, "min": -165.41363525390625, "p10": -38.14254608154297, "median": 37.316429138183594, "p90": 161.54872436523436, "max": 306.94610595703125, "pos_frac": 0.734375, "sample": [206.9044952392578, 138.4476318359375, -21.563262939453125, 11.377914428710938, -42.1683349609375, -46.906158447265625, 5.60400390625, -12.311126708984375, -23.13385009765625, 161.31121826171875, 94.67306518554688, -13.14752197265625, 9.272796630859375, 14.45831298828125, 125.91133117675781, -107.67840576171875, 154.28759765625, 102.44306945800781, 40.72319793701172, 117.47156524658203, 150.0123291015625, 146.32150268554688, 37.88916015625, 203.279296875, 6.1825408935546875, -33.44317626953125, 134.21969604492188, -114.83045959472656, 136.43844604492188, 74.56805419921875, 152.51885986328125, -37.28326416015625, 130.18051147460938, 60.93586730957031, 32.3780517578125, -57.31231689453125, 6.9436492919921875, 68.13909912109375, 143.16104125976562, -75.44522094726562, 124.84298706054688, -12.415283203125, 130.31362915039062, -20.329315185546875, 42.30340576171875, 170.74819946289062, 41.8070068359375, 26.329757690429688, -105.24299621582031, 1.8595123291015625, 49.63261413574219, 138.4876708984375, 22.2852783203125, 150.33370971679688, -82.13076782226562, 59.44969177246094, 131.5785369873047, 162.1029052734375, 15.177886962890625, 132.1959228515625, 192.82562255859375, -165.41363525390625, 135.93670654296875, -14.170883178710938, -17.606826782226562, 195.92648315429688, 164.1156463623047, 48.174560546875, -37.48248291015625, 0.0, 5.444480895996094, 11.406608581542969, 182.21533203125, 76.49658203125, 23.98632049560547, -11.21194839477539, 126.03802490234375, 75.11256408691406, 8.977973937988281, 141.73846435546875, -39.68269348144531, -7.654172897338867, 14.92404556274414, 76.80598449707031, 131.88153076171875, 6.095512390136719, 150.93629455566406, 141.84634399414062, 139.83847045898438, 8.734375, 213.379150390625, 41.47998046875, -44.449424743652344, -13.1220703125, 14.1168212890625, 31.03411865234375, 94.02699279785156, 13.046875, -6.2120513916015625, -4.847412109375, -10.256759643554688, 125.8115234375, 51.38970947265625, 142.47589111328125, 200.34857177734375, 175.91946411132812, -21.711669921875, 111.44874572753906, 306.94610595703125, 19.312423706054688, 145.702880859375, 36.0576171875, 3.8346290588378906, 36.74369812011719, 159.66348266601562, 20.8092041015625, -11.0811767578125, 137.49984741210938, 21.166259765625, 179.01800537109375, 4.950231552124023, 106.6805419921875, 18.848342895507812, 28.7783203125, -93.42111206054688, -34.9715576171875, 82.07811737060547, -85.19651794433594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000298.npy"}
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 44.44041442871094, "std": 72.52891540527344, "min": -162.47311401367188, "p10": -32.08462219238281, "median": 22.987337112426758, "p90": 138.66035156249998, "max": 184.8663330078125, "pos_frac": 0.7109375, "sample": [102.91592407226562, 142.6434326171875, 19.061126708984375, 103.01978302001953, 22.71429443359375, 0.596221923828125, 144.98345947265625, -115.5272216796875, 129.1297607421875, -162.47311401367188, 163.61209106445312, 44.218719482421875, 21.500831604003906, 113.31484985351562, -27.86968994140625, 137.318115234375, -6.4996337890625, 20.876319885253906, 12.47064208984375, -132.81529235839844, 133.10440063476562, -2.794342041015625, 72.109375, -95.06867980957031, 145.88348388671875, -4.9912567138671875, -53.57867431640625, -33.528564453125, 122.11068725585938, 39.91753005981445, 104.569580078125, -4.391754150390625, 144.24293518066406, -2.555450439453125, -88.2666015625, 23.260379791259766, 117.01620483398438, -12.19195556640625, -87.19200897216797, 123.73794555664062, 13.62371826171875, 171.66734313964844, 25.2247314453125, 12.148296356201172, -8.388618469238281, -0.8232421875, 146.53768920898438, -4.612831115722656, -16.535354614257812, -53.64593505859375, -83.03472900390625, 15.37506103515625, 9.275867462158203, 17.299652099609375, 117.25628662109375, -37.29119873046875, 175.28158569335938, -6.0357666015625, 175.4591064453125, 45.989112854003906, 129.0463104248047, 41.056732177734375, 17.926559448242188, 128.65292358398438, -3.392498016357422, 184.8663330078125, 10.2640380859375, 11.965465545654297, 126.01251220703125, 21.830902099609375, 108.0318603515625, 9.777612686157227, 60.96697998046875, 8.062713623046875, 122.04771423339844, 113.87104797363281, -3.2291183471679688, -29.88571548461914, -5.9522857666015625, 133.56170654296875, -2.816497802734375, 87.01434326171875, 44.19386291503906, 15.451751708984375, 75.52975463867188, -6.2198638916015625, 141.792236328125, 41.890869140625, 80.7174072265625, 108.48863220214844, 15.030487060546875, 99.7564697265625, 133.7585906982422, 0.33465576171875, 17.059118270874023, -10.17041015625, 42.342681884765625, 6.925971984863281, 125.44518280029297, -21.245765686035156, -31.465789794921875, 57.641876220703125, 41.84898376464844, -50.1514892578125, 127.65214538574219, 183.78109741210938, 14.028411865234375, 137.2972412109375, 28.190673828125, -2.250732421875, 14.265716552734375, 2.3278560638427734, 128.70684814453125, -36.43724060058594, 7.0479736328125, 12.34521484375, -3.624664306640625, 34.37103271484375, 49.32267761230469, 37.59869384765625, -14.167007446289062, 113.08328247070312, 153.96836853027344, 125.6640625, 26.4635009765625, 133.02474975585938, 25.0816650390625, 102.64205932617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000299.npy"}
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 45.47718048095703, "std": 84.71858215332031, "min": -131.88470458984375, "p10": -65.05167236328126, "median": 35.909706115722656, "p90": 154.2224578857422, "max": 260.70367431640625, "pos_frac": 0.671875, "sample": [154.89505004882812, 152.23922729492188, 22.997905731201172, -89.72734069824219, 13.342048645019531, -27.610809326171875, 119.65432739257812, 72.437255859375, 7.400211334228516, -51.752197265625, 167.567138671875, 143.46255493164062, 151.24850463867188, 1.1156158447265625, -3.4055252075195312, 142.29598999023438, -119.4696044921875, 68.02902221679688, 86.60469055175781, -7.675628662109375, 159.85293579101562, 67.66839599609375, 49.51568603515625, 104.39042663574219, 35.39501953125, 8.25323486328125, 41.15478515625, -19.012786865234375, 17.13421630859375, -18.933555603027344, 101.45538330078125, 108.69735717773438, 139.10037231445312, 128.42266845703125, 164.31942749023438, -86.65608215332031, -131.88470458984375, 14.007476806640625, 3.978549003601074, 192.26336669921875, 79.07992553710938, -20.64984130859375, -39.746734619140625, 1.9896240234375, 124.62057495117188, 138.7154541015625, -74.2027587890625, 142.53082275390625, 121.62149047851562, 178.81454467773438, 14.496147155761719, -125.98606872558594, -114.41047668457031, 33.782196044921875, 33.44614791870117, -2.5396480560302734, 51.684417724609375, 54.03764343261719, 153.9342041015625, -9.26959228515625, 131.70138549804688, 135.65487670898438, -64.71636962890625, 163.14273071289062, 47.2823486328125, -24.842132568359375, 3.862213134765625, 111.17208862304688, 109.62614440917969, 135.3410186767578, 56.8902587890625, 105.47576904296875, -0.4643287658691406, 13.377670288085938, -20.894630432128906, -106.67613220214844, 172.68028259277344, 53.437255859375, -75.18359375, 140.14303588867188, 156.87893676757812, 14.149539947509766, 47.8468017578125, -55.72186279296875, 115.48246002197266, 21.058349609375, -9.76153564453125, 0.0, -16.802963256835938, 36.42439270019531, 178.61785888671875, -11.85919189453125, -131.26239013671875, 109.65673828125, 2.0937347412109375, 142.87326049804688, 90.21612548828125, -1.602294921875, 128.38885498046875, 136.0218505859375, -35.366790771484375, 260.70367431640625, 6.454833984375, 74.55007934570312, -60.11772155761719, 105.3934326171875, 151.98199462890625, -65.83404541015625, -36.19464111328125, -15.73321533203125, 75.49264526367188, -8.594655990600586, 49.011993408203125, -23.733718872070312, 60.63836669921875, 123.32379913330078, 41.676177978515625, 0.238311767578125, -103.39739990234375, -0.130889892578125, 184.3050537109375, 1.485107421875, -100.3807373046875, -29.234725952148438, 138.7476806640625, 159.171875, -7.598728179931641, 9.79364013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000300.npy"}
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 44.51064682006836, "std": 79.48170471191406, "min": -193.14120483398438, "p10": -34.70006713867187, "median": 27.799758911132812, "p90": 158.2257583618164, "max": 216.56683349609375, "pos_frac": 0.7265625, "sample": [13.4796142578125, 179.30926513671875, 137.11090087890625, 96.40582275390625, 19.98810577392578, 144.56866455078125, 32.995033264160156, -42.84620666503906, 99.75999450683594, 8.916475296020508, 3.615039825439453, 1.299560546875, 127.99008178710938, 6.425811767578125, 0.6927490234375, 6.9936676025390625, -13.820266723632812, -46.304901123046875, 129.1840057373047, 168.61422729492188, 15.567253112792969, 13.304389953613281, 121.3968505859375, 23.443756103515625, 36.35249328613281, 19.43988037109375, -1.579071044921875, 51.529541015625, 10.93548583984375, 72.849609375, 131.77554321289062, -7.554595947265625, 0.0, 166.434814453125, 138.95834350585938, 76.86026763916016, -13.9461669921875, 35.80902099609375, -29.085220336914062, 131.7701416015625, -17.761837005615234, 129.5829620361328, 163.6926727294922, 22.97674560546875, -158.65985107421875, -0.620361328125, -54.269775390625, 123.56863403320312, -30.486846923828125, 22.29388427734375, 42.44636917114258, -10.693603515625, -50.87944030761719, 160.0414581298828, 187.06903076171875, 31.182891845703125, 40.34031677246094, 165.29177856445312, 0.9400177001953125, 216.56683349609375, 63.747039794921875, 122.04754638671875, 19.46319580078125, 36.95765686035156, -97.99856567382812, -69.32925415039062, 58.4041748046875, -193.14120483398438, -136.181884765625, 29.164413452148438, 105.30658721923828, 76.20619201660156, -18.34326171875, 32.26477813720703, 10.131103515625, 27.210723876953125, 146.0340118408203, 1.1168212890625, 157.44760131835938, -83.53713989257812, 5.7965240478515625, -14.157806396484375, 182.1561737060547, 111.02020263671875, 167.12371826171875, 29.938720703125, -5.576564788818359, -27.650527954101562, 122.3546142578125, -121.460693359375, 142.96697998046875, 36.64453125, -18.611968994140625, 3.1042022705078125, 125.35797119140625, -38.614410400390625, 93.61174011230469, 20.0025634765625, 92.30789947509766, 116.64739990234375, 129.31317138671875, 143.53851318359375, 162.21182250976562, -10.047714233398438, 29.43914031982422, 151.83990478515625, 33.18231201171875, -22.00975799560547, 165.16429138183594, -15.127586364746094, 21.523162841796875, 6.24896240234375, -10.75628662109375, 4.26019287109375, 4.6172332763671875, 112.6455307006836, -101.9444580078125, 31.36848258972168, 4.414470672607422, 140.04446411132812, 174.984619140625, -33.022491455078125, -19.553466796875, 77.25919342041016, 26.542739868164062, 28.3887939453125, 113.62081909179688, -11.9769287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000301.npy"}
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 54.50590133666992, "std": 81.30390930175781, "min": -158.01211547851562, "p10": -27.343832397460925, "median": 47.83683776855469, "p90": 152.23945312499998, "max": 189.70785522460938, "pos_frac": 0.734375, "sample": [154.14198303222656, 59.3660888671875, 121.97029113769531, 160.487548828125, 22.54071044921875, 120.25379180908203, 0.0, -8.300796508789062, -0.162567138671875, -68.11314392089844, -142.485107421875, 114.00216674804688, 71.08660125732422, 142.25027465820312, -7.715728759765625, 73.4207992553711, 71.58184814453125, 46.84539794921875, -18.048965454101562, -95.2492446899414, 166.67196655273438, 135.81683349609375, 173.0030517578125, -15.11236572265625, 121.91302490234375, 137.93759155273438, 188.6866455078125, 18.404739379882812, -111.0665283203125, 114.7208480834961, 38.619903564453125, 134.85385131835938, 130.7943115234375, 74.18142700195312, 88.02479553222656, 110.6907958984375, 177.07980346679688, 92.8974609375, 49.789306640625, 47.771820068359375, 127.27395629882812, -2.0281143188476562, -158.01211547851562, -5.04547119140625, 97.06498718261719, 88.71864318847656, 185.29800415039062, 19.498306274414062, 169.02197265625, 148.53402709960938, -15.090301513671875, 143.52317810058594, 14.069190979003906, 5.139167785644531, -8.441429138183594, -64.72329711914062, 112.16015625, 4.012603759765625, 122.79425048828125, 38.94659423828125, 122.45266723632812, 104.55753326416016, 45.8927001953125, 3.0613327026367188, 36.7637939453125, 12.85723876953125, 8.610198974609375, 12.514007568359375, -5.536491394042969, -101.79637908935547, 112.66043853759766, 157.06268310546875, -16.45550537109375, -34.92144775390625, 145.8828125, 126.09759521484375, -15.695121765136719, 122.9715576171875, 116.69844055175781, 107.28145599365234, 25.861740112304688, 38.7572021484375, 138.74261474609375, -117.56793212890625, 151.46072387695312, -6.34698486328125, -142.51809692382812, 71.66663360595703, -24.096282958984375, -111.73007202148438, 14.998037338256836, -116.01712036132812, -3.394805908203125, 51.7537841796875, 3.2103214263916016, 47.90185546875, 104.72666931152344, 149.8416748046875, 149.31939697265625, -61.439369201660156, 3.437408447265625, 26.1802978515625, 33.660743713378906, -17.707500457763672, 130.56396484375, 0.675445556640625, 32.95135498046875, 189.70785522460938, 58.795963287353516, 115.15989685058594, -0.3913116455078125, 7.451957702636719, 49.468505859375, -17.06195068359375, 120.86953735351562, 42.75791931152344, 184.60855102539062, 2.930744171142578, 130.82821655273438, 40.85710144042969, 16.94830322265625, 187.8668212890625, 113.89390563964844, 145.40655517578125, 102.82334899902344, -10.633026123046875, 154.05648803710938, -11.707077026367188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000302.npy"}
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 50.16551971435547, "std": 83.5586929321289, "min": -167.81423950195312, "p10": -40.48591003417969, "median": 47.048667907714844, "p90": 152.00573120117187, "max": 205.64208984375, "pos_frac": 0.7734375, "sample": [14.091217041015625, 29.00347137451172, -146.78073120117188, -121.92973327636719, 122.63175964355469, -8.370010375976562, 122.03753662109375, 57.4078369140625, 136.058837890625, 120.14924621582031, -42.0540771484375, 134.43661499023438, 152.91268920898438, 139.9921875, 36.858428955078125, -22.802703857421875, -16.797439575195312, 125.1202392578125, -2.1244029998779297, 159.95428466796875, 112.53486633300781, 23.532135009765625, 122.04674530029297, 59.12066650390625, 103.3072509765625, -40.190643310546875, 102.66764831542969, -10.486328125, 165.24276733398438, 11.403564453125, 157.03634643554688, 139.529052734375, -41.17486572265625, 138.6009979248047, -13.977508544921875, 11.875762939453125, 4.8805694580078125, 60.781036376953125, 0.0, 12.2156982421875, -104.60538482666016, 23.83159637451172, -14.665748596191406, 10.876914978027344, 11.01132583618164, 0.894989013671875, 141.37257385253906, 49.18902587890625, 105.27627563476562, 121.54193115234375, 114.25384521484375, -128.23887634277344, 124.9359130859375, -167.81423950195312, 62.141204833984375, 143.1227569580078, 157.993408203125, 90.21269226074219, 73.56741333007812, 127.8624267578125, -0.14243698120117188, -2.1493682861328125, 20.657455444335938, -9.81243896484375, -71.42483520507812, 7.0545654296875, 101.86300659179688, 127.66893005371094, 146.87789916992188, 88.61515045166016, 157.81329345703125, -130.04852294921875, 19.522735595703125, 5.8210296630859375, 137.22537231445312, 115.92219543457031, 0.004364013671875, 89.66879272460938, 143.34320068359375, 205.64208984375, 69.8198471069336, -163.7679443359375, 115.94166564941406, -133.44155883789062, 7.211332321166992, 21.907073974609375, 121.59817504882812, 41.31443786621094, 20.4791259765625, 83.290771484375, 13.164154052734375, 3.964111328125, 154.2655792236328, 1.36822509765625, 159.0103759765625, 1.55126953125, 1.5156784057617188, 89.5732650756836, -31.816436767578125, 5.709808349609375, -32.697235107421875, -151.09591674804688, 55.00994873046875, -2.6710052490234375, 5.944862365722656, 108.69151306152344, 151.61703491210938, 83.79742431640625, 138.27093505859375, 111.56278991699219, 52.689666748046875, 59.479827880859375, 96.35840606689453, 124.78668212890625, -10.41943359375, 204.009765625, 123.81025695800781, -113.78610229492188, 158.21389770507812, 8.021560668945312, 160.60845947265625, 19.847198486328125, 182.1463623046875, 10.700347900390625, 30.500244140625, 8.854400634765625, 3.743927001953125, 44.90830993652344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000303.npy"}
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 43.65080261230469, "std": 79.9356460571289, "min": -155.07904052734375, "p10": -52.782614135742186, "median": 26.24285125732422, "p90": 147.92494506835936, "max": 185.60809326171875, "pos_frac": 0.7109375, "sample": [0.0, 30.444839477539062, -13.384490966796875, 50.782989501953125, -41.142059326171875, -1.72235107421875, -133.17465209960938, 143.4085693359375, 144.68487548828125, -27.50018310546875, 87.66288757324219, 140.42803955078125, 77.42604064941406, 92.41656494140625, -42.6842041015625, 41.77008056640625, 16.29632568359375, 154.09063720703125, 26.530960083007812, 13.5279541015625, 12.055145263671875, 18.345176696777344, 17.737716674804688, -13.504287719726562, 114.62115478515625, 139.60214233398438, 129.80654907226562, 44.250953674316406, 7.20159912109375, 0.440948486328125, -15.20733642578125, 3.0920352935791016, -10.015579223632812, 1.3902359008789062, 35.23896789550781, 25.954742431640625, 42.59912872314453, -20.41474151611328, 129.32550048828125, 127.09310913085938, -4.420379638671875, 105.56459045410156, 121.95442199707031, -52.3974609375, 178.79531860351562, 21.883056640625, -5.765647888183594, 130.4434814453125, -88.66580200195312, 43.02783203125, 38.44929504394531, -23.654891967773438, 102.35966491699219, -155.07904052734375, 0.0, 151.2833251953125, -78.79954528808594, 109.55623626708984, 5.8233642578125, -19.62091064453125, -111.32049560546875, 23.124160766601562, -119.71778869628906, 57.0960693359375, 19.977813720703125, 63.51661682128906, -37.5250244140625, 50.3917236328125, 185.60809326171875, 133.19287109375, -115.47128295898438, 23.511688232421875, 40.75852966308594, -98.152587890625, -53.681304931640625, 167.3316650390625, 157.34829711914062, 2.146636962890625, 150.95297241210938, -114.15179443359375, 139.22061157226562, 117.41717529296875, -22.48773956298828, -5.474765777587891, 124.0380859375, 19.486007690429688, 151.18087768554688, -4.700042724609375, 143.38858032226562, -139.7720947265625, 8.923294067382812, 1.1412220001220703, 24.696868896484375, 148.0997314453125, 157.02894592285156, 9.606435775756836, -59.479888916015625, 130.860107421875, 142.10926818847656, 150.29043579101562, 34.24267578125, 35.02825927734375, 84.6829605102539, 11.4736328125, 74.05036163330078, 141.69528198242188, 58.19810485839844, 169.72000122070312, 122.10206604003906, 147.85003662109375, 117.98797607421875, 51.04449462890625, -58.488861083984375, 9.528228759765625, 125.77934265136719, 19.253814697265625, -32.029541015625, -14.270950317382812, 131.29067993164062, -14.966171264648438, 23.56342315673828, 112.6857681274414, 1.76068115234375, 135.59286499023438, -19.537521362304688, 139.541015625, 21.49371337890625, 167.30770874023438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000304.npy"}
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 63.78484344482422, "std": 83.30851745605469, "min": -192.66424560546875, "p10": -19.954967498779276, "median": 50.201416015625, "p90": 161.17313232421876, "max": 297.53271484375, "pos_frac": 0.8125, "sample": [7.911552429199219, 10.716777801513672, -14.298599243164062, 205.10177612304688, 37.77056884765625, 4.96990966796875, 120.35456848144531, -63.86454772949219, 111.02125549316406, 20.68724822998047, 83.35100555419922, 45.20635986328125, 0.61236572265625, 48.747314453125, 19.599212646484375, -84.13165283203125, -42.863006591796875, 27.595672607421875, 60.81117248535156, 0.6478614807128906, 92.65542602539062, 151.08685302734375, 1.1893844604492188, 125.30232238769531, 6.09295654296875, 17.506317138671875, -116.09515380859375, -4.8394012451171875, 34.489959716796875, 127.23246765136719, 23.296295166015625, 125.33001708984375, 133.14529418945312, 91.51800537109375, 158.57122802734375, 19.981613159179688, -0.4722900390625, -4.1760711669921875, 138.6357421875, -2.562896728515625, -141.55186462402344, 82.63673400878906, -6.26849365234375, 119.50878143310547, 154.6478271484375, 127.94544982910156, 52.47308349609375, 163.0872802734375, 146.57716369628906, 42.585052490234375, -7.826805114746094, -117.49601745605469, 172.62213134765625, 163.18124389648438, 41.51206970214844, 80.364013671875, 297.53271484375, 110.23487854003906, 188.83367919921875, 172.52767944335938, 23.14373779296875, -0.4362831115722656, -9.78436279296875, 155.79110717773438, 37.664459228515625, 118.55915832519531, 158.5932159423828, 133.2656707763672, 100.59600830078125, 136.85504150390625, -33.153160095214844, 128.66171264648438, -67.73703002929688, 37.428924560546875, 133.36184692382812, -192.66424560546875, 95.06883239746094, 134.3884735107422, 5.993133544921875, 57.61846923828125, 44.97636413574219, 158.9168701171875, 165.89450073242188, 10.458671569824219, 111.54458618164062, 116.74334716796875, -33.377227783203125, 209.74832153320312, 2.6480484008789062, 7.1837005615234375, 162.6002197265625, 104.9990234375, 105.86184692382812, 39.67291259765625, 129.45925903320312, 135.5255889892578, 52.05183410644531, 138.805908203125, 183.96630859375, 51.655517578125, 34.146575927734375, -3.8173828125, 131.30364990234375, 132.01580810546875, 203.06353759765625, 26.762237548828125, 32.6761474609375, 23.13067626953125, 17.216064453125, 22.46417236328125, 160.52047729492188, 0.568511962890625, 34.148681640625, 3.92645263671875, 22.708724975585938, -92.14754486083984, 160.5615234375, 4.666999816894531, 10.252395629882812, -78.6859130859375, 137.03250122070312, 54.5103759765625, 196.12078857421875, 141.91636657714844, 114.86248779296875, -89.47160339355469, -3.26165771484375, 153.7576904296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000305.npy"}
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 55.05732727050781, "std": 80.36931610107422, "min": -157.297607421875, "p10": -24.840673828124995, "median": 49.76231384277344, "p90": 155.56255798339845, "max": 194.80926513671875, "pos_frac": 0.734375, "sample": [44.174224853515625, -111.5811767578125, -108.08600616455078, 194.80926513671875, -157.297607421875, 186.77239990234375, 9.173025131225586, 61.94544982910156, -12.951019287109375, 143.70477294921875, 29.03647804260254, 26.344985961914062, 148.5769805908203, 117.89097595214844, 17.924842834472656, -3.87432861328125, -40.089263916015625, 131.43399047851562, -4.749305725097656, 24.97430419921875, 51.7369384765625, 33.95782470703125, -56.13836669921875, 56.87957763671875, -146.67904663085938, 173.59429931640625, 169.2010498046875, 132.6917724609375, 116.26226806640625, -0.805419921875, 60.681610107421875, 99.72543334960938, 36.14244079589844, 135.68502807617188, 156.19857788085938, 134.551025390625, 178.85430908203125, 132.81857299804688, 125.820068359375, 115.0198974609375, 7.7906341552734375, 120.21866607666016, -8.519287109375, 8.880752563476562, 113.50735473632812, 98.26902770996094, 96.63789367675781, 131.4566192626953, 54.689056396484375, -89.0477294921875, -12.38824462890625, 147.09764099121094, -21.36322784423828, 187.866943359375, 41.95855712890625, -145.6229248046875, 85.11121368408203, 66.166015625, -13.90484619140625, 191.79095458984375, -2.8480224609375, 50.2847900390625, 102.69223022460938, 24.366806030273438, 129.03683471679688, 4.072416305541992, 6.664459228515625, 2.8207616806030273, 112.74185180664062, -63.413848876953125, 123.19131469726562, -123.40057373046875, 118.06416320800781, -17.63372802734375, 185.62982177734375, -1.6664581298828125, -9.541656494140625, 27.19457244873047, 18.395187377929688, 24.93768310546875, 13.397140502929688, 59.162933349609375, 121.05569458007812, 132.11477661132812, 153.9998779296875, 14.122833251953125, -27.04833984375, 64.4814453125, 95.1263427734375, 192.028564453125, 67.5208740234375, 30.903854370117188, 22.151321411132812, -23.89453125, 61.677581787109375, 122.1693115234375, 120.64913940429688, -12.353729248046875, 156.34518432617188, -35.51616668701172, 149.948974609375, 45.265716552734375, 142.68165588378906, 99.19671630859375, 65.323974609375, 14.658187866210938, 6.050193786621094, -14.808700561523438, -4.3023681640625, -3.667724609375, -8.013111114501953, 4.373046875, 149.24290466308594, 64.62266540527344, 14.53924560546875, 180.63787841796875, 14.972017288208008, -3.2217559814453125, -3.85223388671875, 4.355560302734375, 100.55462646484375, 124.5484619140625, -0.32781219482421875, -98.07122802734375, 127.53031158447266, 49.239837646484375, 165.96237182617188, 155.28997802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000306.npy"}
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 59.65784454345703, "std": 78.13683319091797, "min": -187.10250854492188, "p10": -18.9888671875, "median": 59.798030853271484, "p90": 148.33937377929686, "max": 200.79498291015625, "pos_frac": 0.7734375, "sample": [141.58782958984375, 141.87246704101562, 21.293014526367188, 175.51492309570312, -129.65753173828125, -6.4794158935546875, 83.83740234375, 112.74386596679688, 125.30731201171875, 141.62384033203125, -169.13641357421875, -5.255405426025391, 98.05311584472656, -1.415252685546875, -15.5355224609375, 126.93836975097656, 142.99786376953125, 125.38703918457031, 11.502941131591797, 21.5660400390625, 60.057830810546875, -41.27496337890625, 123.45693969726562, 9.953630447387695, 183.85018920898438, -27.958526611328125, 88.43548583984375, 12.69757080078125, 132.55271911621094, 168.0341796875, 109.24995422363281, 24.018524169921875, 30.09051513671875, 108.91515350341797, 120.91145324707031, -20.684585571289062, 139.7207794189453, 67.52725219726562, 22.93768310546875, -8.408485412597656, 25.50262451171875, 105.20773315429688, 131.7356719970703, -13.82806396484375, 147.73162841796875, 104.81507873535156, 146.0363311767578, -11.443527221679688, 0.6469268798828125, 26.58367919921875, 92.95487976074219, 5.966411590576172, 145.6129608154297, 9.555984497070312, 72.89010620117188, 132.5677947998047, 117.21807861328125, 155.5258331298828, 17.63763427734375, 3.16094970703125, 15.28253173828125, 77.06298065185547, -67.50607299804688, 117.85540771484375, -6.4327850341796875, -4.0408477783203125, 134.67356872558594, 47.738197326660156, -7.493824005126953, 6.824199676513672, -5.242549896240234, 143.0968475341797, -55.21575927734375, -52.98626708984375, 30.12774658203125, -29.83343505859375, 90.9122314453125, 138.4050750732422, 63.65008544921875, 27.0164794921875, 149.9698028564453, -18.262130737304688, 59.538230895996094, 160.15969848632812, 168.47567749023438, 121.56283569335938, -11.620479583740234, 129.15621948242188, 16.478404998779297, 64.9923095703125, 16.51854705810547, -41.040924072265625, 84.52923583984375, 174.96163940429688, 107.39323425292969, 31.663818359375, 142.2730255126953, -143.49200439453125, -8.639511108398438, 132.4088134765625, 3.68243408203125, 184.841064453125, 33.0120849609375, 169.05294799804688, -187.10250854492188, 3.378326416015625, 4.75946044921875, 43.806671142578125, 39.169830322265625, 98.04033660888672, -95.67584228515625, 26.1595458984375, 200.79498291015625, 130.53829956054688, 37.153106689453125, 111.38800048828125, -0.464508056640625, -6.126739501953125, 48.56101989746094, 138.9354248046875, 84.125244140625, 5.546577453613281, 125.49899291992188, 134.19070434570312, 165.37271118164062, 3.644214630126953, 149.7574462890625, 120.33609008789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000307.npy"}
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 41.72602844238281, "std": 81.75587463378906, "min": -183.39010620117188, "p10": -55.802362060546855, "median": 23.26734161376953, "p90": 147.51065673828123, "max": 211.67788696289062, "pos_frac": 0.7421875, "sample": [153.8651580810547, 35.781959533691406, -93.31121826171875, 152.0274200439453, 71.68475341796875, -101.35084533691406, -49.958221435546875, 53.84039306640625, 9.313812255859375, 92.5910873413086, 4.2271728515625, 174.74111938476562, 1.3258438110351562, -77.19049072265625, -4.555999755859375, 7.04229736328125, 123.82599639892578, 9.273040771484375, 127.28570556640625, 113.21112060546875, 134.80303955078125, 24.8780517578125, 1.3837890625, -82.14342498779297, 14.852325439453125, 112.52401733398438, -160.94888305664062, -10.509986877441406, 13.670562744140625, 151.80621337890625, 124.11524963378906, 22.0042724609375, 93.9700927734375, 97.79283142089844, -18.08782958984375, -9.17755126953125, 134.79238891601562, 2.8747787475585938, 111.00126647949219, 22.84228515625, 0.00611114501953125, 40.20354461669922, 193.88128662109375, -1.10809326171875, -37.86436462402344, -9.5999755859375, 0.998260498046875, 146.10235595703125, -4.691308975219727, 126.09346008300781, 5.9024505615234375, 32.342323303222656, 2.9428024291992188, 118.77550506591797, -9.438606262207031, 32.430267333984375, 118.11419677734375, 26.404861450195312, -183.39010620117188, 185.90924072265625, 79.40785217285156, 7.2319488525390625, -6.859619140625, -18.305023193359375, 211.67788696289062, 122.04942321777344, -3.7975406646728516, -132.32266235351562, 21.292709350585938, 0.69256591796875, 95.6705322265625, -101.18983459472656, 150.79669189453125, 79.79556274414062, 9.5294189453125, 48.330322265625, 170.07687377929688, 193.9808349609375, 151.84878540039062, -1.6507377624511719, 111.48225402832031, -25.69842529296875, 2.510284423828125, 6.974884033203125, 73.0260009765625, -123.18508911132812, 47.738563537597656, 11.061981201171875, -69.43869018554688, 120.06707763671875, 21.460235595703125, 15.260231018066406, 32.56280517578125, 10.039993286132812, 126.6435317993164, 116.56000518798828, 164.96163940429688, 139.1209716796875, 33.08418273925781, 136.3785400390625, 25.425079345703125, 75.3109130859375, 185.323486328125, -74.03138732910156, 85.41260528564453, 23.692398071289062, 14.79498291015625, -49.3884162902832, 107.1649169921875, 123.93585205078125, -146.50286865234375, 13.669677734375, -25.584869384765625, 1.5992355346679688, 2.4003829956054688, 105.98965454101562, -34.837371826171875, 10.779327392578125, -5.512451171875, 6.718475341796875, -42.794281005859375, 105.47383117675781, 134.98770141601562, 138.2784423828125, -100.60501861572266, 70.83047485351562, 114.2423095703125, 75.16339111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000308.npy"}
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 58.5731315612793, "std": 87.89601135253906, "min": -139.33966064453125, "p10": -46.871415328979445, "median": 46.98078918457031, "p90": 169.2660705566406, "max": 305.9710388183594, "pos_frac": 0.7265625, "sample": [35.08721923828125, 173.1771240234375, 147.27020263671875, 121.12384033203125, -2.5687713623046875, -18.171112060546875, -26.33148193359375, -109.53076934814453, 142.44140625, 44.841033935546875, 167.2947998046875, 11.066535949707031, -13.643768310546875, 6.9250640869140625, -89.3563232421875, 99.7661361694336, 11.31707763671875, 18.87439727783203, -9.887969970703125, 43.247528076171875, -9.389633178710938, 122.64337158203125, -14.05560302734375, -92.45167541503906, 11.23834228515625, 106.43060302734375, 1.11285400390625, 305.9710388183594, 131.30856323242188, 47.642333984375, -8.495559692382812, 179.50018310546875, 68.40959167480469, -1.717559814453125, -5.33062744140625, -11.913833618164062, 93.66032409667969, -80.15452575683594, 208.182373046875, -78.0423583984375, 200.17117309570312, -87.2974853515625, -4.27386474609375, 76.69671630859375, 132.10568237304688, -78.10515594482422, 3.247314453125, -33.104248046875, -109.21223449707031, 36.296722412109375, 140.81724548339844, -24.096942901611328, 16.575653076171875, -99.16461181640625, 160.150146484375, 129.06912231445312, 134.38275146484375, 15.11102294921875, 84.313720703125, 14.38714599609375, 167.62286376953125, 6.835235595703125, 44.467041015625, 242.110595703125, -13.704750061035156, 179.27691650390625, 11.173324584960938, 107.05078887939453, 22.601234436035156, 46.319244384765625, 56.915618896484375, 32.09394836425781, -5.351310729980469, 185.5279541015625, 1.0982666015625, 161.31680297851562, 151.14236450195312, 94.27215576171875, 12.88775634765625, 173.444580078125, 134.49343872070312, 139.33566284179688, 6.556648254394531, 80.38176727294922, 129.44668579101562, 153.44512939453125, -4.067237854003906, 102.0487289428711, 40.50067138671875, -33.5124397277832, 141.03280639648438, -139.33966064453125, 66.84861755371094, -16.995817184448242, 173.1002197265625, 131.97232055664062, 18.173385620117188, 139.5144500732422, 74.671142578125, 10.079299926757812, -116.9088134765625, 54.60447692871094, 143.43603515625, 164.49359130859375, 149.50265502929688, -0.926116943359375, 196.7318115234375, 34.68792724609375, 56.5528564453125, -123.92996215820312, 159.05389404296875, 60.18634033203125, 21.892364501953125, 151.27130126953125, 79.249267578125, 145.30908203125, 55.82391357421875, 173.70501708984375, -32.965789794921875, 150.58261108398438, 180.08554077148438, 58.081199645996094, 119.51152038574219, -80.03125, 104.23248291015625, 40.728424072265625, -2.495136260986328, 94.54852294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000309.npy"}
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 48.63544464111328, "std": 74.68162536621094, "min": -135.60324096679688, "p10": -30.1950698852539, "median": 31.568206787109375, "p90": 145.15955505371093, "max": 221.568115234375, "pos_frac": 0.75, "sample": [221.568115234375, 17.705352783203125, -1.1815071105957031, 189.98486328125, 36.63592529296875, 33.52117919921875, -23.084945678710938, 8.578531265258789, 131.25987243652344, 78.10287475585938, 118.47968292236328, 178.1361083984375, 75.23188781738281, 131.11825561523438, 22.94061279296875, 37.777557373046875, 3.85943603515625, -4.099607467651367, 37.78656005859375, -85.36610412597656, -1.960418701171875, 1.530059814453125, 19.568771362304688, 57.760284423828125, 118.03528594970703, -99.57279968261719, 125.79621887207031, -40.004302978515625, 128.0426788330078, 0.08575439453125, -3.204252243041992, 7.138702392578125, -28.25604248046875, 137.59811401367188, -2.94561767578125, -2.698394775390625, 60.774871826171875, 171.62295532226562, -52.373321533203125, -106.39154052734375, 121.84524536132812, -1.7925262451171875, 10.4326171875, 21.79034423828125, 107.78826904296875, 34.1787109375, 103.51708984375, 28.521995544433594, 50.51823425292969, 2.110015869140625, -10.824310302734375, 56.08537292480469, 158.58615112304688, 101.139404296875, 91.60507202148438, 143.9310302734375, 184.96173095703125, 14.070205688476562, -4.264808654785156, -135.60324096679688, -2.749256134033203, 168.64605712890625, 23.777503967285156, 20.88275146484375, -4.497650146484375, -114.46045684814453, 7.0125885009765625, 17.7930908203125, 108.31024169921875, 4.322700500488281, -23.903038024902344, 16.778823852539062, 83.25920104980469, 29.615234375, 46.17987060546875, 100.6812744140625, 136.38104248046875, 171.94146728515625, 58.21148681640625, -9.842697143554688, -22.984153747558594, 111.90519714355469, 117.63555908203125, 0.4112091064453125, -51.46441650390625, 113.98458862304688, 140.07855224609375, 4.209930419921875, 121.72116088867188, -12.67816162109375, 70.32366943359375, -58.6072998046875, 139.66781616210938, -94.58807373046875, 58.3450813293457, 3.1934738159179688, 10.793838500976562, 146.8670654296875, 56.449493408203125, 42.62745666503906, 53.184959411621094, 159.51016235351562, 9.548309326171875, -34.71946716308594, 144.42776489257812, -11.576141357421875, 7.7418212890625, -49.152923583984375, -93.70802307128906, 106.87222290039062, -12.190353393554688, 142.01861572265625, 53.21087646484375, 106.5174560546875, 7.96368408203125, 23.18255615234375, 18.85357666015625, 19.013172149658203, 4.9275360107421875, 108.90821838378906, 41.330535888671875, 6.217475891113281, 141.06060791015625, 158.79608154296875, 208.58181762695312, 124.08139038085938, 150.86203002929688, 115.54194641113281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000310.npy"}
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 52.008338928222656, "std": 89.7458724975586, "min": -199.45156860351562, "p10": -54.32983779907222, "median": 33.58699417114258, "p90": 159.3593826293945, "max": 216.2769775390625, "pos_frac": 0.734375, "sample": [6.701301574707031, -3.0263824462890625, 130.35353088378906, 125.8081283569336, 152.64300537109375, 13.688180923461914, 30.415817260742188, -4.5643463134765625, 153.32989501953125, 125.97454833984375, 145.6342315673828, 8.247608184814453, -31.781757354736328, -23.35235595703125, 36.06535339355469, -4.240478515625, 156.818359375, 67.62411499023438, -84.62454986572266, 49.681854248046875, 182.91839599609375, 145.17645263671875, 1.8353729248046875, 120.76651000976562, -5.6491241455078125, -19.1063232421875, 165.28843688964844, 121.30807495117188, 6.282672882080078, 13.1259765625, -103.3987808227539, -182.22772216796875, -199.45156860351562, 116.11434936523438, 27.916717529296875, 165.47601318359375, -134.9342041015625, -94.4090576171875, 139.712646484375, 53.8341064453125, -81.61524200439453, 106.57354736328125, 63.737884521484375, 4.43243408203125, -1.868661880493164, 53.47589111328125, -16.184890747070312, 126.62654113769531, -125.34234619140625, 31.96269989013672, 118.89989471435547, 29.78387451171875, 2.05438232421875, 29.173248291015625, 124.8642578125, -42.63609313964844, 85.30525970458984, -14.33392333984375, -21.71197509765625, 173.43862915039062, 166.90411376953125, 144.03549194335938, 151.7266845703125, 124.96187591552734, 130.68597412109375, -89.67840576171875, 33.63946533203125, 8.4351806640625, 20.51404571533203, 33.534523010253906, -112.69622802734375, -13.075855255126953, 183.80752563476562, 12.521392822265625, 21.226409912109375, 98.07642364501953, 165.4703369140625, 128.09890747070312, 27.77315902709961, 155.50390625, 216.2769775390625, 1.034912109375, -111.28890991210938, 64.49811553955078, 0.3556327819824219, -30.683349609375, 171.6356201171875, 171.356689453125, 144.3643798828125, 20.420928955078125, 129.1217803955078, 106.43057250976562, 114.19807434082031, 53.77801513671875, 180.48748779296875, -133.3096160888672, 174.00897216796875, -7.406524658203125, -133.62464904785156, 125.35475158691406, 8.377872467041016, 155.70779418945312, -19.895606994628906, 24.8089599609375, 10.207183837890625, -4.931427001953125, 155.0709991455078, 152.0364990234375, 8.841840744018555, 45.0264892578125, 135.38111877441406, 115.283935546875, 127.05810546875, 25.931915283203125, -37.75419616699219, 6.639076232910156, 110.97877502441406, 146.7619171142578, 147.9987335205078, 2.8883438110351562, -21.64773941040039, -3.056793212890625, 5.967857360839844, 198.48260498046875, 105.63975524902344, 137.75595092773438, 82.767822265625, -32.34587097167969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000311.npy"}
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 53.271507263183594, "std": 83.42378997802734, "min": -174.30697631835938, "p10": -40.550897216796876, "median": 46.49403762817383, "p90": 154.08541259765624, "max": 206.57806396484375, "pos_frac": 0.734375, "sample": [29.67205810546875, -4.486457824707031, -19.770721435546875, 138.41586303710938, 145.2849578857422, -174.30697631835938, 143.51083374023438, 127.8037109375, -38.60839080810547, -40.229034423828125, -66.39166259765625, -23.15044403076172, 94.72067260742188, 119.86441040039062, 78.70318603515625, 166.36148071289062, 116.92399597167969, 138.0486297607422, -15.010787963867188, -124.19186401367188, 6.3101806640625, 128.2029266357422, 139.45172119140625, -0.115753173828125, 7.595283508300781, 46.685569763183594, 56.97808837890625, 205.13235473632812, -41.301910400390625, -13.62491226196289, 162.54605102539062, -5.356685638427734, 1.7835693359375, 164.8463134765625, 73.00921630859375, 34.029205322265625, 101.1395263671875, 0.9147529602050781, 101.30387878417969, 116.82376098632812, 177.45028686523438, 157.3321075439453, 84.1957015991211, 118.861083984375, 122.48619842529297, 121.93598175048828, 61.9586181640625, 170.1302490234375, 133.59390258789062, 24.183334350585938, 142.02810668945312, 145.0792236328125, 18.063934326171875, 2.777374267578125, -39.18865966796875, 28.523773193359375, 46.30250549316406, 19.55364990234375, 1.4196014404296875, 206.57806396484375, 9.901542663574219, -29.841156005859375, 151.52001953125, -119.04541015625, 83.92384338378906, -34.65155029296875, 172.9010009765625, 128.46533203125, -55.90788269042969, 130.18748474121094, 29.892364501953125, 133.67312622070312, 28.963790893554688, -5.457696914672852, 90.18707275390625, 91.19329833984375, 2.4408645629882812, 154.28634643554688, 0.9367828369140625, 117.6563720703125, 147.88046264648438, 153.99929809570312, -39.087554931640625, 129.16407775878906, -3.370494842529297, 0.0, 60.2957763671875, -84.6893310546875, -131.00527954101562, -103.48387145996094, 18.2998046875, 76.83538818359375, 4.8428192138671875, 37.88533020019531, 106.23348999023438, 1.3359603881835938, -115.790771484375, 16.423152923583984, 124.4154052734375, 2.17657470703125, 55.8609619140625, 66.08224487304688, 139.064208984375, 89.85783386230469, -22.058349609375, 117.39971923828125, 138.59429931640625, -4.194580078125, 181.10861206054688, 39.75726318359375, 142.56640625, 19.500823974609375, 37.98797607421875, -5.105552673339844, -88.61871337890625, 102.85908508300781, 107.91259765625, 3.824676513671875, -26.412506103515625, 179.471923828125, -119.85035705566406, 161.021484375, -29.9814453125, 135.29241943359375, 119.38074493408203, 25.673828125, 12.811832427978516, -67.42210388183594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000312.npy"}
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 49.99053955078125, "std": 85.91024780273438, "min": -160.04446411132812, "p10": -51.33896636962889, "median": 33.097190856933594, "p90": 162.00122528076173, "max": 233.0308837890625, "pos_frac": 0.71875, "sample": [172.58010864257812, 149.95663452148438, -30.7935791015625, -27.9290771484375, -29.59918212890625, -1.4911575317382812, 61.96392822265625, 140.0671844482422, 166.601806640625, -12.739013671875, -9.709213256835938, 140.40338134765625, -97.3009033203125, 149.0769500732422, 106.14952850341797, 162.67735290527344, 148.48831176757812, 42.686614990234375, -7.681610107421875, -27.5159912109375, 9.856918334960938, 12.93438720703125, 3.8277549743652344, 147.345947265625, 125.17071533203125, -7.769195556640625, -6.01153564453125, 25.378036499023438, 117.40435791015625, 188.3798828125, 173.1044921875, 214.7449951171875, 71.35111999511719, 95.78753662109375, -47.24993896484375, 69.07318115234375, 114.39671325683594, 76.22579956054688, 183.87582397460938, 30.29155731201172, -75.25772094726562, 186.89093017578125, 150.06004333496094, 16.855743408203125, -114.94573974609375, 49.355621337890625, 22.783203125, -59.80938720703125, 122.83921813964844, 149.04034423828125, 32.75614929199219, 92.55976867675781, -102.04318237304688, 20.02806282043457, 77.69268798828125, 1.0472335815429688, 161.71145629882812, 52.74494934082031, 28.74089813232422, 3.1128082275390625, 127.38274383544922, 3.6935882568359375, 130.3611602783203, 19.010299682617188, 16.92017364501953, 54.67205810546875, -29.470077514648438, 112.21835327148438, -108.61822509765625, -38.78216552734375, 171.00999450683594, -17.19757080078125, 22.60070037841797, 162.77752685546875, 19.541259765625, -5.57269287109375, 33.430572509765625, 144.29251098632812, -114.94314575195312, 110.71125793457031, 9.838287353515625, -112.636962890625, 103.42626953125, -47.70878601074219, 37.300994873046875, 134.6347198486328, 99.20156860351562, -8.61907958984375, 6.4700775146484375, 6.170295715332031, 26.8663330078125, 233.0308837890625, 119.66383361816406, 63.63818359375, 135.06124877929688, 161.3024444580078, 26.357162475585938, 171.3641357421875, -40.823822021484375, -160.04446411132812, 61.1544189453125, 45.646080017089844, 8.198522567749023, 150.94122314453125, 95.600830078125, -92.59016418457031, -0.37334632873535156, -2.119903564453125, 150.80136108398438, 121.33493041992188, 28.235321044921875, 57.52430725097656, 58.438995361328125, 3.93878173828125, -2.9154739379882812, -10.479820251464844, 128.3333740234375, 32.76380920410156, -110.36627197265625, -111.16908264160156, -0.0660858154296875, -159.30938720703125, 109.90177917480469, 71.79601287841797, 20.662811279296875, 158.24655151367188, 27.49456787109375, 168.45948791503906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000313.npy"}
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 49.62617492675781, "std": 86.42066192626953, "min": -177.90786743164062, "p10": -66.3202911376953, "median": 38.663246154785156, "p90": 151.3843231201172, "max": 214.44210815429688, "pos_frac": 0.7109375, "sample": [30.909286499023438, 107.15322875976562, 123.36668395996094, 46.191314697265625, -13.770278930664062, 20.95916748046875, -103.07063293457031, 105.23294067382812, -91.52787017822266, -132.76055908203125, -22.120468139648438, 70.08625793457031, 145.20245361328125, 160.43727111816406, 167.99609375, -67.3739013671875, -99.28117370605469, 188.87567138671875, 138.22647094726562, 156.21206665039062, 73.22677612304688, -9.9261474609375, 47.75379943847656, 123.85331726074219, -1.708709716796875, 9.808372497558594, 13.508445739746094, 174.11419677734375, 6.464599609375, 180.24853515625, 5.513744354248047, -14.633270263671875, 10.99432373046875, -116.6473388671875, 143.87973022460938, 63.79290771484375, 19.644439697265625, 151.374755859375, 41.96343994140625, 150.52630615234375, -9.774642944335938, 124.43846130371094, -77.166259765625, 118.95697021484375, 5.282497406005859, 41.39710998535156, 188.88592529296875, 130.0733642578125, 7.424407958984375, -0.718994140625, 148.31405639648438, 0.47431182861328125, 25.551040649414062, 99.4786376953125, -0.9715652465820312, 110.06765747070312, 29.8768310546875, 18.99053955078125, -19.407958984375, -121.27967834472656, 118.09812927246094, 151.1318817138672, 128.73475646972656, 0.0, 2.33953857421875, -70.07675170898438, -169.2757568359375, 129.72531127929688, 22.21221923828125, -21.053619384765625, -31.177642822265625, 21.4754638671875, 99.61065673828125, 140.589599609375, 165.00540161132812, -65.86874389648438, -21.71428680419922, 167.1690673828125, -29.998321533203125, -177.90786743164062, 140.98480224609375, 60.36627197265625, 140.94326782226562, 86.83340454101562, 0.0, 147.79244995117188, -4.1680908203125, 95.09735107421875, 139.16925048828125, 13.662261962890625, 29.46575927734375, 136.8211669921875, 138.34747314453125, 34.20330810546875, 113.872802734375, 151.40664672851562, 104.83763885498047, 22.86175537109375, 45.7935791015625, 7.72344970703125, 9.376468658447266, -39.2647705078125, -4.43133544921875, -29.160430908203125, 109.44491577148438, 9.9376220703125, 98.10330963134766, 127.41603088378906, 45.555694580078125, 113.87966918945312, 175.41757202148438, 35.92938232421875, -61.02655029296875, -5.9634857177734375, 4.67474365234375, 214.44210815429688, 145.605712890625, -16.74022674560547, 20.229068756103516, 107.074951171875, 136.57424926757812, 63.66728973388672, -80.19854736328125, 88.36196899414062, -171.10531616210938, 173.17538452148438, 95.99696350097656, -4.444246292114258], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000314.npy"}
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 57.08453369140625, "std": 83.77558135986328, "min": -170.5590362548828, "p10": -31.2066291809082, "median": 30.092365264892578, "p90": 165.18873291015623, "max": 267.0880432128906, "pos_frac": 0.765625, "sample": [-1.61773681640625, 29.755722045898438, 3.077159881591797, 5.0229949951171875, 30.770172119140625, -37.90252685546875, -22.78875732421875, 13.480636596679688, 173.53591918945312, -11.87933349609375, 30.1304931640625, -13.3482666015625, 129.61553955078125, 21.004974365234375, 130.66348266601562, -35.46739196777344, 45.17646789550781, 161.89968872070312, 52.511322021484375, -23.960586547851562, -60.19062042236328, -23.05535888671875, 1.92718505859375, -1.4802017211914062, 183.20840454101562, 12.07318115234375, 123.88357543945312, 175.94760131835938, 204.33526611328125, 1.1302375793457031, 25.722103118896484, 43.857696533203125, 154.35382080078125, 162.78955078125, 29.004409790039062, 183.08489990234375, 160.17343139648438, 149.35531616210938, 107.83659362792969, -4.414031982421875, -0.15545654296875, -35.54193115234375, 21.75860595703125, 157.3551025390625, -52.15313720703125, 15.229385375976562, 8.46209716796875, 5.983673095703125, 13.3953857421875, 6.1077880859375, 147.32400512695312, -17.672687530517578, 150.34759521484375, 167.17559814453125, 148.40487670898438, 148.94459533691406, 18.075714111328125, -30.31096649169922, 39.4503173828125, 90.89905548095703, 14.000869750976562, 0.108367919921875, -9.322391510009766, 102.88020324707031, 147.57064819335938, 150.882568359375, 99.70883178710938, 121.9961166381836, 10.33526611328125, 267.0880432128906, -93.6199951171875, 36.16716003417969, 135.0709228515625, 21.196014404296875, 156.84329223632812, -42.33648681640625, -154.38331604003906, 178.6893310546875, 3.491476058959961, 171.29159545898438, 149.35308837890625, 34.442535400390625, -3.2130584716796875, 146.13446044921875, 193.3138427734375, 5.828155517578125, 38.940643310546875, 117.10063171386719, 20.00439453125, 181.36212158203125, 18.740554809570312, 35.790802001953125, -148.26315307617188, -33.2965087890625, 157.16610717773438, 174.63821411132812, 86.15826416015625, 55.354122161865234, -170.5590362548828, 159.285888671875, 30.054237365722656, -11.597412109375, -2.0289306640625, 114.31652069091797, 10.697891235351562, -76.36163330078125, -67.99868774414062, 47.7247314453125, 25.75473403930664, 22.2958984375, 159.62637329101562, 14.5849609375, 138.95252990722656, 212.2100067138672, 17.17864990234375, 112.289306640625, 164.33721923828125, 7.9817657470703125, -9.59100341796875, 13.85565185546875, 57.395729064941406, -2.8511505126953125, 142.20504760742188, 31.985755920410156, 139.06838989257812, 91.4046630859375, 10.702529907226562, 64.38368225097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000315.npy"}
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 58.968963623046875, "std": 84.6307144165039, "min": -147.12265014648438, "p10": -28.503381347656248, "median": 40.471588134765625, "p90": 171.411213684082, "max": 267.60760498046875, "pos_frac": 0.734375, "sample": [109.5775146484375, 94.82861328125, 59.720947265625, 36.135772705078125, -19.039794921875, -26.364974975585938, -11.19439697265625, 175.04132080078125, 92.05873107910156, 175.9146728515625, 23.920928955078125, -39.2115478515625, 148.4718475341797, 32.65058898925781, -76.06523132324219, 10.084966659545898, 127.28860473632812, -33.1846923828125, 68.22509765625, 7.069366455078125, 146.90838623046875, 161.44613647460938, -28.062042236328125, 127.4022216796875, 137.67398071289062, 251.4730224609375, 43.607574462890625, 136.94125366210938, 47.24676513671875, 191.98728942871094, -9.070960998535156, 145.5992431640625, -24.740264892578125, 138.00770568847656, 176.2764892578125, 10.776161193847656, 83.44747161865234, 155.79849243164062, 167.7807159423828, 20.170867919921875, 4.32904052734375, 97.45578002929688, 4.67449951171875, 56.615325927734375, 14.647171020507812, 55.267539978027344, -26.687095642089844, 166.518798828125, 40.23870849609375, 1.305419921875, 82.11911010742188, 4.167327880859375, -147.12265014648438, -92.2001953125, -29.533172607421875, 204.069580078125, 119.08033752441406, 92.78193664550781, -4.1065673828125, 0.0713348388671875, -18.451248168945312, 63.756439208984375, 182.42709350585938, 145.20318603515625, 22.640716552734375, 23.0103759765625, -69.29693603515625, 267.60760498046875, 37.317298889160156, 119.15557861328125, -2.8022613525390625, -2.098602294921875, -33.602783203125, -27.505126953125, 4.8625640869140625, 153.57699584960938, 111.4256591796875, 151.9783935546875, 4.8073883056640625, 145.06442260742188, 131.60699462890625, 134.25082397460938, 170.2115478515625, -5.479705810546875, -6.88484001159668, 66.3729248046875, 29.88067626953125, 191.65460205078125, -88.59809112548828, 121.42315673828125, -105.89364624023438, 80.024658203125, 0.0, -83.20880126953125, 18.554779052734375, 174.68487548828125, 174.21043395996094, 22.12517547607422, -0.276031494140625, 187.6551513671875, 109.94131469726562, 18.3013916015625, -133.6783447265625, 19.347915649414062, -0.8695068359375, 101.2940673828125, 155.46795654296875, -5.263664245605469, 105.94447326660156, 129.8841552734375, 184.61737060546875, 22.413818359375, 161.17132568359375, -25.90496826171875, 19.79913330078125, 27.406982421875, -11.764785766601562, -133.48094177246094, 40.7044677734375, 7.9862823486328125, 41.440582275390625, 123.96803283691406, 120.60575866699219, 40.165374755859375, 111.06014251708984, 17.592147827148438, 138.48812866210938, -10.293869018554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000316.npy"}
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 54.164005279541016, "std": 82.07059478759766, "min": -155.4583740234375, "p10": -28.946684265136717, "median": 34.056495666503906, "p90": 157.2610321044922, "max": 249.61285400390625, "pos_frac": 0.71875, "sample": [137.41102600097656, -130.93295288085938, 0.0, 6.740318298339844, 200.00799560546875, -29.634422302246094, 166.6719207763672, 77.42652893066406, 11.435901641845703, 142.02740478515625, 16.730865478515625, 157.089599609375, 45.838531494140625, -14.212051391601562, 157.96417236328125, -138.81674194335938, -29.526809692382812, 147.36776733398438, 133.339111328125, 11.016914367675781, 48.8031005859375, 11.654144287109375, 135.64617919921875, 121.27162170410156, 2.1445770263671875, 143.85914611816406, 12.054252624511719, 1.060516357421875, -9.25543212890625, 32.56884765625, -10.254886627197266, -49.49937438964844, 33.16636657714844, -5.97479248046875, -26.445068359375, 109.47569274902344, 116.09431457519531, 152.54701232910156, 143.44866943359375, -38.9620361328125, -64.70413208007812, 30.537841796875, 169.3509521484375, -137.985107421875, -134.88861083984375, 152.1104736328125, 176.47064208984375, 157.66104125976562, 11.921463012695312, 0.0, 111.887451171875, 94.2037353515625, 13.922225952148438, 192.43927001953125, -116.20278930664062, -11.4188232421875, 33.88249206542969, 39.350006103515625, 3.1389617919921875, 20.172576904296875, 16.11175537109375, 159.15863037109375, 154.388916015625, 117.61883544921875, 0.0, 14.494354248046875, 19.727813720703125, -1.246185302734375, 37.0107536315918, 82.57614135742188, 151.09939575195312, -5.97747802734375, -23.484283447265625, -155.4583740234375, 170.88998413085938, 82.23112487792969, 161.60205078125, -67.89666748046875, 31.97087287902832, -16.443161010742188, 95.98634338378906, 128.04823303222656, 118.5216064453125, -11.856315612792969, 61.95379638671875, 112.65216064453125, 50.04096984863281, -28.69805908203125, 115.62823486328125, 152.4669189453125, -1.9368743896484375, -10.023681640625, 165.91488647460938, 12.232917785644531, 6.5284881591796875, -0.28214073181152344, -0.0945587158203125, 137.61151123046875, 121.38790893554688, 249.61285400390625, 5.645713806152344, 120.55961608886719, 152.05645751953125, -22.123016357421875, -33.985260009765625, 40.97174072265625, 22.801116943359375, 37.799560546875, 0.28371238708496094, 84.910888671875, 111.21266174316406, 5.829557418823242, 139.65771484375, 6.777927398681641, 100.92904663085938, 144.71102905273438, 123.9510726928711, 56.6801872253418, 165.98968505859375, 8.501667022705078, -7.368623733520508, 34.230499267578125, -12.175308227539062, 154.7030487060547, 92.10218811035156, 127.08697509765625, -4.053951263427734, 138.03952026367188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000317.npy"}
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 55.01261901855469, "std": 78.08081817626953, "min": -166.44935607910156, "p10": -29.96503295898437, "median": 49.78248596191406, "p90": 153.69674987792968, "max": 196.37506103515625, "pos_frac": 0.7265625, "sample": [100.46064758300781, 24.229110717773438, 142.58697509765625, 139.955078125, 28.942962646484375, -9.20440673828125, 36.123321533203125, 47.20526123046875, 153.16714477539062, 44.10111999511719, 2.3702392578125, -16.49087905883789, 154.9324951171875, 39.03028869628906, 178.52276611328125, 26.32684326171875, 135.17181396484375, -17.695709228515625, 114.28357696533203, 103.74644470214844, -22.6058349609375, -18.585784912109375, 137.10037231445312, 157.99960327148438, -19.482879638671875, 38.09619140625, 125.7020263671875, 144.8931121826172, -120.76116943359375, 31.162506103515625, 1.3964385986328125, -55.24261474609375, 86.09689331054688, 76.93566131591797, 76.71846008300781, -28.4522705078125, 9.1348237991333, 62.124237060546875, 147.640869140625, 179.6975860595703, 93.27511596679688, 110.20315551757812, 78.62554931640625, 58.310089111328125, -1.1748199462890625, 7.958709716796875, -145.98092651367188, 121.36029815673828, -0.358184814453125, 87.64666748046875, -162.25927734375, 14.15810775756836, 90.33821105957031, 168.01528930664062, 126.00201416015625, 47.34284973144531, -17.603286743164062, 124.51992797851562, 149.21734619140625, -18.83648681640625, 104.50140380859375, 117.61788940429688, 175.33639526367188, 51.15771484375, -78.37126159667969, 48.74560546875, -48.725006103515625, 2.989044189453125, 162.03091430664062, -3.8528289794921875, 135.57818603515625, 30.635223388671875, 6.46051025390625, 91.21233367919922, 125.234375, 130.71566772460938, 87.83804321289062, 134.5603790283203, 56.811614990234375, -7.727630615234375, 48.421913146972656, 84.61965942382812, 32.5546875, 141.60235595703125, 88.48138427734375, -33.49481201171875, 78.84942626953125, 174.22802734375, -40.544288635253906, -8.337905883789062, -9.94317626953125, 46.097686767578125, 36.6754150390625, 112.91781616210938, 54.021484375, 51.78118896484375, 98.73007202148438, -62.30220031738281, 116.59628295898438, -17.365264892578125, 123.35330200195312, 196.37506103515625, 170.936767578125, 36.642120361328125, -38.101287841796875, 140.9776611328125, 50.819366455078125, 3.0184478759765625, -6.4056854248046875, -53.730621337890625, 158.05072021484375, 35.7152099609375, 60.96630859375, 157.63194274902344, 22.961257934570312, 150.23825073242188, 30.876026153564453, 147.7701416015625, -24.606597900390625, 121.43386840820312, -138.474853515625, -3.506134033203125, -1.7748336791992188, 31.48492431640625, 159.86624145507812, -8.885406494140625, -166.44935607910156, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000318.npy"}
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 59.310020446777344, "std": 71.37281799316406, "min": -162.70692443847656, "p10": -19.384843444824213, "median": 46.56308174133301, "p90": 149.76571655273438, "max": 206.9532470703125, "pos_frac": 0.8203125, "sample": [109.70323181152344, 2.718008041381836, 61.22869873046875, 47.643043518066406, 30.790618896484375, 206.9532470703125, 111.96463012695312, -46.2279052734375, 104.27153015136719, 26.418853759765625, -11.745956420898438, 14.12379264831543, 107.14277648925781, 171.08380126953125, 94.10150146484375, 164.8277587890625, 6.58807373046875, 0.0, 74.38125610351562, 56.23118591308594, 11.270671844482422, 30.111671447753906, 108.68978881835938, 12.517471313476562, 1.650146484375, 133.18338012695312, 5.93267822265625, 48.00401306152344, -17.266693115234375, 111.09213256835938, -2.335113525390625, 8.612991333007812, 131.47760009765625, 5.18182373046875, 118.5201644897461, 62.12115478515625, 6.149139404296875, 68.72479248046875, 143.47265625, 22.004180908203125, 3.8063697814941406, 143.5576934814453, 43.17999267578125, -111.18301391601562, -0.7755661010742188, 107.26947021484375, 171.79876708984375, 8.982330322265625, -49.796783447265625, 149.6378173828125, 143.52490234375, 122.28657531738281, 11.989471435546875, 15.891510009765625, 145.2906494140625, 24.6646728515625, 30.46075439453125, -75.20822143554688, 53.133644104003906, 141.48770141601562, -162.70692443847656, 64.62300872802734, 115.36815643310547, -79.71426391601562, 65.453857421875, -24.70055389404297, 74.93888854980469, 26.735986709594727, 78.423828125, 128.21676635742188, 2.1150588989257812, 111.94951629638672, 89.15838623046875, 147.88963317871094, -3.9183197021484375, -11.659820556640625, 121.885498046875, 121.4300537109375, 150.06414794921875, -24.327194213867188, 14.832267761230469, 146.96969604492188, 134.1474151611328, 27.479339599609375, 36.0338134765625, 0.7281494140625, 167.95846557617188, 205.05245971679688, 96.13449096679688, 18.59771728515625, -12.589813232421875, 26.825042724609375, 102.68179321289062, -53.05987548828125, 131.67132568359375, 55.78404235839844, 31.214492797851562, 165.50918579101562, -45.4383544921875, 163.441162109375, 33.94976806640625, 18.441009521484375, 171.68975830078125, 43.940277099609375, -40.36473083496094, 12.99365234375, 132.019775390625, 67.518798828125, 145.43609619140625, -5.957218170166016, 38.885833740234375, 17.333829879760742, -7.448902130126953, 142.06417846679688, -27.1968994140625, 124.57090759277344, 124.25314331054688, 182.270751953125, -29.648048400878906, 25.1119384765625, 147.458740234375, 13.750274658203125, 45.48311996459961, 1.4454917907714844, 150.24264526367188, 51.24920654296875, 29.92870330810547, 155.75], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000319.npy"}
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 54.52128219604492, "std": 87.6669921875, "min": -161.78448486328125, "p10": -38.28731079101562, "median": 49.72245407104492, "p90": 160.37841339111327, "max": 206.870361328125, "pos_frac": 0.734375, "sample": [120.19309997558594, 88.84812927246094, -14.391143798828125, -149.13345336914062, 131.7384033203125, 155.35430908203125, -89.91717529296875, 6.69189453125, 152.73355102539062, 154.226806640625, 19.49371337890625, 122.65869140625, 28.992767333984375, 12.54461669921875, 167.78201293945312, 93.00485229492188, -19.00439453125, 0.3815650939941406, 15.1051025390625, 23.247779846191406, 18.51678466796875, 145.15176391601562, 42.562744140625, 6.2574920654296875, 1.3298187255859375, -79.79808044433594, 173.64169311523438, 84.53611755371094, 71.29510498046875, 30.976951599121094, 107.51840209960938, 121.6348876953125, 161.74014282226562, 16.149154663085938, 53.5458984375, 144.88626098632812, -12.3037109375, 161.7713623046875, 11.25360107421875, -1.1526260375976562, 5.729104995727539, 51.74280548095703, 119.04705810546875, -25.342987060546875, 169.34877014160156, 184.71444702148438, 121.390625, -21.921417236328125, 78.10064697265625, 103.45909118652344, -34.7769775390625, 4.0023193359375, -141.30816650390625, -37.608978271484375, 169.460693359375, -23.15948486328125, 92.72244262695312, 17.437957763671875, 111.06251525878906, 158.74612426757812, 131.0125732421875, 155.968505859375, 5.273773193359375, 148.36846923828125, 158.53689575195312, -43.64324951171875, 151.81884765625, -17.91912841796875, 168.74911499023438, 144.584716796875, 35.33204650878906, -5.947174072265625, 26.068878173828125, -11.62506103515625, -161.78448486328125, 98.20155334472656, 11.23870849609375, 153.3370361328125, 7.406585693359375, 136.0106658935547, -133.52066040039062, 45.477813720703125, 2.260984420776367, -2.99517822265625, -24.403976440429688, 46.91674041748047, 115.44186401367188, 154.79449462890625, 97.47576904296875, -1.8672561645507812, -73.45133972167969, 54.5262451171875, 196.92083740234375, -11.1341552734375, 128.25653076171875, -29.158748626708984, 12.586135864257812, 83.32605743408203, -24.772659301757812, 166.27618408203125, 87.22088623046875, -39.870086669921875, 88.59979248046875, 159.79481506347656, 145.1470947265625, 206.870361328125, 139.2591094970703, -19.388259887695312, 110.18670654296875, -126.7548828125, 74.51103210449219, 57.60443115234375, -140.70211791992188, -4.847198486328125, 165.727294921875, 30.03582763671875, 150.81011962890625, 64.32081604003906, 139.08766174316406, 114.98121643066406, -35.932708740234375, 29.2677001953125, 29.270614624023438, -131.00009155273438, 185.34786987304688, 47.70210266113281, -155.43885803222656, 130.05633544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000320.npy"}
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 47.468170166015625, "std": 80.99230194091797, "min": -158.20993041992188, "p10": -37.88440093994141, "median": 28.138439178466797, "p90": 155.66167449951172, "max": 192.82659912109375, "pos_frac": 0.7265625, "sample": [148.730224609375, -77.23037719726562, 93.50607299804688, -151.98431396484375, 26.902557373046875, 173.9166259765625, 45.957275390625, 3.928314208984375, -59.88423156738281, 174.55252075195312, 130.12893676757812, 16.819107055664062, -85.3447036743164, 100.61881256103516, 95.22122192382812, 7.73260498046875, -6.196952819824219, 144.06613159179688, 171.38970947265625, 5.531768798828125, 90.01368713378906, -76.69277954101562, 2.475543975830078, -10.51593017578125, 166.05645751953125, 117.902099609375, -0.9663848876953125, 5.61944580078125, 112.78404235839844, 1.605926513671875, 128.8455810546875, 96.05996704101562, 106.84498596191406, 16.11370849609375, 41.006591796875, -57.64874267578125, -34.9761962890625, 88.69064331054688, 4.8962249755859375, 137.90020751953125, 80.26193237304688, 29.33538818359375, -7.6499176025390625, 65.81182861328125, 10.22259521484375, 168.2756805419922, -39.054168701171875, 151.9890594482422, 192.82659912109375, 114.94534301757812, 11.743873596191406, 19.915328979492188, -12.741764068603516, 42.333404541015625, -0.8850479125976562, 9.918228149414062, 153.60238647460938, -9.1280517578125, 154.62893676757812, -22.60797119140625, 22.948837280273438, 115.13349914550781, 29.07042694091797, 138.70571899414062, 66.56790161132812, 158.07139587402344, -34.09832763671875, 66.62115478515625, -72.58419036865234, -6.2462310791015625, -18.33331298828125, 57.18792724609375, -37.38307189941406, 171.81109619140625, -158.20993041992188, 119.48603820800781, 171.58682250976562, 187.96023559570312, 34.190216064453125, 130.9619140625, 21.286895751953125, -33.34423828125, 116.43798828125, 97.21119689941406, 8.445358276367188, 154.26483154296875, -12.888702392578125, 158.8625946044922, -89.41494750976562, 130.85525512695312, 17.816364288330078, 15.890420913696289, 114.11323547363281, -0.6517410278320312, 6.943141937255859, 186.85696411132812, 6.523681640625, 76.89590454101562, 27.206451416015625, 86.9212646484375, 131.68582153320312, -5.74798583984375, 101.1978759765625, 45.35772705078125, -8.020355224609375, 124.25820922851562, 5.548526763916016, 6.089385986328125, 1.2364578247070312, 1.14892578125, -13.677932739257812, -122.00469970703125, 10.008825302124023, -27.381103515625, 39.573631286621094, 175.83493041992188, -155.02529907226562, 150.95004272460938, 35.268646240234375, 77.64836883544922, 4.77587890625, 70.31317138671875, 2.450775146484375, -2.5801010131835938, 148.48300170898438, 148.57391357421875, 0.0, -111.84066772460938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000321.npy"}
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 57.300254821777344, "std": 80.05221557617188, "min": -268.70068359375, "p10": -30.67400207519531, "median": 50.13685607910156, "p90": 159.58057861328123, "max": 200.44964599609375, "pos_frac": 0.734375, "sample": [-51.926666259765625, -29.49395751953125, -64.2965087890625, 94.23164367675781, 39.35069274902344, 14.989997863769531, 53.923614501953125, -42.52471923828125, 144.0799560546875, 81.37811279296875, -15.546142578125, 3.31683349609375, 148.37326049804688, 11.37249755859375, 151.14971923828125, 124.7293701171875, 25.414154052734375, -7.27008056640625, 126.50302124023438, 73.7861328125, 27.36083984375, -19.40313720703125, 173.1118927001953, 1.241180419921875, 0.34175872802734375, 35.288787841796875, -30.438629150390625, 104.90650939941406, 6.311553955078125, 143.69229125976562, 68.10299682617188, 139.539306640625, 0.6565093994140625, -12.50714111328125, 134.0594024658203, 31.83001708984375, 27.804527282714844, 143.05670166015625, -24.168426513671875, 129.841064453125, 143.08969116210938, 122.67091369628906, -49.100921630859375, 15.468017578125, -5.91326904296875, 62.248992919921875, 81.57088470458984, -33.26054382324219, 113.78244018554688, -135.4298553466797, 130.95236206054688, 4.60528564453125, 148.3114471435547, 195.0897216796875, -9.287017822265625, 136.81784057617188, 105.26609802246094, -31.22320556640625, 40.01469421386719, 125.02078247070312, 163.51690673828125, 4.5301513671875, 195.00198364257812, 200.44964599609375, -21.722076416015625, 128.81246948242188, 92.56410217285156, 24.965240478515625, 116.27008056640625, 13.023941040039062, 179.9195556640625, 157.47909545898438, 124.4752197265625, 61.2431640625, 10.986572265625, 0.0, -63.5316162109375, 47.567901611328125, 75.75872802734375, -35.64030456542969, 162.17071533203125, 54.61841583251953, 142.71734619140625, 158.47052001953125, -24.1473388671875, 57.63818359375, 65.31158447265625, -41.51190185546875, -0.560394287109375, 121.29621887207031, 173.59661865234375, -4.727783203125, -16.14886474609375, -43.78277587890625, -18.34747314453125, 0.446258544921875, 100.98968505859375, 1.1652984619140625, -268.70068359375, 75.13714599609375, 148.62954711914062, 72.75360107421875, 52.705810546875, 177.52871704101562, 136.0775146484375, -32.49236297607422, -2.0179824829101562, 194.1655731201172, 192.4453125, 6.261970520019531, -12.1153564453125, 136.0480499267578, 20.510459899902344, 46.52091979980469, -6.349761962890625, 11.10165786743164, -24.228904724121094, 19.669647216796875, 20.79815673828125, 118.66276550292969, 87.413818359375, 181.62380981445312, 89.43283081054688, -7.58685302734375, 19.6669921875, 151.68157958984375, 71.55093383789062, 169.80926513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000322.npy"}
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 37.749839782714844, "std": 89.34197998046875, "min": -227.619384765625, "p10": -75.38298034667967, "median": 23.110292434692383, "p90": 152.07295989990234, "max": 194.87696838378906, "pos_frac": 0.671875, "sample": [13.737861633300781, 17.948089599609375, -128.52000427246094, -121.76336669921875, 130.91189575195312, 141.9815216064453, -35.31565856933594, 88.59619140625, 35.48016357421875, 187.15667724609375, -73.29713439941406, 120.02435302734375, 26.390869140625, -16.97979736328125, 5.9288330078125, 115.77176666259766, 164.54713439941406, -7.0373382568359375, 90.47721099853516, -29.43927001953125, 187.47491455078125, 109.98080444335938, 23.80194091796875, 131.99200439453125, 22.58416748046875, 169.9211883544922, 21.88922119140625, 135.08242797851562, -101.29742431640625, 149.1519775390625, -16.921875, -186.08306884765625, -7.460823059082031, 16.718765258789062, 0.08824348449707031, 110.10858154296875, 23.29596710205078, -11.821662902832031, 20.35125732421875, 18.23356056213379, 29.666030883789062, 136.25729370117188, -29.66436767578125, 98.87982177734375, 137.71734619140625, -8.176605224609375, -20.749801635742188, -227.619384765625, 157.01361083984375, 24.88599395751953, 9.241827011108398, 1.3408775329589844, -83.28347778320312, 9.068382263183594, 57.51310729980469, 84.19061279296875, 54.218170166015625, 145.85275268554688, 194.87696838378906, -26.555770874023438, 82.38958740234375, 89.88043212890625, 131.4251708984375, -22.183319091796875, 161.18145751953125, 24.18438720703125, 138.50442504882812, -17.71715545654297, -50.323486328125, 96.7353515625, 37.339599609375, 2.650360107421875, -70.58680725097656, 1.728546142578125, 130.62290954589844, 86.92811584472656, 179.1978759765625, -80.24995422363281, 174.12158203125, 4.779541015625, 128.30502319335938, -147.79794311523438, 14.600048065185547, 35.385528564453125, 18.9464111328125, 181.18399047851562, -44.158050537109375, 150.37313842773438, -142.66226196289062, -1.3800201416015625, 143.92568969726562, 37.54725646972656, -50.86866760253906, 110.71585845947266, -2.36785888671875, -9.574176788330078, -97.51204681396484, -15.3338623046875, -46.828857421875, -6.709747314453125, 17.58331298828125, -105.17953491210938, -12.946197509765625, 8.529251098632812, 31.109588623046875, 52.479461669921875, 22.924617767333984, -12.59954833984375, 138.6956787109375, 178.6063232421875, -2.6290817260742188, 60.5469970703125, 153.03025817871094, 146.15467834472656, -134.65687561035156, -142.8795166015625, 83.09559631347656, 26.293609619140625, 117.49407958984375, 24.42022705078125, 17.92595100402832, -6.55975341796875, 108.4598388671875, 151.66268920898438, 104.82937622070312, -33.59088134765625, 2.9008522033691406, 191.51666259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000323.npy"}
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 64.52811431884766, "std": 76.34659576416016, "min": -106.29885864257812, "p10": -15.739653015136716, "median": 44.06002616882324, "p90": 162.0561492919922, "max": 208.98651123046875, "pos_frac": 0.7890625, "sample": [91.71507263183594, 165.09991455078125, 16.43133544921875, 132.90884399414062, 22.5074462890625, 129.60272216796875, 3.2412872314453125, 3.997152328491211, 142.20779418945312, 32.875038146972656, 33.243896484375, 2.0537147521972656, -52.5494384765625, 168.6192626953125, 6.577880859375, 171.78900146484375, 131.474609375, 15.705745697021484, -14.5025634765625, 0.0, 8.040771484375, 6.831512451171875, 136.6202392578125, -1.3139266967773438, 3.653623580932617, 34.576385498046875, 143.99685668945312, -56.04627227783203, 137.47048950195312, -58.567604064941406, 6.761526107788086, 127.31034088134766, 154.07359313964844, 162.49072265625, 136.84815979003906, 130.950927734375, -13.053924560546875, 54.13311767578125, 83.10565185546875, 0.0, -6.806358337402344, 149.286865234375, 164.9837646484375, 116.50555419921875, 42.9556884765625, 136.07363891601562, 0.0, -3.600311279296875, 114.80584716796875, -18.23223876953125, -91.34718322753906, 44.56312561035156, 65.67561340332031, 21.660400390625, 118.0025634765625, 98.27349853515625, 11.3045654296875, 132.29974365234375, 170.00323486328125, 4.84637451171875, -45.046295166015625, 178.99459838867188, 9.268020629882812, 138.42764282226562, 109.1607437133789, -1.8863048553466797, 135.76980590820312, 139.62762451171875, 14.37200927734375, -103.52728271484375, 161.86990356445312, 64.94808197021484, -106.29885864257812, 135.51519775390625, -10.704833984375, -14.671401977539062, 33.939857482910156, 93.89752197265625, 10.249801635742188, 208.98651123046875, 189.29638671875, 7.548290252685547, -23.091018676757812, 69.32504272460938, 187.2357177734375, 156.32443237304688, 19.9493408203125, 20.207061767578125, -14.169891357421875, 2.37030029296875, 144.4127655029297, 38.357696533203125, 15.571510314941406, 4.1156005859375, -27.483016967773438, 55.0020751953125, 43.55692672729492, 163.20779418945312, 138.26373291015625, 28.11785888671875, -41.81798553466797, -4.871417999267578, 7.916656494140625, 46.41485595703125, -49.63189697265625, 203.1162109375, 184.1920166015625, 155.52349853515625, 70.410888671875, 7.32769775390625, 156.54632568359375, 116.38182067871094, -22.10497283935547, 134.68650817871094, 74.74301147460938, 135.18284606933594, 161.2611083984375, 134.8256378173828, -0.6209793090820312, 153.5988311767578, 139.29000854492188, 1.04986572265625, 140.39999389648438, 146.3502655029297, 147.86764526367188, 10.14324951171875, 5.89593505859375, 2.374359130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000324.npy"}
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 53.808387756347656, "std": 95.09180450439453, "min": -150.30984497070312, "p10": -50.129112243652344, "median": 31.853370666503906, "p90": 176.34833068847655, "max": 273.98876953125, "pos_frac": 0.671875, "sample": [74.5377197265625, 152.67904663085938, -122.83848571777344, 20.30224609375, 124.2769775390625, 22.34405517578125, 61.243438720703125, 155.18548583984375, 43.63865661621094, -2.709136962890625, -8.000244140625, 13.013702392578125, 148.0665283203125, 149.31613159179688, -8.239990234375, 5.05449104309082, 164.96470642089844, 122.3323974609375, 83.722900390625, 182.89627075195312, 145.33078002929688, 179.0142364501953, 116.0277099609375, -3.9811439514160156, 137.29812622070312, 3.1463394165039062, 14.153560638427734, 80.48062133789062, 135.99192810058594, 27.56988525390625, -108.90522766113281, 6.745018005371094, 85.88752746582031, 165.18443298339844, 170.50555419921875, 112.13716125488281, 22.14385986328125, 170.0543670654297, 35.338470458984375, -80.02037048339844, 135.19125366210938, -150.30984497070312, 161.95606994628906, -1.755584716796875, -81.54701232910156, -1.302276611328125, 273.98876953125, -134.6388702392578, 191.87953186035156, -5.209686279296875, 0.1279449462890625, 98.5421142578125, 167.00701904296875, 9.99481201171875, 9.274200439453125, -124.84404754638672, -10.148101806640625, -122.45940399169922, -11.133880615234375, -50.01039123535156, 74.59736633300781, 132.6175537109375, -0.7457504272460938, -8.281524658203125, 104.54891967773438, 48.25132751464844, 6.08843994140625, -126.41986083984375, 198.08877563476562, -11.59722900390625, 118.66445922851562, 21.00786590576172, 250.1396484375, -12.6470947265625, -9.78509521484375, 0.0585174560546875, -12.723175048828125, -12.146286010742188, 106.54776000976562, 152.02105712890625, -142.9957275390625, -10.754058837890625, -39.167633056640625, 179.70770263671875, -105.40240478515625, 69.70584106445312, 130.55233764648438, 39.30029296875, 46.988861083984375, -21.35174560546875, -50.4061279296875, 126.85603332519531, 176.398681640625, -31.058853149414062, 1.6862068176269531, -149.33566284179688, -17.424957275390625, 176.32675170898438, 51.5584716796875, 140.88265991210938, 178.14175415039062, 37.27998352050781, 157.8901824951172, 206.129150390625, 167.99795532226562, -37.067413330078125, 5.707857131958008, 17.535850524902344, 28.368270874023438, -2.062744140625, -20.828460693359375, 160.421630859375, 49.58329772949219, 144.0277099609375, 8.33502197265625, 131.98370361328125, 136.03305053710938, 194.5352783203125, 15.374908447265625, 0.9509201049804688, -19.952499389648438, -32.525482177734375, 210.5780029296875, -25.013671875, -15.129684448242188, 210.29843139648438, 82.16824340820312, 155.86740112304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000325.npy"}
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 53.90275573730469, "std": 90.15849304199219, "min": -242.63824462890625, "p10": -54.47074584960937, "median": 62.92803192138672, "p90": 155.5963134765625, "max": 187.8493194580078, "pos_frac": 0.7109375, "sample": [150.3961181640625, 138.84280395507812, 135.02357482910156, 183.16200256347656, 130.1059112548828, 67.6839599609375, -47.8389892578125, -34.539886474609375, 137.718994140625, -112.51579284667969, 162.76712036132812, -12.610626220703125, 142.7561492919922, 147.60646057128906, 24.230995178222656, 2.9279441833496094, 5.485591888427734, 159.94717407226562, 187.8493194580078, 10.6409912109375, 73.434326171875, 145.63450622558594, 6.404022216796875, -22.836212158203125, 65.75527954101562, 113.01394653320312, 125.31626892089844, 93.89764404296875, 20.910449981689453, 17.934646606445312, -126.67781829833984, 155.1407470703125, 0.2880859375, 68.75755310058594, -134.69528198242188, -98.49053955078125, 21.85216522216797, 94.51337432861328, -32.657867431640625, 140.62771606445312, 85.9573974609375, 117.69862365722656, -20.352569580078125, 148.4481201171875, 42.045562744140625, 104.78205871582031, -41.158294677734375, 145.03094482421875, -33.9351806640625, 171.9256134033203, 130.0224609375, 104.56672668457031, 150.12051391601562, 17.76000213623047, 153.36825561523438, -131.96292114257812, -37.09437561035156, 14.313385009765625, 134.88247680664062, -10.362590789794922, 133.80410766601562, 118.57478332519531, 22.3277587890625, 89.09173583984375, 0.0, -242.63824462890625, 14.497718811035156, -1.9447021484375, 134.95712280273438, -70.62310791015625, 15.744613647460938, 168.78421020507812, 184.9609375, 89.99008178710938, 51.439483642578125, -56.30059814453125, 34.4381103515625, -92.513427734375, 31.925460815429688, 125.64450073242188, -1.955841064453125, 166.36602783203125, -3.6854248046875, 33.24214172363281, 13.375038146972656, 145.10565185546875, -36.34649658203125, -116.49063110351562, 116.65943908691406, 13.402196884155273, 142.89556884765625, 17.06195068359375, 170.61090087890625, -9.225677490234375, 94.677734375, -12.3031005859375, 113.25138092041016, 156.6593017578125, 24.62603759765625, 137.46926879882812, 163.9783935546875, 0.0, 139.633056640625, 64.44758605957031, 61.408477783203125, 136.75967407226562, -27.506195068359375, 154.43995666503906, 109.53189086914062, 0.0, 138.19064331054688, -160.4259033203125, -25.935546875, -107.2786636352539, 66.08908081054688, -124.98321533203125, 164.17431640625, -9.10919189453125, 152.0750732421875, 18.066646575927734, 87.6864242553711, 25.37762451171875, -53.6865234375, -36.287681579589844, 141.58763122558594, 158.84494018554688, 51.47650146484375, 139.62258911132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000326.npy"}
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 61.80979537963867, "std": 87.38983154296875, "min": -164.07464599609375, "p10": -27.242212677001948, "median": 40.94671630859375, "p90": 171.402880859375, "max": 226.59912109375, "pos_frac": 0.78125, "sample": [36.86975860595703, 137.00233459472656, 130.4810791015625, 10.858802795410156, 202.22552490234375, -151.97988891601562, 116.86434936523438, 149.0126953125, 8.104080200195312, 119.9259033203125, 24.778684616088867, 188.50823974609375, -6.62615966796875, -146.23287963867188, 22.892555236816406, -19.3929443359375, -64.09274291992188, 10.172622680664062, 181.76101684570312, 178.10427856445312, 118.8137435913086, 107.9577865600586, 21.02386474609375, 71.33425903320312, 129.98355102539062, -106.70925903320312, 11.873764038085938, 116.76780700683594, 10.037261962890625, 131.67735290527344, 42.110015869140625, 172.80691528320312, -5.103816986083984, -17.285850524902344, 72.84040832519531, 118.65228271484375, 174.89297485351562, 9.21661376953125, 226.59912109375, 111.08279418945312, -21.50732421875, -6.261350631713867, -8.497451782226562, 120.57685852050781, 34.35240173339844, -13.48529052734375, 172.4716796875, 39.665130615234375, 122.66667175292969, 10.291656494140625, 156.23651123046875, 5.27655029296875, -20.654449462890625, -164.07464599609375, 128.5042724609375, 23.801177978515625, 143.68450927734375, 136.8711395263672, 183.2972412109375, 132.69735717773438, 39.25103759765625, 132.30078125, 119.85148620605469, 45.85357666015625, 170.30975341796875, 154.74481201171875, 177.1627197265625, 20.387542724609375, 89.9656982421875, -23.034088134765625, 33.85992431640625, 4.316497802734375, -11.442138671875, -30.13280487060547, 38.44110107421875, 63.893768310546875, 31.082870483398438, 65.74005126953125, 176.228515625, 11.49114990234375, -26.003387451171875, 3.587310791015625, 29.190673828125, 122.2568359375, 19.012603759765625, 170.94482421875, 16.845611572265625, 0.0, 160.833984375, 63.44615936279297, 25.038604736328125, 164.95947265625, 193.05880737304688, 25.868011474609375, 135.5406494140625, 26.657630920410156, 146.69381713867188, 34.63470458984375, 147.14097595214844, 150.949951171875, 65.96292114257812, 17.339035034179688, -153.4909210205078, -44.653419494628906, 113.050048828125, 123.41344451904297, -102.39834594726562, 39.783416748046875, 64.53509521484375, 139.2422332763672, -140.86749267578125, 135.04541015625, 174.68017578125, 10.0709228515625, -38.840240478515625, 145.64886474609375, -5.0590972900390625, 168.56564331054688, 1.3614044189453125, -5.3288726806640625, 164.48355102539062, 153.7431640625, 141.7364501953125, 6.4165191650390625, -122.548095703125, 153.0679931640625, 22.961029052734375, -60.9178466796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000327.npy"}
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 55.737403869628906, "std": 81.7020034790039, "min": -153.40472412109375, "p10": -25.420745849609375, "median": 35.56200408935547, "p90": 159.87423706054685, "max": 224.92413330078125, "pos_frac": 0.78125, "sample": [31.056861877441406, 130.78652954101562, 153.9422149658203, 39.79425048828125, 194.73785400390625, 27.16619873046875, 0.270294189453125, 131.72132873535156, 133.04165649414062, 127.1951904296875, 120.51187133789062, 89.20628356933594, 4.3287353515625, 123.96627807617188, 69.54891204833984, 0.2713451385498047, 10.193603515625, -8.076629638671875, 49.0972900390625, 26.306106567382812, 35.92478942871094, 95.57505798339844, 102.77490234375, -26.454208374023438, 187.63404846191406, 39.25274658203125, -2.565826416015625, 45.782867431640625, 111.6756591796875, 126.21204376220703, -3.984771728515625, 128.8717041015625, 37.28240966796875, 166.47210693359375, 203.25201416015625, 103.26903533935547, 178.86085510253906, 7.9329833984375, 180.79727172851562, -0.5350055694580078, 4.1241455078125, 142.18722534179688, 89.31362915039062, 0.5163116455078125, 6.9373779296875, -8.070436477661133, -25.317291259765625, 34.56878662109375, -24.53662109375, 145.4945068359375, 40.11260986328125, 9.0552978515625, -25.662139892578125, 174.614501953125, 9.160537719726562, 34.2755126953125, 10.9517822265625, 11.292259216308594, 70.26226043701172, -33.50584411621094, 25.868392944335938, 26.998321533203125, 166.53170776367188, 157.4036865234375, 130.16954040527344, 165.63885498046875, 129.3450927734375, -74.22857666015625, 142.42889404296875, -102.25382995605469, 15.893417358398438, 129.9610595703125, 128.96875, 133.12184143066406, 25.296539306640625, 216.75408935546875, 185.26138305664062, 23.337158203125, 0.8917236328125, 139.24359130859375, -153.40472412109375, 10.338844299316406, 147.96841430664062, 45.13041687011719, -121.88070678710938, 9.17236328125, 128.9497528076172, 130.36343383789062, 128.44984436035156, -23.9378662109375, -16.64093017578125, 172.383056640625, 5.285800933837891, -1.449462890625, 5.868896484375, -22.309341430664062, 3.3861846923828125, 22.796905517578125, -120.82064819335938, 31.731664657592773, 54.07777404785156, 35.19921875, 122.0293197631836, 3.3441162109375, -2.19622802734375, 149.31478881835938, 148.696044921875, -152.37567138671875, -2.550079345703125, -25.992034912109375, 39.010528564453125, 21.487274169921875, 150.33499145507812, -45.2786865234375, 30.443023681640625, 23.255935668945312, -0.1931610107421875, 140.23446655273438, -5.039482116699219, -142.47267150878906, 12.49835205078125, 157.05410766601562, 55.97047424316406, -85.35140991210938, 68.03192138671875, 224.92413330078125, 129.5517120361328, 43.49457550048828], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000328.npy"}
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 46.553245544433594, "std": 93.45848083496094, "min": -179.04547119140625, "p10": -50.06610717773437, "median": 27.98560333251953, "p90": 166.36212768554685, "max": 222.1628875732422, "pos_frac": 0.65625, "sample": [-13.653411865234375, 8.1785888671875, -5.7042694091796875, 41.724456787109375, 2.21337890625, -15.219310760498047, 6.8217926025390625, 27.540435791015625, 3.6657371520996094, 35.850372314453125, 150.68849182128906, -16.98279571533203, -161.449462890625, 123.01937866210938, 109.98722839355469, 183.89877319335938, -7.89117431640625, -48.612762451171875, 148.94493103027344, 123.40563201904297, 89.89108276367188, 21.0338134765625, 162.71234130859375, -11.22027587890625, 132.68533325195312, 161.80369567871094, -141.79132080078125, 127.3984375, 181.29981994628906, 117.48370361328125, 1.29656982421875, 23.85443115234375, -3.1820068359375, 163.77398681640625, -40.58880615234375, 42.516571044921875, 110.41459655761719, 120.91400146484375, 18.503158569335938, 125.35311889648438, 6.38775634765625, 176.168212890625, 0.315399169921875, -44.053466796875, 128.00860595703125, -99.0230712890625, -16.305587768554688, 162.36688232421875, 14.44403076171875, -178.7421875, -4.808538436889648, 176.60986328125, -53.457244873046875, -21.2373046875, 121.54278564453125, 61.42942810058594, 51.429229736328125, 32.314117431640625, 29.569931030273438, -137.8656005859375, 2.7608108520507812, 155.9970703125, -2.7215576171875, -44.66719055175781, -0.0532684326171875, -5.140634536743164, -0.5880928039550781, -22.675079345703125, 206.78146362304688, 50.45795440673828, 28.430770874023438, -0.463897705078125, 143.1073760986328, 216.2802734375, 15.087003707885742, 96.54763793945312, 66.60598754882812, -25.218551635742188, -112.9530029296875, 154.79434204101562, 142.59071350097656, -16.292205810546875, -179.04547119140625, 123.27442932128906, -14.882904052734375, 178.94700622558594, 118.07200622558594, 2.360595703125, 120.4425048828125, 0.58154296875, 192.787109375, -0.8483810424804688, -119.90098571777344, 179.95730590820312, 136.80606079101562, -135.52960205078125, 222.1628875732422, 172.401123046875, -3.1176986694335938, 135.82369995117188, -4.8787994384765625, 11.916290283203125, 123.79946899414062, 31.21875, -0.7636489868164062, 1.8295516967773438, 217.0380859375, 60.580078125, 147.9256134033203, 216.85107421875, -17.772735595703125, 123.89002990722656, 135.87625122070312, -7.966552734375, 36.4031982421875, -88.79785919189453, 25.605010986328125, -19.807281494140625, 41.7105712890625, -13.9559326171875, 90.91481018066406, 137.107421875, 0.9902362823486328, 76.04734802246094, 142.37777709960938, 54.614532470703125, -88.248046875, -160.3201904296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000329.npy"}
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 39.121177673339844, "std": 86.61874389648438, "min": -145.58978271484375, "p10": -90.11002349853514, "median": 27.53826904296875, "p90": 164.48461303710937, "max": 267.52252197265625, "pos_frac": 0.7265625, "sample": [-12.441604614257812, 141.27601623535156, 18.880462646484375, -124.351318359375, 87.53577423095703, 131.44021606445312, 31.294219970703125, 184.16510009765625, 133.64698791503906, 48.64979553222656, -74.94207763671875, 53.30195617675781, 7.1226806640625, -145.58978271484375, 90.65582275390625, -35.062255859375, 164.64410400390625, 30.121490478515625, 3.06341552734375, -2.411163330078125, 164.89952087402344, -137.72874450683594, -35.09332275390625, 9.07464599609375, 1.8399658203125, 120.5264892578125, 23.982666015625, -14.98172378540039, 9.6068115234375, -2.74951171875, 61.179901123046875, 37.557952880859375, 177.9688720703125, 112.90736389160156, 8.377510070800781, -6.5196533203125, -1.18194580078125, -100.11697387695312, -87.06069946289062, 17.02392578125, 45.38031005859375, 175.68768310546875, 36.53202819824219, 42.366546630859375, 28.1131591796875, 128.04428100585938, 17.21800994873047, 198.21380615234375, 14.306404113769531, 100.14073181152344, 2.5907516479492188, 160.641845703125, 65.99673461914062, -133.55477905273438, 6.52386474609375, 24.40008544921875, -126.90338897705078, 36.84039306640625, 32.706024169921875, 37.86248779296875, 16.803176879882812, 177.03741455078125, 13.402164459228516, 108.07485961914062, -117.81852722167969, 2.56500244140625, 137.52835083007812, -99.4508056640625, 5.947048187255859, 10.989936828613281, 44.814544677734375, -97.22511291503906, 185.39788818359375, 111.91607666015625, -3.11956787109375, 109.96383666992188, -136.23458862304688, 179.43682861328125, -26.54876708984375, 5.254791259765625, 30.080322265625, 98.46661376953125, -8.320823669433594, -12.6810302734375, 26.96337890625, 4.0690155029296875, 35.17889404296875, 160.25848388671875, 164.416259765625, 34.019317626953125, 178.1832733154297, 9.20037841796875, 111.79029083251953, 42.460662841796875, 12.296539306640625, -22.737564086914062, 18.611785888671875, 13.873126983642578, 107.78361511230469, 143.49777221679688, 181.995849609375, 35.737884521484375, -126.81765747070312, 95.38746643066406, -121.59963989257812, 154.3936767578125, -14.542236328125, 25.444442749023438, -31.488327026367188, 95.21795654296875, 148.52139282226562, -113.51630401611328, -0.5322799682617188, -47.66839599609375, 0.1661376953125, 188.71087646484375, 39.16548156738281, 43.424522399902344, 0.342376708984375, 142.1817169189453, -22.64044189453125, 267.52252197265625, 82.19143676757812, -19.5255126953125, -26.6309814453125, 143.43011474609375, 96.21788787841797, 32.656097412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000330.npy"}
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 49.680728912353516, "std": 83.81196594238281, "min": -175.9131317138672, "p10": -37.33964920043945, "median": 29.523452758789062, "p90": 169.23526611328126, "max": 198.37725830078125, "pos_frac": 0.75, "sample": [-0.555999755859375, 192.62591552734375, 170.69427490234375, -108.30856323242188, 197.65374755859375, 97.64956665039062, 46.9635009765625, -10.625396728515625, 47.76318359375, 115.38589477539062, -4.1212615966796875, -82.44013977050781, 120.89041137695312, -79.5244140625, 190.5665283203125, -19.011871337890625, 6.559967041015625, 4.223390579223633, -9.92730712890625, -48.927520751953125, 14.835304260253906, 45.502166748046875, -31.473648071289062, 16.31658935546875, 8.564529418945312, 162.8781280517578, 143.36239624023438, 33.41600036621094, 7.621185302734375, 132.2987518310547, 34.576324462890625, 44.86175537109375, -14.285369873046875, 30.19329833984375, 28.853607177734375, 168.62777709960938, 3.2340087890625, 101.19595336914062, 176.90505981445312, 133.13320922851562, -46.26580047607422, 14.048736572265625, -3.4605636596679688, -167.24098205566406, 4.781339645385742, 7.881145477294922, 164.1244659423828, 23.714508056640625, 89.1099853515625, 0.0, 34.626678466796875, -2.142852783203125, 137.72030639648438, -2.43597412109375, 9.039764404296875, 82.78729248046875, 14.59820556640625, 153.45120239257812, 40.822662353515625, 58.04150390625, 26.575164794921875, 53.34735107421875, -38.554656982421875, -79.00111389160156, 182.20657348632812, 60.14558792114258, 170.65274047851562, 119.34312438964844, 7.119194030761719, 43.1083869934082, 4.55316162109375, 21.074432373046875, 157.40765380859375, -24.67779541015625, 46.74846267700195, 26.23027801513672, -38.52410888671875, 5.497039794921875, 9.759071350097656, -36.83202362060547, 60.3267822265625, 31.302886962890625, 125.06036376953125, 37.43931579589844, -175.9131317138672, 192.5546875, 5.19061279296875, 164.18112182617188, -16.208892822265625, 30.976959228515625, 186.24600219726562, 127.78924560546875, 117.21207427978516, 189.05926513671875, 52.79779052734375, 172.5399627685547, 162.7461395263672, 5.977294921875, -17.716064453125, -8.518386840820312, -68.8126220703125, -6.4405517578125, 156.04698181152344, 126.37135314941406, -147.22586059570312, 26.062408447265625, 198.37725830078125, 1.466684341430664, 136.68145751953125, -1.8691959381103516, 2.1473388671875, 15.86669921875, 144.31130981445312, 11.772705078125, 120.81031036376953, 134.50576782226562, 9.3602294921875, 69.8948974609375, 18.456863403320312, 134.15200805664062, -154.82763671875, 23.008277893066406, 185.57608032226562, 146.94273376464844, 110.24276733398438, -31.05084228515625, 141.73721313476562, 14.993659973144531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000331.npy"}
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 62.460174560546875, "std": 89.27235412597656, "min": -186.6382598876953, "p10": -40.67750701904296, "median": 57.41398239135742, "p90": 166.39013061523437, "max": 242.35293579101562, "pos_frac": 0.7578125, "sample": [100.44389343261719, 34.6185302734375, -2.0125503540039062, 193.19464111328125, 131.2234344482422, 140.77264404296875, 108.11604309082031, 58.648765563964844, 193.65130615234375, 137.13681030273438, -1.7476348876953125, 9.915628433227539, 8.778385162353516, -5.0130615234375, 108.43171691894531, 112.87347412109375, -35.86480712890625, 21.423233032226562, 150.31005859375, -67.51008605957031, 132.50442504882812, -31.02252197265625, 46.70465087890625, 43.358154296875, 0.6866531372070312, 44.33477783203125, -108.10870361328125, 4.9544677734375, 13.372085571289062, 11.445068359375, 155.494873046875, 173.72845458984375, 5.9659423828125, 4.126129150390625, 161.2556915283203, 124.83405303955078, 143.9952392578125, 178.81683349609375, 98.9818115234375, 121.64314270019531, 64.17439270019531, 8.907257080078125, 139.0206298828125, 1.2396049499511719, 149.81640625, 38.73974609375, 168.21524047851562, 151.70021057128906, -3.808990478515625, 25.53137969970703, 155.20950317382812, 187.08065795898438, -113.00128173828125, -24.65435791015625, 189.44451904296875, 132.833740234375, -90.9312744140625, 100.96629333496094, 157.28460693359375, -180.65936279296875, 47.24028778076172, 165.60794067382812, 149.93295288085938, -1.7597599029541016, 13.960426330566406, 8.915130615234375, 119.2540512084961, 22.94903564453125, 13.410757064819336, -45.068603515625, 4.47442626953125, 146.830810546875, 119.01078796386719, 21.367752075195312, -4.7577056884765625, 176.35687255859375, 88.7161865234375, 127.28515625, -56.94886779785156, 120.11915588378906, -38.79560852050781, 125.73574829101562, -27.637191772460938, 2.3885498046875, 120.009765625, -114.60736083984375, 112.8917236328125, 20.696258544921875, 7.3790283203125, 123.48957824707031, 159.51025390625, 114.31999206542969, -103.29651641845703, 160.57449340820312, 75.55609130859375, -4.376220703125, 148.65060424804688, -24.832626342773438, -15.712478637695312, 49.841896057128906, 168.80703735351562, 161.00872802734375, -118.9112548828125, -49.9921875, 242.35293579101562, 193.459228515625, 133.43646240234375, 19.9317626953125, 6.756317138671875, 16.71538543701172, 164.11285400390625, 94.61349487304688, 129.12277221679688, 186.88436889648438, 56.17919921875, 139.81976318359375, -26.050796508789062, 153.18191528320312, -10.170948028564453, 125.12405395507812, 120.66961669921875, 180.32586669921875, -1.226806640625, -115.20814514160156, 0.0, 12.22821044921875, -186.6382598876953, 88.11727905273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000332.npy"}
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 45.80164337158203, "std": 79.51985168457031, "min": -143.22174072265625, "p10": -48.82442474365234, "median": 32.619834899902344, "p90": 151.51417236328126, "max": 178.76405334472656, "pos_frac": 0.703125, "sample": [124.34966278076172, -105.66970825195312, -32.85960006713867, 171.71978759765625, 176.00332641601562, -43.742431640625, 85.4571533203125, 100.20770263671875, 41.78987121582031, 70.7713623046875, -55.22248840332031, 11.0750732421875, -19.271240234375, 160.01873779296875, 149.2841796875, 144.01455688476562, 121.76358795166016, 20.792877197265625, 111.64175415039062, 45.964111328125, -23.550888061523438, 32.42118835449219, 12.24505615234375, 151.8753662109375, 143.76502990722656, 96.4156265258789, 57.715240478515625, 82.78924560546875, 139.01373291015625, -2.652099609375, 28.50390625, 178.76405334472656, -12.244476318359375, 74.489990234375, -143.22174072265625, 4.386146545410156, 53.85955810546875, -12.359725952148438, 140.97732543945312, 138.75482177734375, 55.87017822265625, 18.883224487304688, -1.20379638671875, 14.2882080078125, 109.1116943359375, 145.44139099121094, 4.014913558959961, 162.92050170898438, 23.87249755859375, -76.41517639160156, -67.00201416015625, 72.29708862304688, 0.0, -5.536041259765625, 142.77137756347656, 10.76995849609375, 156.55731201171875, 170.75372314453125, 147.90048217773438, 61.40092468261719, -2.02252197265625, 45.35881805419922, -46.0823974609375, 142.24627685546875, 93.3619384765625, -40.98112487792969, 153.18621826171875, 178.7276611328125, 74.57261657714844, 32.8184814453125, -117.04476928710938, 80.43673706054688, 32.994293212890625, -67.994384765625, 7.854957580566406, 138.44435119628906, -1.24578857421875, -2.39105224609375, 4.85687255859375, 130.71046447753906, 40.573448181152344, 41.36415100097656, 31.839385986328125, 162.8834228515625, -56.470062255859375, 150.30198669433594, 155.44036865234375, -7.192626953125, -89.3218994140625, 152.0487518310547, 39.06387710571289, 93.39828491210938, -6.566917419433594, 109.29627990722656, -25.48157501220703, -18.15301513671875, 22.492889404296875, 22.44378662109375, -130.85765075683594, 117.23844909667969, -16.453216552734375, 124.31952667236328, -137.81478881835938, -13.827987670898438, 121.0316162109375, 139.5100860595703, 1.73663330078125, 52.583892822265625, 151.359375, 10.021759033203125, 132.282958984375, 3.7942028045654297, 131.20712280273438, 12.96728515625, 21.91629981994629, -6.99627685546875, -126.87232971191406, 81.60901641845703, -13.242431640625, 25.723052978515625, -8.050140380859375, 3.55316162109375, -118.28657531738281, 15.413246154785156, -2.19110107421875, 48.32823181152344, 17.756927490234375, 2.3505992889404297], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000333.npy"}
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 54.48392105102539, "std": 91.9120864868164, "min": -195.6454315185547, "p10": -31.523242187499996, "median": 36.38191604614258, "p90": 171.32452239990235, "max": 293.61712646484375, "pos_frac": 0.734375, "sample": [130.9625244140625, 134.81826782226562, -28.837188720703125, -11.59698486328125, 0.0, -8.204620361328125, 13.789077758789062, -17.426971435546875, 36.381813049316406, 3.2979202270507812, 27.36444091796875, 23.232315063476562, 13.82843017578125, 126.05778503417969, -121.80569458007812, 137.63217163085938, -4.57415771484375, 173.22299194335938, 37.79490661621094, -16.34661865234375, 123.6621322631836, 67.1707763671875, 209.25404357910156, 171.01808166503906, -91.5799560546875, 36.38201904296875, 20.670394897460938, 26.7469482421875, -3.774200439453125, 17.388565063476562, 83.9083251953125, -10.553104400634766, 0.09430313110351562, 135.668701171875, 137.85952758789062, 43.19384765625, -3.2769393920898438, 129.46319580078125, -174.96078491210938, 165.34783935546875, -0.8485336303710938, 5.62530517578125, 21.2689208984375, 19.824951171875, 24.024269104003906, 51.83990478515625, 11.148681640625, -83.532958984375, 137.52879333496094, -4.47308349609375, 93.61752319335938, -130.50662231445312, 160.8867645263672, 127.2757568359375, 197.34951782226562, 141.34930419921875, -4.370777130126953, 123.112060546875, 9.4227294921875, 20.25537109375, 14.399894714355469, 7.744468688964844, 198.9583740234375, 71.80636596679688, -38.47636413574219, -128.48587036132812, 48.237274169921875, -168.0908203125, 2.326812744140625, -30.64715576171875, 185.4176025390625, 41.59844970703125, -18.308433532714844, 189.36203002929688, 146.42864990234375, 134.34622192382812, 101.25999450683594, 18.39513397216797, 12.0509033203125, 137.94606018066406, 35.732330322265625, 76.21412658691406, -4.806396484375, 133.77491760253906, 124.13594055175781, 181.8875274658203, 71.98280334472656, 54.481292724609375, 133.683349609375, 24.2198486328125, 148.14068603515625, 141.23175048828125, 147.28248596191406, -18.456314086914062, 2.44183349609375, 31.4044189453125, 0.0546722412109375, -21.491607666015625, 162.27755737304688, 183.8077392578125, 172.03955078125, 20.676788330078125, 11.137809753417969, -134.07855224609375, 152.13388061523438, -33.56744384765625, 98.2857437133789, 131.25753784179688, 175.78189086914062, 146.95565795898438, -195.6454315185547, 202.1824951171875, 163.3507080078125, 0.4726524353027344, 132.9130859375, -157.57351684570312, -11.5435791015625, 98.51492309570312, -27.469070434570312, 56.88818359375, 38.0474853515625, 293.61712646484375, 149.7706298828125, 89.48709106445312, 71.71063232421875, -11.376419067382812, -50.503662109375, 202.13494873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000334.npy"}
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 49.880889892578125, "std": 92.432373046875, "min": -175.9429931640625, "p10": -64.95077667236328, "median": 39.464725494384766, "p90": 165.27733459472657, "max": 218.75527954101562, "pos_frac": 0.703125, "sample": [-106.28291320800781, -150.23883056640625, 39.36078643798828, -45.69978332519531, 117.871337890625, -99.87954711914062, 186.56829833984375, -11.5699462890625, 26.28997802734375, 143.72251892089844, 81.59695434570312, -155.2242431640625, 43.5806884765625, -12.87491226196289, -3.5050277709960938, 21.77880859375, 2.3354263305664062, -2.3581714630126953, -30.439178466796875, -67.72067260742188, -17.122573852539062, 181.07699584960938, -3.561065673828125, 6.6757354736328125, 15.607452392578125, -2.1280517578125, 70.93505859375, 68.20975494384766, -34.091217041015625, 142.22503662109375, -175.9429931640625, 194.64151000976562, -24.509719848632812, 73.65408325195312, 51.41778564453125, 198.63223266601562, 7.44940185546875, 136.9095458984375, 159.11354064941406, 60.19750213623047, 164.36795043945312, 151.89944458007812, 157.94940185546875, 128.24156188964844, 187.341552734375, 165.07363891601562, 64.05035400390625, 134.0633544921875, 169.24261474609375, 169.36090087890625, -6.827125549316406, 132.0586395263672, 17.739944458007812, 125.78968048095703, 135.15750122070312, 138.17794799804688, 197.95388793945312, 125.6702880859375, 3.4568328857421875, 39.038818359375, 39.56866455078125, -148.71505737304688, -26.439346313476562, -63.93194580078125, -119.7684326171875, 157.058837890625, 101.59245300292969, 156.31326293945312, 97.30036163330078, 18.40093994140625, 46.99688720703125, 1.2979583740234375, 11.934616088867188, 132.81753540039062, 200.14968872070312, 142.1060791015625, 4.772682189941406, -67.32804870605469, 10.5609130859375, 26.94139289855957, 88.04084777832031, 134.50216674804688, -12.75946044921875, 115.81103515625, 167.5734100341797, -16.298316955566406, -56.0185546875, 5.9511260986328125, 89.04827880859375, 44.98883056640625, -40.650299072265625, 1.9657440185546875, 136.90728759765625, 3.3157424926757812, 0.417999267578125, -16.264175415039062, 14.907745361328125, 22.11553192138672, 86.58572387695312, -4.4262542724609375, 188.62930297851562, -142.90908813476562, 29.96978759765625, 82.01915740966797, 147.60955810546875, 97.12782287597656, 0.0, 158.98483276367188, 165.75262451171875, 136.39865112304688, -107.29998779296875, 145.7350616455078, 147.48477172851562, -125.97779846191406, -138.7572021484375, 218.75527954101562, 8.210968017578125, -3.4445571899414062, 8.219770431518555, -31.629486083984375, 13.904296875, 114.03242492675781, 56.695343017578125, 111.26075744628906, -58.50221252441406, -15.83660888671875, 104.14328002929688, 100.32472229003906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000335.npy"}
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 46.15558624267578, "std": 81.03245544433594, "min": -155.564208984375, "p10": -35.87300567626953, "median": 23.407705307006836, "p90": 162.17188568115233, "max": 256.8560791015625, "pos_frac": 0.6640625, "sample": [62.1160888671875, 6.82269287109375, 75.35293579101562, -15.97607421875, 147.82827758789062, 31.45672607421875, -21.218292236328125, 131.1820526123047, 132.52987670898438, 157.07015991210938, -14.492691040039062, 24.76666259765625, 53.910667419433594, 83.37529754638672, 2.9262351989746094, 61.28662109375, -2.5201416015625, 183.34275817871094, 8.190910339355469, 161.33413696289062, 187.3947296142578, 18.863815307617188, 115.73359680175781, 141.81890869140625, -54.31756591796875, -16.8389892578125, 9.292022705078125, 19.420921325683594, 13.40911865234375, 37.731781005859375, 152.46200561523438, 31.73419189453125, -3.7752914428710938, 146.20947265625, 181.06390380859375, 164.1266326904297, 56.652374267578125, 215.57406616210938, 17.079124450683594, 9.609375, -74.99455261230469, 141.10752868652344, -42.536285400390625, 150.01971435546875, 53.02532958984375, -0.59112548828125, -0.15387725830078125, 136.96408081054688, -37.84901428222656, 23.870655059814453, 88.07373809814453, -0.501708984375, 19.568878173828125, 105.90594482421875, 36.4488525390625, -52.901336669921875, -15.787811279296875, -14.26739501953125, 110.26105499267578, -3.7785110473632812, 98.1875, 32.790252685546875, -82.09661865234375, 79.92025756835938, -19.226837158203125, 106.5958480834961, 49.763214111328125, -40.192291259765625, 112.06716918945312, 140.22634887695312, 4.19500732421875, -7.07147216796875, 148.9281005859375, 98.70347595214844, -2.6160011291503906, -17.152587890625, -7.99615478515625, 3.4265975952148438, 22.94475555419922, -36.58769226074219, 3.3660659790039062, 40.584014892578125, 46.92582702636719, -128.08108520507812, 134.92807006835938, 37.266143798828125, 123.52178955078125, 45.22125244140625, 22.498558044433594, 171.3028564453125, -23.76788330078125, -128.3690185546875, 171.65232849121094, 169.9586181640625, 71.47393798828125, 126.64450073242188, 183.8553466796875, -4.507225036621094, -1.48675537109375, 112.50421142578125, -5.735860824584961, 168.2051239013672, 7.263916015625, -31.78857421875, -132.34478759765625, 256.8560791015625, 16.526763916015625, 148.8920440673828, 197.39157104492188, -155.564208984375, 5.106006622314453, -11.899009704589844, -0.9777679443359375, 159.88253784179688, 26.315994262695312, 10.65169906616211, 16.1630859375, -27.9300537109375, -105.3355712890625, 0.0, -18.9537353515625, 27.27886962890625, 19.638671875, 164.7479248046875, -35.56671142578125, -26.30987548828125, -14.662612915039062, 55.34947204589844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000336.npy"}
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 52.70410919189453, "std": 87.4167709350586, "min": -135.0175018310547, "p10": -69.9939598083496, "median": 36.34611511230469, "p90": 167.74244079589843, "max": 200.7957763671875, "pos_frac": 0.71875, "sample": [37.408843994140625, -27.524307250976562, 10.181610107421875, 0.7542724609375, 13.994186401367188, -3.685638427734375, -17.8631591796875, 186.42446899414062, 68.88589477539062, 200.7957763671875, 2.6922149658203125, 98.28106689453125, 21.206268310546875, 49.54046630859375, 180.972900390625, 168.41024780273438, -23.031219482421875, 29.84276580810547, -8.79071044921875, 16.720458984375, -120.98959350585938, -67.80316925048828, -42.46513366699219, -39.10162353515625, 33.509010314941406, 4.7490386962890625, 5.17401123046875, 159.4936981201172, 0.2841224670410156, 154.06915283203125, 30.20068359375, 176.40420532226562, 83.34759521484375, 163.10934448242188, 163.6109619140625, 134.2154541015625, 32.48626708984375, 9.867835998535156, 105.72943115234375, 119.33840942382812, -11.731513977050781, -31.660667419433594, -17.282608032226562, 149.26507568359375, 60.35546875, 152.592529296875, -10.543701171875, -122.85111999511719, -31.5009765625, 33.327362060546875, 10.874557495117188, 139.54559326171875, -20.099716186523438, 130.94390869140625, 130.44122314453125, -78.88080596923828, 104.80160522460938, 109.9261474609375, 98.2681884765625, 38.62371826171875, 0.2076263427734375, -10.519973754882812, 42.170379638671875, 179.3359375, 139.40451049804688, -21.980606079101562, 144.2828369140625, 143.06358337402344, -135.0175018310547, 76.1195068359375, 95.996337890625, 175.91159057617188, 42.252227783203125, 30.245330810546875, 128.7867889404297, 3.5775146484375, -5.724456787109375, -27.70764923095703, 167.45623779296875, 131.78050231933594, 14.20191764831543, 72.821044921875, 114.52776336669922, 163.42193603515625, 135.8633575439453, 123.0206298828125, -17.430862426757812, 128.17251586914062, -84.89424133300781, -103.2206039428711, 35.28338623046875, 154.27333068847656, 33.29353332519531, 195.22900390625, -96.78120422363281, 188.99859619140625, -77.40451049804688, -7.811676025390625, 123.3388442993164, 71.69528198242188, 94.28978729248047, -14.89849853515625, -113.95722961425781, 173.8660888671875, -64.50604248046875, -94.86920928955078, 182.431396484375, 77.57667541503906, 2.6149864196777344, 2.572538375854492, -28.36016845703125, 16.64385986328125, 155.736328125, 173.00177001953125, -75.10580444335938, 169.59912109375, 1.217041015625, 3.9181442260742188, -81.5382080078125, 137.67994689941406, 140.79751586914062, 119.00051879882812, -116.42315673828125, 47.50755310058594, 16.59515380859375, 161.639404296875, 155.26243591308594, 92.73468017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000337.npy"}
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 61.00089645385742, "std": 84.03160095214844, "min": -173.19322204589844, "p10": -23.07061767578124, "median": 38.71543884277344, "p90": 174.73597717285156, "max": 222.81060791015625, "pos_frac": 0.7734375, "sample": [177.94754028320312, 1.153360366821289, 7.27276611328125, 139.33181762695312, 164.39471435546875, 183.3876953125, 130.9454345703125, 126.66448974609375, 2.7733840942382812, -37.3621826171875, 78.83967590332031, 148.43942260742188, 34.67822265625, 18.601985931396484, 160.539794921875, 15.2989501953125, 144.28094482421875, 46.653778076171875, -6.9506378173828125, -35.4814453125, 170.2017822265625, 36.17919921875, -1.5973129272460938, 187.529296875, 4.0399169921875, 60.14323425292969, 173.93374633789062, 109.08379364013672, -10.10357666015625, 2.1079978942871094, 40.75518798828125, 9.7275390625, -3.603424072265625, 18.084365844726562, 58.431243896484375, 129.13223266601562, 120.8035888671875, 163.92349243164062, 26.84796142578125, 38.71430969238281, 88.219482421875, 32.62841796875, 56.11968994140625, 158.30001831054688, 176.60784912109375, 5.195884704589844, -32.072784423828125, 151.9876708984375, 77.86270141601562, 50.62255859375, 92.61265563964844, -20.308013916015625, 170.02413940429688, -135.1537322998047, 37.036834716796875, -3.900299072265625, -10.023040771484375, 185.52383422851562, -110.10015869140625, 170.35910034179688, 140.18142700195312, 222.81060791015625, 151.71533203125, 46.29168701171875, 77.25421142578125, 38.71656799316406, -13.05072021484375, -13.460052490234375, -134.53271484375, -5.6655426025390625, 11.160175323486328, 131.75839233398438, 184.14791870117188, 20.80206298828125, -29.516693115234375, 3.339885711669922, 8.97391128540039, 162.94100952148438, 169.22901916503906, -1.1523513793945312, 86.6556396484375, 1.0457229614257812, 134.17520141601562, -4.364265441894531, 131.70082092285156, 26.0352783203125, -19.50275421142578, 9.64495849609375, 90.069580078125, 142.8870849609375, 127.185302734375, 187.5772705078125, 17.382919311523438, -7.442169189453125, -128.8074188232422, 164.39199829101562, 23.227996826171875, 199.8267822265625, -102.01939392089844, 74.9087905883789, 29.347549438476562, -29.553482055664062, 17.11347198486328, 63.14508056640625, 14.936782836914062, 171.5650634765625, 124.78939819335938, 30.342987060546875, -31.546798706054688, 192.5556640625, 49.128173828125, 5.483329772949219, 0.43089866638183594, 33.34564208984375, -6.590576171875, 180.278564453125, 43.958892822265625, 111.98277282714844, 128.51187133789062, 25.898681640625, -173.19322204589844, 2.8555755615234375, 26.6922607421875, -11.998985290527344, 195.15484619140625, 190.12139892578125, -38.219207763671875, 157.7674102783203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000338.npy"}
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 64.14511108398438, "std": 88.79486083984375, "min": -185.4259033203125, "p10": -31.386869812011717, "median": 63.533203125, "p90": 163.32140197753907, "max": 228.14218139648438, "pos_frac": 0.7578125, "sample": [78.8839111328125, -48.634613037109375, 116.36820220947266, 124.22650146484375, 81.17374420166016, -15.981109619140625, 30.071441650390625, -83.46685791015625, 124.73199462890625, -2.8648681640625, 193.59872436523438, 19.06787109375, 7.70233154296875, 5.297767639160156, 152.46182250976562, 17.673492431640625, 95.8067626953125, 101.95367431640625, 105.92960357666016, -59.2608642578125, 139.5472869873047, 13.800567626953125, 63.679779052734375, 121.53607177734375, 0.0, 199.66607666015625, -30.93035888671875, 63.386627197265625, 0.0, 29.269390106201172, 65.86852264404297, -31.751663208007812, 197.61016845703125, 145.0752410888672, 8.518310546875, -153.5183563232422, 166.0672607421875, 171.121337890625, 46.320037841796875, -6.9552001953125, 91.26998901367188, 150.88388061523438, -31.23052978515625, 149.308349609375, 103.85943603515625, 100.36260986328125, 138.7554931640625, 0.9658889770507812, -3.9215850830078125, 153.39566040039062, 28.697479248046875, 7.17327880859375, 227.22720336914062, 117.58692932128906, 153.85671997070312, 145.1759033203125, 14.823753356933594, 26.23663330078125, 152.202392578125, 156.3729248046875, 163.56918334960938, -1.0766830444335938, -19.174957275390625, -67.29214477539062, -1.4508419036865234, 181.1990966796875, 5.317047119140625, 124.34429931640625, 130.1326904296875, 32.1248779296875, 44.184295654296875, 19.179443359375, -124.41978454589844, 125.114013671875, 139.01876831054688, 17.88995361328125, 141.4744873046875, 166.939453125, -49.031524658203125, 135.95530700683594, 4.039604187011719, 149.69027709960938, 163.2152099609375, -11.52532958984375, 157.4903564453125, 228.14218139648438, 68.30032348632812, 0.0, 54.36981201171875, 74.89541625976562, 151.2462158203125, 145.96841430664062, 148.942138671875, 158.83590698242188, 162.82894897460938, -11.08697509765625, 64.3175048828125, -6.35028076171875, 8.5650634765625, 159.39407348632812, 126.056884765625, -86.01800537109375, 29.3021240234375, 6.327301025390625, 10.880828857421875, 142.82003784179688, 142.82310485839844, 9.902008056640625, -21.249267578125, 2.0208511352539062, 183.1708984375, 145.92510986328125, -185.4259033203125, 156.44488525390625, 24.42864990234375, 10.764572143554688, 10.859634399414062, -0.1159820556640625, 167.4852294921875, -108.68853759765625, -171.06138610839844, 194.91546630859375, 43.182037353515625, 34.109405517578125, 155.61819458007812, -6.0394439697265625, -131.90472412109375, 152.71145629882812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000339.npy"}
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 51.972259521484375, "std": 87.27645874023438, "min": -148.49981689453125, "p10": -45.09812164306639, "median": 33.748863220214844, "p90": 159.50027923583986, "max": 228.72943115234375, "pos_frac": 0.71875, "sample": [206.65724182128906, 101.4826889038086, -35.04718017578125, 146.33389282226562, 1.8286972045898438, 158.1868438720703, -41.848297119140625, -18.667118072509766, -55.91424560546875, 154.9801788330078, -134.69635009765625, 137.8813018798828, -75.37628173828125, 1.8972091674804688, 0.0, 162.68563842773438, 22.433441162109375, -52.68104553222656, 121.60103607177734, 162.2016143798828, 152.29522705078125, 14.542938232421875, -36.81855773925781, 117.01663208007812, -63.59422302246094, -0.7482757568359375, -27.2105712890625, 142.86248779296875, 11.326766967773438, 159.1102752685547, 196.9891357421875, 4.862735748291016, 64.74642944335938, 159.39059448242188, 26.9619140625, -14.944839477539062, 0.5714321136474609, -116.27902221679688, 120.621337890625, 16.123519897460938, 154.93634033203125, 139.98971557617188, -114.02908325195312, 47.333984375, 34.23069763183594, 162.28440856933594, -16.28863525390625, 77.59452819824219, 8.01226806640625, -16.501022338867188, 118.39324951171875, 192.16268920898438, -2.781219482421875, -98.8304443359375, 9.308870315551758, -12.518341064453125, 142.6860809326172, 115.73764038085938, 35.271453857421875, -33.96833801269531, 8.126800537109375, -148.49981689453125, 169.79432678222656, 0.0, 198.06362915039062, 18.255889892578125, 53.3560791015625, 18.775123596191406, 153.93878173828125, 36.2974853515625, 16.50823974609375, -7.068185806274414, 141.54754638671875, -109.45353698730469, 150.20004272460938, 32.56597900390625, 46.136444091796875, -103.62451171875, 162.30364990234375, 11.581298828125, 157.3785400390625, 223.6915283203125, -85.1334228515625, 33.26702880859375, -13.843963623046875, 131.54476928710938, 82.88723754882812, 45.34974670410156, 148.5482177734375, 23.496231079101562, 42.74560546875, -16.532861709594727, 105.09542846679688, -14.7255859375, 142.11865234375, 144.77734375, -32.923370361328125, 4.779388427734375, 53.246124267578125, 51.3271484375, 228.72943115234375, 8.943880081176758, 151.82791137695312, -28.391891479492188, 158.80917358398438, -34.299224853515625, 14.161369323730469, -143.61614990234375, 145.38528442382812, 122.45260620117188, 139.34555053710938, 38.669578552246094, 13.971240997314453, 115.50032043457031, 162.8416748046875, 159.75621032714844, 25.41864013671875, 18.324920654296875, 13.439445495605469, 125.1385498046875, 38.02555847167969, 72.3656005859375, -23.39068603515625, 20.902362823486328, 47.841217041015625, 19.532150268554688, 139.90335083007812, -9.827716827392578], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000340.npy"}
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 41.93954849243164, "std": 92.28716278076172, "min": -160.99594116210938, "p10": -76.95588531494137, "median": 19.353408813476562, "p90": 159.37617187499998, "max": 223.041015625, "pos_frac": 0.6640625, "sample": [5.703971862792969, -16.5067138671875, 117.36703491210938, 154.82913208007812, 139.73324584960938, -39.04229736328125, 112.19509887695312, 18.3773193359375, 31.055892944335938, 5.2960968017578125, 143.39849853515625, 39.62480163574219, 150.02432250976562, 12.00851821899414, 193.8319091796875, 187.98956298828125, -67.59744262695312, 55.5526123046875, -1.735595703125, 4.730628967285156, 148.06890869140625, 14.2315673828125, 134.703857421875, -98.79225158691406, 194.0032958984375, 56.82685852050781, 12.046783447265625, 6.3765106201171875, 9.741226196289062, -3.36871337890625, -0.105072021484375, 43.884765625, -17.577850341796875, -7.239471435546875, -126.45611572265625, 30.94952392578125, -62.47119140625, -110.60214233398438, -0.5707607269287109, 12.5240478515625, 2.1439208984375, 49.772274017333984, 173.34048461914062, 75.68954467773438, 126.41755676269531, 6.109832763671875, 154.73281860351562, 121.19007873535156, -2.0859317779541016, 128.34335327148438, 109.01881408691406, 153.5687255859375, 191.48480224609375, -22.53125, 147.40078735351562, 126.78817749023438, 118.11563110351562, 124.85009765625, 13.123184204101562, -17.4512939453125, 26.86334228515625, -127.455078125, 223.041015625, -10.240859985351562, 108.18399047851562, -129.16819763183594, -145.811279296875, 151.33477783203125, 168.55894470214844, 4.668548583984375, 77.38671875, 158.9326171875, 114.28317260742188, -20.947402954101562, 135.26797485351562, 150.24612426757812, 146.9913330078125, 20.233261108398438, 18.473556518554688, -151.7220916748047, 66.92733764648438, 6.000946044921875, -11.444534301757812, -58.35125732421875, 205.05844116210938, 31.535797119140625, -22.481353759765625, 39.61566162109375, 24.388916015625, 161.4779510498047, -117.03810119628906, 89.42233276367188, 189.09817504882812, 142.04978942871094, -115.12234497070312, -14.261962890625, -66.38957977294922, -1.5449752807617188, -10.40325927734375, 28.7606201171875, -27.52667236328125, 154.80859375, 16.9171142578125, -17.6077880859375, -130.1485595703125, 44.18690490722656, 36.96588134765625, -25.64923095703125, -160.99594116210938, -30.225540161132812, 17.895416259765625, -64.5460205078125, -8.1424560546875, 163.67755126953125, -144.44985961914062, -4.549591064453125, 137.1251220703125, 160.4111328125, 140.73660278320312, 60.15238952636719, 147.73785400390625, 8.570060729980469, 35.17810821533203, -102.2906494140625, 194.82061767578125, 13.7281494140625, -10.6888427734375, 12.720916748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000341.npy"}
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 53.795799255371094, "std": 81.27725219726562, "min": -177.18002319335938, "p10": -25.443661499023428, "median": 25.293941497802734, "p90": 163.75351409912108, "max": 222.93011474609375, "pos_frac": 0.734375, "sample": [-6.785003662109375, 24.566513061523438, 25.360015869140625, 22.52777099609375, 92.00405883789062, 212.76461791992188, 20.886070251464844, 103.67344665527344, 7.006645202636719, 38.94598388671875, -12.524658203125, 17.797088623046875, -4.7704925537109375, 138.8723602294922, 116.63981628417969, 140.3721923828125, -8.347526550292969, 67.65234375, -73.84648895263672, 3.7219085693359375, 81.25556945800781, 7.178497314453125, 13.20928955078125, 56.78816223144531, -50.40013885498047, 9.707229614257812, 2.5458831787109375, -43.33099365234375, 34.718231201171875, 134.61741638183594, 25.227867126464844, 147.856201171875, 113.7237548828125, 65.87127685546875, -10.5050048828125, 16.17730712890625, 15.612884521484375, -12.064132690429688, -31.195846557617188, 33.0257568359375, 67.70106506347656, 181.15875244140625, 20.729827880859375, -70.71743774414062, -16.501296997070312, -22.978439331054688, -37.700225830078125, 60.578521728515625, 56.6236572265625, 7.148101806640625, 130.7443084716797, -4.70941162109375, -34.13677978515625, 164.83819580078125, 2.7970733642578125, 197.33648681640625, 157.041748046875, -1.4095458984375, 80.94744873046875, 68.72454833984375, 161.85289001464844, 168.72805786132812, 120.940185546875, 133.05435180664062, 125.37718200683594, -125.86962890625, 7.791259765625, 0.0, -18.91949462890625, 40.01678466796875, 180.77618408203125, 7.9368896484375, 180.72866821289062, 4.59880256652832, 27.731422424316406, 138.25823974609375, 167.33914184570312, 53.199562072753906, 183.43434143066406, 4.152482986450195, -20.227294921875, 147.50570678710938, 37.462158203125, 114.76609802246094, -68.04644775390625, -19.860328674316406, -177.18002319335938, -16.09918212890625, 127.03744506835938, 112.32707977294922, 159.7742919921875, 122.1507568359375, -0.22925567626953125, 7.745361328125, 19.99597930908203, 117.74882507324219, 133.62335205078125, 20.184738159179688, 8.964401245117188, -114.31338500976562, 20.273681640625, 144.30838012695312, -12.480133056640625, 222.93011474609375, 16.537994384765625, 163.2886505126953, -86.06439208984375, 12.8304443359375, 88.660400390625, -109.47696685791016, 165.9915771484375, 135.05172729492188, 52.99385070800781, 147.6764373779297, 143.6824951171875, 159.123779296875, 18.683578491210938, 16.206024169921875, 141.65301513671875, 148.59750366210938, -13.253257751464844, -13.6143798828125, -21.195755004882812, 198.8720703125, 130.28713989257812, -5.4048004150390625, 186.57672119140625, 13.9156494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000342.npy"}
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 46.99864959716797, "std": 87.02061462402344, "min": -214.506103515625, "p10": -46.04381713867187, "median": 27.264484405517578, "p90": 159.14291229248047, "max": 216.57040405273438, "pos_frac": 0.71875, "sample": [-32.71551513671875, 142.089599609375, 148.28248596191406, 5.206744194030762, 41.51190185546875, 37.87249755859375, 125.70147705078125, 23.567230224609375, -56.99543762207031, 26.7001953125, 154.9024200439453, -44.1524658203125, -12.31591796875, -24.6287841796875, 109.71102905273438, 180.67156982421875, 166.44613647460938, 48.597137451171875, -11.735076904296875, -7.017843246459961, 209.02093505859375, 24.9146728515625, 161.4881591796875, 155.22894287109375, 110.52523040771484, 54.72001647949219, -4.5944366455078125, 112.566650390625, 14.9852294921875, 137.41079711914062, 54.990386962890625, -141.276123046875, 31.995849609375, -9.979034423828125, -124.81578063964844, 18.860275268554688, 15.050018310546875, 150.46652221679688, -50.45697021484375, 20.78009033203125, -43.490509033203125, 185.41802978515625, 45.91845703125, 64.76995849609375, 0.0, -113.22833251953125, 21.294921875, 150.0896453857422, 37.749061584472656, 212.49835205078125, -131.63485717773438, 7.748207092285156, 45.472145080566406, 11.285974502563477, 153.8323974609375, -36.157623291015625, 117.3323974609375, 70.56088256835938, 1.0338363647460938, 67.34938049316406, 133.02462768554688, 10.062728881835938, 9.271350860595703, 4.84033203125, 141.55616760253906, 149.28956604003906, -23.166290283203125, 155.24136352539062, 20.406280517578125, 80.21231079101562, 96.18449401855469, -19.189910888671875, 58.219207763671875, -5.2609100341796875, -0.8310165405273438, 86.84883880615234, 13.36376953125, 216.57040405273438, 131.36439514160156, 160.4557647705078, 34.26062774658203, 55.41900634765625, -69.8668212890625, 5.845733642578125, 134.20225524902344, -21.224288940429688, 2.7083740234375, 35.761505126953125, 11.622222900390625, -30.9044189453125, 0.18131446838378906, 149.57696533203125, 152.09622192382812, 206.70303344726562, 9.28076171875, -25.0821533203125, 130.8525848388672, 81.10638427734375, 145.42742919921875, 27.828773498535156, 169.5172882080078, 146.30397033691406, -6.654937744140625, 139.06756591796875, -214.506103515625, 158.58026123046875, -19.858612060546875, -4.327014923095703, -79.74859619140625, -59.9730224609375, 11.17844009399414, 136.94650268554688, 162.39501953125, 1.552001953125, -92.49484252929688, 7.6856842041015625, -8.133148193359375, -105.27987670898438, 35.54386901855469, 194.36575317382812, 97.68153381347656, 0.062084197998046875, 30.773345947265625, 11.717384338378906, 9.273193359375, 171.02072143554688, -17.115264892578125, -145.428466796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000343.npy"}
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 57.225677490234375, "std": 85.69425201416016, "min": -180.42822265625, "p10": -34.16360168457031, "median": 34.04804229736328, "p90": 174.89001159667967, "max": 216.61154174804688, "pos_frac": 0.7421875, "sample": [14.88751220703125, -1.4620361328125, 131.50289916992188, 91.30181884765625, -22.0413818359375, 156.9175262451172, 150.44129943847656, 24.32843017578125, 146.10166931152344, -19.828521728515625, 131.456787109375, 30.947769165039062, 126.53819274902344, -11.292829513549805, 79.56130981445312, 185.043212890625, 17.51123046875, 37.75140380859375, -160.3782958984375, 112.9437255859375, -98.67684936523438, 2.764007568359375, 0.962646484375, -34.065155029296875, 193.29705810546875, 80.7724380493164, -14.957832336425781, 176.3892822265625, 171.84129333496094, 10.994537353515625, 179.1461181640625, -40.548095703125, -6.5438232421875, 139.83035278320312, -180.42822265625, 34.72917175292969, 150.60150146484375, 139.4414520263672, 12.168670654296875, -9.867767333984375, -155.54937744140625, 21.416046142578125, 28.905715942382812, -11.722465515136719, -6.103202819824219, 20.215835571289062, -59.50457763671875, 131.96835327148438, 187.3936767578125, 36.96490478515625, -34.393310546875, -40.98554992675781, -1.044036865234375, -50.44860076904297, 122.10855102539062, 114.8866195678711, 183.11659240722656, 22.874847412109375, 193.514404296875, 174.24746704101562, 152.58670043945312, 152.5011749267578, 216.61154174804688, 123.37948608398438, 161.4282989501953, -5.744529724121094, 40.385284423828125, 201.961669921875, -15.470504760742188, 20.443756103515625, -53.714996337890625, 15.153472900390625, 25.66748046875, 204.94174194335938, 82.31402587890625, 33.26158142089844, 19.08465576171875, 79.33294677734375, 27.6788330078125, 97.28648376464844, 32.13884735107422, 21.954315185546875, 51.30938720703125, 143.73963928222656, -6.085113525390625, 97.406005859375, 2.44683837890625, 153.15780639648438, 21.689041137695312, 51.75886535644531, 192.42584228515625, 106.6236343383789, 41.42567443847656, -63.2322998046875, -5.6089630126953125, 165.50119018554688, 57.28179931640625, -1.1237144470214844, 19.830642700195312, -16.777420043945312, -7.868122100830078, 35.70729064941406, 16.394287109375, 206.320068359375, -13.429931640625, -87.98016357421875, 154.3178253173828, 40.795257568359375, 20.454734802246094, 5.1529541015625, 4.49908447265625, 118.533203125, 10.669692993164062, 171.5401611328125, 107.05513000488281, 32.3858642578125, 71.077392578125, 98.75505065917969, 30.62896728515625, 176.64199829101562, 148.0069122314453, 33.366912841796875, 67.60903930664062, 158.16552734375, 70.653076171875, -23.11859130859375, 170.21124267578125, -144.55349731445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000344.npy"}
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 46.14301300048828, "std": 76.33617401123047, "min": -188.65545654296875, "p10": -27.09519538879394, "median": 22.063522338867188, "p90": 159.05288848876953, "max": 197.37258911132812, "pos_frac": 0.7578125, "sample": [45.92169189453125, 121.84188079833984, 68.74420166015625, -141.5463104248047, 182.009521484375, 64.92778015136719, -7.6708984375, -40.710662841796875, 14.61114501953125, 0.40608978271484375, 72.16856384277344, 11.001792907714844, -0.7856369018554688, 8.377037048339844, 91.94790649414062, 21.960479736328125, 171.73910522460938, 15.24176025390625, 13.6695556640625, -3.2197265625, 10.348480224609375, -25.31890869140625, 130.39837646484375, 158.04930114746094, 3.626532554626465, 39.16070556640625, 139.2965087890625, 33.236907958984375, -69.8311767578125, 1.0595741271972656, -10.758865356445312, 29.762664794921875, -23.993473052978516, 13.706817626953125, 1.64581298828125, 63.824737548828125, 58.68670654296875, 2.200592041015625, 18.738250732421875, 22.499908447265625, 29.25543212890625, -7.67888069152832, -31.239864349365234, 141.63653564453125, 12.2799072265625, 143.24671936035156, 32.969512939453125, 1.586273193359375, 47.221885681152344, 195.375732421875, 134.97914123535156, 143.18524169921875, 56.499237060546875, 167.94325256347656, 173.60520935058594, 52.377685546875, 22.16656494140625, -37.40911865234375, 15.252639770507812, -8.239540100097656, 163.7822265625, 34.94383239746094, 93.893798828125, 21.9027099609375, 20.784011840820312, -76.74122619628906, -188.65545654296875, 16.135498046875, -123.734375, -22.160186767578125, 169.8689422607422, 175.7904052734375, 151.20323181152344, 23.047988891601562, 25.023544311523438, -20.109878540039062, 197.37258911132812, -37.962791442871094, 13.731719970703125, -3.6453704833984375, 157.73391723632812, 161.94635009765625, 85.31489562988281, -86.10235595703125, 78.21881103515625, 4.13201904296875, -43.54582214355469, 191.02301025390625, 86.66375732421875, 43.712974548339844, 171.8543243408203, 140.07815551757812, 23.39459228515625, -11.4814453125, 91.470458984375, 16.394287109375, 89.52999877929688, 144.92636108398438, 154.04177856445312, -6.593048095703125, 153.470947265625, -41.51422119140625, 133.71890258789062, -98.97421264648438, 16.601531982421875, 28.580612182617188, 3.7694034576416016, 20.414215087890625, -8.08929443359375, -10.71722412109375, -8.453933715820312, 126.81302642822266, 18.408660888671875, 27.729690551757812, 161.39459228515625, 139.4224853515625, 11.745269775390625, 97.11624145507812, 0.0, 11.608154296875, 3.0598907470703125, 19.97503662109375, 8.443885803222656, -0.3028717041015625, 147.30413818359375, 8.47705078125, 105.404541015625, 111.72686767578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000345.npy"}
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 52.456329345703125, "std": 86.92726135253906, "min": -198.12005615234375, "p10": -41.34219970703125, "median": 32.797645568847656, "p90": 174.1942169189453, "max": 196.30087280273438, "pos_frac": 0.7265625, "sample": [-51.16096496582031, -31.037599563598633, 1.2961788177490234, -40.84397888183594, 77.66059875488281, 46.6671142578125, 178.0091552734375, 179.62008666992188, 134.39031982421875, -6.8883056640625, 144.8065948486328, -24.374710083007812, -17.209228515625, 25.2432861328125, 147.27783203125, -145.07760620117188, -83.02137756347656, 5.222198486328125, 26.07843017578125, 187.88095092773438, 24.038619995117188, -6.7236785888671875, 8.653213500976562, -50.885101318359375, 23.81234359741211, 138.181884765625, 12.594528198242188, 195.046630859375, -12.918670654296875, 0.6302490234375, 41.31304931640625, 43.29639434814453, -114.41191864013672, 28.986541748046875, -2.635528564453125, 163.01397705078125, 59.434112548828125, 174.74871826171875, 10.468240737915039, 17.65081787109375, 79.75784301757812, 130.3272705078125, -136.73019409179688, 24.71575927734375, -92.58488464355469, 88.38726806640625, -12.0255126953125, -15.187728881835938, -17.506439208984375, 36.43721389770508, 147.11953735351562, 14.7337646484375, 134.90432739257812, -53.162193298339844, 18.704193115234375, 69.45939636230469, 189.9324951171875, 6.5533294677734375, -4.201057434082031, 135.65817260742188, 173.95657348632812, 191.39971923828125, 32.97064208984375, 12.133659362792969, 28.070449829101562, 62.726890563964844, 175.9017333984375, -17.560409545898438, 90.20687866210938, 112.54864501953125, 69.1796875, 73.8857421875, 101.69227600097656, 189.99139404296875, 167.0164794921875, 164.52532958984375, -116.83920288085938, 174.752197265625, 58.86134338378906, 32.27313232421875, 138.09490966796875, -172.0164794921875, 162.92800903320312, 4.123687744140625, -6.6273956298828125, -9.032978057861328, -11.557395935058594, 5.7685699462890625, -64.85368347167969, 176.007568359375, 146.304931640625, 38.997802734375, -16.23902130126953, 160.18801879882812, -2.3210678100585938, -15.685478210449219, 130.72874450683594, -198.12005615234375, 2.71600341796875, 162.14959716796875, 1.3245925903320312, 7.7375946044921875, 128.8373565673828, 126.46893310546875, 186.6970977783203, 51.25303649902344, 132.00462341308594, 127.31198120117188, 12.93701171875, 33.192169189453125, 0.0, 65.28256225585938, -4.6717681884765625, 39.29425048828125, 139.13479614257812, 118.45956420898438, 8.738525390625, 140.97869873046875, 32.62464904785156, -27.60382080078125, 134.97305297851562, 11.884552001953125, -42.50471496582031, 130.50970458984375, 157.74530029296875, 133.45167541503906, 8.674291610717773, 196.30087280273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000346.npy"}
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 49.73553466796875, "std": 81.82565307617188, "min": -212.94882202148438, "p10": -22.67614517211913, "median": 31.77630615234375, "p90": 162.80133056640625, "max": 190.40594482421875, "pos_frac": 0.7109375, "sample": [15.217422485351562, 81.79452514648438, 143.28179931640625, 3.9259185791015625, 68.85302734375, 72.3365478515625, 3.3276824951171875, 159.23727416992188, 16.762474060058594, 155.6631317138672, -212.94882202148438, 87.20849609375, 154.6027374267578, 86.09231567382812, 173.59625244140625, -74.30572509765625, -41.86932373046875, -0.8250274658203125, 147.109375, 174.90492248535156, 142.80889892578125, -30.361083984375, 79.66360473632812, 12.417125701904297, 24.916168212890625, 39.0877685546875, 150.86630249023438, 17.56292724609375, 10.782905578613281, -4.19842529296875, -1.2906494140625, 63.5513916015625, 187.93711853027344, 39.01460266113281, 152.18093872070312, 18.24054718017578, 163.28326416015625, -14.804893493652344, 45.800323486328125, 40.611183166503906, 13.828941345214844, 162.59478759765625, 185.49099731445312, 26.871124267578125, 30.505218505859375, 103.8643798828125, -15.308670043945312, 13.86346435546875, 17.738449096679688, -129.46609497070312, -3.447021484375, -3.6410675048828125, 74.81687927246094, -36.27647399902344, 83.71142578125, 124.55009460449219, -0.396881103515625, 151.6763916015625, 92.906005859375, -79.60942077636719, 134.4517822265625, 14.383636474609375, -1.3234634399414062, -20.291770935058594, 31.6865234375, 111.12663269042969, 100.77084350585938, 44.27323913574219, 187.16766357421875, 40.34907531738281, 31.8660888671875, -143.584228515625, -14.41424560546875, 13.690145492553711, 116.300048828125, 0.310394287109375, -40.729652404785156, -16.18939971923828, -16.920074462890625, 3.314849853515625, 5.245491027832031, 127.9232177734375, 168.69723510742188, -14.4010009765625, 13.51230239868164, -155.37698364257812, 122.61884307861328, 190.40594482421875, 29.548583984375, 86.87654113769531, -11.248260498046875, 67.1639404296875, -59.186187744140625, 151.364501953125, 2.755035400390625, -14.382049560546875, 75.64523315429688, -1.4590721130371094, 164.98391723632812, -19.396331787109375, -177.34207153320312, -4.86212158203125, 187.50155639648438, 153.65310668945312, 0.0, 38.98980712890625, 20.535858154296875, 34.71942138671875, 141.8275146484375, 168.29962158203125, 31.17120361328125, -2.9909210205078125, 154.4617919921875, 136.62094116210938, -10.7647705078125, -4.402824401855469, 41.81402587890625, -5.4786376953125, 138.39898681640625, 42.491943359375, 66.93115234375, 170.10687255859375, 182.5297088623047, 55.817535400390625, 17.62969207763672, -28.23968505859375, 117.1607666015625, 25.731849670410156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000347.npy"}
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 42.766021728515625, "std": 79.73297119140625, "min": -201.69671630859375, "p10": -34.11373748779297, "median": 13.546687126159668, "p90": 154.9334945678711, "max": 205.4818572998047, "pos_frac": 0.6640625, "sample": [41.531005859375, 4.200347900390625, 2.9204444885253906, 3.52996826171875, -7.9339599609375, 7.52667236328125, 147.11328125, -8.27438735961914, 105.43185424804688, 136.32711791992188, -4.254730224609375, 153.907958984375, 80.97589111328125, 139.44204711914062, 63.76055908203125, 13.83331298828125, 188.9144287109375, 125.6463623046875, -67.49713897705078, 0.0, 13.460205078125, 51.607757568359375, 102.76026153564453, -2.2767410278320312, -42.30645751953125, 183.97816467285156, -4.669960021972656, 3.9553070068359375, -9.634109497070312, 123.80740356445312, 65.09681701660156, 120.23628234863281, -10.202728271484375, -7.281776428222656, 77.71734619140625, 0.2913818359375, 155.5587158203125, -4.521368026733398, 135.1224365234375, 165.13958740234375, 13.633169174194336, -21.0533447265625, -79.57373046875, 27.092987060546875, -1.96783447265625, 150.3275146484375, 38.931304931640625, 15.049148559570312, 22.643035888671875, 124.89250946044922, 18.957778930664062, 123.5682373046875, -85.185302734375, 0.34520721435546875, -12.547943115234375, 62.185577392578125, 167.90176391601562, 191.39453125, 166.56256103515625, -108.09332275390625, -9.02301025390625, -12.6068115234375, -35.67694091796875, 126.120849609375, 0.48760414123535156, 122.87614440917969, 16.9739990234375, -2.535350799560547, 5.7340545654296875, 45.8328857421875, 151.98382568359375, -107.14886474609375, -145.14256286621094, -4.484683990478516, -104.683837890625, 54.78521728515625, -16.36700439453125, 8.133895874023438, 105.08479309082031, 6.739013671875, 194.6629638671875, 9.1964111328125, 61.086151123046875, 205.4818572998047, -0.58837890625, 176.32275390625, -6.2169036865234375, -29.39556884765625, -33.484710693359375, 20.85235595703125, -9.195043563842773, 154.66554260253906, 138.24383544921875, -9.949409484863281, 201.29238891601562, -53.43829345703125, 10.079032897949219, 2.698638916015625, -50.23162841796875, 7.2420806884765625, 155.92343139648438, 8.428573608398438, 116.64349365234375, -12.177263259887695, 9.359123229980469, -0.7694320678710938, 149.876708984375, 63.278228759765625, -16.382110595703125, 153.2352294921875, 117.53828430175781, -18.407501220703125, 85.86654663085938, -35.58146667480469, 9.956382751464844, 21.023719787597656, 170.64341735839844, 35.84185791015625, 116.26937103271484, 32.00187683105469, 132.37283325195312, 2.8629226684570312, -4.817253112792969, 13.205841064453125, -5.579193115234375, 35.53977966308594, -201.69671630859375, 83.183349609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000348.npy"}
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 59.23157501220703, "std": 88.39400482177734, "min": -214.74136352539062, "p10": -26.235729980468747, "median": 31.693828582763672, "p90": 177.71277465820313, "max": 241.80145263671875, "pos_frac": 0.78125, "sample": [137.50132751464844, 167.76316833496094, 1.229827880859375, -214.74136352539062, -94.02255249023438, -0.6361427307128906, 241.80145263671875, 133.69625854492188, -118.50410461425781, 57.3509521484375, 46.74766540527344, -78.44343566894531, 144.08595275878906, 181.5535888671875, 117.979736328125, 146.13861083984375, 195.1795654296875, 16.432479858398438, 130.33145141601562, 183.602294921875, 77.30990600585938, 2.79693603515625, -22.84881591796875, -22.877197265625, -5.0222930908203125, 13.939361572265625, -21.0201416015625, 2.5941085815429688, 184.533447265625, 179.45452880859375, 67.92457580566406, 136.94459533691406, 134.37600708007812, -47.5574951171875, 34.473876953125, 12.184066772460938, 195.7910614013672, -3.0443115234375, 19.944427490234375, 50.889251708984375, 8.327552795410156, 0.9580764770507812, 153.8447265625, 3.448333740234375, 13.787429809570312, 48.01104736328125, -9.88623046875, 214.80517578125, 33.35443115234375, 18.299163818359375, -49.469451904296875, -0.5897922515869141, -38.442352294921875, 173.32183837890625, 165.11447143554688, 17.989181518554688, 7.313209533691406, 7.0179290771484375, -115.36447143554688, 89.42630004882812, 0.0, 209.72335815429688, 5.65802001953125, 32.5780029296875, 142.07510375976562, 29.423828125, 157.19143676757812, 166.68865966796875, 157.52789306640625, 6.534017562866211, 167.8448944091797, -27.8778076171875, -105.57119750976562, 215.69244384765625, 3.7838134765625, 49.8873291015625, -5.256866455078125, 126.37704467773438, 4.574066162109375, 18.88269805908203, 18.127532958984375, 114.23685455322266, 198.4385986328125, 95.51971435546875, 176.96630859375, 6.705963134765625, 68.9263916015625, 29.84689712524414, 157.9488525390625, 45.64178466796875, 56.43263244628906, 17.16961669921875, 156.5172882080078, 10.47054672241211, -21.71502685546875, 6.576629638671875, 153.8809356689453, 115.65293884277344, 30.127288818359375, -128.593994140625, 37.1170654296875, 30.809654235839844, 81.02461242675781, -2.550567626953125, -125.21875, 26.71826171875, -4.58746337890625, 5.388698577880859, 14.411056518554688, 139.66439819335938, 28.97015380859375, 147.91427612304688, 27.898712158203125, 135.4283905029297, 158.36695861816406, 184.43663024902344, 156.22561645507812, -25.531982421875, 118.143798828125, 199.70748901367188, 13.10089111328125, 66.96258544921875, 144.77590942382812, 132.15780639648438, -57.396209716796875, -1.171722412109375, 14.123641967773438, 143.0379180908203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000349.npy"}
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 51.766422271728516, "std": 92.04623413085938, "min": -163.053955078125, "p10": -71.03358612060546, "median": 34.748558044433594, "p90": 170.08748474121094, "max": 220.2843017578125, "pos_frac": 0.765625, "sample": [170.97698974609375, 53.13538360595703, 0.2688713073730469, -5.8589324951171875, 9.19198226928711, 42.115692138671875, 16.287918090820312, 188.51187133789062, -0.7159900665283203, 135.11312866210938, 29.5054931640625, 15.141860961914062, 166.19387817382812, -10.905029296875, 28.298309326171875, 169.4263916015625, 152.8495330810547, 169.70626831054688, 18.57733917236328, -40.97084426879883, 154.46168518066406, 15.856979370117188, 3.069549560546875, 127.98893737792969, 113.45368194580078, 142.30577087402344, 49.60252380371094, 118.41659545898438, 202.07351684570312, 191.45697021484375, 121.93145751953125, 66.86517333984375, 14.93325424194336, -1.7814960479736328, -47.440826416015625, 75.99395751953125, 15.091018676757812, 21.669113159179688, 58.4071044921875, 36.1536865234375, 21.788848876953125, 124.09391784667969, 130.83541870117188, -146.4158935546875, 166.46771240234375, -15.055389404296875, 7.8933258056640625, 61.92022705078125, -163.053955078125, 3.9137420654296875, 44.57379150390625, 148.94891357421875, 214.30899047851562, 118.18302917480469, 27.602508544921875, 78.80785369873047, 180.03921508789062, 147.4342041015625, 61.402984619140625, 198.28256225585938, 112.52783203125, -87.85211181640625, 167.74058532714844, 11.7999267578125, -21.5491943359375, -59.89164733886719, 43.25311279296875, 24.934463500976562, -153.07965087890625, 10.055660247802734, 137.19276428222656, 140.35150146484375, 2.6122970581054688, -93.2362060546875, -13.183685302734375, 141.91268920898438, -8.648162841796875, 31.010826110839844, 35.326934814453125, 66.69268798828125, 187.78604125976562, 159.20843505859375, 129.88616943359375, 28.025619506835938, 32.458465576171875, 150.4453125, 20.0682373046875, 178.23614501953125, -113.12931823730469, 11.693084716796875, 186.68814086914062, 34.17018127441406, 26.083404541015625, 36.80607604980469, 159.88177490234375, 124.31826782226562, 32.628662109375, 145.6326904296875, 220.2843017578125, 160.30093383789062, 161.6177978515625, -70.48464965820312, 52.505859375, 3.6539535522460938, -157.5645751953125, -112.33609771728516, -112.2353515625, -0.14687347412109375, -148.53240966796875, 52.911163330078125, -153.3363494873047, -81.77081298828125, 12.710784912109375, 185.7159423828125, 54.38105773925781, 49.65423583984375, 11.357208251953125, 26.680313110351562, 173.70358276367188, -72.31443786621094, -13.62432861328125, 24.468971252441406, -24.31817626953125, -7.01873779296875, 167.584716796875, -49.03990173339844, 48.619873046875, 4.485343933105469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000350.npy"}
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 35.83001708984375, "std": 90.45166778564453, "min": -214.05386352539062, "p10": -46.97607421875, "median": 15.41300106048584, "p90": 166.4476593017578, "max": 236.90582275390625, "pos_frac": 0.6328125, "sample": [0.0, 149.66778564453125, 13.09716796875, 202.81442260742188, -1.710693359375, 188.30963134765625, 21.17267608642578, -59.981300354003906, 15.090133666992188, -3.5589599609375, 28.935195922851562, -44.602447509765625, 16.538665771484375, -6.3932342529296875, 165.80172729492188, -142.5006561279297, 59.885929107666016, 115.10836029052734, -32.805335998535156, 20.930419921875, 22.248153686523438, 142.8374786376953, -11.904205322265625, 207.96014404296875, 61.94181823730469, 161.78485107421875, 9.843957901000977, -18.98974609375, -9.04132080078125, 1.3520965576171875, 179.53436279296875, 184.6255645751953, 64.83322143554688, 25.973377227783203, -42.623382568359375, -46.703704833984375, -2.696338653564453, 144.72320556640625, -24.111907958984375, -36.46843338012695, -101.24334716796875, 158.33419799804688, 15.236801147460938, 50.2264404296875, -78.75732421875, 145.94752502441406, 8.992134094238281, 16.75213623046875, -22.707275390625, 15.754852294921875, -72.56475830078125, -29.93890380859375, 19.121429443359375, -8.003524780273438, 19.973388671875, -0.4041748046875, 159.43939208984375, -38.315765380859375, -12.073577880859375, 10.446624755859375, 184.60067749023438, 232.54000854492188, 15.589200973510742, 202.1279296875, 158.8264617919922, -13.729827880859375, 60.38214111328125, 16.735870361328125, 130.15164184570312, -90.40869140625, -5.7009735107421875, 123.11385345458984, -33.141265869140625, 13.72015380859375, 167.954833984375, -175.52569580078125, 30.57781982421875, -11.063751220703125, 13.950958251953125, 49.92840576171875, 0.0, 0.452880859375, 23.331588745117188, 150.24429321289062, 148.0865936279297, -0.7874698638916016, -0.0684967041015625, 29.48986053466797, 128.65985107421875, -23.306304931640625, 31.748046875, 7.545745849609375, 63.125152587890625, -14.923057556152344, 0.0, 188.001708984375, 171.8638916015625, 48.931884765625, 13.173988342285156, 13.338226318359375, 3.5695648193359375, -214.05386352539062, 121.59791564941406, 165.00381469726562, 72.30262756347656, 4.2413177490234375, 236.90582275390625, 32.226287841796875, 161.32167053222656, -2.8525543212890625, -1.6208457946777344, 8.566268920898438, -11.886474609375, 33.74299621582031, 0.6170787811279297, 155.36904907226562, 20.863296508789062, -47.611602783203125, -138.04660034179688, -153.9544677734375, 16.6737060546875, 73.43130493164062, -184.9000244140625, 42.667808532714844, -44.04217529296875, -56.53099060058594, 224.61968994140625, 41.348907470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000351.npy"}
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 58.37958908081055, "std": 86.41031646728516, "min": -141.6287841796875, "p10": -38.939111328124994, "median": 48.85932922363281, "p90": 175.98504028320312, "max": 217.3309326171875, "pos_frac": 0.75, "sample": [124.73239135742188, 29.23638916015625, 165.943603515625, -50.406219482421875, 9.62310791015625, 199.12689208984375, 123.76618194580078, 93.7691421508789, 127.99343872070312, 4.138130187988281, -37.275177001953125, 150.01312255859375, -133.7887420654297, 17.982208251953125, 151.4960479736328, 4.305473327636719, 34.90142822265625, 185.4185791015625, 148.56207275390625, 122.38290405273438, 204.8909912109375, -12.721944808959961, -42.821624755859375, 184.67160034179688, 50.3438720703125, -113.0322265625, -17.738128662109375, 105.60902404785156, 18.53656005859375, 127.72917938232422, -1.4997520446777344, 22.135848999023438, 181.81561279296875, 17.380905151367188, 131.2684326171875, -122.95366668701172, 13.049537658691406, 116.3765869140625, -19.640106201171875, 11.146636962890625, 75.0733642578125, -141.6287841796875, -27.428848266601562, 93.98501586914062, -27.89892578125, 24.08251953125, 0.0, 71.219970703125, 21.3828125, 176.3911590576172, -2.470794677734375, 188.54159545898438, 131.25128173828125, 31.1287841796875, 153.36996459960938, 71.84268188476562, 26.522445678710938, 0.0, 70.99578857421875, 88.079345703125, 116.88816833496094, 2.3158493041992188, 18.133834838867188, 108.32818603515625, -131.54690551757812, -17.89410400390625, 9.50997543334961, 58.45472717285156, -107.37237548828125, 4.039703369140625, 116.48284912109375, 202.01226806640625, 72.65774536132812, 23.223814010620117, 6.854036331176758, -125.70526123046875, 181.2371826171875, 22.287181854248047, -70.49444580078125, -92.234619140625, 217.3309326171875, 179.6205596923828, 175.8109893798828, 24.089599609375, -18.802276611328125, 149.4970703125, 9.516792297363281, -47.0731201171875, 0.8919601440429688, 126.37387084960938, 44.146728515625, -10.863937377929688, 57.815185546875, 17.27297019958496, 56.55021667480469, 47.374786376953125, 162.83700561523438, 137.55661010742188, -9.062530517578125, 186.1549835205078, 53.39164733886719, 85.64956665039062, 141.8544464111328, 117.60107421875, 160.3406219482422, -32.8604736328125, 149.72213745117188, 163.59695434570312, 66.34788513183594, -7.872474670410156, 23.40423583984375, 131.79002380371094, 166.4896240234375, 2.768890380859375, -65.38783264160156, -10.10113525390625, 34.64017105102539, 173.15780639648438, 137.96240234375, 182.9841766357422, -25.170364379882812, -35.62451171875, 138.79856872558594, 78.45639038085938, 98.78682708740234, 122.89656829833984, 19.11566162109375, 144.7244873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000352.npy"}
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 60.6271858215332, "std": 85.60364532470703, "min": -171.92605590820312, "p10": -25.246051025390624, "median": 47.44512939453125, "p90": 161.59404296875, "max": 249.421630859375, "pos_frac": 0.765625, "sample": [36.18211364746094, 144.41632080078125, 88.16461181640625, -15.492645263671875, 111.34756469726562, 172.28704833984375, -111.62000274658203, 161.05889892578125, 147.47683715820312, 16.1619873046875, 33.291961669921875, 31.474990844726562, 8.5562744140625, 122.11697387695312, -6.7391204833984375, 18.207427978515625, -16.50665283203125, -26.78375244140625, 44.078704833984375, -18.88421630859375, 65.6099853515625, 102.2491455078125, 0.14546966552734375, -171.92605590820312, 143.26332092285156, 12.17852783203125, 73.43551635742188, 136.73974609375, 228.7830810546875, -0.22154998779296875, 184.06463623046875, 140.69735717773438, 162.84271240234375, -44.660400390625, 171.31634521484375, -147.51107788085938, 187.33273315429688, 106.24298095703125, 85.5472412109375, 249.421630859375, 21.1036376953125, -97.02214050292969, 41.1541748046875, 160.37246704101562, 38.15406799316406, -9.15399169921875, 160.37472534179688, 238.33114624023438, 209.10543823242188, 134.4711456298828, 145.02442932128906, 68.86598205566406, -36.471214294433594, 133.47158813476562, 15.246231079101562, 219.63864135742188, 1.9410724639892578, -20.577476501464844, 128.63613891601562, 62.91612243652344, 187.43942260742188, 86.84161376953125, -32.69044494628906, -108.41558837890625, 1.5722808837890625, 118.9637451171875, -131.21218872070312, -16.4783935546875, 94.95513916015625, 29.035888671875, 81.98973083496094, 78.86628723144531, 142.0835418701172, 155.07339477539062, 154.86016845703125, 128.47808837890625, 12.617866516113281, 146.35867309570312, 92.93563842773438, 9.403861999511719, 149.1651611328125, 25.880294799804688, 62.65301513671875, 27.725982666015625, 50.811553955078125, 138.24249267578125, 35.168243408203125, 135.21058654785156, 124.26080322265625, 4.949775695800781, -2.943988800048828, -24.5870361328125, 139.0096435546875, 56.174774169921875, 18.173362731933594, -22.8289794921875, -3.7301254272460938, 43.52687072753906, 12.411079406738281, -14.994110107421875, 74.10498046875, 13.387359619140625, 27.005859375, 3.4763545989990234, 166.37286376953125, 34.922210693359375, 94.66748046875, 53.23939514160156, 155.6450653076172, -64.56607055664062, 32.39916229248047, -0.7623519897460938, 28.25809097290039, 65.47137451171875, 6.193996429443359, 0.0, 209.261474609375, 92.05667114257812, -31.16522216796875, -11.228118896484375, 2.082855224609375, -154.18295288085938, -12.716217041015625, 19.64947509765625, 160.40016174316406, 121.62567138671875, 144.76336669921875, 103.02978515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000353.npy"}
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 35.61540603637695, "std": 91.96826934814453, "min": -177.21307373046875, "p10": -75.54358749389648, "median": 11.524726867675781, "p90": 157.39442596435546, "max": 235.97677612304688, "pos_frac": 0.6640625, "sample": [8.293952941894531, -50.27693176269531, -146.65101623535156, 151.48594665527344, 47.849456787109375, 215.1068115234375, 95.95050048828125, 6.747821807861328, 2.6245384216308594, 4.3719024658203125, 32.577301025390625, 40.87274169921875, -52.29325866699219, 16.902618408203125, 155.34246826171875, -10.352401733398438, -13.662818908691406, 0.27161407470703125, 8.695556640625, -3.080535888671875, 117.87718200683594, 130.5187225341797, 93.05557250976562, -75.24005126953125, 32.58049774169922, 12.646003723144531, -10.845035552978516, 0.2831268310546875, 146.46315002441406, 207.98873901367188, -47.90257263183594, -2.5560302734375, 0.12180900573730469, 59.19482421875, 182.00640869140625, 28.95257568359375, 13.60086441040039, -0.42144775390625, 153.16567993164062, 148.96820068359375, 6.2960968017578125, -55.05621337890625, 95.08067321777344, -18.2657470703125, 157.76502990722656, 154.99960327148438, 4.537689208984375, 8.620506286621094, -171.38992309570312, 3.8474273681640625, -6.96539306640625, 204.18270874023438, 60.6629638671875, 180.89541625976562, 183.08517456054688, 184.31982421875, 134.049560546875, 51.035797119140625, 34.906005859375, 84.827392578125, 62.9664306640625, -28.001998901367188, -14.626731872558594, -70.7154541015625, 8.081016540527344, 124.8819580078125, 10.28635025024414, 235.97677612304688, -2.0717849731445312, -4.4720458984375, 157.235595703125, -76.25183868408203, 130.97467041015625, 6.834150314331055, 205.2457275390625, 56.11033630371094, 116.5912094116211, -120.15043640136719, 5.63214111328125, 0.7335586547851562, -8.34771728515625, -56.37005615234375, 195.4415283203125, 41.034881591796875, -20.863616943359375, 44.33306884765625, -1.644622802734375, 148.6669158935547, -12.691070556640625, -110.83659362792969, 21.877044677734375, -50.824989318847656, 155.39981079101562, 10.101799011230469, 141.90342712402344, 29.138168334960938, -9.544403076171875, -2.1359405517578125, -124.9791259765625, -160.62026977539062, 56.53727722167969, 2.764007568359375, -15.394683837890625, 61.43804168701172, 133.441650390625, 58.91888427734375, 1.05267333984375, -5.15631103515625, 22.064163208007812, 191.3626708984375, 153.10513305664062, 86.45205688476562, -78.53589630126953, 159.71234130859375, -140.87184143066406, 75.307373046875, -177.21307373046875, 135.76608276367188, -108.14886474609375, -65.91574096679688, 129.52102661132812, 10.469085693359375, -109.3739013671875, 41.94715881347656, -50.03387451171875, -129.5446319580078, 73.55401611328125, 12.580368041992188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000354.npy"}
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 60.110923767089844, "std": 88.6454849243164, "min": -203.083984375, "p10": -39.08100585937499, "median": 43.13666915893555, "p90": 174.01659698486327, "max": 215.29312133789062, "pos_frac": 0.7734375, "sample": [-2.4900550842285156, -203.083984375, 166.9710693359375, 129.4453125, 2.75872802734375, 10.757026672363281, 56.60223388671875, 4.527374267578125, 75.39437866210938, 55.8211669921875, 163.9696807861328, 8.766021728515625, -11.418609619140625, 136.36599731445312, 4.375299453735352, 156.03009033203125, 189.56256103515625, 155.06161499023438, -5.250312805175781, 1.4996337890625, -49.630706787109375, 20.79473876953125, 13.771896362304688, 188.16995239257812, 14.240997314453125, 190.16896057128906, -14.70950698852539, -8.33642578125, 39.68824768066406, 27.38250732421875, 26.712158203125, 53.0054931640625, 113.05426025390625, 189.4454345703125, 183.50689697265625, 51.546730041503906, 191.9322509765625, 20.572372436523438, -2.30108642578125, -99.654296875, 97.24797821044922, 154.2268829345703, 117.34011840820312, 97.79762268066406, -63.80854797363281, 31.92095947265625, -8.583160400390625, 93.4298095703125, 42.90496063232422, 4.27630615234375, 157.79843139648438, 105.98458862304688, 119.27935791015625, 13.197296142578125, 1.02655029296875, 99.080322265625, 1.9240608215332031, 34.89476013183594, 29.5775146484375, -13.363895416259766, 178.28497314453125, 9.577693939208984, -36.148681640625, -12.309906005859375, 134.07789611816406, -152.42529296875, 24.43421173095703, 215.29312133789062, 92.27780151367188, 163.98281860351562, -0.14803123474121094, 176.99240112304688, -165.56585693359375, 164.11126708984375, 0.6975479125976562, 44.342742919921875, 147.18209838867188, 87.19622802734375, 195.72470092773438, 96.11343383789062, 111.93109130859375, 151.41500854492188, 3.466766357421875, -3.7889404296875, 18.34515380859375, -56.43731689453125, 33.9769287109375, -25.555328369140625, 33.92170715332031, 40.284637451171875, 145.32276916503906, -45.923095703125, -147.36630249023438, 6.73748779296875, -8.177886962890625, 43.368377685546875, 21.833953857421875, 173.7178192138672, 45.63787841796875, 160.06729125976562, 39.517425537109375, -91.71245574951172, 208.603271484375, 127.82545471191406, 183.8778076171875, 163.12554931640625, 163.5998077392578, 124.64004516601562, -56.554901123046875, -19.199859619140625, 9.686859130859375, 35.72319030761719, -73.82429504394531, 168.44650268554688, 137.11480712890625, 141.86517333984375, 154.387451171875, 130.894775390625, 149.6173095703125, 125.1082763671875, 76.23861694335938, -0.9964599609375, 64.55508422851562, 174.7137451171875, 57.85125732421875, -136.40994262695312, 169.3245086669922, 38.532440185546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000355.npy"}
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 47.21686935424805, "std": 79.3420639038086, "min": -180.09194946289062, "p10": -25.962075805664043, "median": 29.063053131103516, "p90": 153.30306549072264, "max": 193.69728088378906, "pos_frac": 0.7109375, "sample": [3.16070556640625, -7.0085601806640625, 152.11563110351562, -2.980743408203125, 147.43850708007812, -54.5091552734375, 19.85589599609375, -45.798797607421875, 7.967611312866211, 3.5013885498046875, -14.579689025878906, 6.8381805419921875, 26.7388916015625, 88.48025512695312, -8.163154602050781, 3.413541793823242, 19.626296997070312, -20.745758056640625, 7.64599609375, 49.694610595703125, -41.821563720703125, 146.72088623046875, 29.579185485839844, 36.512908935546875, -0.04471015930175781, 182.122314453125, -117.79051208496094, 142.1103057861328, 2.6193084716796875, 103.67999267578125, 39.314727783203125, 161.5098876953125, 6.5816650390625, 63.305320739746094, 143.8937530517578, 161.8163604736328, 122.11691284179688, 63.3135986328125, 176.65000915527344, 139.6638946533203, -147.98509216308594, 149.3475341796875, -6.911537170410156, 9.582649230957031, 24.62268829345703, 35.92962646484375, -8.153778076171875, 181.28988647460938, 61.083648681640625, -7.356498718261719, 28.649330139160156, 126.8179931640625, 156.07374572753906, 178.568115234375, -0.07545089721679688, 40.245849609375, -101.12564086914062, 144.73092651367188, 121.69013977050781, 41.13739013671875, 156.10955810546875, -138.6097869873047, 137.45333862304688, 118.65869140625, 149.03221130371094, 138.2257080078125, 98.49359130859375, -11.203857421875, 2.2474822998046875, 4.67291259765625, -10.335128784179688, 23.849456787109375, 138.07958984375, 1.1886672973632812, 4.68060302734375, 37.331642150878906, 109.61161804199219, 0.4077911376953125, 165.38372802734375, -39.97703552246094, 29.645339965820312, -17.98311996459961, 184.96054077148438, 34.098846435546875, -17.001571655273438, 95.1834716796875, 138.33145141601562, 115.32264709472656, 29.476776123046875, -0.7636299133300781, -94.10345458984375, -4.648586273193359, -10.967056274414062, -180.09194946289062, 39.64128112792969, 15.219207763671875, 50.40240478515625, 0.0, 43.255096435546875, 30.15557861328125, 100.28614807128906, 24.992889404296875, -8.866033554077148, -20.25775146484375, 0.4647846221923828, 15.05291748046875, 20.203540802001953, 134.63204956054688, -2.65911865234375, -94.4039306640625, 151.74981689453125, 193.69728088378906, 150.23410034179688, 128.98538208007812, -14.387092590332031, 161.59983825683594, 145.882080078125, 5.472831726074219, 41.223724365234375, 145.73959350585938, -38.13348388671875, 104.63375854492188, -2.3849945068359375, -61.573570251464844, 55.17633056640625, -11.506927490234375, 22.762802124023438, 157.0006103515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000356.npy"}
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 51.845359802246094, "std": 90.15930938720703, "min": -185.8974151611328, "p10": -35.583178710937496, "median": 29.305891036987305, "p90": 176.05886993408203, "max": 250.10430908203125, "pos_frac": 0.7109375, "sample": [88.41867065429688, 156.50872802734375, -15.950252532958984, 1.26123046875, 163.79876708984375, 42.82666015625, 0.0, 3.7035675048828125, 76.85263061523438, 5.815704345703125, 155.2736053466797, -144.12362670898438, 5.358152389526367, 250.10430908203125, 205.95098876953125, 214.3739013671875, 41.0291862487793, -0.33349609375, 132.47743225097656, 25.49530029296875, 15.832626342773438, 16.83673095703125, 107.37513732910156, 157.36184692382812, 175.74778747558594, -93.9454345703125, 230.62997436523438, 22.575576782226562, 26.830501556396484, 183.78543090820312, -67.384765625, 13.115234375, -20.893211364746094, 7.7206268310546875, 211.04281616210938, 43.105709075927734, -18.25469970703125, -129.50180053710938, 7.519256591796875, 121.02855682373047, 120.5138931274414, 206.6776123046875, 14.665130615234375, 124.83819580078125, 68.29653930664062, -34.2421875, 4.6851806640625, -120.11404418945312, 169.19161987304688, 11.626556396484375, -33.28062438964844, 36.199501037597656, 47.747711181640625, 23.742034912109375, 71.26919555664062, 143.75636291503906, 187.38922119140625, 168.70101928710938, 34.92132568359375, -3.94378662109375, 149.333740234375, 53.1231689453125, 18.688720703125, -20.294281005859375, 57.223876953125, -24.58917236328125, 152.38095092773438, 154.40411376953125, 33.7099609375, 37.982208251953125, -8.12786865234375, -100.21961975097656, -12.2108154296875, 5.3238372802734375, 1.4061279296875, 154.0816192626953, 89.72645568847656, -61.112274169921875, 21.746047973632812, -37.523773193359375, -55.78582763671875, 10.495574951171875, 15.997085571289062, 31.781280517578125, 183.69000244140625, 37.367431640625, 164.09747314453125, -34.751495361328125, -0.1235504150390625, -0.859466552734375, -4.457557678222656, 40.19024658203125, 63.29588317871094, -96.63558959960938, 49.914642333984375, 148.8203125, -1.021759033203125, -112.14775848388672, 0.0, 176.78472900390625, -142.46852111816406, 0.0, 39.382568359375, -5.78045654296875, 171.0048828125, 12.056949615478516, 14.562965393066406, 154.80653381347656, 163.81625366210938, 125.18988037109375, 182.577392578125, -185.8974151611328, 162.78555297851562, 82.08120727539062, 2.4427833557128906, 115.92242431640625, -4.059173583984375, 145.83123779296875, 92.6925277709961, 190.51153564453125, 193.3668212890625, 18.481964111328125, -3.8197479248046875, -24.43048095703125, 64.8836669921875, -2.490325927734375, 2.5674209594726562, 126.47343444824219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000357.npy"}
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 59.121253967285156, "std": 93.8873062133789, "min": -186.20559692382812, "p10": -45.373472595214835, "median": 36.550872802734375, "p90": 178.02529907226562, "max": 223.5560302734375, "pos_frac": 0.7578125, "sample": [1.09539794921875, 3.2935104370117188, 121.81282043457031, -62.51567077636719, 10.0357666015625, 47.588401794433594, -36.031982421875, 122.29812622070312, 188.12908935546875, 223.5560302734375, 9.750320434570312, 6.011714935302734, 29.063430786132812, 158.9655303955078, 179.67877197265625, 161.03639221191406, 26.840652465820312, 181.357421875, 177.7105255126953, 35.18035888671875, -144.5897674560547, 148.0989990234375, -36.05625915527344, -126.46435546875, 73.3514404296875, -11.608535766601562, 132.9756622314453, -5.73651123046875, 182.38592529296875, -12.607879638671875, 5.2004852294921875, 114.5830078125, 23.12071990966797, 135.49664306640625, 106.26043701171875, 28.628662109375, 207.11410522460938, -83.96200561523438, 37.74774169921875, 135.3144073486328, -124.07872009277344, 173.57937622070312, 168.91958618164062, 177.74844360351562, 30.50743865966797, 163.90274047851562, 130.34033203125, 158.74525451660156, 161.6566162109375, 3.64703369140625, -31.48187255859375, -13.324745178222656, -8.355377197265625, 26.664867401123047, 157.2366485595703, -10.82373046875, 161.28573608398438, 169.069091796875, -63.211273193359375, 214.07861328125, -72.49002075195312, -186.20559692382812, 180.03515625, -17.54925537109375, 8.665840148925781, 188.19497680664062, 79.43463134765625, 31.673126220703125, 78.22210693359375, 12.454864501953125, 32.62689208984375, 11.502670288085938, 21.35509490966797, 33.961883544921875, 5.223785400390625, -134.84524536132812, 153.2060089111328, 118.39927673339844, -11.929588317871094, 188.64915466308594, 67.97590637207031, -48.667999267578125, 135.3404083251953, 50.0859375, 178.67129516601562, 37.4207763671875, 112.57684326171875, 35.68096923828125, 39.32326889038086, 172.12152099609375, 8.051925659179688, 195.25863647460938, 149.64968872070312, 113.26615905761719, 145.170654296875, 157.96002197265625, -34.188629150390625, 10.745513916015625, 9.35565185546875, 134.408935546875, 204.3597412109375, 1.735382080078125, 126.22920227050781, 174.13333129882812, 74.179443359375, 173.551025390625, 162.07684326171875, -175.00299072265625, -16.992385864257812, -124.66995239257812, -123.06338500976562, 6.009326934814453, 140.21226501464844, 4.899436950683594, 97.93311309814453, -35.929771423339844, 10.097099304199219, 13.636573791503906, -2.4835662841796875, 33.75885009765625, 0.0, 41.2275390625, 66.01296997070312, 47.571990966796875, 149.78253173828125, -43.96153259277344, 32.99957275390625, -1.7908935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000358.npy"}
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 56.422149658203125, "std": 84.4388427734375, "min": -200.4886474609375, "p10": -28.445756912231445, "median": 36.677520751953125, "p90": 165.94965209960935, "max": 230.48403930664062, "pos_frac": 0.765625, "sample": [48.431480407714844, -59.64593505859375, 142.0305633544922, 183.60296630859375, 115.93505859375, -84.910888671875, 95.30847930908203, 16.912704467773438, 37.23065185546875, 51.964141845703125, 27.500164031982422, 80.76913452148438, 12.472942352294922, 1.1471710205078125, -15.28851318359375, -7.800384521484375, 118.5971450805664, 49.93031311035156, 20.1905517578125, 16.76483154296875, 137.02001953125, -12.540847778320312, 150.85467529296875, 177.196044921875, -18.993560791015625, 154.87405395507812, 13.94049072265625, 36.020599365234375, 78.065673828125, -193.64141845703125, 1.2056884765625, 207.68310546875, 3.985931396484375, 81.61473846435547, 2.7707481384277344, 63.851837158203125, -8.951604843139648, 146.17507934570312, 164.85482788085938, 172.45947265625, -27.05889892578125, -124.20155334472656, 144.49737548828125, 3.8174209594726562, 117.3524169921875, 121.127685546875, 11.76788330078125, 58.293853759765625, 139.75234985351562, 132.2723846435547, 152.6217041015625, 3.7640342712402344, 223.02310180664062, -22.62390899658203, 119.5040283203125, 1.7087631225585938, 56.245086669921875, -177.2733154296875, 132.71023559570312, 42.2332763671875, 199.44818115234375, 131.02078247070312, -200.4886474609375, 74.87216186523438, 5.6185302734375, -12.924812316894531, -11.701171875, 48.396209716796875, 131.64682006835938, 91.610595703125, -3.03363037109375, 32.97210693359375, -31.804122924804688, 35.094024658203125, 104.51192474365234, 78.20822143554688, -14.197174072265625, 142.79840087890625, 15.083114624023438, 148.45535278320312, 52.96746826171875, -0.814361572265625, 6.8975830078125, 142.51071166992188, 31.140884399414062, 19.00667953491211, 19.672088623046875, 162.27891540527344, 24.959930419921875, 16.300216674804688, 173.37294006347656, 14.309341430664062, 149.01007080078125, 134.98410034179688, 22.24652099609375, -29.263092041015625, -84.72689819335938, 136.6990966796875, -28.095470428466797, -0.6829986572265625, 29.904388427734375, 185.57669067382812, 36.1243896484375, -65.98995971679688, 16.364883422851562, -45.174102783203125, 175.85162353515625, -30.32794189453125, 53.498748779296875, 168.50424194335938, 141.95101928710938, 101.74219512939453, 7.40789794921875, 61.60791778564453, -8.272857666015625, 183.87896728515625, 161.22137451171875, 117.62889099121094, 133.94996643066406, -1.4470748901367188, -5.432106018066406, 230.48403930664062, 28.744369506835938, 3.724172592163086, 169.0140380859375, 30.059188842773438, 127.93531799316406, -33.977813720703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000359.npy"}
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 48.79966735839844, "std": 80.33026885986328, "min": -214.1014404296875, "p10": -24.584989929199217, "median": 26.341482162475586, "p90": 164.5107894897461, "max": 238.37762451171875, "pos_frac": 0.71875, "sample": [117.72268676757812, 127.87005615234375, -46.763885498046875, 165.77926635742188, -143.76214599609375, -8.485357284545898, -100.81597900390625, 80.93133544921875, 238.37762451171875, -47.62611389160156, 31.817108154296875, 109.5325927734375, -21.264205932617188, 122.07373046875, 51.10089111328125, 97.6163330078125, 16.2581787109375, 150.55328369140625, -26.22650146484375, 51.22886657714844, 180.3355712890625, 63.99137878417969, 13.863525390625, 124.24014282226562, 128.134033203125, -12.6796875, 3.4161376953125, 20.02581787109375, -17.059326171875, 168.62405395507812, 36.69696044921875, 188.0374298095703, 94.9366455078125, 20.25689697265625, 60.82037353515625, 7.22076416015625, -23.881484985351562, 140.9464111328125, -33.45758056640625, 80.74662780761719, 94.34136962890625, 104.03875732421875, 148.73202514648438, 38.31744384765625, 38.371734619140625, 22.372791290283203, 19.383636474609375, 146.1309814453125, -7.5528564453125, 40.652557373046875, -11.795486450195312, 155.87136840820312, 132.56884765625, 104.83329772949219, 198.67098999023438, -13.008537292480469, 174.9176483154297, -42.645751953125, 155.11801147460938, 14.394157409667969, 49.484161376953125, 39.08953857421875, 160.4312744140625, 29.79277229309082, 17.46380615234375, 1.8622055053710938, 177.73849487304688, 18.808563232421875, 93.67955017089844, -146.920166015625, -2.369659423828125, 138.12286376953125, 1.5837726593017578, 168.94239807128906, -17.9920654296875, 26.420391082763672, 62.12969970703125, 174.9981689453125, 148.58984375, 142.34324645996094, -9.022552490234375, 121.66033935546875, 6.356319427490234, -6.989551544189453, -214.1014404296875, 127.56610107421875, 49.301910400390625, 9.81719970703125, -20.676361083984375, 4.767242431640625, 157.05001831054688, -72.39010620117188, 191.99154663085938, 31.03314971923828, -1.3210601806640625, -6.14764404296875, 4.644378662109375, -137.83802795410156, 8.632965087890625, -0.041473388671875, 93.95771789550781, 0.0, 22.510345458984375, 19.89141845703125, 164.14207458496094, 2.827892303466797, 65.0711669921875, 46.9859619140625, -1.021331787109375, 8.607728958129883, 20.05523681640625, -6.39788818359375, -4.126411437988281, 127.78848266601562, 12.291419982910156, 54.48479461669922, -29.44189453125, -18.9664306640625, 0.0, 11.992469787597656, 24.051429748535156, 26.2625732421875, 38.31610107421875, -33.49689483642578, -15.328392028808594, 18.959014892578125, 165.37112426757812, 178.23190307617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000360.npy"}
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 64.7015609741211, "std": 91.10247802734375, "min": -186.24508666992188, "p10": -34.041510009765624, "median": 50.85176467895508, "p90": 173.8239532470703, "max": 246.25091552734375, "pos_frac": 0.7265625, "sample": [56.27592468261719, -9.21588134765625, -134.18234252929688, 33.40850830078125, 33.486541748046875, -7.57000732421875, 1.3229217529296875, -33.5543212890625, 167.08010864257812, 47.336212158203125, 135.23880004882812, 14.49713134765625, 2.7578353881835938, 212.68460083007812, -13.782739639282227, 165.18385314941406, -3.2256603240966797, 15.67938232421875, -35.17828369140625, 1.6984405517578125, 164.56988525390625, 195.25845336914062, -62.85455322265625, 160.08792114257812, -0.04357147216796875, 195.27798461914062, 134.5393829345703, 142.83953857421875, 25.134235382080078, 195.57318115234375, 11.95013427734375, -1.4045181274414062, 128.53289794921875, 48.038108825683594, 14.758171081542969, 149.35386657714844, -146.6990966796875, 94.948974609375, 131.51211547851562, 68.58634948730469, 191.38351440429688, 193.01974487304688, 25.287551879882812, 91.68122863769531, 208.658447265625, 25.935150146484375, -119.07867431640625, -10.379486083984375, 23.692840576171875, -5.281883239746094, -1.162588119506836, 39.49055480957031, 114.72515869140625, 0.0, 36.14190673828125, 12.8489990234375, 156.8653564453125, 61.076812744140625, 3.4286117553710938, -21.649932861328125, 35.27531433105469, 146.01393127441406, -9.493463516235352, 171.48785400390625, 134.91607666015625, -50.732666015625, 47.525238037109375, 105.72083282470703, 32.89764404296875, -57.520538330078125, 82.08984375, -3.97052001953125, -88.84506225585938, -28.983245849609375, 116.14100646972656, 0.0, 114.91937255859375, -41.56758117675781, 81.95478820800781, -10.972320556640625, 213.16342163085938, 161.26437377929688, 185.01107788085938, 73.64547729492188, -186.24508666992188, 147.94113159179688, 1.83477783203125, -31.311553955078125, 246.25091552734375, 142.18942260742188, -57.068511962890625, 53.66542053222656, -134.429931640625, 164.21279907226562, 181.56118774414062, 139.0544891357422, 55.631805419921875, 18.44671630859375, 159.63812255859375, 136.3151092529297, 168.0755615234375, -7.8916015625, 214.31544494628906, 72.16934204101562, 169.52813720703125, 160.94415283203125, 91.26628112792969, 33.75604248046875, 21.2939453125, 166.92327880859375, 167.787841796875, 70.68213653564453, -17.40850830078125, 108.89067077636719, 177.82684326171875, 150.41827392578125, -6.170074462890625, 161.952392578125, 172.10842895507812, 121.80262756347656, 9.474685668945312, 140.1080780029297, -4.1884918212890625, 97.341796875, 32.53227996826172, 17.502235412597656, 157.4500732421875, -140.90362548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000361.npy"}
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 54.975860595703125, "std": 87.60921478271484, "min": -185.08258056640625, "p10": -36.85501403808593, "median": 30.76214599609375, "p90": 162.3897521972656, "max": 257.23004150390625, "pos_frac": 0.75, "sample": [22.220413208007812, 100.9329833984375, -10.640430450439453, 66.58792114257812, 152.50042724609375, 20.539901733398438, 138.29769897460938, 13.107513427734375, 74.23797607421875, 12.618133544921875, 83.30389404296875, -185.08258056640625, -146.68362426757812, -116.73855590820312, 224.23727416992188, 6.672967910766602, 177.02989196777344, -40.4859619140625, 73.75995635986328, 126.48291015625, 97.41680908203125, 149.5849609375, 257.23004150390625, 49.03924560546875, 103.30557250976562, 170.8361358642578, 132.07318115234375, -20.302581787109375, -164.71798706054688, 66.42578887939453, -25.9635009765625, -33.60167694091797, 130.9324951171875, -8.246925354003906, -82.52254486083984, 138.96136474609375, 17.30489730834961, -12.480571746826172, 154.60781860351562, 17.92510223388672, 207.8865966796875, 96.95452880859375, 190.3365478515625, 141.4230499267578, 156.76324462890625, 167.53182983398438, 138.8109588623047, -0.5499267578125, -16.85528564453125, 10.198280334472656, -38.13775634765625, 142.3182373046875, 79.85162353515625, -35.983917236328125, 2.3392162322998047, 3.60614013671875, -29.736572265625, 185.24429321289062, -7.56536865234375, 151.10061645507812, 118.47515869140625, -20.824981689453125, 31.17633056640625, 130.94305419921875, 123.05772399902344, 121.55224609375, 25.14435577392578, 18.342926025390625, 123.57231140136719, 87.19973754882812, 178.5826416015625, 67.87551879882812, 84.3942642211914, 29.8818359375, -5.368629455566406, 199.41015625, 113.21229553222656, 5.270713806152344, 100.06695556640625, 137.6659393310547, -91.22674560546875, 138.98330688476562, -89.6082763671875, 8.875137329101562, -36.305267333984375, -62.203948974609375, -29.911237716674805, 189.6732635498047, 9.281570434570312, 11.618339538574219, 25.103652954101562, 5.795501708984375, -5.97210693359375, 13.869789123535156, 25.379638671875, 122.16508483886719, 129.9737548828125, 26.172042846679688, 12.741867065429688, -10.594772338867188, -10.658294677734375, 153.630126953125, 148.56707763671875, 21.70745849609375, 2.5605239868164062, 44.89923095703125, -147.34625244140625, 13.527252197265625, 110.01019287109375, 20.47016143798828, 173.5684356689453, 9.63343620300293, 127.59286499023438, 89.02166748046875, 31.7132568359375, -10.597715377807617, 112.1749267578125, 183.419189453125, 24.4498291015625, 19.552337646484375, -64.7208251953125, 125.40106964111328, 140.99859619140625, 160.18600463867188, 30.34796142578125, -113.96511840820312, 77.80816650390625, 23.27471923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000362.npy"}
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 64.77784729003906, "std": 90.76455688476562, "min": -200.3194580078125, "p10": -22.620215415954583, "median": 42.14060020446777, "p90": 178.24290466308594, "max": 288.12249755859375, "pos_frac": 0.7734375, "sample": [156.11740112304688, -4.823905944824219, -14.28753662109375, 174.359130859375, 154.478271484375, 95.57867431640625, -0.3991985321044922, -127.0400390625, 35.2969970703125, 91.55536651611328, -16.069366455078125, 7.978462219238281, 153.345947265625, 119.84011840820312, 9.983024597167969, 100.25169372558594, 131.53958129882812, 199.45533752441406, 3.127716064453125, -27.677186965942383, 21.92333984375, 139.67333984375, 195.27017211914062, 95.46746826171875, 22.028823852539062, 185.078369140625, 198.97702026367188, 231.4007568359375, 12.696155548095703, 16.56165313720703, 156.61878967285156, -4.5585784912109375, -17.669063568115234, -6.4150543212890625, 121.55584716796875, 1.9435997009277344, -1.9299545288085938, 106.16632080078125, 159.60960388183594, 2.51629638671875, 8.625038146972656, 23.685028076171875, 100.33090209960938, 131.65921020507812, 116.66725158691406, 84.50885009765625, -13.423080444335938, 108.5350341796875, 176.61959838867188, 29.003402709960938, 144.5057373046875, 47.538299560546875, 288.12249755859375, 166.95123291015625, -137.46636962890625, 121.55320739746094, 5.5285186767578125, 100.9639892578125, 33.6341552734375, 106.34271240234375, 29.38763427734375, 38.63240051269531, 183.63038635253906, 137.99053955078125, 190.5516357421875, -0.7042770385742188, 189.97439575195312, 62.3101806640625, 39.865478515625, 5.984615325927734, 154.5880889892578, 6.220428466796875, 185.06640625, 231.625244140625, 8.815399169921875, 166.00177001953125, -99.53012084960938, -11.738235473632812, 43.80198287963867, 64.63525390625, 1.72393798828125, 9.799835205078125, -40.472869873046875, 5.594482421875, 63.89466857910156, -16.296920776367188, -2.845947265625, -6.26904296875, -118.5830078125, 15.577804565429688, 129.26947021484375, -133.77333068847656, 147.57415771484375, 4.9720001220703125, 56.89128112792969, 177.90841674804688, 162.31163024902344, -20.45294189453125, 169.50762939453125, 103.69522094726562, 23.552398681640625, 126.4957275390625, -154.33750915527344, 123.70953369140625, 179.02337646484375, 163.1441650390625, 27.963027954101562, 65.83113098144531, 170.15626525878906, 76.83804321289062, -200.3194580078125, -65.88442993164062, -32.7862548828125, -1.8447647094726562, 19.917938232421875, 40.479217529296875, 6.1981201171875, 13.1275634765625, 138.17974853515625, 177.33114624023438, 174.23855590820312, 169.76397705078125, 19.79084014892578, 186.559326171875, -31.34185028076172, 17.122344970703125, 26.73162841796875, -28.62213134765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000363.npy"}
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 47.58568572998047, "std": 90.17512512207031, "min": -216.37811279296875, "p10": -41.93712615966796, "median": 24.749488830566406, "p90": 165.66185913085937, "max": 249.93072509765625, "pos_frac": 0.6640625, "sample": [65.27740478515625, -2.4902114868164062, 138.2178955078125, 40.18706130981445, 113.41020202636719, 14.152236938476562, 135.84283447265625, 139.233642578125, 128.3541259765625, 6.9399871826171875, 42.28913879394531, 126.74737548828125, -13.98651123046875, -9.76715087890625, -13.3626708984375, 39.82469177246094, 56.739898681640625, -14.690185546875, 0.317596435546875, 122.16030883789062, -158.7353057861328, -5.47076416015625, 82.17828369140625, -163.487548828125, 39.616241455078125, -78.63223266601562, 190.52706909179688, 145.89175415039062, -9.065757751464844, 147.918701171875, 5.81744384765625, 18.69189453125, 86.03547668457031, 6.589153289794922, -25.74212646484375, -5.211883544921875, 0.7883148193359375, 144.8498992919922, 31.183074951171875, 166.36544799804688, 120.45838928222656, 9.048095703125, -188.7274169921875, 62.67384338378906, 137.70547485351562, 185.34353637695312, -32.02534484863281, 94.10604858398438, -45.814971923828125, -25.713180541992188, 151.012451171875, 165.36032104492188, -2.82470703125, 64.22784423828125, -12.619491577148438, -0.5348358154296875, 187.8065185546875, -1.6856231689453125, -216.37811279296875, 227.93629455566406, 74.57476806640625, -16.14263916015625, -9.47442626953125, -40.27519226074219, 36.993072509765625, 135.23667907714844, -31.367462158203125, 183.74169921875, -1.2173919677734375, 4.1815185546875, 33.4560546875, 158.84580993652344, 83.55810546875, -15.672882080078125, 139.8258056640625, 8.129241943359375, 191.9124755859375, 97.48954010009766, 14.906532287597656, 153.34762573242188, 141.82757568359375, -46.859466552734375, 28.279464721679688, -4.44818115234375, -4.760894775390625, 22.23980712890625, 20.848068237304688, -24.15277099609375, 27.259170532226562, 224.7362060546875, 138.642333984375, 10.876785278320312, -33.426910400390625, -3.9116668701171875, 129.46990966796875, 147.93954467773438, 8.066516876220703, 0.7552375793457031, -62.901641845703125, 249.93072509765625, 35.73333740234375, 168.6511993408203, -111.79107666015625, 213.73809814453125, 154.6304931640625, -53.989776611328125, 119.64443969726562, 103.72235107421875, -0.635162353515625, -15.533050537109375, 140.12042236328125, -1.86810302734375, -17.49852752685547, 191.75038146972656, 60.399864196777344, 7.686492919921875, 5.676513671875, -97.45698547363281, 92.4515380859375, -104.4720458984375, 14.90433120727539, 58.82892608642578, 10.255203247070312, 8.4095458984375, -47.849365234375, 106.56097412109375, 229.1378936767578, 32.441436767578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000364.npy"}
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 60.695674896240234, "std": 86.85017395019531, "min": -174.5223388671875, "p10": -31.89012756347656, "median": 31.949684143066406, "p90": 177.0141830444336, "max": 215.62667846679688, "pos_frac": 0.75, "sample": [0.0, 4.095371246337891, 141.21331787109375, 4.0220947265625, 174.9097900390625, 195.3594970703125, 20.818115234375, 59.609832763671875, 0.0, 22.32275390625, 23.3818359375, 3.307830810546875, 17.3072509765625, 183.6787109375, 15.647346496582031, 11.703857421875, 21.83856201171875, 46.789573669433594, 12.886978149414062, 172.46495056152344, 44.971832275390625, 171.99359130859375, -9.31494140625, 22.69287109375, 35.96705627441406, 57.52885437011719, -20.087905883789062, 160.49423217773438, 144.11900329589844, 161.58413696289062, 161.51296997070312, -11.153160095214844, -31.72772216796875, 131.94308471679688, -72.81341552734375, -56.788818359375, 144.21405029296875, 98.7044448852539, 132.16371154785156, 59.93304443359375, -19.889156341552734, 148.89263916015625, 187.4300537109375, 140.59085083007812, -36.821685791015625, 191.3724365234375, -61.987213134765625, 137.01077270507812, 53.31713104248047, 166.43984985351562, 146.62547302246094, -21.49835205078125, 7.625732421875, -3.7993621826171875, 17.135269165039062, 8.580963134765625, 104.09461975097656, 173.6001434326172, 157.52566528320312, -43.629180908203125, 26.7181396484375, 156.2408447265625, -38.671844482421875, -8.00244140625, 4.513881683349609, 96.3399658203125, 135.5558624267578, 146.24974060058594, 15.8604736328125, 124.52816772460938, 197.89178466796875, -9.132827758789062, 186.16802978515625, 0.2685089111328125, 11.62359619140625, 195.88534545898438, 72.53176879882812, -144.04623413085938, 176.4959716796875, 3.240062713623047, 188.42721557617188, 3.0986862182617188, 215.62667846679688, 3.357452392578125, 13.296689987182617, -15.89312744140625, 153.87020874023438, 59.62385559082031, 4.346675872802734, -18.16881561279297, 90.08818054199219, 10.33444595336914, 53.06409454345703, 178.68917846679688, -7.416007995605469, 7.5594635009765625, 3.163604736328125, -19.554763793945312, 146.31411743164062, -7.8782958984375, 29.158065795898438, -152.05943298339844, -17.46075439453125, -2.4488525390625, 148.5079345703125, 0.0, 27.981552124023438, 3.6898117065429688, 142.64503479003906, 187.95697021484375, 98.90408325195312, 113.54537963867188, 6.317653656005859, -32.269073486328125, -53.26751708984375, -174.5223388671875, 149.2686767578125, 178.2233428955078, -41.491668701171875, 143.26333618164062, 34.741302490234375, -118.7633056640625, 213.33746337890625, 110.48046875, 66.11581420898438, 146.0417938232422, 161.15382385253906, 77.87738037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000365.npy"}
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 60.39002227783203, "std": 87.5918960571289, "min": -206.6290283203125, "p10": -34.53215026855469, "median": 54.38600158691406, "p90": 169.39425659179688, "max": 225.74853515625, "pos_frac": 0.734375, "sample": [2.1110000610351562, 37.85911560058594, 225.74853515625, 152.4394989013672, 131.17245483398438, 146.62509155273438, 172.82504272460938, 125.97943115234375, 91.38458251953125, 194.94686889648438, 176.07183837890625, 150.03074645996094, 122.92862701416016, -42.293670654296875, 32.51570510864258, 131.12307739257812, 122.00068664550781, 100.07475280761719, -145.19775390625, 46.297454833984375, -206.6290283203125, 206.151123046875, -134.7117919921875, 15.795211791992188, 53.5830078125, 139.44192504882812, 132.77198791503906, 79.36029052734375, 155.44900512695312, 107.30712890625, -17.130828857421875, 162.70989990234375, 20.171127319335938, -35.456390380859375, 167.92391967773438, -181.53329467773438, -4.41180419921875, 132.4505615234375, 25.293731689453125, 55.188995361328125, 3.4363174438476562, -103.93553161621094, -49.14208984375, -19.691566467285156, -0.599517822265625, 106.59185791015625, 155.6473846435547, 9.924835205078125, 101.28277587890625, -13.90484619140625, 73.23019409179688, 37.329071044921875, 13.5625, 58.40969467163086, 122.48320007324219, -17.296173095703125, 29.62713623046875, 17.957534790039062, 43.690277099609375, 18.31409454345703, -0.66217041015625, 151.53302001953125, 111.92317199707031, 3.0552425384521484, -0.9299392700195312, 178.223876953125, 58.504337310791016, 100.53948974609375, 111.7152328491211, -73.54177856445312, 141.67727661132812, 181.285888671875, -58.59819030761719, 0.0, -28.376708984375, 179.3131561279297, 59.644287109375, 108.44146728515625, 132.63827514648438, -1.5763664245605469, 29.661972045898438, 44.19953918457031, 195.23434448242188, 79.13453674316406, -2.34716796875, 51.0986328125, 70.35638427734375, 202.03042602539062, -4.000885009765625, -6.750129699707031, -4.047149658203125, -25.82025146484375, 98.97689819335938, 15.393768310546875, 55.31068801879883, 141.63137817382812, 85.44903564453125, -6.36651611328125, 57.237548828125, 165.39015197753906, 153.3350372314453, 18.856216430664062, 5.9913330078125, 198.24496459960938, 162.9451904296875, 26.680099487304688, 140.3376007080078, -19.762638092041016, 207.04534912109375, 2.9994373321533203, 4.85272216796875, 132.57875061035156, 111.13421630859375, 5.54949951171875, -22.752532958984375, -41.43475341796875, 161.49920654296875, -136.48419189453125, 142.268310546875, 36.433319091796875, 141.62448120117188, 18.915771484375, 1.7326202392578125, 160.15663146972656, -42.4095458984375, -13.955795288085938, 175.8099365234375, -34.13604736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000366.npy"}
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 60.84651184082031, "std": 84.10665893554688, "min": -145.64663696289062, "p10": -27.874847412109375, "median": 44.37298774719238, "p90": 177.99139099121092, "max": 207.52713012695312, "pos_frac": 0.75, "sample": [140.73220825195312, 28.62371826171875, -15.9814453125, 190.463623046875, -2.058809280395508, 15.835098266601562, 38.973087310791016, -140.1878662109375, -6.25543212890625, 55.72119903564453, 158.3057861328125, 23.90776252746582, 14.581367492675781, -0.6550445556640625, 129.10397338867188, 196.09600830078125, 19.868606567382812, 176.94729614257812, -0.494476318359375, 15.45758056640625, 5.1082763671875, 204.246826171875, 96.0958251953125, -41.46392822265625, 96.86481475830078, 191.67605590820312, -145.64663696289062, -9.65667724609375, 69.51985168457031, -19.117630004882812, -26.250625610351562, 94.1932373046875, 3.318696975708008, -43.335540771484375, 152.39862060546875, -14.23895263671875, 65.89132690429688, 143.52183532714844, 126.00668334960938, -101.31097412109375, 163.79705810546875, 93.71553039550781, 10.53892707824707, 126.06414794921875, 168.55255126953125, 2.9561004638671875, -76.93865966796875, 0.753692626953125, -2.41754150390625, -14.480789184570312, 52.61279296875, 6.8071746826171875, 37.3668212890625, 55.87005615234375, 152.6129150390625, 162.02069091796875, 148.49551391601562, 108.48601531982422, 91.82917785644531, -27.9224853515625, -68.59622192382812, -28.4066162109375, 198.16531372070312, 114.31500244140625, 19.5225830078125, 182.83413696289062, -27.85443115234375, 138.10269165039062, 207.52713012695312, 51.56668472290039, 64.35995483398438, 12.366317749023438, 32.58058166503906, 114.85189819335938, 183.21377563476562, 163.84368896484375, 16.238494873046875, -25.192596435546875, 16.29412841796875, -70.03561401367188, 41.2242431640625, 181.97267150878906, 161.51171875, 172.93310546875, 16.890869140625, 29.100738525390625, 170.5897979736328, 24.190277099609375, -64.81289672851562, 111.27041625976562, 160.39137268066406, 130.02716064453125, 22.56072998046875, -6.128875732421875, 6.497201919555664, -4.063514709472656, -107.36771392822266, 180.4276123046875, 47.252872467041016, 175.68701171875, -0.37825775146484375, 187.2562255859375, 160.5103302001953, 77.86669921875, -12.216094970703125, 97.72779846191406, 72.79742431640625, 180.43463134765625, 8.3663330078125, 128.742919921875, 74.43806457519531, 108.54753875732422, 31.30731201171875, -123.19355010986328, 54.07121276855469, 3.5694351196289062, -0.9124755859375, 145.1933135986328, 109.50460815429688, 155.7411346435547, 13.034639358520508, 49.828338623046875, 18.63031005859375, 6.796875, 204.49612426757812, 121.07185363769531, -19.749053955078125, 41.49310302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000367.npy"}
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 69.19657897949219, "std": 87.94712829589844, "min": -164.77056884765625, "p10": -21.51478042602539, "median": 57.061309814453125, "p90": 174.147802734375, "max": 282.3336486816406, "pos_frac": 0.796875, "sample": [11.8072509765625, 187.64407348632812, 95.87171936035156, 161.43731689453125, 18.603851318359375, 168.28433227539062, 50.68798065185547, -4.108154296875, 6.1890869140625, -3.6905670166015625, 170.36245727539062, -31.369125366210938, 86.52255249023438, 200.431640625, 121.710205078125, 56.954437255859375, 219.82595825195312, 25.41552734375, 75.45138549804688, 117.3254623413086, -21.280609130859375, -10.794189453125, 49.5780029296875, -22.061180114746094, 4.9256591796875, 25.178558349609375, -14.5623779296875, 5.193267822265625, 12.7308349609375, 170.031494140625, 2.4888839721679688, 52.460205078125, -86.22126007080078, 65.06698608398438, 11.018798828125, -161.62750244140625, 14.55816650390625, 282.3336486816406, 143.10043334960938, 133.9415283203125, 33.060455322265625, 174.02032470703125, 158.4615936279297, 140.49000549316406, 28.7833251953125, 26.241455078125, 142.90365600585938, -8.498733520507812, 26.774805068969727, -154.40480041503906, 6.642852783203125, 148.69955444335938, 7.562765121459961, 127.643310546875, -41.652931213378906, -7.9060211181640625, 192.17681884765625, 143.41854858398438, -1.110198974609375, 123.68980407714844, 36.274932861328125, 65.36898803710938, 76.40484619140625, 159.0499725341797, 181.04519653320312, 144.0669708251953, -48.165496826171875, 98.02299499511719, 34.416717529296875, -8.880401611328125, -2.57586669921875, 191.2249755859375, 183.22195434570312, 38.92561340332031, 31.814773559570312, 138.75001525878906, 154.07501220703125, 99.2210693359375, 170.20846557617188, 145.91519165039062, 8.311630249023438, 126.3035888671875, 161.41542053222656, 57.168182373046875, -8.446449279785156, 169.9193572998047, 9.32659912109375, 60.34520721435547, -46.24131774902344, 125.8736572265625, -0.8632640838623047, 10.513832092285156, 170.75860595703125, 63.092872619628906, 115.8496322631836, 22.755271911621094, 100.19713592529297, 131.42572021484375, 0.34418487548828125, 128.92169189453125, 174.44525146484375, -73.1884994506836, 96.14164733886719, 191.863525390625, -64.38534545898438, 27.8502197265625, -10.00579833984375, 130.69589233398438, 212.04046630859375, 181.27664184570312, 28.46795654296875, 159.11793518066406, 43.842315673828125, 10.260696411132812, 15.820640563964844, 153.25355529785156, 172.93695068359375, 160.81729125976562, 143.09637451171875, 119.67593383789062, 43.420989990234375, -141.0667266845703, -164.77056884765625, 5.44769287109375, 48.86835479736328, 177.5971221923828, -98.94668579101562, 158.81698608398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000368.npy"}
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 49.775001525878906, "std": 91.45663452148438, "min": -185.68771362304688, "p10": -58.87808074951172, "median": 31.174938201904297, "p90": 166.78943328857423, "max": 234.86813354492188, "pos_frac": 0.7265625, "sample": [149.2648162841797, 25.77764129638672, 147.82728576660156, -76.65042114257812, 23.932029724121094, -9.40167236328125, 76.70420837402344, 3.61614990234375, 143.00103759765625, 0.326416015625, 80.25949096679688, 86.567138671875, -45.95931625366211, 63.54144287109375, 192.06314086914062, 175.24639892578125, 127.47416687011719, 142.86659240722656, 21.36871337890625, 136.88165283203125, 196.22726440429688, 2.5045528411865234, 101.47994995117188, 55.346221923828125, -41.90325927734375, -170.11846923828125, 159.22406005859375, -18.032379150390625, 89.89730834960938, 4.251007080078125, 138.708740234375, 25.98968505859375, 168.20993041992188, 128.7845458984375, 12.2344970703125, 147.58828735351562, -60.28343200683594, 69.56683349609375, 144.21072387695312, 87.52854919433594, 7.206756591796875, 2.92999267578125, 161.03622436523438, -35.39862060546875, 15.53375244140625, -127.699462890625, -5.308433532714844, 152.8812255859375, -123.43328857421875, 189.83477783203125, -96.70845031738281, -132.64358520507812, 83.52702331542969, -108.16943359375, 181.24810791015625, 128.52764892578125, -122.98892211914062, -8.839851379394531, 146.40283203125, 6.094512939453125, -133.41912841796875, -2.105377197265625, -135.83938598632812, 166.52992248535156, 173.01947021484375, 70.19218444824219, 62.73321533203125, -1.1566162109375, 157.493896484375, -29.760391235351562, 234.86813354492188, -14.500228881835938, 33.05774688720703, 28.3944091796875, 5.502197265625, 212.52090454101562, 68.88983154296875, -0.388885498046875, -53.703041076660156, 138.43634033203125, -1.6101455688476562, -185.68771362304688, 11.140586853027344, 66.3326416015625, 18.38299560546875, 126.9128189086914, 160.184326171875, 107.90226745605469, -37.785919189453125, 0.0, 29.292129516601562, 146.3346405029297, 35.64567565917969, 161.6565704345703, -12.945755004882812, 3.2023468017578125, 126.26655578613281, -21.899566650390625, 48.6163330078125, 160.34628295898438, 19.888427734375, 153.36520385742188, 3.60009765625, -58.275787353515625, 67.68594360351562, -42.26033020019531, 148.39712524414062, 17.873947143554688, 167.39495849609375, -80.05096435546875, 89.41539001464844, 33.92273712158203, 16.0242919921875, 150.05160522460938, 172.46966552734375, 1.8389892578125, 5.96160888671875, 182.08798217773438, -14.006820678710938, 85.34750366210938, 15.692089080810547, 182.14566040039062, 16.0958251953125, 22.340652465820312, 36.88178253173828, -8.218292236328125, 41.3673095703125, 0.956787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000369.npy"}
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 61.86237716674805, "std": 95.60497283935547, "min": -202.79791259765625, "p10": -25.874060821533202, "median": 35.90531921386719, "p90": 174.61166381835938, "max": 350.0596008300781, "pos_frac": 0.78125, "sample": [121.86927795410156, -161.94424438476562, 167.72341918945312, 35.40399169921875, 39.7989501953125, 5.6024932861328125, -65.522705078125, 2.75494384765625, 135.60023498535156, 22.529953002929688, 76.85508728027344, 111.09567260742188, -10.800407409667969, 19.987060546875, 136.80978393554688, 9.864013671875, 0.892608642578125, 248.55889892578125, 3.2064285278320312, 32.01275634765625, 152.62713623046875, 29.6387882232666, 128.84066772460938, 150.30096435546875, -24.218902587890625, -3.0570449829101562, -29.238998413085938, 153.08230590820312, 7.389614105224609, 29.0196533203125, 28.409454345703125, -38.18231201171875, 145.96925354003906, 34.32004928588867, 19.354782104492188, -0.9618988037109375, 233.056640625, 100.40980529785156, 3.721881866455078, 180.9833526611328, 135.81466674804688, -166.58323669433594, 176.10382080078125, 9.440963745117188, 128.4869384765625, 106.74253845214844, 54.59104919433594, 39.04014587402344, 41.93072509765625, 0.49953460693359375, 123.80055236816406, 17.927261352539062, 9.622795104980469, 140.84727478027344, 101.29534912109375, -18.173690795898438, 141.3413543701172, -6.722412109375, 167.953857421875, -121.83661651611328, -22.97657012939453, 165.6131591796875, -11.64996337890625, -115.18009948730469, 7.087604522705078, 143.1578826904297, 223.6121826171875, 71.11494445800781, 133.0673828125, -62.435394287109375, 4.595428466796875, -10.152618408203125, 63.0157470703125, 169.04507446289062, 180.92929077148438, -25.84429931640625, 187.08926391601562, 181.41778564453125, 126.80999755859375, 164.5904998779297, 180.39447021484375, 155.15306091308594, 45.3199462890625, 217.8265380859375, 149.94879150390625, 159.34503173828125, -25.943504333496094, 7.46337890625, -172.23516845703125, 159.83517456054688, 134.8079833984375, 12.099908828735352, -105.43441772460938, 27.12274169921875, 350.0596008300781, -7.136478424072266, 23.42047119140625, 33.97944641113281, 92.59783935546875, 161.30426025390625, 172.01133728027344, 180.22564697265625, 23.909515380859375, 173.97216796875, -10.606613159179688, 116.38008117675781, 84.94937133789062, 25.963024139404297, 36.406646728515625, 67.47528076171875, 196.51556396484375, -202.79791259765625, 7.8082427978515625, -9.249359130859375, 1.708740234375, 153.2920684814453, 12.68841552734375, 8.156623840332031, 164.52456665039062, 14.703536987304688, 17.20502281188965, -138.73602294921875, 45.723106384277344, 162.55926513671875, -11.761802673339844, 10.422134399414062, 132.9737548828125, -6.760467529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000370.npy"}
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 52.83734130859375, "std": 86.82843780517578, "min": -218.46133422851562, "p10": -30.462162780761712, "median": 40.565059661865234, "p90": 168.22913513183593, "max": 207.01052856445312, "pos_frac": 0.7109375, "sample": [-1.1557731628417969, 102.18376159667969, 52.5069580078125, 151.5218505859375, -13.232683181762695, -0.03704071044921875, 196.47528076171875, -2.2042236328125, 63.83876037597656, 20.775146484375, 1.6931190490722656, -204.83163452148438, 6.1134033203125, 15.412689208984375, 30.243545532226562, -9.666259765625, -93.452880859375, 105.25894165039062, 11.904449462890625, -23.135406494140625, -28.38702392578125, 113.95535278320312, -9.362075805664062, 174.45382690429688, 62.07899475097656, 106.10137939453125, -0.24827003479003906, 48.0777587890625, -28.626663208007812, 47.518402099609375, 139.20236206054688, 173.1715545654297, -60.33750915527344, 12.181472778320312, -218.46133422851562, 75.31600952148438, -42.5501708984375, 1.24896240234375, 131.4285888671875, 24.6923828125, 174.0154571533203, 12.640472412109375, -104.77854919433594, -5.016571044921875, 57.79150390625, 40.181060791015625, 13.49249267578125, 0.0, -52.411163330078125, 9.792709350585938, 181.57762145996094, 155.0082550048828, 101.6796875, -66.31134033203125, -149.04327392578125, -5.3813018798828125, 35.41856384277344, 183.49798583984375, 155.01626586914062, 12.978084564208984, 17.16358184814453, 98.68505859375, 79.71385192871094, 202.80364990234375, 42.81671142578125, 105.2138671875, 84.25106811523438, 142.75640869140625, 171.10470581054688, -19.963165283203125, 31.129756927490234, 157.74920654296875, 130.76531982421875, -107.37567138671875, 79.944091796875, 141.98016357421875, 141.57073974609375, 33.62843322753906, 168.271240234375, 16.247817993164062, -24.721710205078125, -154.50418090820312, 19.696197509765625, 206.33978271484375, 114.61767578125, -22.2958984375, -103.92415618896484, 144.432861328125, 144.75994873046875, 42.95484161376953, -0.6990203857421875, -19.23858642578125, 78.36862182617188, 103.45420837402344, 37.73612976074219, -1.4619522094726562, 150.856201171875, 58.800018310546875, 23.186325073242188, 207.01052856445312, 30.1452693939209, -1.5327997207641602, 103.68902587890625, 40.949058532714844, 148.8373565673828, 144.73187255859375, -1.0319061279296875, 35.184600830078125, 27.79503631591797, 144.70462036132812, 149.65231323242188, -34.7449951171875, 73.58517456054688, 60.370933532714844, 12.41094970703125, 155.2496337890625, 168.21109008789062, 26.07375144958496, 194.747314453125, 154.65176391601562, -3.29180908203125, -26.5709228515625, 48.79296875, -6.2123260498046875, 60.675865173339844, 148.57833862304688, 120.4553451538086, 181.43345642089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000371.npy"}
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 68.8276138305664, "std": 89.946533203125, "min": -178.46359252929688, "p10": -16.84309158325195, "median": 60.76247024536133, "p90": 181.09270935058592, "max": 258.79443359375, "pos_frac": 0.765625, "sample": [183.097900390625, -6.848297119140625, 142.73104858398438, 64.11746215820312, -1.5218524932861328, 143.56198120117188, 130.1822967529297, -8.594146728515625, 131.38406372070312, 176.00706481933594, 41.463233947753906, 62.25407409667969, 137.66348266601562, 140.72496032714844, -25.015304565429688, 134.09246826171875, 14.442123413085938, 182.81158447265625, -11.788976669311523, 19.681182861328125, 84.86752319335938, 34.10985565185547, 117.67172241210938, -9.834136962890625, -11.06121826171875, 127.54373168945312, 209.4500732421875, 102.65309143066406, 159.30908203125, 137.0541229248047, -26.928665161132812, -24.524383544921875, 186.36117553710938, -12.56451416015625, 114.71113586425781, 132.5331573486328, 195.5947265625, -1.9254913330078125, 111.83731079101562, 60.400108337402344, -17.990455627441406, 184.67245483398438, 87.39610290527344, 122.70304107666016, 197.81353759765625, -169.51702880859375, 204.3035888671875, 84.03244018554688, 12.438621520996094, 34.62054443359375, 2.3244705200195312, 6.93023681640625, 125.09686279296875, -0.79791259765625, 26.569427490234375, 32.0943603515625, 116.47897338867188, 160.015869140625, 129.2463836669922, 35.892791748046875, 118.55964660644531, 31.526840209960938, 137.95111083984375, 19.169525146484375, 5.88873291015625, 61.12483215332031, 160.0643310546875, 17.4140625, 173.65087890625, -16.351364135742188, 3.085479736328125, 7.40863037109375, 13.938247680664062, 165.90911865234375, 180.35604858398438, 196.581298828125, -100.88369750976562, 188.16995239257812, 18.65386962890625, -154.88693237304688, 157.40274047851562, 69.00752258300781, 36.80462646484375, 57.84619140625, -147.94149780273438, 52.493438720703125, 68.19357299804688, 14.493118286132812, -5.478404998779297, 170.20652770996094, -12.934867858886719, -13.587081909179688, 160.21697998046875, 66.63131713867188, 45.9163818359375, -178.46359252929688, -75.72560119628906, 35.157958984375, 133.0177459716797, -75.77278900146484, 170.2857666015625, -7.2311553955078125, -7.58697509765625, 2.057647705078125, 180.05152893066406, 147.95050048828125, -161.31842041015625, 180.29888916015625, 258.79443359375, 193.871337890625, 41.0931396484375, 215.49188232421875, 172.87965393066406, 4.3971405029296875, 69.17776489257812, 177.73895263671875, -1.8843746185302734, 112.021240234375, -29.68701171875, 144.71612548828125, 126.73507690429688, 147.74850463867188, 24.116924285888672, 3.8587646484375, -3.0617637634277344, 59.54447937011719, 49.36480712890625, 41.666229248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000372.npy"}
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 65.36766052246094, "std": 92.62096405029297, "min": -211.177978515625, "p10": -34.02947654724121, "median": 54.03874588012695, "p90": 186.397509765625, "max": 211.0401611328125, "pos_frac": 0.78125, "sample": [13.661865234375, 54.248268127441406, 172.525146484375, 29.510299682617188, 0.588104248046875, -16.272216796875, 18.538055419921875, 200.15121459960938, 13.14056396484375, 184.53988647460938, 156.97659301757812, -18.0020751953125, 21.651504516601562, 166.10594177246094, 105.33676147460938, 79.6727294921875, 1.5982131958007812, 59.31072998046875, -33.88409423828125, 162.39346313476562, 67.74966430664062, -0.4052925109863281, 113.75703430175781, 70.679443359375, -46.32220458984375, 49.548370361328125, -5.5920867919921875, -41.00666809082031, 34.437744140625, 197.88241577148438, 3.9589767456054688, -25.84967041015625, 31.24761962890625, 80.71023559570312, 100.0943603515625, 130.951904296875, 161.27926635742188, -18.733123779296875, 172.0791015625, 76.25924682617188, 98.15191650390625, 150.8868408203125, 171.75497436523438, 99.6225814819336, 16.8717041015625, 186.1104736328125, 18.834457397460938, -86.39605712890625, 48.10736083984375, 187.7587890625, -34.36870193481445, 168.87393188476562, -110.6351089477539, -44.4716796875, 109.21368408203125, -120.55792236328125, 200.07244873046875, 10.66827392578125, 28.0960693359375, 53.8292236328125, 7.85749626159668, -211.177978515625, 202.93338012695312, 30.325592041015625, 171.21514892578125, 16.743576049804688, -116.33558654785156, 172.4983367919922, 9.046035766601562, 137.07363891601562, 181.01861572265625, 167.78216552734375, 12.29437255859375, 89.46701049804688, 26.891738891601562, 83.40281677246094, 63.71966552734375, -0.5430908203125, 82.76889038085938, 2.434173583984375, -155.04925537109375, -15.534141540527344, 198.33499145507812, 67.39525604248047, 40.8828125, 172.06903076171875, 150.35720825195312, 201.55181884765625, 206.33218383789062, 108.43856811523438, 8.3702392578125, 182.68421936035156, -4.83123779296875, 21.192840576171875, 174.1853790283203, 172.4163818359375, 22.59619140625, 200.4476318359375, -125.6185302734375, -140.91357421875, 76.6905517578125, 164.0215301513672, 211.0401611328125, 164.38735961914062, 66.05216979980469, 15.52606201171875, 137.3865966796875, -14.273162841796875, 208.40902709960938, 18.224273681640625, 14.686222076416016, 15.003768920898438, 23.344390869140625, 102.90846252441406, -68.96626281738281, -9.136810302734375, 130.85934448242188, 6.3271026611328125, -23.459753036499023, 136.77699279785156, -3.431079864501953, 187.0672607421875, -15.057193756103516, 7.043243408203125, 175.59988403320312, 196.24322509765625, 138.681640625, 13.441070556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000373.npy"}
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 42.60383605957031, "std": 92.35050201416016, "min": -190.589111328125, "p10": -64.8571838378906, "median": 23.093643188476562, "p90": 173.06692352294922, "max": 242.2335205078125, "pos_frac": 0.71875, "sample": [217.8870849609375, 0.0, 7.62542724609375, 223.5147705078125, 4.6749267578125, -135.02468872070312, -7.192649841308594, 26.454940795898438, 173.1320037841797, 113.27098083496094, 16.07447624206543, 210.78643798828125, 19.512603759765625, -57.661376953125, 74.04521179199219, 34.51068115234375, 124.77574920654297, 139.24282836914062, 19.7318115234375, 6.1657562255859375, -135.282470703125, -51.527557373046875, 107.6471939086914, -134.629638671875, 167.80043029785156, 27.09197235107422, 21.482818603515625, 11.770111083984375, 10.344043731689453, -21.367019653320312, 6.515380859375, 7.9054718017578125, -49.912841796875, 61.2110595703125, -8.928955078125, 24.800697326660156, 23.17681884765625, -20.800048828125, 190.59783935546875, 146.57676696777344, 52.2327880859375, 8.674072265625, 242.2335205078125, 0.02657318115234375, 64.42318725585938, 2.482696533203125, 3.1814651489257812, 51.5257568359375, -190.589111328125, -16.577964782714844, 47.61009216308594, -47.195220947265625, 173.03903198242188, 94.27112579345703, 24.55609130859375, 72.15325927734375, -7.7132568359375, 60.30194091796875, 73.60940551757812, -1.2544898986816406, 28.80815887451172, 191.41156005859375, 195.684814453125, -2.711578369140625, -1.309814453125, 83.70105743408203, 51.95013427734375, 110.9366455078125, -35.481475830078125, 108.28434753417969, 21.642303466796875, 59.90716552734375, 140.2403564453125, 127.50437927246094, -173.85012817382812, 17.474639892578125, -81.64739990234375, 11.02008056640625, 16.492950439453125, 155.6111297607422, 112.65509796142578, 9.338348388671875, 54.023193359375, 157.48501586914062, 0.0, 157.14256286621094, 3.2998428344726562, -89.2728271484375, -41.415283203125, -115.00433349609375, 7.585975646972656, 139.89794921875, 156.885498046875, 179.8536834716797, 21.501220703125, -18.470703125, 208.90713500976562, 23.010467529296875, 189.58749389648438, 57.42657470703125, 17.8218994140625, -115.81393432617188, -108.47784423828125, 27.45941162109375, 22.406522750854492, 5.303962707519531, 29.98638916015625, 200.53445434570312, -20.636878967285156, -117.35726928710938, -3.07257080078125, 115.755615234375, 41.2872314453125, -5.0140838623046875, 168.92123413085938, -2.708953857421875, 157.7184600830078, 84.60362243652344, -141.18899536132812, 91.65435791015625, 222.73880004882812, 151.00320434570312, 155.3386688232422, 29.709884643554688, 36.210205078125, -3.528766632080078, 4.3059844970703125, -134.76580810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000374.npy"}
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 55.3155517578125, "std": 91.46284484863281, "min": -189.20596313476562, "p10": -34.72238540649414, "median": 28.148780822753906, "p90": 174.16046752929688, "max": 265.75018310546875, "pos_frac": 0.7578125, "sample": [5.45306396484375, 78.50177764892578, 106.1636962890625, 36.1536865234375, 85.343017578125, 3.743274688720703, 20.88507080078125, -123.62284851074219, 6.497867584228516, 38.805877685546875, 188.71337890625, 0.4941902160644531, 169.26239013671875, 169.63934326171875, 103.73521423339844, -9.2152099609375, 78.63931274414062, 120.6500244140625, 5.578121185302734, 7.116645812988281, 162.18984985351562, 174.79266357421875, 17.641586303710938, 28.87969970703125, 63.84801483154297, 164.92369079589844, -12.458251953125, 9.961753845214844, -37.06932830810547, 105.20587158203125, 16.959739685058594, 23.743019104003906, -4.50743293762207, -101.4625244140625, 11.777275085449219, 172.57638549804688, 114.5826416015625, 177.779541015625, 22.016868591308594, -29.360260009765625, 17.519882202148438, 13.358695983886719, -128.30853271484375, 1.149932861328125, -189.20596313476562, 107.44145202636719, 156.21200561523438, 161.62527465820312, 10.861488342285156, -148.64212036132812, -143.32357788085938, -16.009521484375, 117.53048706054688, -2.5858936309814453, 20.295166015625, -4.0184326171875, 143.4989013671875, -10.699920654296875, -33.716552734375, -133.17337036132812, 187.6771240234375, 170.86257934570312, -15.5660400390625, 221.03302001953125, 79.10154724121094, 22.19207763671875, -51.78041076660156, 4.345634460449219, 265.75018310546875, -6.697813034057617, 79.49006652832031, 13.354257583618164, 27.417861938476562, 83.40119171142578, 34.97003173828125, 187.6995849609375, 50.420074462890625, 1.313568115234375, 53.6195068359375, 31.015457153320312, 143.89816284179688, 105.68264770507812, -30.92596435546875, -44.83677673339844, -2.882232666015625, 126.83493041992188, 2.00189208984375, 138.63772583007812, 125.2523193359375, 159.17852783203125, -152.9278564453125, 178.50973510742188, 208.7369384765625, -14.031875610351562, 214.293212890625, 227.36331176757812, -47.00128173828125, 172.69229125976562, 96.19688415527344, 176.51513671875, -14.222896575927734, -63.758567810058594, 172.29693603515625, -1.7556610107421875, 148.9227294921875, 2.681913375854492, 118.9671630859375, 173.8895263671875, 67.70465087890625, 4.23388671875, -3.231964111328125, 185.0313720703125, 98.99197387695312, 37.212127685546875, 166.70465087890625, -31.585540771484375, 3.6240463256835938, 170.82188415527344, 2.3424224853515625, 16.645477294921875, 7.612396240234375, 23.885154724121094, 50.055511474609375, 136.22149658203125, 12.791900634765625, 147.0768585205078, 8.7669677734375, 101.2891845703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000375.npy"}
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 51.965126037597656, "std": 94.55447387695312, "min": -165.64610290527344, "p10": -55.86816864013671, "median": 39.61370086669922, "p90": 172.59647521972656, "max": 271.65234375, "pos_frac": 0.7109375, "sample": [97.30851745605469, 80.65048217773438, 177.4093017578125, 271.65234375, 130.49899291992188, 231.56884765625, 53.66973876953125, 69.75936889648438, 54.94493103027344, -124.01609802246094, 133.8079833984375, -24.359298706054688, -142.54931640625, 23.8291015625, -21.803955078125, 91.91940307617188, 1.9989852905273438, 46.54974365234375, 133.29376220703125, 177.99362182617188, 112.07156372070312, -155.044677734375, 1.7334365844726562, 8.00189208984375, 66.33175659179688, 26.833740234375, -31.971893310546875, 191.7681884765625, -3.7062835693359375, 163.78851318359375, 155.03402709960938, -8.631973266601562, 177.28765869140625, 44.26495361328125, -11.041030883789062, -17.58935546875, 168.80474853515625, 97.34004211425781, 169.57757568359375, 11.752548217773438, 0.0, 20.17596435546875, 20.822799682617188, 130.24908447265625, -53.92784118652344, 200.30615234375, -90.376220703125, 149.45150756835938, 114.55439758300781, 136.69284057617188, -152.89149475097656, 91.35870361328125, 46.51918029785156, 5.984619140625, 161.72854614257812, 9.81048583984375, 28.601776123046875, 104.60469055175781, -47.377655029296875, -20.961669921875, 112.87228393554688, -2.45220947265625, 224.03460693359375, 209.27688598632812, 30.774383544921875, 193.33831787109375, 168.12399291992188, 55.48363494873047, -14.297149658203125, -27.57427978515625, 83.995849609375, -60.395599365234375, 26.6455078125, -48.191612243652344, -4.841821670532227, 2.15362548828125, -135.9429473876953, 2.7148818969726562, -84.65422058105469, 181.1786651611328, 137.2772216796875, 56.356170654296875, 170.58596801757812, 81.16949462890625, 162.56253051757812, 135.6375274658203, 7.287021636962891, -165.64610290527344, 10.944183349609375, -100.30101013183594, 49.299774169921875, 3.222900390625, 152.221435546875, -23.103824615478516, 12.742515563964844, 141.109130859375, 40.64567565917969, -12.196136474609375, -125.74786376953125, 167.1513671875, 13.030670166015625, 129.77886962890625, -85.689453125, 147.08851623535156, 13.5965576171875, 28.81909942626953, 119.43494415283203, 24.483169555664062, 57.49436950683594, 149.90826416015625, -17.510467529296875, -41.882171630859375, -47.51137161254883, 38.58172607421875, 129.171875, -15.412353515625, 36.03387451171875, -38.75519561767578, 64.1473388671875, -3.2531967163085938, 5.584383010864258, 81.95811462402344, 111.15933227539062, 36.52459716796875, 184.98184204101562, -141.37147521972656, 201.44430541992188, 170.18115234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000376.npy"}
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 49.242523193359375, "std": 87.28379821777344, "min": -161.80609130859375, "p10": -38.66752624511719, "median": 25.698707580566406, "p90": 174.70835113525388, "max": 225.14419555664062, "pos_frac": 0.703125, "sample": [188.43359375, 129.57252502441406, 128.24781799316406, -15.325202941894531, 79.887451171875, 17.39239501953125, 8.47515869140625, 151.326171875, 148.30860900878906, 15.470367431640625, 22.54578399658203, 38.87188720703125, 141.89797973632812, -64.50631713867188, 111.03128051757812, 140.25587463378906, 186.06396484375, -33.60943603515625, 0.6326942443847656, -49.91375732421875, 144.68292236328125, 2.1461944580078125, 7.018096923828125, 56.36196517944336, -15.38709831237793, -13.640571594238281, 68.62542724609375, -8.634628295898438, 173.15403747558594, 28.478347778320312, 115.90693664550781, 12.082489013671875, 104.62983703613281, -135.96047973632812, -9.477157592773438, 160.63150024414062, 0.5556221008300781, -5.455322265625, -8.428863525390625, 43.703125, 14.33953857421875, 58.31697463989258, 14.634452819824219, -69.162841796875, 139.9005584716797, 9.00665283203125, 196.2689208984375, 67.7059326171875, 12.5159912109375, -103.35894775390625, 8.8426513671875, 208.70343017578125, -20.658065795898438, -0.02484130859375, 183.3623046875, 5.460338592529297, -33.69148254394531, -0.6547050476074219, 148.64952087402344, 144.42071533203125, 165.57919311523438, 28.5982666015625, 87.30377197265625, 114.53689575195312, -34.2489013671875, 152.6394500732422, 88.93165588378906, 0.0, -37.96160888671875, 62.05742645263672, 147.39187622070312, -9.320419311523438, 127.84315490722656, 149.34571838378906, -5.43572998046875, 40.983428955078125, 9.477256774902344, -40.314666748046875, -0.997283935546875, 141.20947265625, 23.86285400390625, 215.7308349609375, 8.538818359375, -26.511734008789062, -16.533416748046875, -104.48943328857422, 39.78108215332031, 10.308807373046875, 171.86019897460938, 44.01031494140625, 70.22576904296875, 31.462493896484375, 8.896446228027344, 152.84942626953125, -62.1690673828125, 186.92694091796875, 11.427131652832031, 23.146987915039062, 46.2655029296875, 141.27304077148438, 35.481781005859375, -122.85107421875, -13.487464904785156, 80.76446533203125, 199.4593505859375, -62.694549560546875, 205.59799194335938, 32.32086181640625, 6.489013671875, 154.25302124023438, 5.620155334472656, 11.414970397949219, 28.436538696289062, 13.5618896484375, 188.21054077148438, 147.640380859375, 178.3350830078125, -138.86587524414062, -14.21710205078125, 225.14419555664062, 27.534561157226562, -30.132110595703125, -134.2500457763672, 189.392822265625, -10.380477905273438, -9.354522705078125, -161.80609130859375, 56.31440734863281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000377.npy"}
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 49.26468276977539, "std": 89.06780242919922, "min": -174.849609375, "p10": -45.54178619384765, "median": 25.118621826171875, "p90": 178.20186462402344, "max": 219.82876586914062, "pos_frac": 0.6953125, "sample": [135.82022094726562, 2.47674560546875, 20.45379638671875, 135.56130981445312, -27.302490234375, 6.044811248779297, -28.79620361328125, 9.05322265625, 190.77566528320312, 69.04815673828125, -21.492332458496094, 6.4595794677734375, 29.959197998046875, 142.39915466308594, 140.63461303710938, -49.23902893066406, -174.849609375, 183.3271484375, 198.695556640625, -34.68048095703125, 46.626983642578125, 26.88092041015625, 11.472396850585938, -44.07481384277344, 45.06573486328125, 191.60992431640625, 219.82876586914062, 154.09536743164062, 22.819549560546875, 136.54891967773438, -1.4222869873046875, 148.22091674804688, 127.07310485839844, 21.045989990234375, 7.85650634765625, -136.95428466796875, 3.2643375396728516, 2.9750289916992188, 80.03555297851562, 98.46640014648438, 0.0, 165.17251586914062, -7.07073974609375, 128.3709716796875, 194.27587890625, -48.9647216796875, 9.897735595703125, 209.656005859375, -25.39452362060547, 188.2718505859375, 95.60302734375, 123.35423278808594, -9.590042114257812, 4.1236572265625, 9.208587646484375, -11.028900146484375, -29.53314208984375, 144.53070068359375, -20.53167724609375, 153.31668090820312, 64.10186767578125, 1.7083721160888672, 140.9178466796875, 153.2281494140625, 39.3468017578125, 134.92422485351562, -60.930267333984375, -67.7381591796875, 178.04293823242188, 74.56997680664062, 105.73031616210938, 23.66448974609375, -0.5179443359375, -29.494918823242188, -121.92106628417969, 178.57269287109375, -66.56303405761719, -93.5654296875, 42.04547119140625, -1.7628326416015625, 55.99486541748047, 4.884521484375, 87.4171142578125, 121.89469909667969, 9.754150390625, 111.38650512695312, 186.25604248046875, 8.971649169921875, 19.640411376953125, 86.126953125, 164.16104125976562, 165.5148162841797, -26.96875, -15.311019897460938, 24.633331298828125, -42.915740966796875, -73.5902099609375, 9.601394653320312, 13.29339599609375, -23.458587646484375, -113.9530029296875, 185.72161865234375, -12.813690185546875, 43.327301025390625, -145.4595184326172, -4.3441009521484375, 28.41192626953125, 127.0415267944336, -18.713623046875, 34.723419189453125, 98.5521240234375, 45.742095947265625, 0.8558807373046875, -9.195419311523438, 204.7724609375, 149.4573974609375, 176.98193359375, 175.04115295410156, 185.9075927734375, 19.240737915039062, -154.41921997070312, 105.81905364990234, 142.88656616210938, 51.4063720703125, 33.90449523925781, -18.86553955078125, 25.603912353515625, -2.852294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000378.npy"}
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 51.586612701416016, "std": 96.77461242675781, "min": -187.4210205078125, "p10": -67.26136627197265, "median": 27.102035522460938, "p90": 176.0424331665039, "max": 214.21893310546875, "pos_frac": 0.734375, "sample": [-6.966728210449219, 7.0760650634765625, 3.85296630859375, 8.600128173828125, -0.6334609985351562, -82.44035339355469, 0.46421051025390625, 159.21145629882812, 82.75381469726562, -6.399658203125, 141.37216186523438, 157.137939453125, 38.86195373535156, 65.31332397460938, 14.1923828125, 27.541717529296875, 112.69197082519531, -129.51287841796875, 194.75234985351562, 47.46598815917969, 170.8173828125, 85.240234375, 165.34007263183594, -33.36083984375, 12.273796081542969, -18.661361694335938, 160.7133331298828, 3.14599609375, 11.72613525390625, -10.666580200195312, 41.942413330078125, 176.2590789794922, 178.95489501953125, 59.9859619140625, 48.892822265625, 9.55255126953125, 10.334365844726562, 191.06741333007812, 130.77139282226562, 7.605354309082031, -8.633087158203125, -119.73199462890625, 37.558837890625, 19.798336029052734, 5.832794189453125, 207.62026977539062, 10.004581451416016, -66.71102905273438, -31.8671875, 5.85009765625, 149.15072631835938, 4.154299736022949, 41.1336669921875, -162.44570922851562, 1.378692626953125, 8.22113037109375, 171.75869750976562, 153.75445556640625, 134.10955810546875, 122.01339721679688, 117.56237030029297, -115.43124389648438, 26.662353515625, 71.67474365234375, 143.12075805664062, 214.21893310546875, -3.072540283203125, 179.47259521484375, 20.376693725585938, -29.01434326171875, -68.54548645019531, 0.0, -11.280242919921875, 60.96031188964844, 20.996917724609375, 185.73074340820312, 59.36859130859375, 162.46206665039062, 180.43014526367188, 167.2002716064453, 8.95635986328125, 133.1951446533203, 5.430877685546875, 57.793617248535156, 151.02432250976562, 167.28277587890625, -9.714279174804688, 133.92230224609375, -0.5224151611328125, 146.348876953125, 86.38899230957031, -16.252113342285156, 186.63473510742188, 3.4861278533935547, 126.00506591796875, -12.66827392578125, 2.0224761962890625, 132.90982055664062, -132.7349853515625, 206.72808837890625, 212.75634765625, -187.4210205078125, 17.358673095703125, 5.3270263671875, 173.91888427734375, 79.83065795898438, 61.85791015625, -48.09808349609375, 171.17437744140625, 161.75161743164062, 192.37020874023438, 121.17681884765625, 169.9798583984375, 58.64402770996094, -121.90093994140625, -145.84873962402344, 24.08574676513672, 6.354759216308594, -22.19073486328125, 175.9495849609375, -64.72732543945312, 128.1158905029297, -157.5065460205078, -115.23808288574219, -158.57098388671875, 25.94127082824707, -16.731536865234375, 175.3753662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000379.npy"}
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 50.753353118896484, "std": 91.7546615600586, "min": -152.60369873046875, "p10": -55.922096252441406, "median": 30.36981201171875, "p90": 179.99073181152343, "max": 262.45904541015625, "pos_frac": 0.7109375, "sample": [152.37185668945312, 56.892364501953125, 86.59371948242188, 16.928604125976562, 128.15371704101562, 17.81170654296875, -132.6128387451172, 3.113250732421875, 0.5300712585449219, -56.68341064453125, 113.95431518554688, 36.60052490234375, 157.31671142578125, -127.32172393798828, 54.9677734375, 14.189353942871094, 149.30746459960938, 45.17462158203125, 150.70883178710938, -2.6706924438476562, 0.0, 46.356292724609375, 11.625312805175781, 25.5211181640625, 27.949066162109375, 162.53851318359375, 8.58416748046875, 58.6307373046875, -0.9402618408203125, 10.710296630859375, 50.553741455078125, 216.96624755859375, 15.951095581054688, 142.62118530273438, -0.2181396484375, -152.60369873046875, 139.49974060058594, -8.429157257080078, 24.933380126953125, -19.439430236816406, 46.086273193359375, 65.10853576660156, 176.29180908203125, -40.398162841796875, -18.381202697753906, 17.034881591796875, 17.744918823242188, 186.6177978515625, 184.28607177734375, 2.151702880859375, -91.34738159179688, 60.93194580078125, 19.616470336914062, -2.69952392578125, 26.985443115234375, -98.74585723876953, -17.30062484741211, 16.7890625, 188.42941284179688, 153.76800537109375, -81.23077392578125, 170.03570556640625, 39.3135986328125, 41.796783447265625, 200.996826171875, 171.63922119140625, -114.60311126708984, -11.03997802734375, 18.378875732421875, -0.9294929504394531, 142.3118896484375, 192.87771606445312, 2.56427001953125, 129.82717895507812, 163.07907104492188, -6.757965087890625, 84.49481201171875, -44.09161376953125, 36.188018798828125, -129.14895629882812, 118.24775695800781, 262.45904541015625, -44.25300598144531, 141.4945831298828, 171.00872802734375, 56.180450439453125, 152.70919799804688, 178.14987182617188, 171.771728515625, 47.49769592285156, 36.822784423828125, 225.12124633789062, -71.79150390625, 71.99951171875, 15.856487274169922, 68.15595245361328, 7.716552734375, -41.39471435546875, 191.67041015625, 258.967041015625, 34.8718147277832, -2.7724456787109375, 62.188751220703125, 169.60214233398438, 62.372406005859375, 18.88884735107422, 28.497802734375, 48.94688415527344, -38.087615966796875, -100.70835876464844, -55.9012451171875, -55.97074890136719, 9.202713012695312, 18.90943717956543, 27.078292846679688, 124.82980346679688, -20.95025634765625, 74.7733154296875, 88.8193588256836, 233.81500244140625, 32.2418212890625, -16.27911376953125, -38.361419677734375, 190.21649169921875, -0.9306793212890625, -117.62617492675781, 211.8819580078125, -36.317596435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000380.npy"}
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 60.45320129394531, "std": 84.44866943359375, "min": -125.95564270019531, "p10": -35.192533874511696, "median": 47.18270492553711, "p90": 174.87193298339844, "max": 197.98846435546875, "pos_frac": 0.71875, "sample": [141.54736328125, 195.10833740234375, 175.02499389648438, 51.13105010986328, 193.3416290283203, -57.27862548828125, 177.34803771972656, 190.46246337890625, 121.65379333496094, 6.741828918457031, 190.83596801757812, 19.886688232421875, 16.16448974609375, 152.79470825195312, 129.89071655273438, 112.29757690429688, 0.0, 11.088226318359375, 106.14122772216797, 141.31936645507812, 84.80644226074219, 64.5712890625, 14.910934448242188, 197.98846435546875, 58.884429931640625, 61.134315490722656, 137.26852416992188, -125.95564270019531, 0.0, -19.888427734375, 69.92291259765625, 128.45840454101562, -12.136962890625, 152.25204467773438, -10.158809661865234, 59.57725524902344, 19.722076416015625, 70.85176086425781, 10.987274169921875, 122.25885009765625, 196.3074951171875, 94.80006408691406, -17.24395751953125, 144.0561981201172, 130.0152587890625, 6.307861328125, 172.88369750976562, -11.286815643310547, -81.64668273925781, 158.55862426757812, 158.42974853515625, 170.7955322265625, 43.12257385253906, 2.7633209228515625, 194.06521606445312, 175.56317138671875, 12.60540771484375, 126.86764526367188, 174.80633544921875, 81.2247314453125, -48.70112609863281, 7.6190185546875, 77.65309143066406, 9.194122314453125, -10.125579833984375, 48.38652801513672, 162.60244750976562, -1.625, -18.615631103515625, -81.05050659179688, -1.1841201782226562, -2.0789051055908203, -0.189971923828125, 159.61419677734375, -21.86151123046875, 38.22361755371094, -6.815223693847656, 106.99771118164062, 72.95462036132812, -106.75419616699219, 169.01055908203125, -28.0023193359375, 155.1995086669922, 45.9788818359375, 187.80917358398438, -4.89337158203125, -25.432586669921875, -8.895492553710938, 167.0266571044922, 13.256011962890625, 5.1356201171875, -55.857696533203125, 146.01011657714844, 133.3542938232422, 117.47329711914062, -51.57904052734375, 2.788463592529297, 8.76776123046875, 44.69078063964844, -114.80224609375, 117.03048706054688, 10.632904052734375, -87.89199829101562, 117.76876068115234, -78.779052734375, 111.72547912597656, -26.37249755859375, 8.054431915283203, -12.961257934570312, 124.91876220703125, -10.14990234375, 187.63352966308594, -58.66168212890625, 191.52630615234375, 44.8271484375, 57.67640686035156, 7.270057678222656, 10.823829650878906, 132.27206420898438, 18.045730590820312, 159.72262573242188, 166.7449951171875, 36.73400115966797, -29.40313720703125, -99.1441650390625, 152.67791748046875, 29.894805908203125, 70.160400390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000381.npy"}
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 73.3727035522461, "std": 91.76318359375, "min": -138.72959899902344, "p10": -49.8518714904785, "median": 81.22280883789062, "p90": 178.8710739135742, "max": 221.16571044921875, "pos_frac": 0.765625, "sample": [-123.2391357421875, 30.32684326171875, 85.60614013671875, 76.8394775390625, 113.97897338867188, 163.41799926757812, 154.2410888671875, 19.165130615234375, 20.119997024536133, 123.55982971191406, 135.84268188476562, -108.82565307617188, 121.12834167480469, 15.939323425292969, 184.2650146484375, 101.73312377929688, 29.536882400512695, 56.61634826660156, 134.225341796875, -13.636543273925781, 52.287109375, -138.72959899902344, 171.0163116455078, 5.04913330078125, -15.60711669921875, 192.84307861328125, 47.76991271972656, -31.708175659179688, 146.0697021484375, 117.11408233642578, -11.59552001953125, 147.03970336914062, 112.67471313476562, 35.868873596191406, 156.317138671875, 90.98627471923828, -29.02770233154297, 107.61913299560547, -7.73043155670166, 177.76263427734375, 40.95130157470703, 9.23095703125, 175.48806762695312, 26.918701171875, 170.7754669189453, 175.65423583984375, 190.5331268310547, 50.088134765625, 29.9730224609375, 10.150001525878906, 180.40396118164062, 144.91990661621094, 15.849723815917969, 43.0848388671875, -14.09619140625, 123.12567138671875, 221.16571044921875, 209.74624633789062, -105.46524047851562, -7.1703338623046875, -46.57347106933594, -138.01431274414062, 218.56112670898438, 69.80889129638672, 3.5256805419921875, 91.52615356445312, 178.74981689453125, 73.33306884765625, 195.6683349609375, 29.999114990234375, 200.52322387695312, 26.316818237304688, 46.21095275878906, -4.4412841796875, 177.10189819335938, 114.57757568359375, 143.6829833984375, 20.82098388671875, 129.01806640625, 172.0599365234375, 67.9884033203125, -30.693603515625, 56.5506591796875, 155.9095458984375, -86.88801574707031, -4.596405029296875, -79.85514831542969, 169.62420654296875, -22.02130126953125, 173.2642059326172, 178.3681640625, -28.89202880859375, 11.943038940429688, 180.0286102294922, 172.16683959960938, 85.66920471191406, 131.84494018554688, -100.47805786132812, 5.3405914306640625, -17.80035400390625, 26.055496215820312, 73.5052490234375, 111.12815856933594, 154.35488891601562, 131.46493530273438, -73.68698120117188, 51.30339050292969, 149.24880981445312, -86.3017578125, 150.5139617919922, 159.04705810546875, -15.720695495605469, 148.6731719970703, 130.45065307617188, 74.45777893066406, 134.0186309814453, -127.53021240234375, 105.20484924316406, -57.50147247314453, 132.52285766601562, 175.00375366210938, -21.19674301147461, 208.06781005859375, 131.0472869873047, 186.659912109375, 170.839111328125, 179.1540069580078, -67.19384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000382.npy"}
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 52.91009521484375, "std": 89.71863555908203, "min": -162.8207550048828, "p10": -52.518734741210935, "median": 31.441795349121094, "p90": 172.08319396972655, "max": 224.15188598632812, "pos_frac": 0.7421875, "sample": [-27.68212890625, 40.46551513671875, 152.96054077148438, 114.41156005859375, -45.47314453125, 17.154815673828125, -55.0045166015625, 14.599761962890625, -1.9783935546875, 77.45506286621094, 143.77801513671875, -137.7296905517578, 145.13682556152344, 32.8555908203125, 0.0, -13.046501159667969, 30.2725830078125, -60.909698486328125, 55.483154296875, -71.98837280273438, 21.2713623046875, 184.69705200195312, 156.58509826660156, 94.73091125488281, -122.00579833984375, 184.3340301513672, -101.65298461914062, -5.394733428955078, -36.64460754394531, 10.654327392578125, 85.17623901367188, 2.92578125, -22.688934326171875, -32.84833526611328, 163.00665283203125, -26.96075439453125, 197.580810546875, 159.2275390625, 36.36543273925781, 149.8581085205078, 94.73400115966797, 6.580162048339844, 137.65939331054688, 220.10360717773438, -162.8207550048828, 0.037139892578125, 49.78642272949219, 167.34417724609375, 44.459625244140625, 8.40362548828125, 173.345703125, 1.778900146484375, 13.630796432495117, 153.86761474609375, 116.51992797851562, 22.205291748046875, -11.496597290039062, -84.5398941040039, 0.11514854431152344, -136.55276489257812, 139.21055603027344, 151.3494873046875, 12.431482315063477, 178.83517456054688, -28.08489990234375, 14.883453369140625, 172.46722412109375, -4.794036865234375, 10.803421020507812, 21.365081787109375, 112.28633880615234, 184.2163848876953, 136.70574951171875, 143.034423828125, 85.39804077148438, 52.41444778442383, 0.2816162109375, 11.812057495117188, 28.568527221679688, 152.7596893310547, 3.6243743896484375, 127.97716522216797, 121.900146484375, 55.54066467285156, 184.35205078125, 31.894271850585938, 30.98931884765625, 22.978187561035156, 3.2573394775390625, -160.1593017578125, 174.66702270507812, 36.5613899230957, 58.58308410644531, 62.71729278564453, 18.416595458984375, 4.554773330688477, -42.623748779296875, 200.2459716796875, 148.12815856933594, 140.32891845703125, 147.11984252929688, -59.99951171875, 224.15188598632812, -109.74101257324219, -110.64727783203125, 23.648826599121094, 120.55381774902344, 20.323577880859375, 0.0, 137.33409118652344, 156.86856079101562, -3.5911312103271484, 57.254150390625, 123.8927001953125, 100.65274047851562, -11.617996215820312, -51.453399658203125, 17.110876083374023, 171.91860961914062, 123.52508544921875, 6.9059600830078125, 149.3411865234375, 204.37417602539062, -47.657554626464844, 7.94451904296875, -1.5328826904296875, 94.90678405761719, 156.88833618164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000383.npy"}
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 38.643924713134766, "std": 94.99095916748047, "min": -179.655517578125, "p10": -89.35493087768555, "median": 21.07251739501953, "p90": 167.8202117919922, "max": 224.1441650390625, "pos_frac": 0.6328125, "sample": [200.4423828125, -179.655517578125, 115.97674560546875, -22.338623046875, -25.023223876953125, 137.89199829101562, 56.831024169921875, 224.1441650390625, 20.222885131835938, -35.78080749511719, 155.01673889160156, -19.97616195678711, -43.50831604003906, -93.39910888671875, 36.355987548828125, -124.14166259765625, 160.45672607421875, -3.9767990112304688, 189.69290161132812, -22.752716064453125, 5.160051345825195, 179.94229125976562, 140.6945037841797, 144.71127319335938, -14.397140502929688, 159.38336181640625, 25.63201904296875, 147.74143981933594, 17.85400390625, -89.58294677734375, 4.0970306396484375, -67.95343017578125, -21.646240234375, 90.36099243164062, -24.23699951171875, 110.4725112915039, 24.197662353515625, 35.4893798828125, 152.05966186523438, -15.716361999511719, -5.4793701171875, -132.81185913085938, -0.0953521728515625, 14.153175354003906, 140.53945922851562, 157.1586151123047, 159.39266967773438, -170.0901336669922, 5.090362548828125, 22.598648071289062, 128.670166015625, 191.46246337890625, -86.56210327148438, -9.1826171875, -154.8212890625, 42.839515686035156, 104.96810913085938, 43.60113525390625, -125.36721801757812, -33.590301513671875, 5.224639892578125, -134.52908325195312, 134.68991088867188, 132.68429565429688, 178.41098022460938, 45.331298828125, 93.65771484375, 58.12652587890625, 104.75927734375, -4.198101043701172, -8.937389373779297, 100.69915008544922, 167.5343017578125, 13.681411743164062, 130.69451904296875, 168.48733520507812, -11.800811767578125, 17.149810791015625, -15.454784393310547, 89.4398193359375, 63.52470397949219, -119.68788146972656, 125.73240661621094, 111.4891357421875, 155.48275756835938, 214.0562744140625, 24.81231689453125, -6.416748046875, 86.18804931640625, 39.892608642578125, 184.09686279296875, -12.129325866699219, 155.6639404296875, 61.605072021484375, -5.989898681640625, 2.5107879638671875, -45.15374755859375, 181.5259552001953, 10.015045166015625, -17.678695678710938, -4.0581512451171875, -49.914306640625, 19.978759765625, -119.38522338867188, -134.353759765625, 33.30036926269531, 1.0019683837890625, 29.667133331298828, 18.1577091217041, 21.922149658203125, 197.58419799804688, -149.05474853515625, 59.41162109375, -3.766132354736328, -4.67943000793457, -14.3031005859375, 181.99981689453125, 172.9493408203125, -89.25720977783203, 134.84178161621094, 144.242431640625, -67.24838256835938, 2.7025070190429688, 17.299468994140625, 49.056060791015625, 59.83905029296875, 12.10760498046875, -72.02557373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000384.npy"}
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 53.92945861816406, "std": 89.47311401367188, "min": -176.17144775390625, "p10": -41.28186340332031, "median": 41.01842498779297, "p90": 175.4945526123047, "max": 232.69192504882812, "pos_frac": 0.7265625, "sample": [-42.0555419921875, 111.79167175292969, 172.96444702148438, 138.3743896484375, 47.8118896484375, 28.7664794921875, 163.63241577148438, 37.98724365234375, 175.32083129882812, 148.5353240966797, 29.5216064453125, 0.1549072265625, -56.44734191894531, 22.378616333007812, 25.734512329101562, 102.56614685058594, 49.588623046875, 135.29421997070312, -73.55847930908203, -16.84417724609375, 176.875732421875, -69.29266357421875, 160.95278930664062, -112.23954772949219, 16.895523071289062, 169.24545288085938, 132.78785705566406, 11.080207824707031, 138.0996856689453, 25.268524169921875, 7.7044830322265625, 213.16705322265625, 141.14288330078125, -24.7388916015625, 175.89990234375, 231.005126953125, 162.37591552734375, 34.63670349121094, 94.02086639404297, 225.95225524902344, 51.5614013671875, -33.59062194824219, 54.36940002441406, 89.15946960449219, 64.90452575683594, -131.0546417236328, 143.89039611816406, 146.9754180908203, -128.1806640625, -0.7864990234375, 232.69192504882812, 182.68011474609375, 225.09487915039062, -9.985942840576172, 134.43458557128906, 25.106781005859375, 147.74490356445312, 1.1719799041748047, -169.88278198242188, 84.01580047607422, 13.626129150390625, 193.04815673828125, -13.357330322265625, 168.57550048828125, 3.11328125, 3.8294677734375, -29.2518310546875, 135.293701171875, -2.30084228515625, 60.51141357421875, 83.572021484375, 3.9759140014648438, -34.59326171875, 11.863166809082031, 19.668014526367188, -21.740493774414062, 107.95082092285156, 146.04287719726562, 132.81915283203125, -108.08718872070312, 12.132598876953125, 122.7047119140625, 149.86093139648438, 53.34466552734375, 142.98202514648438, 12.26568603515625, -19.94671630859375, 26.69122314453125, -176.17144775390625, 2.8964996337890625, 55.4281005859375, -4.820804595947266, -2.4839630126953125, 31.981727600097656, 180.394287109375, 49.864105224609375, 0.0, 42.17311096191406, 71.60159301757812, 50.20835876464844, 5.144706726074219, 12.083023071289062, 47.509368896484375, -37.70306396484375, 191.67965698242188, -2.2878799438476562, -112.07855224609375, 119.91671752929688, -9.356109619140625, 198.36572265625, 53.298828125, 0.0, -40.950286865234375, -103.97000885009766, 136.23214721679688, 4.368621826171875, 83.1810302734375, -69.84037017822266, -6.023468017578125, -9.1934814453125, -6.14447021484375, 44.066131591796875, 88.64610290527344, 39.863739013671875, 136.06460571289062, 47.13128662109375, 177.4141845703125, 17.209197998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000385.npy"}
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 65.63438415527344, "std": 94.45350646972656, "min": -244.04010009765625, "p10": -24.227626037597656, "median": 48.74653625488281, "p90": 184.79104766845703, "max": 251.10757446289062, "pos_frac": 0.7734375, "sample": [142.9257049560547, -4.014732360839844, 5.3872528076171875, 136.8624267578125, 68.65545654296875, 4.25335693359375, 70.6043701171875, 181.27505493164062, 123.02667236328125, 156.8053436279297, 149.43482971191406, 251.10757446289062, -38.9041748046875, 34.4339599609375, 200.65377807617188, 39.76713562011719, 33.42889404296875, 30.701690673828125, 1.1847267150878906, 124.98348999023438, 112.20327758789062, 174.41317749023438, 26.9837646484375, 61.55255126953125, -14.2818603515625, -151.20623779296875, 41.41644287109375, -11.310379028320312, 168.7504425048828, 91.17245483398438, 33.446746826171875, -130.7639617919922, 201.0556640625, 36.58453369140625, -6.5629425048828125, 16.81439208984375, 15.484039306640625, -30.627700805664062, 52.62528991699219, 154.8995361328125, 236.67523193359375, 182.66641235351562, 80.89187622070312, 50.06221008300781, 77.76870727539062, 129.9407501220703, 199.6336669921875, -75.00396728515625, 64.01809692382812, 211.6300048828125, 79.473876953125, -24.3836669921875, -95.3157958984375, -36.68604278564453, 205.76824951171875, 150.16311645507812, -8.867881774902344, 0.0, -97.35879516601562, -24.160751342773438, -145.0804443359375, -111.480712890625, 93.06582641601562, 36.152587890625, 1.4969615936279297, -11.860595703125, 53.48834228515625, -1.5376701354980469, 163.2207489013672, -188.7486114501953, 193.14013671875, 22.541656494140625, 151.82086181640625, 207.49093627929688, 206.75592041015625, 117.67903900146484, 6.652042388916016, -4.171630859375, 169.0800018310547, 154.3679656982422, 114.62818908691406, 10.51275634765625, -244.04010009765625, 46.82928466796875, 4.98724365234375, 174.4219970703125, 158.47177124023438, 14.01409912109375, 145.77334594726562, 4.4692535400390625, 48.478485107421875, 19.743144989013672, 1.9720458984375, 0.0, 205.91448974609375, 101.46430969238281, -9.879425048828125, 5.6794586181640625, 94.99737548828125, 137.244140625, 169.34115600585938, 175.89759826660156, 49.01458740234375, 164.2223358154297, 14.565689086914062, 183.1617431640625, 32.7235107421875, -23.48687744140625, -3.8692626953125, 162.22039794921875, 188.59275817871094, 43.18255615234375, 195.08287048339844, 68.14888000488281, 4.88385009765625, 30.253387451171875, 31.938217163085938, 11.45709228515625, 97.47955322265625, 152.83822631835938, 30.21542739868164, 182.78875732421875, 15.016864776611328, -4.820526123046875, 170.30584716796875, 0.0, 150.03732299804688, 98.11628723144531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000386.npy"}
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 61.93959045410156, "std": 93.7007064819336, "min": -201.3564453125, "p10": -39.70298461914062, "median": 44.468177795410156, "p90": 188.07288513183593, "max": 224.2662353515625, "pos_frac": 0.6875, "sample": [77.10255432128906, 14.706344604492188, 134.292724609375, 27.227489471435547, 8.196044921875, 186.75259399414062, 205.75245666503906, -94.89736938476562, -136.19122314453125, -80.77938842773438, 0.0, 195.40939331054688, -10.304229736328125, 210.1748046875, 148.70223999023438, 85.7952880859375, 7.095611572265625, -13.5709228515625, -26.736465454101562, 78.98446655273438, -47.40306091308594, 23.089088439941406, 150.6871337890625, -39.282684326171875, 61.00603485107422, 171.90090942382812, -103.83078002929688, -21.516143798828125, 169.90667724609375, 0.07626914978027344, 161.0863037109375, -75.945068359375, 161.3988037109375, 31.664993286132812, 159.5870361328125, 157.218017578125, -16.09161376953125, 123.875732421875, 83.48509979248047, -7.297393798828125, -40.906951904296875, 143.4694366455078, 15.969512939453125, -36.271820068359375, 191.153564453125, 45.33338928222656, 206.72341918945312, 111.21441650390625, -7.734014511108398, -3.8011703491210938, -4.058311462402344, 191.40460205078125, -40.683685302734375, -13.58599853515625, 39.39862060546875, 26.076873779296875, 14.18218994140625, -111.93711853027344, 151.3148193359375, 152.31390380859375, 100.26058959960938, 35.81951904296875, 68.50735473632812, 13.720733642578125, 145.6310577392578, -126.3353271484375, 54.046875, 38.4078369140625, 7.378509521484375, -12.421371459960938, -19.406723022460938, -20.446380615234375, 52.84112548828125, 198.72817993164062, 156.47134399414062, 133.2581787109375, 128.54466247558594, 41.16478729248047, -8.151229858398438, 123.42909240722656, 12.250289916992188, -1.6766586303710938, 176.6260223388672, 23.548858642578125, 84.36495971679688, 182.9854736328125, 136.00643920898438, 4.924285888671875, 47.580078125, 149.70831298828125, -1.1798095703125, 175.28201293945312, 84.38520050048828, 21.59967041015625, 210.985595703125, 89.63836669921875, 153.80517578125, 13.283966064453125, 160.31549072265625, 153.15403747558594, 7.810140609741211, 54.95660400390625, -29.3050537109375, 152.14212036132812, -6.0742034912109375, -63.397247314453125, -201.3564453125, -26.433948516845703, 147.8724822998047, 224.2662353515625, -6.661895751953125, -7.420806884765625, 65.99425506591797, 105.61660766601562, 205.06256103515625, 149.41085815429688, 160.48968505859375, 43.60296630859375, 211.37789916992188, 14.659385681152344, -27.988433837890625, -136.59317016601562, 219.25411987304688, 202.70614624023438, -0.07775115966796875, 158.87693786621094, -20.72464370727539, 150.262451171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000387.npy"}
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 48.93608856201172, "std": 95.41944885253906, "min": -183.47146606445312, "p10": -53.25442199707031, "median": 24.623963356018066, "p90": 173.3454376220703, "max": 292.6907958984375, "pos_frac": 0.6953125, "sample": [102.1041259765625, -96.07522583007812, 138.9461669921875, 0.0, 292.6907958984375, 163.2935333251953, -6.0909423828125, 192.95120239257812, 44.467529296875, 166.66033935546875, -159.12454223632812, 148.451171875, -140.61679077148438, 0.9686622619628906, 6.5276947021484375, 29.707061767578125, 46.01837921142578, 131.983154296875, 209.77157592773438, -22.77142333984375, 105.28353881835938, 207.795654296875, -91.8631591796875, -15.57586669921875, -12.82763671875, -41.30577087402344, 99.51383209228516, 133.37091064453125, 175.08157348632812, 181.80734252929688, 46.996337890625, 15.678535461425781, 0.0, 16.2490234375, 20.034271240234375, -102.1654052734375, -18.894317626953125, 13.100250244140625, 113.63211059570312, 133.9414520263672, -84.15061950683594, 28.45684814453125, 63.51780700683594, 148.61997985839844, -28.75921630859375, -54.420867919921875, 177.6143798828125, -14.669975280761719, -6.913555145263672, 5.8773345947265625, 109.16969299316406, -11.416748046875, 180.2034912109375, -37.05366516113281, 12.575424194335938, 126.36837005615234, -4.102294921875, 184.01376342773438, 100.38458251953125, -33.60479736328125, -7.38433837890625, 82.81048583984375, 150.3794403076172, 5.0153961181640625, 22.895925521850586, -52.7545166015625, 111.80398559570312, -183.47146606445312, 191.3589630126953, -120.13154602050781, 66.30172729492188, 165.86972045898438, 134.32461547851562, 109.29740142822266, 47.858978271484375, -0.31005859375, 0.0325927734375, 169.75250244140625, 11.217933654785156, -144.5533447265625, -13.250198364257812, 190.15151977539062, 163.99691772460938, -21.282608032226562, 19.94388198852539, 10.496353149414062, 23.36907958984375, 148.7017364501953, 104.39492797851562, 46.651214599609375, -137.89697265625, 44.31517791748047, 2.1611328125, -182.44284057617188, 48.58989334106445, 191.75559997558594, 167.88604736328125, 149.6710205078125, 23.42503547668457, 163.92074584960938, 4.308563232421875, 128.71905517578125, 172.60137939453125, 203.5611572265625, -28.859161376953125, 11.850044250488281, 34.004913330078125, 22.093978881835938, -27.810302734375, 2.061918258666992, 167.51632690429688, 10.490753173828125, -36.82402038574219, 31.607147216796875, 20.514760971069336, -40.23389434814453, 141.65219116210938, 5.950035095214844, 145.8115234375, 98.71597290039062, -108.8988037109375, -30.955245971679688, 31.03045654296875, 9.5076904296875, 25.822891235351562, 138.5048065185547, 139.8482666015625, -5.070655822753906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000388.npy"}
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 62.14739990234375, "std": 92.33099365234375, "min": -175.0894775390625, "p10": -37.69233093261719, "median": 61.36830139160156, "p90": 171.33202209472657, "max": 203.50054931640625, "pos_frac": 0.75, "sample": [2.6415252685546875, -127.61958312988281, -20.141334533691406, 107.28610229492188, 152.18771362304688, 154.5694580078125, 36.40010070800781, 28.61358642578125, -9.334388732910156, 190.623046875, 25.65966796875, 164.52976989746094, 43.02838134765625, 193.6739501953125, 73.7912826538086, 149.09097290039062, 1.33746337890625, 153.85348510742188, 109.95668029785156, 198.37643432617188, -57.80763244628906, 140.22877502441406, 188.731201171875, 100.80715942382812, 4.487129211425781, 155.4407196044922, -165.87188720703125, 194.52728271484375, 37.35108947753906, 172.13821411132812, 29.7861328125, -167.42446899414062, 129.89883422851562, -38.052093505859375, -89.5595703125, -5.195037841796875, -39.04991912841797, 143.722412109375, 189.13043212890625, 192.58782958984375, 151.58921813964844, 0.0, -121.73541259765625, 14.501182556152344, 142.170654296875, -102.44366455078125, 172.2118377685547, 145.88357543945312, 159.93296813964844, 20.157318115234375, -19.332229614257812, 154.33050537109375, 49.90789794921875, 23.095060348510742, 133.33348083496094, 118.68136596679688, 31.650955200195312, 57.861846923828125, -37.53814697265625, -9.104499816894531, 55.85115051269531, 21.543991088867188, 85.90620422363281, 157.35487365722656, -111.739501953125, 144.34683227539062, 93.43807983398438, 149.06825256347656, 130.384765625, 118.48280334472656, 147.74667358398438, 27.5084228515625, 49.0823974609375, 162.80746459960938, 2.617431640625, 203.50054931640625, 88.86544799804688, 94.12158203125, -30.020263671875, -7.57275390625, 86.37100219726562, 67.28997802734375, 22.329811096191406, 58.081939697265625, 16.279739379882812, 163.25189208984375, -164.75347900390625, 130.3112335205078, 105.9814453125, 4.202850341796875, 193.92523193359375, 44.801788330078125, -0.28436279296875, 3.840412139892578, -3.7222213745117188, 86.12273406982422, 9.80999755859375, -175.0894775390625, -27.054595947265625, 170.98651123046875, -35.85400390625, 165.62420654296875, -16.11700439453125, -37.30455780029297, 102.49237060546875, 186.04092407226562, 0.0, 91.02947998046875, 148.7435302734375, 199.81251525878906, 6.447589874267578, 11.8399658203125, -4.066925048828125, 95.5537109375, 66.17933654785156, 161.0300750732422, 64.6546630859375, 9.253608703613281, -18.799423217773438, 19.43640899658203, 108.91438293457031, 160.61915588378906, -146.71981811523438, 17.38409423828125, 116.93360900878906, -3.31170654296875, 114.00053405761719, 165.51898193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000389.npy"}
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 60.26325988769531, "std": 99.47760772705078, "min": -178.8734130859375, "p10": -66.21676330566406, "median": 64.05860137939453, "p90": 178.71559143066406, "max": 209.7447509765625, "pos_frac": 0.734375, "sample": [130.1539306640625, 2.5221405029296875, 66.95193481445312, 15.448150634765625, 111.56759643554688, 147.60655212402344, 1.3861885070800781, 159.89340209960938, 133.78314208984375, 68.4207763671875, -36.8612060546875, 4.5699310302734375, 1.509765625, -57.88677978515625, 36.38153076171875, 128.07757568359375, 0.8232955932617188, 187.67495727539062, 163.31903076171875, 141.56912231445312, 120.45858764648438, -41.94354248046875, 171.90174865722656, 13.965744018554688, 78.49078369140625, 86.1466064453125, -146.8106689453125, 160.47683715820312, 155.59042358398438, 86.6376953125, 169.9224853515625, 141.0523223876953, -112.96577453613281, 48.005165100097656, 165.35940551757812, 39.50982666015625, 114.80781555175781, 23.066390991210938, -3.9553680419921875, 114.50492858886719, -169.24639892578125, 157.01339721679688, -65.04202270507812, 138.3653564453125, 85.19156646728516, -110.17910766601562, -28.81732177734375, -176.8331298828125, 149.68746948242188, -62.146514892578125, 55.12974548339844, 6.4886474609375, -143.96746826171875, 36.99383544921875, 203.15304565429688, 15.387680053710938, 61.226531982421875, 174.2453155517578, 136.09298706054688, -151.96826171875, -109.4710693359375, 76.39627075195312, 177.15074157714844, -8.304168701171875, 196.37521362304688, -151.79629516601562, 206.20028686523438, 86.85879516601562, -17.404541015625, 147.76681518554688, 66.89067077636719, 203.86538696289062, 137.59588623046875, 0.0, -13.227447509765625, -4.448028564453125, 157.3450927734375, -28.892105102539062, 148.86959838867188, 73.71087646484375, -15.107597351074219, 168.1236572265625, 208.028564453125, -178.8734130859375, -16.093658447265625, 169.075927734375, 209.7447509765625, 164.8909454345703, -0.5103759765625, 9.778228759765625, 79.84331512451172, 191.20016479492188, 36.938720703125, 126.66885375976562, 43.382171630859375, 132.67025756835938, -14.091423034667969, 177.83407592773438, 34.06681823730469, -68.95782470703125, -3.6350440979003906, -97.8974609375, 191.01791381835938, 159.63067626953125, -20.726150512695312, 37.38861083984375, 199.2984619140625, 5.65374755859375, 1.7705726623535156, 1.1571044921875, -44.891326904296875, 3.8553619384765625, -4.665931701660156, 135.20220947265625, 51.827796936035156, 27.377395629882812, 168.94094848632812, 161.90744018554688, 67.16511535644531, 180.7724609375, 146.7574920654297, 36.29425048828125, 0.2105712890625, 181.26092529296875, 182.07980346679688, -107.13204956054688, 124.12959289550781, 22.944976806640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000390.npy"}
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 58.550357818603516, "std": 84.61660766601562, "min": -154.87904357910156, "p10": -28.783631896972654, "median": 41.31429672241211, "p90": 164.57178497314453, "max": 357.177978515625, "pos_frac": 0.7421875, "sample": [-8.825508117675781, -66.94831085205078, 5.278863906860352, 86.63019561767578, 65.36614227294922, 9.688751220703125, -2.221435546875, 6.225135803222656, 22.710678100585938, 118.58868408203125, 5.257713317871094, 16.16033935546875, -22.221878051757812, 100.79988098144531, 190.603759765625, 108.75424194335938, 85.81880950927734, 107.70246124267578, -63.10462951660156, -15.789558410644531, -17.785430908203125, 162.82952880859375, 81.12094116210938, 67.89898681640625, -28.564102172851562, 43.592864990234375, 164.2229461669922, 72.58218383789062, 12.208354949951172, 40.187255859375, 1.5482025146484375, -76.1539306640625, 157.35736083984375, -7.7601165771484375, 52.332550048828125, 89.52130126953125, 132.10862731933594, 10.29664421081543, -13.631210327148438, 101.9779052734375, 9.988780975341797, 142.314697265625, 139.37294006347656, 220.12060546875, 161.611328125, -4.586151123046875, 165.3857421875, 13.654647827148438, -12.543102264404297, -3.0445556640625, 213.59799194335938, 104.3870620727539, -63.203773498535156, 82.123291015625, -3.265380859375, 117.75225830078125, 137.61111450195312, 147.1042938232422, 148.45834350585938, -32.11834716796875, 50.238433837890625, -104.27244567871094, 144.32858276367188, 36.88531494140625, 32.910003662109375, 11.748016357421875, 101.91815185546875, 8.570850372314453, 63.367401123046875, 5.8845672607421875, 1.87725830078125, 177.50296020507812, 23.9310302734375, 122.28617858886719, 132.6690673828125, -9.27423095703125, 131.5302734375, 56.664794921875, -7.478759765625, 163.27862548828125, 1.5996246337890625, 59.2117919921875, 159.71282958984375, 357.177978515625, 187.4136962890625, 12.205230712890625, -9.225318908691406, -18.851043701171875, 157.82083129882812, 13.560882568359375, 10.51995849609375, -3.8130054473876953, -154.87904357910156, 159.42779541015625, 40.711761474609375, 59.4005126953125, -34.47651672363281, 131.60118103027344, 187.24169921875, 183.4969482421875, -32.96173095703125, -51.05908203125, 25.996078491210938, -98.56329345703125, 2.679931640625, -2.1526451110839844, 41.916831970214844, 116.89773559570312, 127.82647705078125, 174.16192626953125, 13.273284912109375, 140.92239379882812, 125.28841400146484, -28.074661254882812, 93.3416748046875, 75.91317749023438, 183.03448486328125, 155.16311645507812, 217.9390869140625, 11.544586181640625, -29.295867919921875, 9.809249877929688, 32.17218017578125, 7.183219909667969, -9.745033264160156, 49.548309326171875, -151.66693115234375, 167.841796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000391.npy"}
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 73.48565673828125, "std": 86.4381103515625, "min": -157.61883544921875, "p10": -13.908675384521484, "median": 46.841880798339844, "p90": 195.69058227539062, "max": 243.83401489257812, "pos_frac": 0.796875, "sample": [165.27392578125, 133.75051879882812, 34.27626037597656, 45.464080810546875, -19.503326416015625, 3.8071060180664062, 56.7470703125, 128.04486083984375, 79.4610595703125, 110.58721923828125, 16.062057495117188, 111.13348388671875, 135.98060607910156, 33.82745361328125, -15.214813232421875, 22.504684448242188, -157.61883544921875, 225.96044921875, 89.08534240722656, -27.890533447265625, 195.68572998046875, -6.025821685791016, 6.632110595703125, 53.283843994140625, 177.8822021484375, 211.715087890625, 0.0, -44.587493896484375, 173.53567504882812, 137.1339111328125, 7.883598327636719, 37.198486328125, 195.86767578125, 46.990509033203125, 71.74254608154297, 203.40386962890625, 195.701904296875, 172.79544067382812, 147.96571350097656, 42.54997253417969, 151.1915283203125, 164.41873168945312, 17.97918701171875, 10.677093505859375, -9.098114013671875, 22.073020935058594, -7.6973876953125, 5.530029296875, 120.4053955078125, 170.6682891845703, 40.188018798828125, 177.40357971191406, 202.6527099609375, 126.53007507324219, 195.46865844726562, 142.81167602539062, 28.576171875, 118.33065795898438, 200.50262451171875, -6.951812744140625, -13.468757629394531, 125.56853485107422, 2.227783203125, 193.0347900390625, 19.05228042602539, 46.69325256347656, -28.297073364257812, 117.34521484375, 210.19131469726562, 1.61444091796875, 56.67987060546875, 186.77044677734375, 203.95065307617188, 149.13790893554688, 30.9237060546875, 197.4102783203125, 48.882781982421875, -130.39077758789062, -12.409385681152344, 0.7693748474121094, 10.064895629882812, 165.87344360351562, 5.01971435546875, 106.718505859375, 27.1695556640625, -4.571441650390625, -8.76226806640625, 5.456634521484375, 23.338455200195312, -26.597579956054688, -1.47760009765625, 0.0, 145.96173095703125, 24.28778076171875, 163.5767822265625, 205.92706298828125, 59.86224365234375, 182.6004638671875, 192.73443603515625, 0.738311767578125, 183.90365600585938, 0.14791488647460938, 65.76776123046875, 2.1718597412109375, 38.828369140625, 243.83401489257812, -3.3089981079101562, 203.53179931640625, 20.369766235351562, 70.77217864990234, -119.09004211425781, 178.49884033203125, -23.23548126220703, 10.594223022460938, 162.69708251953125, 30.330322265625, 62.85955810546875, -14.935150146484375, -11.300537109375, 67.33891296386719, 38.876007080078125, 151.20407104492188, -65.3028564453125, 162.63973999023438, -26.051544189453125, 23.63019561767578, 164.49111938476562, 32.540069580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000392.npy"}
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 54.63249969482422, "std": 93.17679595947266, "min": -201.69528198242188, "p10": -42.64376220703125, "median": 34.34819984436035, "p90": 166.31504516601564, "max": 311.6541748046875, "pos_frac": 0.734375, "sample": [23.625789642333984, -71.47518920898438, 6.029045104980469, 104.65414428710938, 116.8915786743164, -64.73101806640625, 153.30584716796875, 93.153564453125, 125.48764038085938, 5.547119140625, 11.3974609375, -3.2872314453125, 77.86752319335938, 1.1245498657226562, 156.29067993164062, 29.1168212890625, -111.87687683105469, 134.54782104492188, 83.07347869873047, 109.1898193359375, 176.38800048828125, 52.468719482421875, 12.4874267578125, 82.7545166015625, -42.70263671875, 181.98294067382812, 168.6604766845703, 162.134765625, 34.6312255859375, 30.049819946289062, 162.68499755859375, -20.558029174804688, 89.27731323242188, 144.095703125, 24.853607177734375, 28.229202270507812, -2.2159271240234375, 184.17990112304688, 25.026397705078125, 148.65725708007812, 28.429443359375, -80.86697387695312, 134.6699981689453, 166.55682373046875, -101.88519287109375, 169.06976318359375, -32.99085235595703, -38.235748291015625, -101.71954345703125, 88.5179443359375, 264.038818359375, 121.60885620117188, 8.746810913085938, 21.574417114257812, 15.778007507324219, 85.51123046875, 0.7823238372802734, -116.27668762207031, 139.84031677246094, 110.82872009277344, -185.16030883789062, 22.03887939453125, 14.87359619140625, -19.678619384765625, 21.748382568359375, 160.7729034423828, 115.55735778808594, 166.21142578125, -158.788330078125, -42.6185302734375, -39.5162353515625, 15.15460205078125, 28.308456420898438, 51.329437255859375, 15.678821563720703, 61.11041259765625, 207.06033325195312, -27.917572021484375, 49.164459228515625, -6.5643768310546875, 9.971805572509766, -10.518562316894531, 152.83001708984375, -168.27601623535156, 34.0651741027832, 119.80421447753906, 156.98062133789062, 103.68324279785156, 148.59735107421875, 159.1378173828125, -28.065170288085938, 181.92355346679688, 161.82870483398438, 167.141357421875, 169.95762634277344, 36.98577880859375, 49.655792236328125, -42.19439697265625, 154.11830139160156, -201.69528198242188, 5.72314453125, 37.518829345703125, -6.28369140625, -5.403816223144531, 311.6541748046875, 114.51687622070312, 54.427490234375, 0.0, 10.518157958984375, 74.13601684570312, 215.6116943359375, -43.647430419921875, 3.3978652954101562, 8.990570068359375, -16.123321533203125, 122.84078979492188, 153.85072326660156, -16.38600730895996, -20.604263305664062, 5.7684326171875, 0.12072181701660156, -27.704132080078125, 161.00457763671875, 163.89413452148438, -1.233978271484375, 96.283203125, 124.82670593261719, 153.56820678710938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000393.npy"}
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 52.98966979980469, "std": 88.51522064208984, "min": -158.82765197753906, "p10": -45.19986267089843, "median": 35.32505798339844, "p90": 176.38988342285157, "max": 233.12432861328125, "pos_frac": 0.6796875, "sample": [175.502197265625, 72.98110961914062, 177.50753784179688, 143.1964111328125, 50.78350830078125, 157.65936279296875, -14.4261474609375, -7.431999206542969, 24.83160400390625, 180.8675537109375, 60.02545166015625, -5.932891845703125, 33.7569580078125, 69.706298828125, -31.2298583984375, -2.3147850036621094, 169.54734802246094, -49.447288513183594, 174.18817138671875, -22.81332778930664, 2.195648193359375, 164.85589599609375, 216.37435913085938, -16.608047485351562, 1.1376285552978516, -104.52691650390625, 42.699066162109375, 95.80046081542969, 111.193603515625, 47.6671142578125, -22.91064453125, 0.0, -51.92113494873047, -59.693939208984375, 175.910888671875, 12.776092529296875, 16.079345703125, 125.80066680908203, 5.223419189453125, 36.893157958984375, 78.44842529296875, 85.44410705566406, 49.37640380859375, -158.82765197753906, 20.99114227294922, 172.78619384765625, 173.39642333984375, 40.280216217041016, 1.406402587890625, 17.50152587890625, -21.46246337890625, -44.968719482421875, 59.555938720703125, -13.75823974609375, -22.87103271484375, 186.1295166015625, 29.420242309570312, 86.083984375, 11.106201171875, 57.077880859375, 96.79298400878906, -54.10650634765625, -0.6141128540039062, -35.68263244628906, 170.12652587890625, 2.8571109771728516, 53.372093200683594, 100.998291015625, 13.852165222167969, 101.80587768554688, 161.0586700439453, 154.2412872314453, 185.58108520507812, 3.2443466186523438, 211.9208984375, -3.959716796875, -63.8726806640625, -32.572784423828125, -7.552459716796875, 233.12432861328125, 30.09731674194336, -129.90386962890625, 63.08612060546875, -27.67596435546875, -49.975677490234375, 19.986846923828125, 37.07635498046875, 29.6883544921875, 8.459869384765625, 104.1278076171875, -1.4450225830078125, -16.21715545654297, 119.54415893554688, 16.095481872558594, -24.750823974609375, -132.32022094726562, 75.72576904296875, 43.00604248046875, 108.43733215332031, 81.1971435546875, -139.9088592529297, 155.9993133544922, 31.356353759765625, -33.58848571777344, 191.97164916992188, 218.751953125, -30.295578002929688, 55.201904296875, 113.28501892089844, 181.55963134765625, 126.46737670898438, 139.84616088867188, 170.06134033203125, 154.71542358398438, 174.49853515625, -45.73919677734375, 211.930908203125, 16.88330078125, -14.19805908203125, 181.53997802734375, 15.7679443359375, 177.60302734375, -83.7977294921875, 141.15853881835938, -39.64965057373047, -12.96978759765625, -37.982025146484375, 154.33367919921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000394.npy"}
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 55.392494201660156, "std": 87.21307373046875, "min": -159.66830444335938, "p10": -25.688793945312497, "median": 26.7847900390625, "p90": 178.91783294677734, "max": 216.980224609375, "pos_frac": 0.765625, "sample": [48.64605712890625, 6.45159912109375, -9.026931762695312, -8.675338745117188, 0.5849609375, -16.23529052734375, 4.204071044921875, 214.81121826171875, 201.69195556640625, 16.616165161132812, 20.170555114746094, 78.53045654296875, 29.25567626953125, 19.82391357421875, 0.5928478240966797, 40.57566833496094, 1.1804733276367188, -116.74459838867188, 79.46708679199219, 6.205474853515625, 25.762847900390625, 168.76358032226562, -20.15380859375, -110.84445190429688, 188.81912231445312, 126.61367797851562, 12.716781616210938, 3.114917755126953, 152.16754150390625, 43.57835388183594, 140.4679718017578, 143.49432373046875, 120.69831848144531, 115.4886474609375, 31.575836181640625, 178.670654296875, 95.09228515625, 51.61614990234375, 0.01338958740234375, -12.829216003417969, 92.3831787109375, 12.799375534057617, -13.0853271484375, 9.686187744140625, 158.510986328125, 140.00900268554688, -11.016937255859375, -68.67343139648438, 137.527587890625, 180.1334228515625, 197.49468994140625, -80.86126708984375, 63.663482666015625, 158.55911254882812, 173.16482543945312, 125.6539306640625, 42.720916748046875, 176.56338500976562, 16.82781982421875, 111.06834411621094, -14.130050659179688, -95.78607177734375, 24.27655029296875, -28.990203857421875, 166.66529846191406, -24.53350830078125, 19.76958465576172, 79.11801147460938, -15.735076904296875, 11.36920166015625, -28.38446044921875, 6.8078460693359375, -10.341064453125, -24.24847412109375, 40.6612548828125, 120.00444030761719, 151.47691345214844, -3.3308563232421875, 18.4088134765625, 5.4534912109375, 26.790618896484375, 20.383399963378906, 216.980224609375, 178.37872314453125, 171.3682403564453, -21.1876220703125, 2.3980255126953125, -76.64363098144531, 26.778961181640625, 43.84869384765625, 9.51687240600586, 42.38407897949219, -150.46151733398438, 125.41300201416016, -14.8028564453125, 18.272262573242188, 187.42913818359375, 194.8956756591797, 189.62660217285156, 177.244873046875, -9.862190246582031, 146.60263061523438, 27.42584228515625, -38.396270751953125, 124.711181640625, 53.845672607421875, 166.22975158691406, 10.466766357421875, 161.45065307617188, -134.67140197753906, -2.8172454833984375, 151.0017852783203, 189.84259033203125, -70.265625, 11.898773193359375, -159.66830444335938, 35.40081787109375, 6.333797454833984, 188.58877563476562, 184.8424072265625, 172.710205078125, 179.4945831298828, 20.68206787109375, 12.66363525390625, 26.39227294921875, 103.10601043701172, 8.648178100585938, 60.35491943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000395.npy"}
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 55.85353088378906, "std": 93.86991119384766, "min": -212.93557739257812, "p10": -46.68001403808592, "median": 46.101409912109375, "p90": 175.27839355468748, "max": 231.6683349609375, "pos_frac": 0.703125, "sample": [136.776611328125, 15.127197265625, -11.50384521484375, 176.403076171875, 159.57052612304688, 16.170074462890625, -6.049102783203125, 17.80584716796875, -39.69091796875, 195.06777954101562, 93.68927001953125, -23.428253173828125, 87.99391174316406, 47.02398681640625, 2.3457489013671875, 77.72769165039062, 193.49148559570312, -96.38092041015625, 7.997901916503906, 144.2264404296875, -32.1141357421875, -11.408866882324219, -12.073486328125, 228.9237060546875, -5.0873565673828125, 8.798965454101562, 81.20709228515625, 144.20901489257812, 60.42811584472656, 131.83236694335938, 106.00042724609375, -124.41224670410156, 76.31576538085938, 153.88037109375, -132.86572265625, 177.80294799804688, 10.741432189941406, -6.5927734375, -6.240573883056641, 115.93052673339844, 43.22969055175781, 147.15292358398438, 80.58999633789062, 159.42642211914062, 192.70361328125, -19.876953125, 14.641523361206055, 7.44964599609375, -7.23876953125, 127.8461685180664, 202.61782836914062, 190.97735595703125, 231.6683349609375, 5.237094879150391, 167.26913452148438, 90.50042724609375, 158.10284423828125, 0.9881744384765625, 16.723419189453125, 22.6422176361084, 7.6742401123046875, -43.0819091796875, -79.54508972167969, -82.39031982421875, 111.9857177734375, -73.21173095703125, 48.948883056640625, -153.9754638671875, 13.375579833984375, 25.117515563964844, 150.51885986328125, -7.55828857421875, -42.643157958984375, -8.768562316894531, 152.06686401367188, 154.9307861328125, 67.194091796875, 152.76031494140625, 25.133026123046875, -158.7388153076172, 74.04742431640625, 121.95394134521484, 35.380577087402344, 147.2713623046875, 55.93951416015625, 174.79638671875, 23.96445083618164, 199.9485626220703, -121.78734588623047, 162.92626953125, -28.50341796875, 89.35064697265625, -55.075592041015625, 75.91279602050781, 88.96913146972656, 126.30397033691406, 184.42364501953125, -5.90447998046875, 18.44598388671875, 45.1788330078125, 170.76148986816406, 109.45280456542969, -2.64837646484375, -212.93557739257812, -28.465744018554688, 168.7835235595703, -11.898574829101562, 193.3829345703125, 166.65643310546875, 176.7757568359375, 0.0, 19.56500244140625, 166.12173461914062, 42.64892578125, 44.67449951171875, 8.677909851074219, 0.0, 49.64387512207031, 132.1735382080078, -18.978927612304688, 53.522674560546875, -142.37078857421875, 86.59933471679688, 159.69874572753906, 156.62771606445312, -93.50180053710938, -40.18756103515625, 128.8465118408203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000396.npy"}
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 53.75213623046875, "std": 93.56927490234375, "min": -193.607666015625, "p10": -58.11501312255859, "median": 43.16307830810547, "p90": 173.73407287597655, "max": 210.481201171875, "pos_frac": 0.703125, "sample": [-9.43524169921875, 6.732208251953125, -13.517845153808594, 28.545364379882812, 183.42578125, 19.68853759765625, -97.70935821533203, 48.8912353515625, -4.565788269042969, 19.16485595703125, 9.446121215820312, 14.3785400390625, 113.92878723144531, -27.61053466796875, 26.635177612304688, 173.23037719726562, 58.933929443359375, 130.8614044189453, 207.60867309570312, 143.381591796875, -145.9533233642578, 26.90838623046875, -23.909912109375, 47.12005615234375, 141.33871459960938, 160.72396850585938, 0.0, 4.61395263671875, 196.82940673828125, 43.2899169921875, 0.0, -77.7154541015625, 136.37911987304688, -4.1588592529296875, -5.31787109375, 2.556365966796875, 7.7183837890625, 162.3601837158203, -79.51019287109375, 150.96563720703125, 181.13082885742188, 78.98686981201172, 108.393310546875, 174.29391479492188, -31.131072998046875, 1.6598587036132812, 149.32418823242188, 161.04901123046875, 210.481201171875, -43.5849609375, 208.65127563476562, -48.04571533203125, 195.14324951171875, -193.607666015625, -118.2730712890625, -40.37530517578125, 46.95965576171875, -133.46388244628906, -1.3029937744140625, -158.79901123046875, -40.67626953125, 110.99089050292969, 170.984130859375, 146.5557403564453, 100.49284362792969, 32.032989501953125, 139.09817504882812, -10.833282470703125, 50.08915710449219, -1.4477615356445312, 63.929107666015625, 192.796875, 76.30746459960938, 98.51808166503906, 154.77557373046875, 165.05548095703125, 137.64974975585938, -56.28961181640625, 102.59335327148438, 51.64617919921875, 158.93341064453125, 29.254547119140625, 183.77337646484375, 176.89987182617188, 40.535247802734375, 0.8318347930908203, 153.78623962402344, -3.908935546875, 13.176239013671875, 9.583526611328125, 136.877197265625, 15.918960571289062, 12.260391235351562, 13.207878112792969, 161.789306640625, 151.68954467773438, 17.426109313964844, 38.83589172363281, 180.05929565429688, 87.42581176757812, 20.7470703125, -139.9605712890625, 45.30328369140625, -15.345382690429688, 193.45578002929688, 169.8758544921875, 135.5841522216797, 3.866912841796875, -14.852798461914062, -20.327972412109375, -25.536514282226562, 173.494140625, 139.22003173828125, -77.19012451171875, 63.4921875, 43.05517578125, -127.0399169921875, -11.879425048828125, 149.24598693847656, 43.27098083496094, -43.570648193359375, -62.37428283691406, 139.44027709960938, 108.84349822998047, 80.23516845703125, -115.37979125976562, 129.302001953125, 108.93153381347656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000397.npy"}
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 41.136085510253906, "std": 99.16651916503906, "min": -178.47171020507812, "p10": -82.02593460083007, "median": 20.006515502929688, "p90": 175.14246215820313, "max": 259.6773681640625, "pos_frac": 0.65625, "sample": [14.3145751953125, 16.935760498046875, 157.26107788085938, -13.5340576171875, 259.6773681640625, -0.52587890625, 60.20970153808594, -53.89990234375, 82.20477294921875, 176.64132690429688, -136.70819091796875, -61.76715087890625, -172.59103393554688, 4.01837158203125, 8.810493469238281, -36.180877685546875, 35.15899658203125, -4.807929992675781, 170.39907836914062, 181.62255859375, 170.97750854492188, -142.8460693359375, -121.5107421875, 133.23443603515625, 37.0391845703125, 126.95669555664062, 15.991531372070312, 161.79306030273438, 158.99884033203125, 88.47660827636719, -107.49142456054688, 6.7443084716796875, 44.08935546875, -178.47171020507812, -73.18533325195312, 151.26023864746094, 10.4061279296875, 147.6876220703125, 0.94549560546875, -124.53313446044922, 1.5892314910888672, -85.6536865234375, 18.175933837890625, 187.51962280273438, -0.2660989761352539, -56.276123046875, 7.1575927734375, 141.193115234375, 21.83709716796875, -9.518524169921875, 32.86431884765625, -18.37689208984375, 99.27619934082031, -15.592453002929688, 138.37045288085938, 45.35296630859375, 13.292098999023438, -126.09341430664062, 25.410171508789062, -19.318710327148438, 21.997802734375, 52.87371826171875, 171.07830810546875, -21.8980712890625, 194.3759765625, 3.17132568359375, 23.24163055419922, -154.900146484375, 127.1173095703125, 181.55909729003906, 29.432968139648438, 207.34820556640625, 36.653961181640625, 213.56631469726562, -15.18865966796875, -19.405242919921875, 9.89478874206543, 170.57122802734375, -32.4764404296875, -13.47039794921875, 182.35498046875, 186.73129272460938, 11.446968078613281, 32.70111083984375, 45.69988250732422, 7.45404052734375, -68.06623840332031, 149.5118865966797, 50.239044189453125, -52.60418701171875, 136.47286987304688, 102.05038452148438, -12.960403442382812, -157.92138671875, 48.431640625, -6.49920654296875, 148.26873779296875, -27.244598388671875, -1.1392822265625, 96.95034790039062, -17.712806701660156, -12.80865478515625, 151.41534423828125, 131.1225128173828, 14.85009765625, 15.896018981933594, 57.18934631347656, 179.0480194091797, 170.3212890625, -38.945587158203125, -80.47118377685547, 194.1551513671875, 154.161865234375, -100.44995880126953, 171.97357177734375, -3.616790771484375, -35.09513854980469, -178.27349853515625, 13.061187744140625, 174.50009155273438, 145.76751708984375, -9.055267333984375, 7.013368606567383, 226.48550415039062, 78.81500244140625, 32.56196594238281, 123.72822570800781, 47.61578369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000398.npy"}
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 51.00968933105469, "std": 88.33368682861328, "min": -180.99063110351562, "p10": -32.82328948974609, "median": 35.003623962402344, "p90": 174.23858032226562, "max": 220.35699462890625, "pos_frac": 0.703125, "sample": [-8.839981079101562, 150.51268005371094, 153.3333740234375, 157.16903686523438, 12.181365966796875, -31.507827758789062, -0.112030029296875, 40.41400146484375, 33.6636962890625, 183.18923950195312, 51.84088897705078, -11.32977294921875, -27.89111328125, 160.6739959716797, -2.4729232788085938, 43.270538330078125, 105.95794677734375, 5.604339599609375, 175.44676208496094, 43.84156799316406, -3.570220947265625, -116.62252807617188, -20.20196533203125, 171.34405517578125, 14.606475830078125, 0.5592041015625, 35.08213806152344, -138.60296630859375, -180.99063110351562, 132.17987060546875, -1.410858154296875, 176.8095703125, 13.061504364013672, 181.5054931640625, 18.968910217285156, -14.970794677734375, 156.74237060546875, -8.707786560058594, 4.532073974609375, 191.2479248046875, 34.92510986328125, 163.8555145263672, 220.35699462890625, 162.245361328125, 14.44476318359375, -12.018821716308594, 173.89447021484375, -77.84732055664062, 5.491598129272461, 155.2567138671875, -46.014129638671875, 35.760108947753906, 125.9107666015625, 118.43746948242188, 195.2176513671875, -13.24200439453125, 19.873981475830078, 11.63653564453125, 17.489120483398438, -9.989704132080078, 36.844451904296875, 175.04150390625, -8.64886474609375, 35.49365234375, 160.6573486328125, 36.182926177978516, 159.30921936035156, 20.427600860595703, 97.53497314453125, 45.6944580078125, 55.98939514160156, 134.57443237304688, -11.55712890625, -134.984619140625, -176.71054077148438, -126.83311462402344, -12.8035888671875, 35.78448486328125, 173.5291748046875, 179.41842651367188, 24.189849853515625, 156.001953125, 74.5679931640625, 146.8040771484375, 135.28469848632812, 63.2547607421875, 13.37213134765625, -26.208572387695312, 36.655517578125, 146.32199096679688, 20.155555725097656, -42.779205322265625, 19.69806671142578, 87.38284301757812, 5.8238983154296875, 19.873275756835938, -76.99642944335938, 2.953075408935547, 178.74594116210938, 7.353912353515625, 78.2501220703125, -3.684600830078125, 69.8382568359375, -51.600372314453125, -6.6776123046875, 176.68081665039062, -22.071014404296875, 211.75570678710938, 131.21725463867188, 102.57940673828125, 185.7198028564453, -145.06698608398438, -35.8927001953125, 41.895965576171875, 24.805625915527344, 24.49249267578125, 113.19041442871094, 135.64825439453125, -9.2232666015625, 151.24200439453125, -26.30926513671875, 40.8704833984375, 130.87161254882812, 92.52848052978516, -31.479766845703125, -1.42724609375, 122.37176513671875, 19.12353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000399.npy"}
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 58.935123443603516, "std": 90.68914031982422, "min": -192.42071533203125, "p10": -46.46702575683594, "median": 40.03338623046875, "p90": 172.85487518310546, "max": 213.96524047851562, "pos_frac": 0.71875, "sample": [5.3976287841796875, -78.16899108886719, -0.125946044921875, 38.75201416015625, 136.4677734375, 19.41980743408203, 128.671875, 128.73789978027344, 69.88307189941406, -107.5796890258789, 5.898963928222656, 90.35638427734375, -192.42071533203125, 23.192691802978516, 180.8005828857422, 17.85272216796875, 6.2541961669921875, 163.81797790527344, 136.63916015625, 153.4290008544922, 206.60894775390625, 15.639617919921875, 35.37968826293945, 63.71441650390625, 163.85995483398438, 106.48440551757812, -46.736083984375, 159.97007751464844, -21.8460693359375, 149.722412109375, 50.796051025390625, 9.124893188476562, 15.400123596191406, 134.90301513671875, 45.20960998535156, 178.70123291015625, 167.60223388671875, 108.7806396484375, 24.068077087402344, 46.19572448730469, -146.60726928710938, 146.91653442382812, 135.34674072265625, 171.77938842773438, 16.40625, 168.73849487304688, -2.47344970703125, 151.32949829101562, 5.4376220703125, 10.013946533203125, 41.31475830078125, 147.95108032226562, 183.33258056640625, -128.02008056640625, -0.9512939453125, -32.15216064453125, 7.984194755554199, 14.858711242675781, 178.21517944335938, -2.356403350830078, -1.4497661590576172, 48.3642578125, 176.49618530273438, 146.89508056640625, 30.7698974609375, -37.7578125, 19.007171630859375, 163.06912231445312, 62.154327392578125, 182.14610290527344, -13.68505859375, 4.758735656738281, -24.45404052734375, 27.45721435546875, -59.94920349121094, -63.085479736328125, 169.69036865234375, 178.82835388183594, -0.738555908203125, -46.351715087890625, 90.49009704589844, 52.525177001953125, 132.860107421875, 131.56689453125, -3.6934852600097656, 172.57391357421875, 142.49053955078125, -21.057281494140625, 37.2105712890625, 81.22984313964844, 94.88898468017578, -25.704345703125, -5.4961395263671875, 2.1724586486816406, 3.26177978515625, 158.85382080078125, -47.700439453125, 133.22264099121094, 3.359048843383789, -120.14111328125, 67.97021484375, 184.27699279785156, 140.633056640625, 24.361114501953125, -36.6510009765625, 15.359619140625, 140.4587860107422, -7.2950439453125, -8.097320556640625, 173.5104522705078, 152.6788787841797, -3.4319839477539062, 166.0093994140625, -15.977813720703125, 164.37164306640625, 167.5028839111328, 161.95196533203125, -22.891845703125, 213.96524047851562, 193.0139923095703, 178.8942413330078, 166.91159057617188, 0.0, -74.8919677734375, -82.63504028320312, 3.1094970703125, -145.2347412109375, 42.82501220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000400.npy"}
{"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 70.60415649414062, "std": 88.19844055175781, "min": -235.75177001953125, "p10": -22.52549362182617, "median": 61.068328857421875, "p90": 181.4239715576172, "max": 244.6971435546875, "pos_frac": 0.796875, "sample": [5.4635162353515625, 5.708038330078125, 24.8387451171875, -10.472686767578125, 20.1109619140625, 126.42756652832031, 13.38592529296875, 136.82354736328125, 88.61520385742188, -5.63641357421875, 160.33428955078125, 21.268890380859375, 58.932708740234375, 173.17776489257812, 132.47348022460938, 35.937469482421875, 96.7232666015625, 151.817626953125, -6.58050537109375, 139.0885009765625, -10.880691528320312, 52.0020751953125, -21.56444549560547, 108.0533447265625, -3.9891357421875, 174.52145385742188, 90.25057983398438, 1.46063232421875, 159.765625, 76.81979370117188, 187.94754028320312, 213.82049560546875, 150.1477813720703, 79.84979248046875, 14.622802734375, 15.447723388671875, 156.32041931152344, 145.76663208007812, 10.360054016113281, 149.56201171875, 244.6971435546875, -20.60919189453125, 168.38438415527344, 182.86163330078125, 192.674560546875, 163.21005249023438, 59.007667541503906, 17.820068359375, 160.0196533203125, 89.09725952148438, -100.5718994140625, 151.249755859375, 30.90875244140625, 10.392082214355469, -235.75177001953125, 82.48089599609375, 10.1014404296875, 202.352783203125, 113.90106201171875, -34.000152587890625, -63.17852783203125, 145.28854370117188, 194.9310302734375, 142.23886108398438, 139.79498291015625, 114.05606079101562, -85.41293334960938, -100.24958801269531, 167.93545532226562, -33.461212158203125, -0.4537391662597656, 3.4319992065429688, 210.21484375, 61.38348388671875, 39.446868896484375, 3.775390625, 165.81651306152344, 94.6055908203125, 52.40338134765625, 79.14730834960938, 99.106201171875, -25.003570556640625, 155.47726440429688, 179.86895751953125, 173.89683532714844, 193.306884765625, 46.150848388671875, -3.86102294921875, 60.753173828125, -24.557769775390625, 96.80914306640625, 32.655120849609375, -10.5936279296875, -17.167633056640625, 81.59703063964844, -25.3118896484375, -27.701751708984375, 186.61846923828125, 213.51329040527344, 22.1358642578125, 46.6973876953125, 0.60430908203125, 180.80783081054688, 169.0684356689453, 183.34048461914062, 26.62274169921875, 17.011688232421875, 26.354736328125, -21.654518127441406, 70.161376953125, 22.73004150390625, 1.9989013671875, 14.4090576171875, 15.034679412841797, -17.220592498779297, 0.32657623291015625, 109.66860961914062, 149.73387145996094, 100.9991455078125, -39.87421417236328, 81.94996643066406, 166.26698303222656, 169.51904296875, 8.471466064453125, 208.4861602783203, -184.4790496826172, 9.652603149414062, 174.2894287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000401.npy"}
{"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 55.365875244140625, "std": 90.60821533203125, "min": -164.8189697265625, "p10": -37.1639820098877, "median": 39.13102722167969, "p90": 171.49954071044922, "max": 250.49661254882812, "pos_frac": 0.7578125, "sample": [-4.49267578125, -2.24896240234375, -137.794189453125, 47.37774658203125, 161.25827026367188, -18.365890502929688, -76.366943359375, -32.7960205078125, 112.70440673828125, -164.8189697265625, -142.58892822265625, 143.47415161132812, 13.96417236328125, -36.64866256713867, 7.420036315917969, 222.03964233398438, 154.6522979736328, 139.85763549804688, 23.012649536132812, -27.73748779296875, 147.41432189941406, 2.443450927734375, 12.686210632324219, 191.2064208984375, -140.3833770751953, 4.469581604003906, 1.1331787109375, 152.59756469726562, -160.08035278320312, 38.247772216796875, 138.040283203125, -6.4927978515625, 55.83587646484375, 122.845458984375, 9.25653076171875, 0.0, 222.73712158203125, 162.64291381835938, -35.599517822265625, 153.77740478515625, 96.91813659667969, 32.48089599609375, 155.318115234375, 15.978622436523438, 122.35858917236328, 157.00326538085938, -94.84226989746094, -11.622894287109375, 182.83056640625, 150.2943115234375, 26.0679931640625, -89.09870910644531, -38.36639404296875, 112.65005493164062, 128.57034301757812, 40.0142822265625, 71.25614929199219, 101.81013488769531, 3.313995361328125, 124.74260711669922, 12.322061538696289, 204.73388671875, 168.40750122070312, 12.88226318359375, 131.24331665039062, -24.242782592773438, -110.12220764160156, 61.09716796875, 17.939117431640625, -150.1829833984375, 171.86117553710938, 159.76947021484375, 41.905242919921875, -2.6326904296875, 146.27780151367188, -7.1285858154296875, 250.49661254882812, 4.697322845458984, 72.2154541015625, 179.14010620117188, 168.9659423828125, 171.40074157714844, 61.04541015625, 89.47058868408203, 91.11370849609375, 40.19285583496094, 176.732666015625, 124.40188598632812, -52.5665283203125, 135.2109375, 79.6358413696289, 95.86647033691406, 12.992301940917969, -0.1726531982421875, 49.37762451171875, 79.78363037109375, 7.1641845703125, 37.349578857421875, 22.78125, 171.73007202148438, 143.7828369140625, 173.02517700195312, 154.26019287109375, -28.769805908203125, 208.87005615234375, -4.4297332763671875, 43.15089416503906, 13.263427734375, 27.63128662109375, 182.83795166015625, 7.558509826660156, 142.2266845703125, 111.37261962890625, 56.81193542480469, -74.0341796875, 6.01568603515625, 0.8440399169921875, 34.865264892578125, 25.656982421875, 5.529541015625, 105.91412353515625, 1.199615478515625, 5.497297286987305, -32.59722900390625, -13.7760009765625, 28.486480712890625, 115.15977478027344, 0.9652462005615234], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000402.npy"}
{"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 77.56963348388672, "std": 80.3834457397461, "min": -149.24578857421875, "p10": -14.130632019042965, "median": 67.74897384643555, "p90": 176.64566650390623, "max": 232.71002197265625, "pos_frac": 0.8125, "sample": [-17.322067260742188, 136.6149139404297, -9.09259033203125, 171.7675018310547, 118.43658447265625, 53.46282958984375, 168.17318725585938, 105.65059661865234, 25.1884765625, 27.374786376953125, 172.30911254882812, 174.125244140625, 47.658447265625, 58.9208984375, 10.256370544433594, 69.97552490234375, 19.31549072265625, 79.84005737304688, 37.09197998046875, 9.202735900878906, 174.6299285888672, -18.68133544921875, -37.037353515625, 33.2423095703125, 173.05580139160156, 149.76536560058594, 180.81610107421875, 51.4345703125, 34.77703857421875, 198.96200561523438, 210.45602416992188, 8.71893310546875, 0.6703567504882812, 130.85055541992188, -6.636651992797852, 85.90520477294922, 16.988861083984375, 33.6263427734375, 0.0, -16.483551025390625, -0.12917709350585938, 163.9791259765625, 206.71875, 108.24577331542969, 170.61099243164062, 169.57879638671875, 160.47186279296875, 22.4412841796875, 150.06033325195312, 210.20318603515625, 15.015380859375, 65.52242279052734, 22.90802001953125, 16.088321685791016, 181.2396240234375, 10.3621826171875, 27.566375732421875, -53.838104248046875, 147.7601318359375, -13.122238159179688, -23.417999267578125, 127.68789672851562, -1.3888969421386719, 185.84967041015625, 159.17552185058594, 13.702720642089844, 200.98214721679688, 1.3746395111083984, 30.230615615844727, 25.779876708984375, 31.79412841796875, 89.0181884765625, 22.607833862304688, 91.4815673828125, -11.60101318359375, 27.087661743164062, 115.08767700195312, 174.85833740234375, 4.402565002441406, 87.05764770507812, -1.8982925415039062, 64.2445068359375, 187.49215698242188, 15.989715576171875, 138.71133422851562, -19.52777099609375, 11.848251342773438, 152.89205932617188, 138.76544189453125, 50.80445861816406, 228.92822265625, 101.34384155273438, 232.71002197265625, 195.22509765625, 120.94767761230469, 195.479248046875, -36.8873291015625, 143.98863220214844, 159.6107177734375, 139.37664794921875, 127.27760314941406, 94.51344299316406, 133.5922088623047, 26.678634643554688, -149.24578857421875, 162.7628631591797, -29.497833251953125, 163.00912475585938, -1.4160385131835938, 151.65567016601562, -54.07965087890625, -3.6329574584960938, 23.036376953125, 72.46733093261719, -23.97698974609375, 44.36912536621094, 83.93971252441406, 14.8262939453125, 117.61752319335938, -7.4820098876953125, -129.28515625, 159.3234100341797, 161.09716796875, 53.455657958984375, 144.85763549804688, 137.82772827148438, 138.218505859375, 99.4947509765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000403.npy"}
{"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 54.6715087890625, "std": 95.4151611328125, "min": -211.78961181640625, "p10": -44.015090179443355, "median": 37.752349853515625, "p90": 183.09978485107422, "max": 240.2476806640625, "pos_frac": 0.6875, "sample": [182.4479522705078, 170.90948486328125, 37.5894775390625, 185.8129119873047, -146.98666381835938, 170.31796264648438, 192.71026611328125, -0.0674285888671875, -43.69739532470703, 0.0, -156.7593994140625, 76.4677505493164, 30.22010040283203, -25.0150146484375, -87.47274780273438, 154.76524353027344, 21.40130615234375, 184.6207275390625, 0.8255844116210938, -52.4638671875, 172.05609130859375, 65.47474670410156, 21.319244384765625, -3.07916259765625, -18.602020263671875, 191.07147216796875, -211.78961181640625, 165.05380249023438, -46.573890686035156, 7.084510803222656, -0.33251953125, -38.9957275390625, 68.53079223632812, 10.5250244140625, 93.07025146484375, -11.379570007324219, 4.049665451049805, -11.28790283203125, 240.2476806640625, 129.58383178710938, 161.48532104492188, -6.4756011962890625, -7.680755615234375, -128.08358764648438, 153.06231689453125, 80.71612548828125, 70.71243286132812, 33.21453857421875, 4.399257659912109, 21.120361328125, 113.13636779785156, 7.4402923583984375, 173.88375854492188, -24.20513916015625, 49.043548583984375, 202.16253662109375, -144.04763793945312, 17.390777587890625, -17.596466064453125, 18.332839965820312, 207.79293823242188, 187.4833526611328, -141.59310913085938, 156.4898681640625, 105.43199157714844, 153.77102661132812, 160.00616455078125, -11.15869140625, -6.870086669921875, 166.23568725585938, 157.5098876953125, 0.9992733001708984, 13.057220458984375, -10.560136795043945, -44.756378173828125, 127.9991455078125, 8.584075927734375, -17.37384033203125, 196.37353515625, 44.64996337890625, 105.61444091796875, 26.613571166992188, 48.0308837890625, 64.0159912109375, 7.851226806640625, -31.530410766601562, -17.26190185546875, -16.916748046875, 187.79928588867188, -79.48455810546875, -5.879486083984375, 142.10879516601562, 165.90390014648438, 40.1739501953125, -148.3943634033203, -21.98504638671875, 43.14445495605469, 18.533493041992188, 176.401123046875, 73.33929443359375, -9.713502883911133, 2.71124267578125, 152.88931274414062, 59.17864990234375, 176.76321411132812, 101.67646026611328, 38.63178253173828, 71.40357971191406, -127.81664276123047, 190.37405395507812, 37.91522216796875, 206.51101684570312, 166.49215698242188, 57.92198181152344, 30.51910400390625, 1.6942901611328125, 78.02516174316406, -11.085136413574219, 84.83328247070312, 153.32571411132812, 31.994659423828125, -8.303436279296875, -0.70758056640625, 139.33587646484375, 150.40447998046875, 115.12553405761719, 171.20187377929688, 204.83697509765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000404.npy"}
{"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 62.308616638183594, "std": 86.19384765625, "min": -160.83917236328125, "p10": -32.855953979492185, "median": 39.572654724121094, "p90": 184.12882537841796, "max": 228.3111572265625, "pos_frac": 0.7578125, "sample": [16.523193359375, 4.597015380859375, 11.081787109375, 0.0, -47.30133056640625, -3.834686279296875, 7.934814453125, 151.8465576171875, 83.90823364257812, 155.98599243164062, 3.6112213134765625, 18.72454833984375, 52.64631652832031, 36.928955078125, 150.28152465820312, -43.76207733154297, -160.83917236328125, 21.7890625, 42.38328552246094, 195.0550537109375, 92.99555969238281, 166.30075073242188, -2.8911094665527344, -14.633575439453125, 149.57479858398438, 9.686813354492188, -23.836807250976562, 76.31640625, 184.02281188964844, 33.171958923339844, 138.39633178710938, 153.35562133789062, 6.9976806640625, -0.475006103515625, 62.175926208496094, 149.90048217773438, 10.262496948242188, -88.41647338867188, -69.40877532958984, 134.9222412109375, -2.660717010498047, 196.98988342285156, 145.06080627441406, 1.0861797332763672, -14.103912353515625, 228.3111572265625, 157.669677734375, 114.73876953125, 123.55120849609375, 72.73947143554688, 8.279541015625, 87.1927490234375, 185.52725219726562, 43.624237060546875, 7.1675872802734375, 1.182952880859375, 142.13609313964844, 160.42935180664062, 146.7222442626953, -35.5706787109375, -98.24762725830078, 106.74763488769531, 92.24224853515625, -16.696945190429688, 14.685638427734375, -7.235969543457031, 64.80517578125, -0.476104736328125, 19.721435546875, 217.39195251464844, 58.53517150878906, 115.85101318359375, 161.45623779296875, 148.524658203125, 199.42697143554688, 209.79638671875, -32.016021728515625, 126.4415054321289, 12.460762023925781, 51.749725341796875, 214.656982421875, 11.72513198852539, -34.8157958984375, -89.66204833984375, 161.92236328125, 24.14693832397461, -0.22984695434570312, 205.95611572265625, 16.277626037597656, 95.51707458496094, 126.6390380859375, 25.425331115722656, -95.29072570800781, 7.775665283203125, 178.01458740234375, 196.10177612304688, 12.23019027709961, 7.1983184814453125, 167.75473022460938, 130.20855712890625, -70.12417602539062, 184.37619018554688, 1.1744384765625, 160.27603149414062, 39.83854675292969, 81.33069610595703, 12.27486801147461, 154.84384155273438, 68.50601196289062, 39.3067626953125, 45.18389892578125, 1.4104652404785156, 182.0545654296875, -3.40301513671875, 0.4922027587890625, -14.545013427734375, -26.688796997070312, 5.864765167236328, -28.368362426757812, 3.8062877655029297, -60.944976806640625, 189.80169677734375, 220.8665771484375, -8.60101318359375, -40.307289123535156, 179.79534912109375, 101.649658203125, 112.86447143554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000405.npy"}
{"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 53.05079650878906, "std": 94.79022979736328, "min": -198.91256713867188, "p10": -49.85897674560547, "median": 32.32908630371094, "p90": 169.0883331298828, "max": 221.67715454101562, "pos_frac": 0.7109375, "sample": [151.9310760498047, -34.75043487548828, -36.824127197265625, 17.667144775390625, -94.69573974609375, 204.6885986328125, 105.43258666992188, 159.37002563476562, 107.52608489990234, 170.95176696777344, -29.039283752441406, 0.7038135528564453, -21.9168701171875, 184.40460205078125, 61.5780029296875, -30.057205200195312, 145.26007080078125, -6.502361297607422, 120.49510192871094, 25.728927612304688, -137.09066772460938, -50.81871032714844, 16.91082763671875, 109.23187255859375, 157.92510986328125, 4.051300048828125, 126.45320129394531, 210.85736083984375, 29.746246337890625, 135.81640625, 57.7666015625, 168.2897186279297, -143.56912231445312, -5.3648681640625, 25.579010009765625, -138.86471557617188, -21.254150390625, -7.550010681152344, -21.124290466308594, 155.72076416015625, -36.9500732421875, -49.447662353515625, -132.49932861328125, -23.28033447265625, 45.31932830810547, 174.53195190429688, 19.61273193359375, 151.42236328125, 135.06155395507812, 166.38754272460938, -1.6044082641601562, 22.478302001953125, 162.96542358398438, 126.11425018310547, -198.91256713867188, -8.411773681640625, 143.01016235351562, 165.36163330078125, 119.4454116821289, 17.74609375, 2.556671142578125, 162.27157592773438, -123.10588073730469, 186.54168701171875, 23.8939208984375, 29.28369140625, 10.016357421875, 4.266265869140625, -43.468231201171875, 156.5645751953125, 221.67715454101562, 160.26580810546875, 11.653671264648438, 37.324951171875, 180.50384521484375, 35.603607177734375, 3.4940643310546875, 151.37762451171875, 15.932086944580078, 197.1849365234375, 163.74755859375, -19.84869384765625, -155.06341552734375, 9.117366790771484, 157.65882873535156, 133.28482055664062, 119.505126953125, -11.723846435546875, 26.2962646484375, 112.46194458007812, 36.68445587158203, -13.795578002929688, 123.19841003417969, 141.6811065673828, -79.9275894165039, 218.9400634765625, 122.30064392089844, 3.4771804809570312, 135.71046447753906, 60.115478515625, 105.78981018066406, 152.655517578125, 38.25225830078125, 148.9408721923828, -26.30902099609375, 57.84574890136719, -134.14141845703125, 194.11404418945312, 5.00750732421875, 34.91192626953125, 15.663665771484375, 105.88540649414062, -38.93438720703125, 128.759765625, 215.13018798828125, -1.5472564697265625, 2.7699737548828125, 45.105369567871094, -11.853057861328125, -23.417572021484375, -66.65798950195312, 21.822723388671875, 9.5006103515625, 36.72467041015625, 195.81451416015625, -67.57177734375, 36.91255187988281, 28.65399932861328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000406.npy"}
{"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 58.27226257324219, "std": 91.27465057373047, "min": -147.46206665039062, "p10": -39.733020019531246, "median": 27.341094970703125, "p90": 177.66053771972656, "max": 226.64306640625, "pos_frac": 0.703125, "sample": [-22.898826599121094, 27.4818115234375, 104.82818603515625, -15.439376831054688, 163.163330078125, 14.932926177978516, 168.6640625, 77.36541748046875, -33.697845458984375, 26.64167594909668, -1.369537353515625, -60.40093994140625, -144.5767059326172, -4.71685791015625, -65.29342651367188, 143.6064453125, -0.6233062744140625, -1.111785888671875, 16.9630126953125, -28.032928466796875, 144.43240356445312, -1.6839447021484375, -9.264739990234375, 131.6400146484375, 176.81842041015625, -67.85716247558594, 65.7684097290039, 130.3165283203125, -9.373207092285156, -13.02520751953125, -147.46206665039062, 20.397842407226562, -38.209442138671875, 36.338897705078125, 104.72163391113281, 183.21044921875, 169.37551879882812, 183.56146240234375, -48.041542053222656, -3.20819091796875, 69.759033203125, 27.20037841796875, 34.991180419921875, 3.0241012573242188, -39.26873779296875, 203.16323852539062, 72.81420135498047, 128.09231567382812, 195.24319458007812, 5.3851318359375, 102.65167236328125, 152.81085205078125, 7.867584228515625, 201.02273559570312, 5.765960693359375, 156.45111083984375, -71.0042953491211, -0.522369384765625, 22.707275390625, 128.36660766601562, 146.6957550048828, 25.33154296875, 9.134918212890625, 26.663625717163086, 159.57708740234375, 54.083648681640625, -78.8094482421875, -20.05975341796875, 126.85861206054688, 177.36642456054688, 24.831146240234375, -32.19657897949219, 148.32199096679688, 170.93020629882812, 195.80731201171875, 18.39977264404297, 6.910675048828125, 19.20227813720703, 0.0, 206.87234497070312, 165.98541259765625, 12.656700134277344, 148.33370971679688, -40.81634521484375, 178.3468017578125, -100.433349609375, 195.96609497070312, 116.90077209472656, 16.704986572265625, 142.221435546875, 172.37600708007812, 0.479827880859375, -28.010528564453125, 119.43763732910156, 72.74609375, 26.210716247558594, -37.75933837890625, 36.436195373535156, 38.786407470703125, 147.53317260742188, 165.72955322265625, -22.571273803710938, 204.90780639648438, -136.22894287109375, 69.13677978515625, -123.90676879882812, 96.09161376953125, 88.07501220703125, 175.30685424804688, 5.906463623046875, 157.15689086914062, 180.2093505859375, 53.1859130859375, -131.5706787109375, 188.44601440429688, 11.06488037109375, 40.76722717285156, -30.189489364624023, 9.544769287109375, 169.0076904296875, 153.52655029296875, 163.4134521484375, 23.5113525390625, 7.848382949829102, -6.2725067138671875, 226.64306640625, 169.62347412109375, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000407.npy"}
{"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 61.60485076904297, "std": 88.81791687011719, "min": -201.47299194335938, "p10": -30.72781219482421, "median": 50.82400894165039, "p90": 173.45792388916016, "max": 206.80975341796875, "pos_frac": 0.734375, "sample": [-3.2742919921875, 18.679168701171875, -57.01677703857422, 64.97454833984375, 11.716453552246094, 164.20675659179688, 64.91366577148438, 194.1591796875, 182.55120849609375, 162.73892211914062, -26.421859741210938, 4.909355163574219, 99.37789916992188, 128.65457153320312, -68.91683959960938, 160.96548461914062, 18.81884765625, -14.69317626953125, -3.3176422119140625, 98.38542175292969, 0.138763427734375, 19.950626373291016, 13.712570190429688, -10.639900207519531, 191.74609375, 157.54922485351562, 192.53045654296875, 180.598876953125, -8.82501220703125, -3.71295166015625, -20.858322143554688, 158.30718994140625, 144.869140625, 95.18865966796875, -38.33319091796875, 96.49806213378906, 201.36610412597656, 134.22508239746094, 164.01104736328125, 66.37820434570312, -21.127349853515625, -194.8082275390625, -201.47299194335938, 11.789794921875, 186.42718505859375, 155.65219116210938, 5.608955383300781, 48.87706756591797, -28.3017578125, 164.30430603027344, 64.56259155273438, 132.234375, -9.779586791992188, -6.0545196533203125, 94.28678894042969, 37.0283203125, 144.40545654296875, 154.88406372070312, -35.217987060546875, 67.66241455078125, 168.8905792236328, 149.4641876220703, 206.80975341796875, 7.075187683105469, -16.931396484375, 171.77218627929688, -44.28643798828125, -28.803451538085938, 148.3875732421875, -67.0953369140625, 183.48959350585938, 3.6119384765625, 143.97592163085938, 12.54443359375, 78.76300048828125, 171.61282348632812, -19.0321044921875, 95.91270446777344, -8.06890869140625, 27.398361206054688, 166.85260009765625, 73.02406311035156, 7.310981750488281, 45.355560302734375, -26.09429931640625, 2.41632080078125, 153.54559326171875, 149.1168212890625, 131.53335571289062, -60.75450134277344, 36.875213623046875, 14.103233337402344, 65.58279418945312, 164.23699951171875, 166.34066772460938, -2.8322296142578125, -162.0306396484375, 66.99078369140625, 177.19076538085938, 43.38726806640625, -62.15971374511719, 173.66209411621094, 19.3389892578125, 52.77095031738281, 137.64276123046875, -9.5126953125, 17.589126586914062, 15.980316162109375, 6.386566162109375, 48.433868408203125, 66.81700134277344, 4.783843994140625, 23.752944946289062, 0.0, 13.312728881835938, 206.573974609375, 61.70521545410156, -0.3519878387451172, 132.8209228515625, -43.453765869140625, -136.09371948242188, 27.828125, 173.37042236328125, 81.90341186523438, 190.90505981445312, 107.52606201171875, 127.17762756347656, 141.9957275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000408.npy"}
{"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 63.026336669921875, "std": 98.24243927001953, "min": -167.65017700195312, "p10": -44.078277587890625, "median": 36.02613830566406, "p90": 183.72865295410156, "max": 242.4522705078125, "pos_frac": 0.75, "sample": [165.01409912109375, 135.93971252441406, 154.91026306152344, 46.15904235839844, 190.2677001953125, -43.18701171875, 132.48626708984375, 189.66180419921875, 5.32659912109375, 43.719085693359375, 19.998191833496094, 4.9531707763671875, 22.988021850585938, 5.62506103515625, 32.140106201171875, 209.82754516601562, 26.00908660888672, 24.05279541015625, 184.23184204101562, -156.29067993164062, -11.794361114501953, -108.09051513671875, -1.4659690856933594, 178.0856170654297, 156.38058471679688, -97.6197509765625, 17.83697509765625, -117.6700439453125, 179.1880340576172, -167.65017700195312, 183.51300048828125, 163.590576171875, -89.5685806274414, 35.83143615722656, -7.11810302734375, 199.39974975585938, 24.394699096679688, 39.312530517578125, 198.22442626953125, 182.6497802734375, 140.3111572265625, 176.46951293945312, -103.06076049804688, 30.489578247070312, 2.8029022216796875, 30.803741455078125, 86.12869262695312, 167.91934204101562, 11.53509521484375, 208.37661743164062, 160.39105224609375, 144.8912353515625, 119.2821044921875, 154.95538330078125, 242.4522705078125, -42.607513427734375, -94.55169677734375, 61.990936279296875, 23.4892578125, 1.4619140625, 224.6060791015625, 129.82440185546875, 20.23638916015625, 87.76568603515625, 163.2047119140625, 77.57051086425781, 20.974639892578125, -40.643524169921875, -4.20611572265625, -38.24259948730469, 163.69699096679688, -29.492996215820312, -146.87506103515625, 182.15989685058594, -8.651500701904297, 98.29925537109375, 195.7342529296875, 229.52752685546875, 63.17485809326172, 27.52728271484375, 6.732242584228516, -42.51298904418945, 113.84552001953125, 0.0, 119.14656066894531, 231.025390625, 33.30720520019531, 183.38232421875, 158.34925842285156, -146.3675537109375, -53.960601806640625, 164.28704833984375, 34.469970703125, -9.042991638183594, 36.22084045410156, -1.9421234130859375, 139.819580078125, 165.84381103515625, 151.56503295898438, 154.17538452148438, 33.060035705566406, -10.720436096191406, 137.91677856445312, 77.27117919921875, 40.08409118652344, -12.202247619628906, 21.618057250976562, -46.15789794921875, -2.3000335693359375, -11.165153503417969, 41.61061096191406, 4.062045097351074, -167.05807495117188, 33.13890075683594, 156.73873901367188, 128.27566528320312, 9.285383224487305, 180.08038330078125, 33.38873291015625, -27.07806396484375, 160.53952026367188, 10.793571472167969, 7.8319244384765625, 131.46353149414062, 2.169677734375, 139.8230438232422, 206.44168090820312, 59.13179016113281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000409.npy"}
{"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 53.081546783447266, "std": 96.907470703125, "min": -187.32217407226562, "p10": -49.61951751708983, "median": 41.515933990478516, "p90": 182.43108520507812, "max": 245.41351318359375, "pos_frac": 0.6875, "sample": [87.1129150390625, -81.54498291015625, 16.003936767578125, -40.3558349609375, -18.96738052368164, 157.66156005859375, 3.281688690185547, 31.58343505859375, -28.080169677734375, 108.94419860839844, 224.58358764648438, -22.253814697265625, 28.72593879699707, -58.0025634765625, 51.981529235839844, -163.41665649414062, -17.23779296875, 58.86328125, 171.488037109375, 102.04547119140625, 187.05230712890625, -187.32217407226562, 83.52499389648438, 13.622200012207031, 164.7150421142578, -8.80194091796875, 138.63096618652344, 98.30001831054688, 126.68367004394531, 148.29718017578125, 33.5687255859375, 114.22964477539062, 9.7789306640625, 162.3267822265625, -0.0636749267578125, 137.94358825683594, 33.69024658203125, 60.08489990234375, 35.31646728515625, 160.10501098632812, 176.35772705078125, 61.135955810546875, -7.4177398681640625, 170.15594482421875, -155.4110107421875, 142.19744873046875, -153.6952362060547, -5.1785888671875, -18.492324829101562, -46.25761413574219, 108.8201904296875, 149.17221069335938, 115.29833984375, -57.463958740234375, -18.250221252441406, 52.5263671875, -5.2447967529296875, 161.9815673828125, 128.08416748046875, 48.99029541015625, -19.568748474121094, 201.87353515625, 40.69287109375, 192.64630126953125, -2.2852611541748047, 10.151947021484375, -29.765350341796875, -1.6922607421875, 126.18744659423828, 76.45423889160156, 141.31065368652344, 157.11001586914062, 49.410980224609375, -13.27777099609375, -146.96328735351562, 0.0, 42.33899688720703, 25.804519653320312, 201.94522094726562, 180.4505615234375, 207.27130126953125, 10.454620361328125, 12.523513793945312, -63.89738464355469, -37.67186737060547, 149.19375610351562, 19.281837463378906, 16.197265625, -29.497634887695312, 27.475868225097656, -147.65626525878906, 69.21163940429688, -29.9248046875, 145.78662109375, -39.8441162109375, 208.47927856445312, -127.34060668945312, 158.83267211914062, 7.675239562988281, 38.74681091308594, 67.7220458984375, -34.22241973876953, 105.24390411376953, -163.49783325195312, 58.864349365234375, 10.230422973632812, 62.842987060546875, 153.63250732421875, 232.76806640625, 138.88333129882812, 190.72052001953125, -20.056243896484375, 128.77017211914062, 114.3965835571289, 1.707611083984375, 61.57666015625, 217.11395263671875, -88.24188232421875, 47.07856750488281, 9.055740356445312, 4.042997360229492, 190.62432861328125, 215.7647705078125, 17.883926391601562, 73.38108825683594, -43.45904541015625, 245.41351318359375, -3.3331222534179688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000410.npy"}
{"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 64.12574768066406, "std": 89.18885803222656, "min": -208.05426025390625, "p10": -19.468673706054688, "median": 41.78045654296875, "p90": 176.81408081054687, "max": 323.009765625, "pos_frac": 0.75, "sample": [129.99583435058594, 174.08782958984375, 33.39013671875, 152.34149169921875, 26.14337158203125, -1.5576629638671875, -1.903045654296875, 26.917945861816406, 48.279052734375, 143.2411346435547, 155.845947265625, 110.81265258789062, -1.5120697021484375, -18.438385009765625, -133.8951416015625, -86.70458984375, 210.3543701171875, -10.01312255859375, 126.00048828125, 86.59530639648438, 41.75592041015625, 61.6195068359375, 153.23011779785156, 97.992919921875, 146.28387451171875, 52.367767333984375, 73.74813842773438, 323.009765625, 142.13665771484375, -19.803741455078125, 183.72213745117188, 50.50825500488281, 35.09192657470703, 19.988662719726562, 140.95724487304688, 41.80499267578125, 183.89410400390625, 155.02989196777344, 50.164794921875, 33.05625534057617, 7.118621826171875, 33.16703796386719, -4.1671905517578125, 70.60411071777344, 1.2414093017578125, -57.506446838378906, -2.18267822265625, 5.075763702392578, -18.1627197265625, 139.30311584472656, 148.88217163085938, 28.013519287109375, -12.558013916015625, 3.9358043670654297, 15.82368278503418, 173.52374267578125, -40.58965301513672, 41.3365478515625, 39.079864501953125, 70.69615936279297, -0.07666778564453125, 45.377044677734375, -9.965057373046875, -92.1868896484375, 152.02749633789062, 155.81356811523438, 127.67572021484375, 175.95059204101562, 26.771163940429688, 198.2855224609375, 8.757671356201172, 137.02098083496094, 23.798919677734375, 178.82888793945312, -0.9382705688476562, 195.87347412109375, 24.781829833984375, 6.847625732421875, 164.05413818359375, 147.96414184570312, 35.470909118652344, -208.05426025390625, 184.2635498046875, 15.668712615966797, 11.307952880859375, -137.03585815429688, 46.69487762451172, -2.542591094970703, 158.27651977539062, 205.76422119140625, 127.99734497070312, -111.79470825195312, 33.05694580078125, 38.611785888671875, 8.242195129394531, 123.65139770507812, 123.22845458984375, 213.44082641601562, 144.39935302734375, -56.887359619140625, 147.06918334960938, -19.3250732421875, -83.24267578125, -14.1339111328125, 8.368309020996094, 38.8192138671875, 133.2357940673828, 49.89453125, 41.46226501464844, 134.57669067382812, -40.86941909790039, 62.4033203125, 163.9983367919922, -8.043678283691406, 3.293487548828125, -13.45404052734375, 72.67276763916016, 219.2979736328125, 152.47955322265625, 153.31971740722656, 148.10894775390625, 207.21376037597656, 133.903076171875, 190.39404296875, 16.000473022460938, -6.0823974609375, -73.64906311035156, -9.213623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000411.npy"}
{"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 52.3337287902832, "std": 95.86219787597656, "min": -193.6507568359375, "p10": -53.06874847412109, "median": 32.48543930053711, "p90": 176.8030822753906, "max": 256.6656799316406, "pos_frac": 0.6796875, "sample": [-138.534912109375, 173.007080078125, 5.8886260986328125, 190.9832763671875, 63.93450927734375, 112.59823608398438, 38.590972900390625, 99.11175537109375, -35.753173828125, -20.652435302734375, 14.09112548828125, 36.178314208984375, -3.1883163452148438, -14.874185562133789, 23.33514404296875, -193.6507568359375, 0.004058837890625, -28.8978271484375, -93.14170837402344, 30.143592834472656, 6.5962066650390625, 2.467742919921875, 18.4664306640625, 85.87432861328125, 171.2611846923828, 166.76983642578125, 181.57534790039062, 35.337425231933594, -5.7708892822265625, 26.61688995361328, 128.8659210205078, 132.38226318359375, -22.237091064453125, 172.14212036132812, 191.97271728515625, 166.44091796875, 173.49993896484375, 18.29033660888672, 169.028564453125, 73.0650634765625, 140.87855529785156, -27.61712646484375, 179.52920532226562, -132.27484130859375, 97.77015686035156, -20.43988037109375, -3.0259933471679688, -27.70062255859375, -157.32138061523438, 51.120635986328125, 96.277099609375, 186.96917724609375, 209.79788208007812, -17.53057861328125, 178.62869262695312, 12.717575073242188, -9.352630615234375, 76.28692626953125, -101.72032165527344, -50.115325927734375, -78.0609130859375, -8.069931030273438, 217.22525024414062, 77.60302734375, -3.9577789306640625, 176.35440063476562, 138.75875854492188, 256.6656799316406, 50.211700439453125, 62.89239501953125, -111.8096923828125, 72.82652282714844, -50.72442626953125, -16.58087158203125, 22.2978515625, 34.82728576660156, 204.5166778564453, 170.7222900390625, 21.730621337890625, -174.2393798828125, 40.87091064453125, 111.47351837158203, 170.80145263671875, -1.852294921875, 177.85000610351562, 4.2269134521484375, 167.51901245117188, 165.64793395996094, 191.61569213867188, -60.27458190917969, 18.066558837890625, 175.19024658203125, 155.5588836669922, -130.61007690429688, -6.9065399169921875, 40.06695556640625, -15.797622680664062, 138.11590576171875, -12.74053955078125, 135.0469970703125, -5.219635009765625, 0.10256004333496094, 211.04833984375, 16.917236328125, 154.239990234375, 6.813945770263672, -29.804412841796875, 107.52144622802734, 8.57958984375, 126.9403076171875, 165.07861328125, 27.782791137695312, 0.0, 55.60420227050781, -13.54681396484375, 108.96743774414062, 7.2389678955078125, 43.663612365722656, -107.94271850585938, 4.4655303955078125, 23.036544799804688, 145.48788452148438, -58.53883361816406, 54.81549072265625, -12.106460571289062, -15.799163818359375, 140.78518676757812, 170.83102416992188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000412.npy"}
{"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 48.91182327270508, "std": 86.13528442382812, "min": -180.83843994140625, "p10": -42.89609985351562, "median": 21.121110916137695, "p90": 171.58596649169922, "max": 249.16848754882812, "pos_frac": 0.7265625, "sample": [24.44514274597168, -12.866409301757812, -1.2693214416503906, -3.2237319946289062, 6.40863037109375, 155.88571166992188, 60.387603759765625, 19.872802734375, -180.83843994140625, -39.957244873046875, 186.19610595703125, 41.2724609375, 189.0970458984375, -50.59515380859375, 32.534393310546875, 18.40479278564453, 10.01324462890625, 184.82403564453125, 66.09286499023438, -127.12284851074219, 52.32208251953125, 15.175655364990234, 83.25221252441406, 16.16131591796875, 19.103408813476562, -25.732177734375, 8.191764831542969, 198.393310546875, -4.501382827758789, 111.87682342529297, 151.739013671875, -39.591217041015625, -50.23191833496094, 101.17669677734375, -43.34210205078125, 160.62355041503906, 171.0745391845703, 152.72323608398438, 155.29025268554688, 15.678825378417969, 127.56906127929688, -4.4978485107421875, -100.17913818359375, 5.03265380859375, 86.93682861328125, -30.6846923828125, -110.7580337524414, 54.00135803222656, 191.42083740234375, 74.68167114257812, -34.2828369140625, 4.862762451171875, -19.371490478515625, 100.6820068359375, 24.74981689453125, 134.2333984375, 173.0068359375, -25.929946899414062, 44.847869873046875, 71.83367919921875, 178.50790405273438, 23.15924072265625, 15.821884155273438, 10.080329895019531, 36.97760009765625, 210.05950927734375, -67.56681823730469, 47.89361572265625, 162.7811279296875, -64.75656127929688, 156.76788330078125, 47.28165054321289, -23.17469024658203, 249.16848754882812, 13.235057830810547, -12.354583740234375, -3.229217529296875, 23.4244384765625, -19.96481704711914, 102.49188232421875, 7.8914337158203125, 158.9185791015625, -51.64923095703125, 155.2056884765625, 148.02749633789062, -23.905548095703125, 18.79681396484375, 161.68212890625, 5.164648056030273, -58.563720703125, 167.517333984375, 22.26858901977539, -42.7049560546875, 149.8565673828125, 19.9736328125, -20.144573211669922, 9.83355712890625, 8.092788696289062, 158.73828125, 12.681163787841797, 10.104602813720703, -29.585960388183594, 6.9206695556640625, 132.09423828125, 94.7413330078125, 78.81201171875, 0.4508018493652344, 10.71734619140625, 137.29306030273438, 46.194122314453125, 26.878334045410156, 15.92547607421875, 69.95702362060547, 163.6348876953125, -16.65789794921875, -85.82635498046875, 68.83349609375, 172.779296875, -143.92813110351562, 213.8055419921875, 3.04193115234375, 192.67947387695312, 5.9718780517578125, -0.14271163940429688, 187.41351318359375, 160.78665161132812, 28.91302490234375, 17.5211181640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000413.npy"}
{"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 58.029144287109375, "std": 101.31095886230469, "min": -214.17279052734375, "p10": -55.414721679687496, "median": 37.33708190917969, "p90": 185.80841674804688, "max": 310.25494384765625, "pos_frac": 0.7109375, "sample": [124.93145751953125, -146.89735412597656, 99.24203491210938, 161.26101684570312, 0.7146148681640625, 151.19952392578125, 76.3900146484375, 62.35247802734375, 31.586578369140625, -132.1773681640625, -9.745437622070312, 8.407463073730469, 183.73568725585938, 22.477783203125, 8.500244140625, -2.1015586853027344, -91.8597412109375, -123.34503173828125, 160.80703735351562, -9.314414978027344, -214.17279052734375, -72.87017822265625, 220.81109619140625, 8.837158203125, 104.183837890625, -154.32814025878906, -49.86480712890625, 17.301963806152344, 150.1494140625, 152.36178588867188, -31.801517486572266, -28.3182373046875, 149.0569610595703, 164.8555145263672, -54.3619384765625, 185.6813507080078, 153.06466674804688, -5.7287750244140625, 206.4427490234375, -38.30799102783203, 27.357330322265625, 169.95391845703125, 57.629638671875, -12.84539794921875, 135.0894775390625, 164.55868530273438, 2.87298583984375, 6.08673095703125, 57.111572265625, 209.40798950195312, -6.606719970703125, -66.97136688232422, -3.5223398208618164, 211.93252563476562, 151.8326416015625, -11.540618896484375, 158.92054748535156, 186.1049041748047, 185.61572265625, -125.88809204101562, 138.98001098632812, 2.8053836822509766, 1.39532470703125, 159.85186767578125, 8.7584228515625, 61.27252197265625, 102.02352142333984, 127.4158935546875, 58.070831298828125, -0.31878662109375, 158.38031005859375, -8.812904357910156, 34.11041259765625, 200.3072509765625, 149.23779296875, 182.04779052734375, 161.27825927734375, 3.356719970703125, 38.66766357421875, 133.722900390625, 2.4460601806640625, -41.0780029296875, -136.74673461914062, 210.36312866210938, 310.25494384765625, 171.1055145263672, 168.32125854492188, 115.62796783447266, 4.618415832519531, 149.60867309570312, -199.0406494140625, 11.6573486328125, 36.006500244140625, 195.95086669921875, 13.47589111328125, 187.2366943359375, -53.63343811035156, -4.99481201171875, -19.829727172851562, 14.396507263183594, -64.36591339111328, 68.735107421875, 232.01473999023438, 40.34980010986328, 7.875846862792969, 13.117225646972656, 160.1990509033203, 89.47663879394531, 114.32072448730469, 16.6300048828125, -47.89251708984375, 71.74050903320312, 206.1468505859375, 74.33465576171875, -4.106719017028809, 15.225616455078125, 129.63101196289062, 192.67962646484375, 24.4552001953125, 39.42732238769531, 22.364151000976562, -0.632659912109375, -57.8712158203125, -6.38916015625, 66.57650756835938, 63.828948974609375, 185.50982666015625, -10.205215454101562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000414.npy"}
{"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 69.9118423461914, "std": 93.64462280273438, "min": -181.09515380859375, "p10": -30.628085327148437, "median": 55.40068054199219, "p90": 185.53296813964843, "max": 255.24761962890625, "pos_frac": 0.7421875, "sample": [17.752254486083984, 142.3055877685547, 16.411598205566406, 40.62330627441406, -38.5118408203125, -9.546859741210938, 154.39950561523438, -1.76995849609375, 147.4588623046875, 10.663131713867188, 68.33039093017578, -1.7168769836425781, 186.51351928710938, 111.6181640625, -27.755844116210938, -11.146987915039062, 178.90716552734375, 4.6763916015625, 153.67147827148438, 51.35716247558594, 159.22756958007812, 21.281402587890625, 194.58425903320312, 3.17767333984375, 155.0544891357422, 54.92144775390625, 77.40424346923828, 169.99517822265625, 170.51751708984375, 127.81634521484375, -139.2025909423828, 201.7340087890625, 192.57867431640625, 24.27545166015625, -150.22933959960938, 21.69415283203125, 54.58251953125, 182.96661376953125, -10.242416381835938, 112.94155883789062, 123.3657455444336, 26.234161376953125, 0.074005126953125, 17.951446533203125, 46.22185516357422, -23.966903686523438, 168.35150146484375, 35.8226318359375, 185.11273193359375, 148.0941619873047, -12.384483337402344, 26.806365966796875, -126.56597900390625, 55.879913330078125, 159.595947265625, 180.62893676757812, -34.92572021484375, 181.37142944335938, 44.510162353515625, -30.33056640625, 72.25341796875, 163.19166564941406, -106.30250549316406, 130.22958374023438, 255.24761962890625, -23.13006591796875, 29.228919982910156, 56.090171813964844, -31.322296142578125, -0.5729694366455078, -28.303237915039062, 198.6968994140625, 150.51129150390625, 31.49969482421875, -1.8054656982421875, -8.689468383789062, 169.03021240234375, 204.55465698242188, 128.12632751464844, -44.94830322265625, 85.98941040039062, 9.497314453125, 120.22848510742188, -61.97210693359375, 183.2279052734375, -26.33343505859375, -98.47679901123047, 126.09093475341797, 111.96920776367188, -28.345367431640625, -19.336395263671875, 7.201416015625, 94.68890380859375, 161.9259033203125, 124.1915283203125, -31.347061157226562, 56.716552734375, 114.15020751953125, 169.86328125, -181.09515380859375, 206.55911254882812, 44.58689880371094, 147.2882080078125, 13.604232788085938, 30.41705322265625, 193.48580932617188, 60.503692626953125, -21.316497802734375, 172.17755126953125, 148.01463317871094, 155.87548828125, 54.171607971191406, 151.06488037109375, 138.1717529296875, -124.19139862060547, 223.4559326171875, 25.055862426757812, 193.42076110839844, 196.51611328125, -6.850067138671875, 5.4205322265625, 21.70965576171875, 147.5645751953125, 2.7277374267578125, 165.18728637695312, 148.1042022705078, -21.086585998535156, 223.4891357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000415.npy"}
{"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 66.81501770019531, "std": 95.38325500488281, "min": -198.77801513671875, "p10": -42.984341812133785, "median": 55.16859436035156, "p90": 182.57577514648438, "max": 237.54901123046875, "pos_frac": 0.75, "sample": [173.173828125, 210.95236206054688, 186.9376220703125, 28.021697998046875, 3.2061996459960938, 3.1486129760742188, -42.54510498046875, 10.613128662109375, -14.670974731445312, 139.5685577392578, -38.22357177734375, 237.54901123046875, 63.78312683105469, 59.18382263183594, 60.868865966796875, 15.634178161621094, -82.73275756835938, 191.34112548828125, 217.552490234375, 108.23951721191406, 134.3400421142578, 159.4839630126953, 45.889923095703125, 180.3376007080078, -5.0521240234375, 41.664642333984375, -50.46014404296875, 147.1470947265625, -103.24324035644531, 31.777816772460938, 128.2910614013672, 166.25143432617188, 153.9678192138672, -5.28985595703125, -0.2216033935546875, 182.40274047851562, -147.91026306152344, 169.11761474609375, 21.276016235351562, 23.14208984375, 33.5767822265625, 39.1173095703125, -20.209762573242188, -198.77801513671875, -44.00922775268555, 185.8486328125, 27.833572387695312, 140.70848083496094, 140.26968383789062, 185.9912109375, 42.188720703125, 181.75865173339844, 14.39532470703125, 26.46752166748047, -74.86239624023438, 161.29275512695312, 182.97952270507812, 126.951416015625, -6.336578369140625, -11.480926513671875, 140.5848388671875, 48.656005859375, -145.63961791992188, -23.4228515625, 207.63949584960938, 127.11618041992188, 1.0123367309570312, 143.56927490234375, -54.18157958984375, 146.6654510498047, 0.594085693359375, -30.586563110351562, 49.61822509765625, 160.76220703125, 211.6314697265625, 146.40467834472656, 101.54046630859375, 160.94140625, -37.15510559082031, 2.845062255859375, 107.1568603515625, 28.10626220703125, 168.37677001953125, -20.20318603515625, -108.47036743164062, 117.74942016601562, 51.563262939453125, 94.0546875, -3.4225502014160156, 24.787002563476562, 69.90245056152344, 148.43605041503906, 69.85066986083984, 117.55972290039062, -172.16180419921875, 140.95840454101562, 66.1976318359375, 1.7971038818359375, 66.4429931640625, 171.68434143066406, -88.78231811523438, 148.75970458984375, 47.9073486328125, 25.167022705078125, 58.77392578125, -12.441650390625, 198.934814453125, 137.0374298095703, 216.86526489257812, -17.702529907226562, 4.663215637207031, 213.27813720703125, 49.468292236328125, 23.5902099609375, -16.007843017578125, 163.12435913085938, 160.01953125, 161.22567749023438, 87.77274322509766, -9.758472442626953, 31.0887451171875, -113.87054443359375, 176.17599487304688, -25.82004165649414, 159.12384033203125, 17.668869018554688, 148.28793334960938, 170.59373474121094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000416.npy"}
{"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 57.086097717285156, "std": 96.03541564941406, "min": -177.59976196289062, "p10": -49.50348281860351, "median": 43.91331481933594, "p90": 179.70335693359374, "max": 228.23223876953125, "pos_frac": 0.7109375, "sample": [-2.493602752685547, 36.20948791503906, 36.75694274902344, 106.689453125, -30.8134765625, 108.84085083007812, 32.59428405761719, -44.029327392578125, 191.41033935546875, 191.654541015625, 12.935211181640625, 43.3494873046875, 2.339188575744629, 147.8971710205078, -154.7020263671875, 7.956737518310547, 11.239013671875, -7.7064971923828125, -13.38140869140625, 160.1187744140625, 173.96728515625, -29.06744384765625, 193.78482055664062, -103.17672729492188, -47.84416961669922, -152.03756713867188, 114.74810791015625, 4.873954772949219, -41.99803161621094, 208.14395141601562, 67.99263000488281, -10.471710205078125, -12.2855224609375, 98.3702621459961, 122.44670104980469, 180.01205444335938, 23.43267822265625, -129.03656005859375, 150.0376434326172, 128.63731384277344, 168.54544067382812, 27.503082275390625, 113.76835632324219, 0.128021240234375, 147.13900756835938, 123.86572265625, 16.726669311523438, 71.8133544921875, 138.44772338867188, 67.37187194824219, 25.625734329223633, -0.1226348876953125, -27.1456298828125, -100.45779418945312, -59.333953857421875, 45.35661315917969, -2.17041015625, 164.20504760742188, 72.02572631835938, -14.64276123046875, 111.822509765625, 19.607757568359375, 3.557952880859375, 15.287033081054688, 170.70941162109375, 1.1761932373046875, 168.9072265625, -30.31964111328125, -177.59976196289062, 2.4117431640625, -24.886993408203125, 177.1512451171875, 128.6727294921875, 100.94015502929688, 54.42132568359375, -97.99884033203125, 109.72247314453125, 169.8175048828125, 0.0, -1.37554931640625, 92.3619384765625, 0.0, 44.477142333984375, 156.6470947265625, 124.48953247070312, 145.4729461669922, 67.90420532226562, 3.5479736328125, 150.796875, -20.120269775390625, -90.9632568359375, 179.68310546875, 168.68528747558594, 163.49078369140625, 179.44659423828125, 71.098388671875, -173.78334045410156, 107.25723266601562, 224.691162109375, 14.028457641601562, 179.7506103515625, -105.51573944091797, -53.375213623046875, 77.5299072265625, 165.84442138671875, 182.91798400878906, 193.99655151367188, -40.289031982421875, -143.48550415039062, 14.90313720703125, 6.1624755859375, -9.572647094726562, 11.047624588012695, 170.54962158203125, 191.92123413085938, 178.87966918945312, -14.6920166015625, 49.61643981933594, 183.95144653320312, 6.245830535888672, 46.90570068359375, 24.345596313476562, 37.784996032714844, -7.45294189453125, 115.22738647460938, 135.88623046875, 228.23223876953125, 212.423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000417.npy"}
{"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 49.68887710571289, "std": 90.65829467773438, "min": -204.71316528320312, "p10": -56.14183044433594, "median": 33.71337890625, "p90": 166.41517944335936, "max": 244.21408081054688, "pos_frac": 0.71875, "sample": [-30.10992431640625, 157.015380859375, 34.462066650390625, 52.555023193359375, -88.29864501953125, -112.74507141113281, -0.4375, 193.13690185546875, -50.210174560546875, -204.71316528320312, 36.937408447265625, 119.15021514892578, 31.937713623046875, -113.99362182617188, 56.931732177734375, 70.32463073730469, 140.00424194335938, -97.27633666992188, 137.32098388671875, 16.3304443359375, 155.18917846679688, 118.15048217773438, -17.58673095703125, 0.07135581970214844, -1.85986328125, 8.66973876953125, -23.423049926757812, 142.4359130859375, 51.72796630859375, 192.08428955078125, 142.92840576171875, 38.5968017578125, 167.0132598876953, 131.42620849609375, 16.702560424804688, 90.76373291015625, -25.911346435546875, 148.0792999267578, -47.502593994140625, 3.318145751953125, -0.7446060180664062, 181.5589599609375, 18.565093994140625, 156.85031127929688, -1.4810028076171875, 18.896636962890625, 2.1786365509033203, 88.51365661621094, 29.298675537109375, -21.807952880859375, 0.83026123046875, 171.6728515625, 124.66839599609375, 29.95733642578125, 19.135650634765625, -10.4051513671875, 9.23876953125, 20.502288818359375, 67.03108215332031, 119.61395263671875, -74.38623046875, 142.48452758789062, 168.81483459472656, 28.476425170898438, 32.964691162109375, 8.804557800292969, 142.2322998046875, -82.8655776977539, -147.519287109375, 8.014381408691406, -23.237030029296875, 12.721336364746094, 143.3912353515625, 52.149658203125, 88.94821166992188, 184.6723175048828, 144.34585571289062, -12.317962646484375, 88.9000244140625, -56.145355224609375, 34.557952880859375, 155.52667236328125, -7.84521484375, 64.76904296875, 118.84628295898438, 166.00161743164062, 155.43426513671875, -46.40576171875, 2.89605712890625, -96.04254150390625, 13.796234130859375, -39.13966369628906, -22.9228515625, 142.1756591796875, 158.06922912597656, 17.223968505859375, 178.41122436523438, 107.36808776855469, -0.8074111938476562, 162.76707458496094, -56.14031982421875, 0.5924568176269531, 9.055648803710938, 14.290603637695312, -120.48492431640625, 38.9486083984375, 66.41390991210938, 149.595458984375, -178.1746826171875, 174.16876220703125, 37.515987396240234, 18.199127197265625, 51.69657897949219, -1.059885025024414, 165.31549072265625, 63.74822235107422, 244.21408081054688, 171.37570190429688, 47.30278778076172, 227.56008911132812, 76.01788330078125, -19.631385803222656, -10.48004150390625, 66.92095947265625, -76.17726135253906, 166.1588592529297, 242.222900390625, 12.611557006835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000418.npy"}
{"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 67.8760986328125, "std": 103.09896850585938, "min": -239.84815979003906, "p10": -59.90984191894529, "median": 54.90784454345703, "p90": 182.2153564453125, "max": 278.34197998046875, "pos_frac": 0.7578125, "sample": [108.269287109375, 14.037826538085938, 6.64947509765625, 179.49725341796875, 29.0897216796875, 25.014434814453125, 6.143669128417969, 17.128875732421875, 70.00711059570312, -119.85513305664062, -81.08354949951172, 84.20263671875, -133.90020751953125, 175.17864990234375, 2.6378631591796875, 30.85455322265625, 215.38568115234375, 7.396900177001953, 134.15414428710938, 168.84921264648438, 180.49986267089844, 166.77813720703125, 185.44732666015625, 40.00291442871094, 192.2393798828125, -0.9423828125, 53.44590759277344, 3.476959228515625, -20.790618896484375, -143.76513671875, 172.618896484375, 160.98095703125, 146.87649536132812, -4.312957763671875, 68.30436706542969, -27.837203979492188, 86.21893310546875, -0.3943367004394531, 158.91925048828125, -158.03025817871094, 179.47201538085938, 4.0694427490234375, 173.637451171875, 147.77670288085938, 278.34197998046875, 10.465446472167969, 218.83753967285156, -100.8046875, 56.51904296875, 206.91415405273438, 128.5126953125, 105.9122314453125, -21.40601348876953, 52.145851135253906, 40.432373046875, 172.55618286132812, 216.65643310546875, 154.61624145507812, 3.6591796875, -13.029632568359375, -126.82286071777344, 37.772369384765625, -23.014366149902344, 3.116405487060547, 238.81121826171875, -157.60928344726562, 164.49478149414062, 186.67774963378906, 11.361831665039062, 38.281402587890625, -21.030921936035156, 155.52410888671875, 149.75942993164062, 24.14520263671875, 121.15748596191406, 0.0, -101.11497497558594, -54.07373046875, 16.696975708007812, 0.0, 35.870697021484375, 146.1068878173828, 151.11083984375, 181.61387634277344, -6.3068389892578125, 160.07371520996094, 28.784881591796875, 178.48779296875, 105.34658813476562, 21.004562377929688, 180.78062438964844, 37.326904296875, 183.458984375, 56.369781494140625, -7.6822509765625, 16.6903076171875, 137.09803771972656, 180.84591674804688, 160.362548828125, 181.682373046875, 201.65261840820312, 11.88037109375, 227.31005859375, 17.7169189453125, 176.29144287109375, -41.646209716796875, 155.41188049316406, 159.56056213378906, 60.27423095703125, 153.11666870117188, -85.50982666015625, -73.52743530273438, 187.27255249023438, 158.33311462402344, -239.84815979003906, 99.8543701171875, -157.42282104492188, -16.275177001953125, 159.6500244140625, -22.4962158203125, 120.0445556640625, 159.28146362304688, -10.153289794921875, 20.482452392578125, 9.041549682617188, 59.934814453125, 155.0042724609375, 35.06813049316406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000419.npy"}
{"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 65.18153381347656, "std": 93.77689361572266, "min": -187.74424743652344, "p10": -37.65085945129395, "median": 45.1121826171875, "p90": 180.19489746093748, "max": 231.79611206054688, "pos_frac": 0.7890625, "sample": [-65.01405334472656, 15.632064819335938, 30.92181396484375, -180.35687255859375, -25.272315979003906, -31.380939483642578, 37.365020751953125, 168.2490234375, 3.614044189453125, 96.6663818359375, 177.04733276367188, 4.395751953125, -134.3685302734375, 147.03921508789062, 135.38055419921875, 11.415145874023438, 157.08340454101562, -136.04495239257812, 147.59283447265625, 128.504638671875, 38.616424560546875, 58.779022216796875, -5.389415740966797, 38.37811279296875, 4.9785308837890625, 179.7506103515625, 6.45623779296875, 21.245155334472656, 141.32757568359375, 32.58685302734375, 7.1943206787109375, 219.21337890625, 43.86524963378906, 161.46124267578125, 14.104095458984375, -4.326263427734375, 141.68655395507812, 192.1064453125, 197.42669677734375, 14.8502197265625, 207.9853515625, 58.39324951171875, 22.12220001220703, 159.44134521484375, 24.71856689453125, 33.91326904296875, 122.92886352539062, -6.72027587890625, -3.8954620361328125, 107.62034606933594, -10.912384033203125, 114.67608642578125, 154.0664825439453, 171.32366943359375, 138.6458282470703, 146.91055297851562, 0.0, 46.35911560058594, -37.80982971191406, -142.64340209960938, 142.48385620117188, 141.63702392578125, 10.69940185546875, 4.217002868652344, 13.42901611328125, 32.067291259765625, 4.483612060546875, 200.81695556640625, 215.7730712890625, 147.1888427734375, 151.44631958007812, -74.21818542480469, 27.882568359375, 92.1185302734375, -40.070159912109375, 187.27520751953125, 123.90823364257812, 21.40924072265625, 174.36233520507812, 31.01336669921875, -16.135650634765625, 12.142890930175781, 231.79611206054688, -49.60699462890625, 160.9757843017578, 85.0, 69.57872009277344, -37.58272933959961, 176.10433959960938, 183.8070831298828, 87.14529418945312, 22.74267578125, 181.2315673828125, 174.71206665039062, 9.39202880859375, 13.559173583984375, 107.64030456542969, 59.955780029296875, 209.04049682617188, -9.49755859375, 198.08718872070312, 123.79409790039062, -187.74424743652344, -0.9105072021484375, 27.03802490234375, 169.65802001953125, 98.5684814453125, 7.61102294921875, -70.03256225585938, -95.58790588378906, 94.31234741210938, 119.45883178710938, 132.84413146972656, 169.8570556640625, 165.80551147460938, 21.409820556640625, 125.51312255859375, -180.59530639648438, 21.531768798828125, 216.86941528320312, 156.57073974609375, 87.1763916015625, -37.107574462890625, -7.342052459716797, 5.6840057373046875, 124.28561401367188, 0.02857208251953125, 76.62301635742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000420.npy"}
{"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 71.39762878417969, "std": 89.46037292480469, "min": -166.04409790039062, "p10": -29.280377578735347, "median": 82.5430908203125, "p90": 181.09841918945312, "max": 214.60501098632812, "pos_frac": 0.75, "sample": [180.93426513671875, 111.95442199707031, 129.22621154785156, 111.405029296875, 143.97958374023438, 92.56478881835938, 124.04315185546875, 11.438201904296875, -160.67257690429688, -9.117012023925781, 156.3214111328125, -32.0159912109375, 10.93280029296875, 126.96417236328125, 33.630943298339844, 135.91537475585938, 45.754638671875, -149.07839965820312, 40.11395263671875, 135.0001678466797, -50.446311950683594, -79.927734375, 81.43653869628906, 12.972114562988281, 151.85165405273438, 211.8726806640625, 167.5858154296875, 43.500213623046875, 11.039661407470703, 100.42955780029297, 0.0, -96.19195556640625, 84.69659423828125, 5.0869140625, 12.261611938476562, 83.26544189453125, 179.1551513671875, 179.14669799804688, 27.152297973632812, 164.28680419921875, 191.91616821289062, 161.12835693359375, 45.5679931640625, 26.284202575683594, -6.125553131103516, 21.039405822753906, 137.25399780273438, -22.3353271484375, 203.3643798828125, 141.16452026367188, 142.171142578125, -12.880233764648438, 101.4554443359375, 105.75439453125, 9.40679931640625, 185.6873779296875, -102.3918685913086, 167.66055297851562, 9.9990234375, 94.77934265136719, 40.0665283203125, 96.41370391845703, 213.5906982421875, 196.62551879882812, 167.8719482421875, 87.08318328857422, 169.83193969726562, 164.33389282226562, 166.34779357910156, 156.2537384033203, 163.4637908935547, 188.10736083984375, 108.701171875, 173.11483764648438, 111.35794067382812, 60.251800537109375, -21.186019897460938, 26.537628173828125, 171.83096313476562, 12.313987731933594, 162.7298583984375, 8.265609741210938, 43.5428466796875, 188.6141357421875, -2.43280029296875, 90.43121337890625, 38.445587158203125, -23.505508422851562, 155.417236328125, -6.507133483886719, 137.174072265625, 140.48143005371094, 206.93606567382812, 115.05389404296875, -31.272762298583984, 53.295867919921875, 131.31570434570312, -110.06005859375, -166.04409790039062, 110.31239318847656, -23.6239013671875, -31.82263946533203, 38.924346923828125, -9.9058837890625, 17.670013427734375, 181.4814453125, 6.73846435546875, -10.3233642578125, 136.69802856445312, -69.16635131835938, -28.426498413085938, 170.26553344726562, 43.52801513671875, 151.08767700195312, -23.018707275390625, 186.02230834960938, 121.4814682006836, 0.0, -40.49932861328125, -26.312427520751953, 214.60501098632812, 81.82073974609375, 28.01239013671875, 0.0, 6.73358154296875, -7.04107666015625, 200.59527587890625, -1.101776123046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000421.npy"}
{"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 48.600624084472656, "std": 98.46440887451172, "min": -195.96197509765625, "p10": -81.6297149658203, "median": 39.93986511230469, "p90": 179.18242340087892, "max": 219.16677856445312, "pos_frac": 0.6953125, "sample": [17.11590576171875, -18.19470977783203, 157.5605010986328, -34.3310546875, 171.58566284179688, -18.27136993408203, 164.89837646484375, -158.6226806640625, 34.279693603515625, -0.137969970703125, 133.1067657470703, 6.694244384765625, 41.126678466796875, 167.37591552734375, 118.83047485351562, -192.48843383789062, 9.60028076171875, 14.86297607421875, -13.937942504882812, 133.137451171875, 126.37091827392578, 38.424346923828125, 117.9747314453125, 25.941482543945312, -46.108184814453125, 163.1978759765625, -147.8190460205078, 129.36135864257812, 139.1599884033203, 191.33502197265625, 28.37615966796875, 3.818206787109375, -21.980880737304688, 77.82974243164062, 8.337272644042969, 139.92843627929688, 6.831892013549805, -124.97058868408203, 191.4261932373047, 9.70751953125, 6.4912109375, 172.91500854492188, -27.907623291015625, 42.107421875, 41.38909912109375, -173.29547119140625, -11.455093383789062, 219.16677856445312, 49.597808837890625, 164.38217163085938, 41.50531005859375, 191.47500610351562, 134.85525512695312, 21.79861068725586, 121.13424682617188, -16.136199951171875, 47.668060302734375, 14.998025894165039, 15.983154296875, 70.19989013671875, 143.7757568359375, -90.51264190673828, -37.43458557128906, -155.19082641601562, 198.93402099609375, 5.215423583984375, 116.31844329833984, 12.394256591796875, 2.3992843627929688, 196.97036743164062, 103.89266967773438, -195.96197509765625, 171.92823791503906, -35.81105041503906, 64.96917724609375, -6.738002777099609, 153.791259765625, -77.82274627685547, 183.06301879882812, 7.683086395263672, 192.621826171875, -1.9915084838867188, 143.27947998046875, 43.3865966796875, 190.457275390625, 47.719696044921875, -104.36325073242188, 12.04107666015625, 179.28903198242188, -0.4030609130859375, -15.161445617675781, -21.8702392578125, 45.325225830078125, -25.46539306640625, 143.73641967773438, 118.76249694824219, -60.61224365234375, -3.67413330078125, 55.4720458984375, 128.77981567382812, 124.59886169433594, 17.073471069335938, 204.61868286132812, 159.35614013671875, -111.2742919921875, 151.57904052734375, 65.6031494140625, -154.20501708984375, -17.54559326171875, 103.37127685546875, 0.0, -126.55059814453125, 179.13673400878906, 152.9313507080078, -123.96533203125, 45.9603271484375, 86.32878112792969, 142.21563720703125, 38.7530517578125, -1.3358230590820312, -30.377410888671875, 108.58454132080078, 4.286979675292969, 181.6155548095703, 42.10035705566406, 34.16593933105469, -1.2536582946777344, 197.70864868164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000422.npy"}
{"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 69.95562744140625, "std": 97.56948852539062, "min": -167.6171875, "p10": -39.82080993652344, "median": 61.82878875732422, "p90": 188.9058639526367, "max": 216.44534301757812, "pos_frac": 0.765625, "sample": [1.5380401611328125, 188.6544647216797, -14.781997680664062, 15.597412109375, 171.23214721679688, 154.62167358398438, 12.287145614624023, -13.713409423828125, 2.7743072509765625, 154.09332275390625, 110.11090087890625, 158.73281860351562, 28.865814208984375, 144.58615112304688, -125.83889770507812, -14.050048828125, 165.48223876953125, -94.64923095703125, 192.67205810546875, 13.347274780273438, 14.399955749511719, -33.600433349609375, 23.008651733398438, 152.84976196289062, -153.9791259765625, 125.45016479492188, 146.6652374267578, 13.351516723632812, 165.32969665527344, 192.64968872070312, 6.3518524169921875, 24.97308349609375, 72.75994873046875, 44.159454345703125, 123.172607421875, 144.40850830078125, 178.5029296875, -1.5753326416015625, -23.21307373046875, 93.15972900390625, 140.80398559570312, 182.13644409179688, -14.3021240234375, 177.04931640625, -1.503509521484375, 33.97688293457031, 9.108917236328125, 206.65106201171875, -39.474365234375, -36.24371337890625, 189.49246215820312, -39.69097900390625, -59.83551025390625, 62.556549072265625, 182.5576171875, -37.896392822265625, -120.4534912109375, -117.48291015625, 186.9449462890625, 176.77403259277344, 35.355743408203125, 0.9273834228515625, 155.48728942871094, 15.471687316894531, -13.020332336425781, 129.66636657714844, 17.88688850402832, 176.62603759765625, 12.003517150878906, 33.11216735839844, 152.06756591796875, 165.09835815429688, 130.66500854492188, 149.08969116210938, 44.47613525390625, 66.3292236328125, -64.7396240234375, 176.52508544921875, 209.95852661132812, 74.90829467773438, 17.582733154296875, 183.93292236328125, 15.91448974609375, 160.23861694335938, 197.7186279296875, 61.10102844238281, 140.07693481445312, 25.893646240234375, 3.9446563720703125, 200.68014526367188, 151.33885192871094, 161.3134765625, 200.115966796875, 3.97723388671875, 209.41693115234375, 141.95645141601562, 2.5876216888427734, -75.00357055664062, 190.32888793945312, -4.53924560546875, 1.9417266845703125, 146.92636108398438, 38.77496337890625, 196.69906616210938, 216.44534301757812, -40.123748779296875, -11.290054321289062, 133.63397216796875, 192.4415740966797, -17.2078857421875, -147.05548095703125, 50.70098876953125, -99.27053833007812, 38.0830078125, 18.070865631103516, 177.4730224609375, 77.36105346679688, 152.6026611328125, 181.75765991210938, 64.5220947265625, 143.60211181640625, 154.48529052734375, 155.17953491210938, -135.77017211914062, 31.848785400390625, 105.10948181152344, -3.0323925018310547, -167.6171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000423.npy"}
{"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 66.21212005615234, "std": 87.0085678100586, "min": -175.05880737304688, "p10": -38.981817626953124, "median": 67.69316101074219, "p90": 172.49555511474608, "max": 224.925048828125, "pos_frac": 0.7578125, "sample": [78.85365295410156, 51.57749938964844, 0.74017333984375, -7.52850341796875, 86.6151123046875, 12.793975830078125, -11.439323425292969, -5.80633544921875, 34.57636260986328, 9.4481201171875, 2.2032012939453125, -130.60302734375, 224.925048828125, 67.43548583984375, -5.748870849609375, 104.0903549194336, 17.817138671875, 199.17808532714844, 172.24420166015625, 154.29119873046875, -6.533939361572266, 6.08740234375, 199.55984497070312, 151.43511962890625, 109.16192626953125, 120.00029754638672, -2.2650489807128906, -50.245269775390625, 137.748291015625, 132.07846069335938, 117.9103012084961, 201.89505004882812, 159.99241638183594, -85.468017578125, 127.14404296875, 147.96092224121094, 34.96832275390625, 153.62240600585938, 178.87322998046875, -63.6826171875, 12.973724365234375, 165.17391967773438, 154.50216674804688, -175.05880737304688, -0.0951995849609375, 197.1187744140625, 187.25924682617188, 75.57244873046875, -159.6405792236328, 42.2061767578125, -74.23228454589844, -2.55535888671875, -5.401611328125, 173.08204650878906, 92.36688232421875, 161.44912719726562, 67.95083618164062, 159.05435180664062, 72.09390258789062, 89.34979248046875, 6.860809326171875, -7.267120361328125, 170.62762451171875, 157.05694580078125, 199.57748413085938, 142.72695922851562, 70.90933227539062, 37.6658935546875, -59.55938720703125, -45.0068359375, -40.311279296875, 74.78164672851562, 146.21885681152344, 197.253662109375, 210.09637451171875, 169.51702880859375, 18.109092712402344, 47.28400421142578, 140.40658569335938, 109.26992797851562, 46.876190185546875, 123.47494506835938, -2.0021514892578125, 3.0458908081054688, 175.46836853027344, 74.19329071044922, -27.596160888671875, 0.0, -47.15159606933594, 32.14640808105469, 12.638275146484375, 136.72964477539062, 154.763916015625, -9.732757568359375, 59.670127868652344, 163.8723907470703, 106.95986938476562, 156.689453125, 159.28326416015625, 137.33328247070312, 16.167362213134766, -105.49200439453125, 134.3740234375, 148.10946655273438, 129.36807250976562, 1.5006351470947266, -9.054927825927734, 12.63458251953125, -38.41204833984375, 6.055681228637695, 135.0948028564453, -57.151283264160156, 166.80401611328125, 155.76937866210938, -35.78948974609375, 8.386920928955078, 15.980953216552734, 27.130237579345703, 27.488616943359375, 69.1109619140625, 25.249542236328125, 94.44776916503906, -34.104644775390625, 29.932769775390625, 77.7783203125, 16.007431030273438, 1.7919578552246094, 194.0157470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000424.npy"}
{"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 58.04555892944336, "std": 92.43417358398438, "min": -214.5191650390625, "p10": -39.265625, "median": 27.250160217285156, "p90": 183.3983947753906, "max": 230.8770751953125, "pos_frac": 0.7265625, "sample": [153.930419921875, 163.98251342773438, 77.0479736328125, 62.94139862060547, -15.0235595703125, 17.248504638671875, 43.56793212890625, 153.6769561767578, 12.902755737304688, -19.664932250976562, 168.00808715820312, 22.54559326171875, 179.06890869140625, 89.04132080078125, 142.2534942626953, -0.683197021484375, 131.32064819335938, 146.8773193359375, 1.4995059967041016, 22.45159912109375, 84.7315673828125, 144.39085388183594, 21.020553588867188, -2.8735427856445312, 163.83905029296875, 107.40424346923828, 4.23326301574707, 22.6531982421875, -95.7313232421875, -4.391349792480469, 49.88067626953125, 4.86859130859375, -19.8858642578125, 15.69879150390625, 192.52178955078125, -20.515966415405273, -11.088043212890625, 199.11904907226562, 7.428802490234375, 14.185844421386719, -20.983123779296875, 198.99856567382812, 148.4702606201172, 31.8548583984375, 6.491455078125, 6.148002624511719, -39.40052795410156, -0.237213134765625, -154.60256958007812, 229.81951904296875, 80.0843505859375, 149.228271484375, -102.4967041015625, 54.64544677734375, 177.760009765625, -2.5023651123046875, 153.9517364501953, 230.8770751953125, 40.160888671875, 26.39654541015625, -39.20780944824219, 27.020614624023438, 7.790699005126953, 43.27061462402344, 188.828369140625, -47.168975830078125, 102.70945739746094, -4.196647644042969, 171.88034057617188, 145.1884765625, 149.04815673828125, 9.6090087890625, 206.8994140625, 7.2209320068359375, -20.186920166015625, 12.247901916503906, -12.47280502319336, 29.3009033203125, -18.860702514648438, 112.83308410644531, 0.0037384033203125, 173.783447265625, 31.316131591796875, 134.5361328125, 184.8795928955078, -108.6545639038086, 15.695960998535156, -39.64936828613281, 27.479705810546875, -25.7716064453125, 18.709228515625, -25.119396209716797, 35.533111572265625, -101.04911804199219, -44.9744873046875, -71.85830688476562, 117.6346435546875, 6.5058135986328125, 145.67144775390625, -132.23712158203125, 48.954254150390625, 228.70123291015625, 17.392730712890625, 165.2814178466797, 60.091705322265625, 115.4466552734375, 174.392333984375, 5.34661865234375, 6.063751220703125, 160.0006103515625, 0.0, -12.468994140625, 165.4178466796875, 186.3245849609375, 222.95822143554688, -4.506349563598633, 17.14252471923828, 169.24363708496094, -214.5191650390625, 59.60758972167969, 197.81683349609375, 182.7635955810547, 134.4136962890625, 166.29295349121094, 188.81588745117188, -64.51087188720703, -7.1319427490234375, 1.1630783081054688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000425.npy"}
{"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 61.614166259765625, "std": 85.28872680664062, "min": -155.71859741210938, "p10": -17.760881042480467, "median": 33.55778503417969, "p90": 180.13297882080076, "max": 213.55267333984375, "pos_frac": 0.7578125, "sample": [151.6244354248047, 15.190628051757812, 2.162860870361328, -35.90179443359375, 191.53460693359375, 24.946701049804688, 128.11663818359375, 2.528390884399414, -137.44320678710938, 3.6733856201171875, -147.75247192382812, -47.01324462890625, -48.16654968261719, 134.86712646484375, 160.43186950683594, 171.55465698242188, 153.56982421875, 29.40399169921875, 146.6474151611328, -0.1696929931640625, -26.623291015625, 55.20489501953125, 191.92864990234375, 21.655540466308594, 3.3937530517578125, 22.022064208984375, 178.9290008544922, 191.06396484375, 153.00067138671875, 182.9422607421875, 127.3311996459961, 4.386993408203125, 185.87515258789062, 130.8231201171875, 162.34408569335938, 7.045408248901367, 0.6223602294921875, 25.284629821777344, 175.46383666992188, 186.2428741455078, 43.01043701171875, 191.69874572753906, 9.373504638671875, -6.486083984375, 167.629150390625, 169.66925048828125, -2.98175048828125, 131.07484436035156, 17.7393798828125, 14.138427734375, -131.2206268310547, 26.24407958984375, 61.455047607421875, 4.103057861328125, 28.02608871459961, 170.00662231445312, -45.04315185546875, 40.361610412597656, 168.793701171875, 131.89926147460938, 137.23431396484375, 145.19647216796875, -35.2476806640625, -2.471160888671875, -3.3083343505859375, 82.14129638671875, 60.85520935058594, 141.44317626953125, -0.2273712158203125, -10.359375, 68.66094970703125, 26.669113159179688, -14.101333618164062, 24.064971923828125, -0.26941871643066406, 205.21795654296875, -5.160797119140625, -1.88458251953125, 56.15501403808594, 146.08370971679688, -32.53434753417969, 160.13589477539062, 69.45074462890625, 79.22256469726562, 55.70008850097656, 31.010223388671875, 36.1053466796875, 9.285652160644531, 9.30263900756836, 148.65603637695312, 119.4879150390625, 21.511627197265625, 1.856466293334961, -18.310928344726562, 191.56982421875, 42.80828857421875, 71.68780517578125, 183.87103271484375, -0.7371826171875, -11.687101364135742, 102.18529510498047, 5.039794921875, 0.0, 11.204307556152344, 1.92999267578125, 201.27066040039062, 76.2923583984375, 3.0956268310546875, 5.953369140625, 66.39851379394531, 148.0965576171875, 29.60589599609375, -126.767822265625, -5.6866455078125, -8.644660949707031, 145.011962890625, -14.108062744140625, 171.06256103515625, 157.62075805664062, 116.01849365234375, 134.528564453125, 4.943807601928711, -155.71859741210938, 117.60736083984375, 213.55267333984375, 58.96659851074219, -17.525146484375, 187.36048889160156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000426.npy"}
{"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 82.20220947265625, "std": 90.28668212890625, "min": -167.2431640625, "p10": -23.56473388671875, "median": 85.55368041992188, "p90": 189.14630279541015, "max": 318.65435791015625, "pos_frac": 0.796875, "sample": [18.38904571533203, 197.297119140625, 18.289566040039062, -28.3970947265625, -60.825592041015625, 12.844589233398438, 149.3836669921875, 102.8040771484375, 119.23886108398438, 30.453536987304688, 122.19931030273438, 46.34834289550781, 194.69625854492188, 18.39403533935547, 163.07537841796875, 180.10757446289062, 163.8457794189453, 7.864479064941406, 155.19271850585938, 200.31637573242188, 196.34988403320312, 51.02717590332031, -19.933517456054688, 127.488037109375, 162.25787353515625, 152.23416137695312, 141.94082641601562, 90.40032958984375, 69.40718078613281, 51.192909240722656, 223.31536865234375, 179.77218627929688, 111.1639404296875, 38.3802490234375, 1.6947479248046875, -1.1764373779296875, 131.7307586669922, 7.749744415283203, -4.023834228515625, -35.978912353515625, 180.8985137939453, 7.031982421875, -98.23651123046875, 61.03971862792969, -129.53668212890625, 150.39231872558594, -0.5036525726318359, 189.04515075683594, -25.156005859375, 163.57037353515625, 187.44247436523438, 33.25123596191406, 156.3912353515625, 80.70703125, 204.1318359375, 183.1463623046875, 214.19876098632812, -10.220428466796875, 12.802520751953125, -11.211639404296875, 148.56430053710938, 132.26193237304688, 204.13528442382812, 20.37493896484375, 96.47149658203125, 55.81333923339844, -17.46466064453125, 158.66793823242188, 156.17596435546875, -77.61492919921875, -167.2431640625, 182.59396362304688, 124.00257110595703, -24.534027099609375, 124.12841796875, 20.709909439086914, -23.149322509765625, 158.33775329589844, 130.12283325195312, 31.8619384765625, 217.18356323242188, -132.16461181640625, 157.2100372314453, -34.556976318359375, 25.487808227539062, 93.20877075195312, 149.18910217285156, 62.454193115234375, 78.76718139648438, 156.30450439453125, 155.12118530273438, 179.8680419921875, -29.2137451171875, 54.1094970703125, 170.9765625, 146.990966796875, 189.38232421875, 54.10528564453125, 0.740264892578125, 131.79208374023438, -22.3896484375, 14.151901245117188, 43.88487243652344, -30.304214477539062, 7.360908508300781, 318.65435791015625, -10.285598754882812, 158.04908752441406, 4.6380462646484375, 175.115966796875, 190.91632080078125, 112.28500366210938, 147.0636444091797, -4.856292724609375, 50.08293151855469, -10.688400268554688, 162.11334228515625, 92.00013732910156, 1.94793701171875, 4.3695068359375, 177.34384155273438, 29.173049926757812, -1.4837188720703125, 34.42866516113281, 32.61981201171875, 166.42401123046875, 206.59402465820312, 175.80479431152344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000427.npy"}
{"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 77.61674499511719, "std": 89.1983871459961, "min": -158.68792724609375, "p10": -13.843463897705073, "median": 67.2984619140625, "p90": 180.80141448974607, "max": 274.9434509277344, "pos_frac": 0.796875, "sample": [25.685951232910156, -2.510498046875, -74.38471984863281, 173.78448486328125, 194.49981689453125, -21.067962646484375, 55.6390380859375, 63.1168212890625, -101.78341674804688, 12.084762573242188, 274.9434509277344, 23.248016357421875, 115.37918090820312, 117.20587158203125, -8.30450439453125, 173.87326049804688, 20.92303466796875, -11.28369140625, 17.71881675720215, -5.63677978515625, 190.65338134765625, -6.380157470703125, -158.68792724609375, 207.70907592773438, 169.8902587890625, 196.7265625, 152.46322631835938, 151.4212646484375, 132.40769958496094, 237.00335693359375, 31.633865356445312, 186.0472412109375, 2.387296676635742, 96.89486694335938, -17.527740478515625, 14.34210205078125, -2.3114776611328125, 144.60028076171875, 38.110107421875, 147.48892211914062, 60.62451171875, -5.892967224121094, -152.14559936523438, 149.0467071533203, 167.20794677734375, 207.32412719726562, -62.77520751953125, 45.299591064453125, 3.6016693115234375, 16.187850952148438, 156.33470153808594, 119.5050048828125, 135.3427734375, 114.91386413574219, 26.023651123046875, 135.64385986328125, 162.49583435058594, 116.50286865234375, -41.74224853515625, 229.21929931640625, -137.48599243164062, 206.71902465820312, 159.0126953125, 49.057525634765625, 161.80889892578125, 6.3088226318359375, 143.19036865234375, 165.2699432373047, 119.87252807617188, 49.0667724609375, 20.425880432128906, 37.822509765625, 163.18704223632812, 161.91683959960938, 250.37213134765625, 16.611053466796875, 49.47564697265625, -0.12255859375, 75.43034362792969, 22.270736694335938, 53.10474395751953, 17.90826416015625, 160.71319580078125, 71.4801025390625, -5.0072174072265625, 5.8155670166015625, -18.417930603027344, 25.497802734375, 107.6212158203125, 140.9110870361328, 257.83575439453125, 179.2267303466797, 160.37969970703125, 3.49627685546875, 89.58990478515625, -138.33859252929688, -19.690765380859375, 170.73373413085938, 82.859619140625, 0.0, 24.811004638671875, 172.9335479736328, 155.6229248046875, 18.111602783203125, 159.00604248046875, 20.886329650878906, -12.264488220214844, 116.0247802734375, 133.69020080566406, 79.85595703125, 55.074310302734375, 149.42782592773438, 36.95655059814453, 62.757225036621094, 170.12033081054688, 81.7666015625, 15.563507080078125, 101.84548950195312, 128.42733764648438, -2.308013916015625, 45.234336853027344, 28.70123291015625, 184.47567749023438, -20.686491012573242, -4.352996826171875, 154.44015502929688, 139.0738525390625, 133.09751892089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000428.npy"}
{"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 60.91999435424805, "std": 104.90223693847656, "min": -203.61280822753906, "p10": -66.59468688964843, "median": 42.938175201416016, "p90": 192.4625518798828, "max": 225.19009399414062, "pos_frac": 0.6796875, "sample": [164.65342712402344, -196.13812255859375, -9.1796875, 135.1753387451172, 174.1683807373047, 167.99560546875, 41.025428771972656, 29.10906982421875, -51.15434265136719, 138.2894287109375, 6.25396728515625, -90.59536743164062, 30.18817138671875, 163.78013610839844, 198.02685546875, 164.84751892089844, 1.6189632415771484, 161.7725830078125, -44.515411376953125, 25.1839599609375, 129.37925720214844, 141.7265625, 5.839111328125, 171.05026245117188, -65.814208984375, 187.100341796875, 128.57574462890625, 204.3557891845703, 17.210845947265625, 89.611572265625, -137.25674438476562, 151.49598693847656, 145.73867797851562, 168.34616088867188, 32.51399230957031, 31.470855712890625, -132.5596923828125, -8.89508056640625, 170.76040649414062, 225.19009399414062, 174.20831298828125, 161.28079223632812, -93.13114929199219, -3.636077880859375, 107.11752319335938, -7.6722869873046875, 210.458740234375, 115.42218017578125, 8.439895629882812, -41.39434814453125, 7.819021224975586, 198.67135620117188, 205.65338134765625, 176.97509765625, 13.991256713867188, 56.428192138671875, 216.99783325195312, -115.83091735839844, -108.74319458007812, 163.41848754882812, -3.2490234375, 44.850921630859375, 175.83120727539062, 191.02767944335938, -32.66607666015625, 16.35462188720703, 52.1881103515625, 62.4322509765625, 21.31714630126953, 92.68252563476562, -24.641937255859375, -71.89812469482422, -40.0355224609375, -68.41580200195312, 158.27542114257812, -203.61280822753906, 184.868408203125, -15.481742858886719, 17.732757568359375, 175.30348205566406, -3.4239139556884766, 0.0, 172.31248474121094, 211.64773559570312, 177.8389892578125, 84.14627075195312, 175.0310516357422, 51.78400421142578, -2.602264404296875, 6.022634506225586, 203.71188354492188, -32.36566162109375, 132.41868591308594, -16.423240661621094, 15.010696411132812, 55.2861328125, 21.65723419189453, 4.572853088378906, -1.9171142578125, 121.6744384765625, -90.96636962890625, -20.597412109375, -38.190216064453125, 17.953567504882812, -36.42364501953125, 108.37326049804688, 138.5528564453125, 195.9542236328125, -168.03929138183594, 107.4044189453125, 198.15872192382812, 160.1444091796875, -50.32856750488281, 187.39414978027344, 193.49847412109375, -20.55078125, 20.640974044799805, 3.1387481689453125, 204.3262939453125, 192.01858520507812, -38.375396728515625, 152.15826416015625, -17.608810424804688, -30.374893188476562, 92.01425170898438, -51.42002868652344, 119.89910888671875, -153.061767578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000429.npy"}
{"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 52.7877197265625, "std": 98.23112487792969, "min": -181.80908203125, "p10": -66.97367248535156, "median": 33.78564453125, "p90": 179.54716491699216, "max": 232.25015258789062, "pos_frac": 0.703125, "sample": [-92.05886840820312, -28.895767211914062, -137.22569274902344, 5.921295166015625, 134.60882568359375, 134.9228515625, -155.7161102294922, 171.30364990234375, -13.524322509765625, 29.647159576416016, 153.39190673828125, 19.633724212646484, 34.368133544921875, 42.53944396972656, 77.45309448242188, -181.80908203125, 189.7008056640625, 23.904388427734375, 1.6008243560791016, 161.478271484375, 78.99568176269531, -109.06512451171875, -25.955108642578125, 138.41213989257812, 12.947998046875, 33.203155517578125, -45.066650390625, 0.0, 184.17520141601562, 56.9320068359375, 108.01030731201172, -23.099853515625, 154.2786865234375, 41.22901916503906, 18.452362060546875, 0.0, 73.36370849609375, -4.875968933105469, 3.56072998046875, 143.04434204101562, -97.73150634765625, 199.51358032226562, -70.12847900390625, 157.79769897460938, 198.00369262695312, 115.92939758300781, 98.12416076660156, 21.76397705078125, -24.056304931640625, 232.25015258789062, 142.95980834960938, -11.48248291015625, 112.37989807128906, 19.1444091796875, 168.41949462890625, 158.76443481445312, 20.849105834960938, -65.62161254882812, 166.53948974609375, -158.18115234375, 160.27674865722656, 175.4790496826172, 188.03778076171875, 171.7105712890625, 115.42501831054688, 60.85302734375, 170.42620849609375, 150.44332885742188, 11.714834213256836, 191.819091796875, 30.3951416015625, 32.51832580566406, 216.27593994140625, 152.94808959960938, -39.31523132324219, -50.433189392089844, -4.59326171875, -2.575164794921875, 1.2750396728515625, 67.06646728515625, 157.5391845703125, 40.84014892578125, -34.277374267578125, 12.554248809814453, -14.407577514648438, 52.16485595703125, 4.805023193359375, 214.43606567382812, 0.0, 6.730194091796875, 42.868919372558594, -27.444091796875, 91.23074340820312, 15.602203369140625, 205.87091064453125, 137.60952758789062, 33.08489990234375, 171.67245483398438, 8.30778694152832, -127.18971252441406, 206.5004119873047, 193.05047607421875, 46.13800048828125, -42.65733337402344, -134.67901611328125, 88.13174438476562, 145.6766357421875, 4.355583190917969, 187.86834716796875, 134.34487915039062, -128.36767578125, 4.3643341064453125, 61.970062255859375, -12.806137084960938, 38.94664001464844, 177.563720703125, 10.795074462890625, -1.273773193359375, -1.1604766845703125, -154.11642456054688, -10.03546142578125, -157.1734619140625, 71.9666748046875, 139.3690185546875, 136.39828491210938, 29.1014404296875, -24.1942138671875, 159.97933959960938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000430.npy"}
{"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 50.63361358642578, "std": 91.5191879272461, "min": -164.40359497070312, "p10": -43.29888000488281, "median": 31.079116821289062, "p90": 184.34313354492187, "max": 251.5479736328125, "pos_frac": 0.6953125, "sample": [-9.377899169921875, -164.40359497070312, 64.99403381347656, 159.3957977294922, 145.581787109375, 68.95089721679688, 22.155868530273438, 114.04562377929688, 7.826507568359375, -76.0264892578125, -2.3892822265625, 26.3638916015625, 5.18988037109375, 64.76007080078125, -42.97021484375, 63.865203857421875, -39.156036376953125, 126.78642272949219, 121.02876281738281, 12.79348373413086, -3.780059814453125, 141.59286499023438, 63.41532897949219, -115.6181640625, 109.24154663085938, 154.63272094726562, 94.556884765625, -58.942474365234375, 162.13504028320312, 207.99093627929688, 55.14396667480469, 106.69581604003906, 58.60206604003906, -118.27698516845703, -52.07566833496094, 153.37728881835938, 111.96117401123047, -22.094512939453125, -6.773811340332031, 116.49101257324219, -37.0044059753418, -152.762939453125, 7.803422927856445, 1.070556640625, 143.68447875976562, 54.248626708984375, 190.9114227294922, -19.515792846679688, -42.878631591796875, -151.42898559570312, 4.338510513305664, 34.99346923828125, 1.4228858947753906, 154.50787353515625, -0.196685791015625, 162.23455810546875, 48.699951171875, 202.9229736328125, 168.64739990234375, -20.882781982421875, -1.2928466796875, 14.824417114257812, 175.4596405029297, 33.789581298828125, 20.036956787109375, 116.76123046875, 67.667236328125, 186.10137939453125, -23.359474182128906, -21.356170654296875, 183.589599609375, -12.318824768066406, -1.847412109375, 192.76873779296875, -2.799285888671875, 231.37295532226562, 5.067779541015625, 218.69366455078125, 44.64833068847656, 31.116546630859375, 196.60751342773438, 251.5479736328125, 44.9176025390625, -76.725341796875, -1.1338062286376953, -8.7257080078125, 152.31544494628906, 32.619781494140625, -44.9981689453125, 19.489036560058594, 215.381103515625, 148.34320068359375, 182.78221130371094, -30.284011840820312, 3.1865692138671875, 229.77783203125, -44.065765380859375, 201.991455078125, -0.6061344146728516, 10.91229248046875, 38.94036865234375, 54.469146728515625, 10.43072509765625, 5.19757080078125, 31.04168701171875, 4.6825714111328125, 153.68124389648438, 61.45166015625, 130.36550903320312, -15.16839599609375, 35.100990295410156, -13.917938232421875, 6.2758941650390625, 148.03823852539062, 25.228851318359375, 105.27430725097656, 2.6383514404296875, 4.8497314453125, -111.18043518066406, 34.04908752441406, -27.94470977783203, -22.33148193359375, 12.510200500488281, 1.231964111328125, -117.08648681640625, 214.4727020263672, 151.93728637695312, 36.10112762451172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000431.npy"}
{"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 66.13349151611328, "std": 92.439453125, "min": -188.3057861328125, "p10": -35.624472045898436, "median": 43.1055908203125, "p90": 189.27211303710936, "max": 232.4835205078125, "pos_frac": 0.703125, "sample": [-13.684326171875, 172.18942260742188, 4.647972106933594, 167.858642578125, 217.29736328125, 205.35382080078125, 164.81350708007812, 79.97467041015625, -43.60986328125, -159.93243408203125, 201.936279296875, -6.2582550048828125, 174.82220458984375, 39.17803955078125, -9.467117309570312, 13.584968566894531, 38.43231201171875, 184.6352081298828, -37.55519104003906, 6.037818908691406, 135.05860900878906, -188.3057861328125, 52.671356201171875, -1.2146339416503906, -33.521392822265625, 7.077239036560059, -26.41973876953125, 171.96237182617188, 70.4520263671875, -8.250579833984375, 28.09307861328125, 176.69834899902344, 202.25778198242188, 182.02647399902344, -35.3369140625, -6.798545837402344, 149.42147827148438, -36.295440673828125, 188.37322998046875, 21.52728271484375, 136.31808471679688, 142.23806762695312, 36.127227783203125, 2.555570602416992, -3.2685623168945312, 0.468719482421875, 79.3994140625, 15.719215393066406, -127.36618041992188, -4.1239013671875, 134.19537353515625, 71.34616088867188, 6.225440979003906, 196.1004638671875, -18.47333526611328, 13.497711181640625, -85.82210540771484, 161.88308715820312, -128.81063842773438, -4.099967956542969, -7.149444580078125, 25.804428100585938, 157.87486267089844, -12.299713134765625, 161.79852294921875, 5.566394805908203, 197.06549072265625, 148.58578491210938, 147.57476806640625, 146.84100341796875, -42.56501770019531, 75.48591613769531, 5.90020751953125, -17.532180786132812, 16.938865661621094, 201.00457763671875, 147.9409942626953, 207.78164672851562, 33.589691162109375, 143.65811157226562, 42.34228515625, 4.603584289550781, 232.4835205078125, -15.924072265625, 155.96844482421875, 113.69354248046875, 178.0088653564453, 132.32699584960938, -28.253570556640625, 138.271484375, 11.532669067382812, -13.301326751708984, 0.0, 191.3695068359375, 173.99069213867188, -22.312103271484375, 10.1202392578125, -38.03765869140625, -3.080810546875, 67.76739501953125, 63.15093994140625, -5.9047698974609375, -38.38481140136719, 147.73541259765625, 129.13839721679688, 131.2287139892578, 205.80990600585938, 160.04449462890625, 203.7242431640625, 168.9626007080078, 0.39947509765625, 148.05145263671875, 43.868896484375, 206.8809814453125, -4.405673980712891, 45.8648681640625, 77.65628051757812, 109.73275756835938, 5.600393295288086, 162.44091796875, 102.31182861328125, 154.16055297851562, -55.071746826171875, 4.687896728515625, 63.83018493652344, 128.3256378173828, -5.402851104736328, -38.62542724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000432.npy"}
{"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 52.58147430419922, "std": 93.76206970214844, "min": -205.5887451171875, "p10": -39.24098815917968, "median": 46.417293548583984, "p90": 154.72247924804685, "max": 324.20526123046875, "pos_frac": 0.7421875, "sample": [127.54048156738281, -160.60659790039062, -186.50570678710938, 198.94415283203125, 91.23095703125, 36.60935974121094, 131.4063720703125, -14.87957763671875, -14.377607345581055, 38.54144287109375, 24.197723388671875, 20.005859375, 150.74868774414062, 86.9384765625, 55.985107421875, 22.5111083984375, -11.012176513671875, 49.78746795654297, 65.24114990234375, 18.680206298828125, 41.54071044921875, 12.992401123046875, 140.9088897705078, -35.96403503417969, -0.17712020874023438, -12.67337417602539, 189.52078247070312, 5.042972564697266, 199.22491455078125, 165.70263671875, 151.05560302734375, 2.1082763671875, 146.9517822265625, 52.845703125, 96.03195190429688, 196.87741088867188, 151.71188354492188, 138.18930053710938, -121.63627624511719, 63.4554443359375, -12.651557922363281, 19.2862548828125, -0.146575927734375, 35.44618225097656, 58.09791564941406, 88.50570678710938, 127.03793334960938, -0.8621177673339844, 29.31756591796875, 147.11480712890625, 139.4179229736328, 52.26087951660156, 126.47859191894531, 7.9326171875, 33.84130859375, -117.7083740234375, 49.5098876953125, 146.0757598876953, 16.634410858154297, 10.165901184082031, -21.43121337890625, 87.55247497558594, -70.2410659790039, 124.52857971191406, 174.23934936523438, 72.94076538085938, 141.57467651367188, -205.5887451171875, 101.22544860839844, 142.02239990234375, -45.402923583984375, 18.792037963867188, 163.04550170898438, -8.172599792480469, 6.07061767578125, 33.8741455078125, 116.62174987792969, 140.80535888671875, 150.22866821289062, -180.55770874023438, -120.06268310546875, 168.3597869873047, 98.64695739746094, -93.79415893554688, 27.4703369140625, -3.2000732421875, 158.8106689453125, -6.584690093994141, 74.47329711914062, 43.211883544921875, 147.29808044433594, -0.9561386108398438, -187.09469604492188, 25.817745208740234, 21.951217651367188, 119.95494842529297, 145.52352905273438, 176.69813537597656, 130.8062286376953, -127.5972900390625, 48.1644287109375, -13.44708251953125, 3.5230026245117188, 324.20526123046875, -17.4683837890625, 125.22967529296875, 163.4893798828125, 152.1363067626953, 152.97039794921875, 1.3561019897460938, 48.538116455078125, 12.77703857421875, 62.007110595703125, -27.398849487304688, 30.682418823242188, 107.61331176757812, 44.67015838623047, -7.697299957275391, 145.36524963378906, 206.62374877929688, 33.2225341796875, 125.0281982421875, -162.43992614746094, 74.58293151855469, -21.291114807128906, -36.60015869140625, 137.32138061523438, 4.95306396484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000433.npy"}
{"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 64.64707946777344, "std": 92.78485870361328, "min": -186.07498168945312, "p10": -29.32683181762695, "median": 37.649925231933594, "p90": 178.2259521484375, "max": 248.66033935546875, "pos_frac": 0.7109375, "sample": [75.11509704589844, -11.2657470703125, -10.46160888671875, 7.110931396484375, 175.07313537597656, 23.420120239257812, 97.4499282836914, 138.70103454589844, -20.128173828125, 243.36279296875, -9.057151794433594, 162.4281768798828, -85.85333251953125, 1.60784912109375, -10.957202911376953, 11.11883544921875, -23.13031005859375, 136.93765258789062, 99.65117645263672, 0.287109375, 12.760772705078125, 231.0770263671875, -6.96575927734375, 248.66033935546875, 174.822998046875, -82.82635498046875, 0.0, -16.59984588623047, 46.45642852783203, 134.62203979492188, 208.342529296875, 160.11050415039062, -22.021224975585938, 165.99639892578125, 146.3199462890625, 153.1013641357422, -12.43328857421875, 55.569976806640625, 28.52679443359375, 205.35195922851562, 46.637908935546875, 125.00320434570312, 188.49893188476562, -48.11932373046875, 34.44715881347656, -41.48809814453125, 130.14971923828125, 140.49090576171875, 29.38018798828125, 127.6182861328125, 180.04666137695312, 177.44564819335938, 121.21524047851562, 239.658935546875, 28.55584716796875, 129.81857299804688, 160.245361328125, 13.235565185546875, -40.85394287109375, 173.730712890625, 12.148223876953125, 136.72271728515625, 209.5172119140625, -18.80266571044922, 18.3394775390625, 58.196990966796875, 91.24496459960938, 171.47756958007812, 24.471771240234375, 18.950191497802734, 199.8764190673828, 7.9835052490234375, -38.49192810058594, 186.55227661132812, -62.86346435546875, 1.2038345336914062, -0.752838134765625, 200.11669921875, 6.11810302734375, 134.18582153320312, 162.04220581054688, -32.125640869140625, 66.90481567382812, 11.18890380859375, 40.852691650390625, 164.1221923828125, 170.105224609375, -183.998779296875, 134.89663696289062, -21.617828369140625, 71.86813354492188, -45.3173828125, -125.84906005859375, 0.0, 160.54931640625, 32.47808074951172, 27.19329833984375, 176.15185546875, 175.95037841796875, 13.638618469238281, -18.598419189453125, -6.043365478515625, -23.843381881713867, 186.1365966796875, -9.420074462890625, 10.401477813720703, -7.143085479736328, -28.127342224121094, 157.95660400390625, 84.47932434082031, 41.801971435546875, 2.6972579956054688, 0.0, 134.95619201660156, 128.0508575439453, -6.996246337890625, 21.344985961914062, 161.14151000976562, 83.59722900390625, 78.46865844726562, -186.07498168945312, 15.806549072265625, 26.52838134765625, -7.5821075439453125, -119.41485595703125, 176.60049438476562, 120.45829772949219, 154.4144287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000434.npy"}
{"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 55.98232650756836, "std": 90.2076187133789, "min": -183.9681396484375, "p10": -34.01533050537109, "median": 37.242401123046875, "p90": 177.38624877929686, "max": 231.23602294921875, "pos_frac": 0.7578125, "sample": [148.74197387695312, 9.217681884765625, 5.091300964355469, 59.55272674560547, 34.07940673828125, 188.40440368652344, 161.61070251464844, -130.41624450683594, 136.9409637451172, -5.878021240234375, -10.924945831298828, -22.174026489257812, -183.9681396484375, 46.64801788330078, 4.691923141479492, -20.03448486328125, 76.46832275390625, 71.54873657226562, 168.98922729492188, 99.36465454101562, 18.1768798828125, 28.223215103149414, 177.30859375, 0.0, 3.799074172973633, -12.486083984375, 70.66665649414062, 25.369651794433594, 139.42416381835938, 14.506988525390625, 149.89328002929688, 161.19683837890625, 134.7484130859375, -24.842567443847656, -7.933540344238281, 211.30441284179688, 90.85249328613281, 178.64892578125, 47.562744140625, 7.858123779296875, 178.39920043945312, 177.81109619140625, 177.56744384765625, 26.869430541992188, 152.62628173828125, 25.602813720703125, 153.51492309570312, -37.050262451171875, 171.76454162597656, 30.67572021484375, 150.81866455078125, 96.88943481445312, 202.72879028320312, 156.20770263671875, 24.146867752075195, 3.8950347900390625, 42.5289306640625, 28.386306762695312, 106.69111633300781, -96.93939208984375, 11.548233032226562, 204.0799560546875, 9.255104064941406, -155.3764190673828, 19.682350158691406, 11.696563720703125, 40.4053955078125, 178.74325561523438, -131.43267822265625, -66.91647338867188, 28.768417358398438, -2.7349395751953125, -8.80340576171875, 192.72616577148438, 189.19903564453125, 7.965087890625, 149.57200622558594, 169.28594970703125, -113.63916015625, 120.3812255859375, 4.21014404296875, 151.89926147460938, 63.613525390625, 153.19895935058594, -121.866943359375, 14.0870361328125, 81.33453369140625, 4.502363204956055, 0.1150665283203125, -15.979461669921875, -8.021408081054688, 148.89013671875, 101.89846801757812, 231.23602294921875, 114.65570068359375, 168.42982482910156, 123.17436218261719, 126.23204040527344, -25.59760284423828, 64.65283203125, -8.046981811523438, 43.812408447265625, 45.34850311279297, 3.6528072357177734, 21.375732421875, 13.84466552734375, 40.433624267578125, 7.3099517822265625, 141.4890594482422, 6.503181457519531, -10.174346923828125, -61.59765625, -102.65289306640625, -63.777366638183594, -1.90460205078125, 216.10516357421875, 8.979156494140625, 22.9063720703125, 110.86917114257812, -149.06619262695312, 47.535125732421875, 138.37606811523438, 159.29779052734375, -1.8855438232421875, -32.71464538574219, 153.80303955078125, 45.02569580078125, 110.45327758789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000435.npy"}
{"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 39.97752380371094, "std": 87.97541046142578, "min": -183.35641479492188, "p10": -50.66243438720703, "median": 20.590160369873047, "p90": 171.28223419189453, "max": 224.134765625, "pos_frac": 0.6875, "sample": [90.44769287109375, -55.359130859375, 210.55337524414062, 141.48158264160156, 2.4185714721679688, 20.573020935058594, 156.3992156982422, 78.27569580078125, -36.59552001953125, 103.3526611328125, -106.16748046875, 59.915008544921875, -92.16241455078125, 34.0692138671875, -4.204345703125, 198.7191162109375, 30.98260498046875, -21.470970153808594, -44.54600524902344, -85.332275390625, 1.722747802734375, 104.80818176269531, 145.4913330078125, 137.8577880859375, 181.38232421875, -34.44158935546875, 179.84600830078125, 0.643035888671875, 148.50363159179688, 20.6072998046875, 54.05784606933594, 19.837615966796875, -4.811042785644531, 24.194122314453125, 170.02061462402344, 50.58525848388672, -18.129234313964844, 44.86571502685547, -1.075408935546875, -49.27568054199219, 121.37262725830078, 10.4951171875, 18.162452697753906, 61.84478759765625, 177.12338256835938, 31.2728271484375, 5.114227294921875, 5.569488525390625, -25.94635009765625, 168.02325439453125, 24.517486572265625, -23.129058837890625, 29.062088012695312, -12.20709228515625, 152.2603759765625, -34.339561462402344, 191.326904296875, -183.35641479492188, 5.69610595703125, 174.22601318359375, 154.32896423339844, 32.93223571777344, 14.43450927734375, 86.50242614746094, 52.16448974609375, 146.9362030029297, -12.998825073242188, -137.3482666015625, 99.49716186523438, 142.53480529785156, 0.79559326171875, 222.97882080078125, 96.645263671875, -127.42469787597656, 23.222946166992188, 57.096527099609375, 19.71478271484375, 128.888916015625, 11.429931640625, -2.604583740234375, 75.41482543945312, -27.762710571289062, 15.654083251953125, -4.8972625732421875, 63.20697021484375, 80.89620971679688, -4.0472259521484375, -16.92626953125, 35.26555633544922, -58.4678955078125, 175.26510620117188, 15.725433349609375, -4.379579544067383, -60.97784423828125, 14.86376953125, 22.205810546875, 224.134765625, 2.2397308349609375, 4.943071365356445, -29.685531616210938, -16.856964111328125, -174.80853271484375, 18.007545471191406, 97.87069702148438, 116.12933349609375, 11.9920654296875, -156.08016967773438, 144.6412811279297, 221.46820068359375, -13.963272094726562, 9.644111633300781, 28.1220703125, -24.917755126953125, 117.17153930664062, 21.811767578125, 128.7471923828125, 9.48883056640625, 212.12738037109375, -32.656063079833984, 8.6160888671875, -53.898193359375, 48.255584716796875, -157.87554931640625, 95.13943481445312, -28.12249755859375, 183.88433837890625, 37.20503234863281, -21.54736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000436.npy"}
{"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 67.55149841308594, "std": 96.9585952758789, "min": -167.19503784179688, "p10": -23.187962341308587, "median": 55.33978271484375, "p90": 184.745654296875, "max": 253.28143310546875, "pos_frac": 0.7421875, "sample": [25.6671142578125, 165.70797729492188, 44.6673583984375, 152.3162841796875, -166.6634521484375, 224.328369140625, 4.085205078125, 7.9154815673828125, 80.3623046875, 15.09375, 179.01934814453125, 19.895858764648438, 138.0085906982422, 71.39242553710938, 12.317459106445312, 159.41464233398438, -19.204864501953125, 4.199310302734375, 45.40643310546875, 156.45938110351562, 38.222198486328125, 193.29220581054688, 253.28143310546875, 172.46107482910156, 11.0877685546875, 29.13458251953125, -2.75238037109375, 149.44424438476562, 156.02944946289062, 101.90255737304688, -14.46771240234375, -3.200592041015625, 7.497772216796875, 74.34945678710938, -2.057342529296875, 46.138267517089844, -141.8121337890625, -30.673049926757812, 7.6342620849609375, 30.78570556640625, 151.33555603027344, 62.1739501953125, 174.90695190429688, 166.66897583007812, -43.80712890625, 5.159820556640625, 31.317047119140625, 14.880859375, 191.2525634765625, 155.70635986328125, 142.54164123535156, 208.25123596191406, 132.55142211914062, 171.26580810546875, 8.625480651855469, 32.1845703125, -27.45220947265625, -15.319145202636719, 196.19351196289062, 185.94903564453125, 165.2943115234375, -3.457489013671875, 182.1790771484375, 201.09078979492188, 219.7528076171875, 122.42849731445312, 0.6396751403808594, 65.85296630859375, -16.166061401367188, -21.360427856445312, 63.123985290527344, 10.171051025390625, -39.739044189453125, 182.31910705566406, -167.19503784179688, 162.31414794921875, -3.09078311920166, 101.84181213378906, -163.60147094726562, 0.0, 129.93301391601562, -132.19631958007812, 8.208877563476562, -9.309206008911133, 184.22991943359375, 139.66519165039062, -87.50997924804688, 85.94839477539062, -148.27685546875, 159.81088256835938, 195.95135498046875, 148.63031005859375, -2.7824573516845703, 133.887451171875, 18.85418701171875, -4.95928955078125, 158.599853515625, 25.61847686767578, 84.25534057617188, 119.59317779541016, 186.1019744873047, 194.160888671875, 115.24661254882812, 182.62615966796875, 33.135009765625, -3.8836746215820312, 95.44992065429688, 123.26778411865234, 171.154052734375, -136.60035705566406, 3.4191741943359375, -1.434295654296875, 150.126220703125, -164.43756103515625, -7.779449462890625, 156.87359619140625, 48.505615234375, 158.73536682128906, 136.88150024414062, -1.5836639404296875, 114.42655944824219, -13.816802978515625, 28.76080322265625, 73.78314208984375, 0.0, 13.705780029296875, 143.1587677001953, 234.98519897460938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000437.npy"}
{"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 30.17676544189453, "std": 94.8612060546875, "min": -202.169677734375, "p10": -111.61148300170899, "median": 10.802135944366455, "p90": 162.5090118408203, "max": 236.3629150390625, "pos_frac": 0.6015625, "sample": [19.892601013183594, -173.8731689453125, 122.43209838867188, -10.735855102539062, -13.994033813476562, 147.86895751953125, 50.270843505859375, 38.838409423828125, 0.0, 11.656423568725586, 183.29136657714844, -72.12547302246094, 17.260398864746094, 162.42898559570312, 41.74639892578125, 10.30999755859375, 19.48406982421875, 163.44447326660156, 145.91650390625, -19.762916564941406, -45.18896484375, -0.8621826171875, 5.88238525390625, -1.1720123291015625, -111.5126953125, -111.84198760986328, -22.32855224609375, 124.98562622070312, 9.89862060546875, 34.60650634765625, 10.959733963012695, 187.443115234375, 143.124755859375, -14.025222778320312, 29.2880859375, 162.69573974609375, -152.9599609375, -21.549217224121094, 39.165130615234375, -173.885498046875, -202.169677734375, -137.87240600585938, 100.542236328125, 134.67123413085938, -9.251419067382812, -13.688629150390625, 2.997802734375, 24.84033203125, -161.66799926757812, -3.37762451171875, -1.56182861328125, -145.2841796875, -1.1329574584960938, 16.635818481445312, 9.2509765625, 34.75177001953125, 120.6617431640625, -53.34527587890625, 70.57675170898438, 113.13119506835938, 10.643646240234375, -13.542840957641602, -0.9336395263671875, -153.80435180664062, 103.34417724609375, -8.396293640136719, -20.520965576171875, 13.131591796875, 167.7230224609375, -125.0286865234375, 64.3697509765625, -52.20457458496094, 174.76858520507812, -15.643852233886719, 25.48162841796875, 84.4195556640625, 133.80458068847656, -113.68180847167969, 159.43307495117188, 159.00265502929688, -27.551971435546875, 154.89186096191406, -78.1263656616211, -3.3597412109375, 204.22979736328125, -5.8452606201171875, 145.55113220214844, 75.12985229492188, -112.765380859375, 6.54083251953125, 115.24115753173828, 16.67034149169922, 2.150087356567383, -9.723419189453125, 178.61236572265625, 147.84970092773438, 7.480064392089844, 171.48193359375, 229.96038818359375, -27.945709228515625, -8.78472900390625, -34.982940673828125, 37.23004150390625, 121.8546142578125, 140.2559814453125, -30.3685302734375, 17.864852905273438, 30.92237091064453, -30.586456298828125, 151.64007568359375, 182.09017944335938, 10.644537925720215, -60.21356201171875, 21.888317108154297, 0.0, 1.5487060546875, 132.5491485595703, 9.753562927246094, -27.315895080566406, 76.37068176269531, 178.27093505859375, -124.76385498046875, 9.159191131591797, 160.5570068359375, 236.3629150390625, -3.3798446655273438, 15.41680908203125, 30.023452758789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000438.npy"}
{"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 54.19435119628906, "std": 92.64398956298828, "min": -211.0230712890625, "p10": -46.771522521972656, "median": 27.17554473876953, "p90": 175.40992736816406, "max": 223.60064697265625, "pos_frac": 0.703125, "sample": [-111.1883544921875, -111.79022216796875, -153.99142456054688, 206.17691040039062, -57.92449951171875, 172.28280639648438, 16.679946899414062, 17.78162384033203, 64.69300842285156, 162.6660614013672, -3.2292709350585938, 16.299652099609375, 3.596731185913086, 4.31878662109375, 88.44917297363281, -143.33059692382812, 9.8984375, 121.26223754882812, -1.7966461181640625, 72.03787231445312, 192.9110107421875, 158.81973266601562, 10.962860107421875, 26.541748046875, 4.0467529296875, 154.26296997070312, 62.857269287109375, 154.51486206054688, 178.91708374023438, 18.225229263305664, -111.09516906738281, 121.13137817382812, 13.137657165527344, 206.030029296875, -21.0736083984375, 125.33282470703125, 206.33389282226562, 57.178985595703125, 223.60064697265625, 31.329940795898438, -33.23638916015625, 18.878780364990234, 158.06661987304688, 70.70307922363281, 4.9610595703125, 132.2727508544922, 150.35562133789062, 10.72332763671875, 4.2111358642578125, 6.027435302734375, -1.20294189453125, 16.9461669921875, -83.52365112304688, 14.612548828125, 35.986114501953125, 148.9034423828125, 144.63832092285156, -26.199378967285156, 89.31098937988281, 0.0, 27.809341430664062, 0.0, -34.536956787109375, 11.926513671875, -40.4378662109375, 172.03536987304688, 24.7208251953125, -46.54266357421875, 11.008316040039062, -36.641387939453125, 2.2193603515625, 173.9068603515625, 25.207901000976562, -24.753326416015625, -7.021278381347656, 58.740333557128906, -4.534137725830078, 116.23265838623047, -42.04486083984375, 47.985137939453125, -35.69378662109375, 115.04810333251953, 155.33006286621094, 26.038818359375, 52.67045593261719, 114.55390930175781, -56.20085144042969, -4.43194580078125, -89.49091339111328, -6.002838134765625, 172.51766967773438, -144.57763671875, 0.0, 181.42550659179688, 198.6295166015625, 83.42865753173828, 196.10494995117188, 157.39578247070312, 41.53985595703125, -47.30552673339844, 47.2474365234375, -78.50503540039062, 144.95970153808594, -3.223388671875, 58.521759033203125, 10.7945556640625, 113.50497436523438, 136.49459838867188, 140.0400390625, -8.81060791015625, 170.50953674316406, 74.8272705078125, 75.56044006347656, 171.18359375, 155.30169677734375, -0.8853740692138672, -211.0230712890625, 194.86386108398438, 117.550537109375, 188.4912109375, 169.89401245117188, 139.41131591796875, 19.75066375732422, -19.7021484375, 191.48365783691406, 0.0, 181.035888671875, 162.04702758789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000439.npy"}
{"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 44.480892181396484, "std": 95.16773986816406, "min": -184.87109375, "p10": -79.10150222778319, "median": 27.121688842773438, "p90": 171.0048919677734, "max": 244.02813720703125, "pos_frac": 0.6875, "sample": [-33.648468017578125, 0.0, 31.03369140625, -149.5228271484375, -18.77027702331543, 81.24246215820312, -158.03933715820312, 117.57803344726562, -10.432472229003906, 100.65544128417969, -21.9361572265625, 25.130859375, -108.37274169921875, 4.102081298828125, 61.47247314453125, -133.07534790039062, 111.30262756347656, 162.05239868164062, 135.40866088867188, 144.18466186523438, 52.50445556640625, 191.65106201171875, -3.5990676879882812, 25.839141845703125, -76.18264770507812, 2.5248260498046875, -184.87109375, 123.78343200683594, 50.193023681640625, 105.34466552734375, 191.02032470703125, 217.55096435546875, 11.081809997558594, -17.245849609375, -16.896484375, 59.98388671875, 111.13101196289062, -161.48040771484375, 0.0, 50.64306640625, -171.20462036132812, -2.28900146484375, 9.1849365234375, 119.13257598876953, 179.09310913085938, -28.201011657714844, -105.98550415039062, 0.66192626953125, 17.271743774414062, 150.51438903808594, 205.21234130859375, -13.513877868652344, 142.36233520507812, 0.19117164611816406, 39.424072265625, 188.45382690429688, -2.9748077392578125, -134.2742462158203, -56.91766357421875, -31.18389892578125, -49.843170166015625, -0.829681396484375, 7.340576171875, -40.27935791015625, 244.02813720703125, 146.6744384765625, 208.63723754882812, 140.53538513183594, 149.71014404296875, 45.737640380859375, 0.1578388214111328, 28.1026611328125, 13.456924438476562, 151.7103729248047, 161.88458251953125, 7.4705047607421875, -1.78851318359375, -75.75283813476562, 17.726348876953125, 16.82098388671875, 47.78877258300781, 128.083251953125, 158.02639770507812, 67.549560546875, 196.4156951904297, -102.29583740234375, 32.169952392578125, 188.881591796875, 38.19970703125, -104.62310028076172, 196.33856201171875, -20.298175811767578, 155.3785400390625, -29.92242431640625, 147.32119750976562, 101.38839721679688, 116.52177429199219, 30.17180633544922, 65.70561218261719, 41.85107421875, 167.53851318359375, -85.91216278076172, 21.32685089111328, 20.635696411132812, 50.51336669921875, 70.5916748046875, 6.72772216796875, 188.78472900390625, 94.57340240478516, 67.97256469726562, 12.978240966796875, 55.496612548828125, 21.867218017578125, 164.7460479736328, -40.03411865234375, 158.28433227539062, 148.25579833984375, 22.751312255859375, 20.832763671875, 116.53805541992188, -38.56158447265625, 183.35504150390625, 26.140716552734375, -100.9677734375, -6.081634521484375, 16.89093017578125, -26.304847717285156, 150.16244506835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000440.npy"}
{"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 66.13367462158203, "std": 86.384521484375, "min": -179.4072723388672, "p10": -14.56371383666992, "median": 51.72760009765625, "p90": 180.2550827026367, "max": 223.49102783203125, "pos_frac": 0.78125, "sample": [-12.418533325195312, 30.2744140625, -134.8486328125, 84.89556884765625, 0.32312583923339844, -13.679244995117188, 191.81256103515625, 1.9200439453125, 4.6208038330078125, -10.2542724609375, -48.10472106933594, 223.49102783203125, 29.42438507080078, 135.6734619140625, 29.414886474609375, 149.27020263671875, 164.56353759765625, 159.20928955078125, 116.16681671142578, 69.48173522949219, 53.694580078125, 139.249267578125, 16.345369338989258, 19.120025634765625, 161.8914794921875, 70.89706420898438, 156.9461669921875, 169.4999542236328, 56.72607421875, 0.69952392578125, 12.799835205078125, 156.663330078125, 90.45408630371094, -14.276496887207031, 151.93362426757812, 121.784423828125, 15.76959228515625, 172.27999877929688, 140.4872283935547, -105.38909149169922, 168.38296508789062, 153.8561248779297, -80.2553939819336, -4.891845703125, 201.514404296875, -1.6237602233886719, 48.385498046875, 39.495941162109375, 148.73855590820312, 155.382080078125, 143.7942657470703, 7.3681640625, 51.88665771484375, 19.28937530517578, 17.83928680419922, 14.68798828125, -1.437255859375, 152.34786987304688, 33.61619567871094, -15.23388671875, -1.44390869140625, 34.190765380859375, 18.42005157470703, 152.03546142578125, -1.2730712890625, 0.0, 3.6230506896972656, 103.25349426269531, 63.998741149902344, 210.21481323242188, -62.4105224609375, -28.94094467163086, 0.19885635375976562, 14.224136352539062, 0.0, 146.8806610107422, 42.2843017578125, 56.5272216796875, 160.318359375, 69.54949951171875, 15.199615478515625, 160.63262939453125, 25.962371826171875, -65.82879638671875, 51.56854248046875, 118.6807861328125, 150.9678192138672, 98.87113952636719, -0.03118896484375, 30.88104248046875, 3.2626953125, 55.5882568359375, 39.029693603515625, 139.19656372070312, 151.28976440429688, 18.09771728515625, 179.6642608642578, -100.63236236572266, 182.219482421875, 54.7176513671875, 181.6336669921875, -179.4072723388672, 60.090545654296875, 3.874053955078125, -101.03802490234375, 8.964187622070312, 205.149658203125, -120.06640625, 164.58311462402344, -12.633514404296875, 30.726585388183594, 187.11465454101562, 184.62802124023438, 183.8104705810547, 195.45913696289062, 82.8134765625, 86.20028686523438, -3.6890869140625, -31.114501953125, 9.931854248046875, 219.115478515625, -9.112716674804688, 154.89556884765625, 162.3885498046875, 184.23251342773438, 134.42527770996094, 143.48507690429688, 35.7052001953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000441.npy"}
{"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 49.97093200683594, "std": 102.0716552734375, "min": -212.4197998046875, "p10": -64.34014892578124, "median": 26.74493408203125, "p90": 178.71946868896484, "max": 257.6688232421875, "pos_frac": 0.7109375, "sample": [147.15347290039062, 15.78094482421875, 20.506271362304688, -3.6895790100097656, 220.003662109375, 173.15380859375, 1.2104835510253906, -8.630523681640625, 179.04595947265625, 4.71575927734375, 104.35067749023438, 24.172317504882812, -91.4389877319336, -9.145339965820312, 159.713134765625, -155.09515380859375, 20.158737182617188, 21.661285400390625, 18.773773193359375, -138.6929473876953, 13.04071044921875, 40.50078201293945, -0.7509307861328125, -3.9093265533447266, 133.44384765625, -6.093963623046875, 139.43521118164062, -75.10504150390625, 200.08200073242188, 185.3876495361328, -153.7890625, 4.3280181884765625, -38.97760009765625, 156.49957275390625, 75.20268249511719, 155.79736328125, -7.104827880859375, 25.022003173828125, -155.3341064453125, 6.94012451171875, 21.168365478515625, -58.60331726074219, 104.83013916015625, 12.911643981933594, -24.095184326171875, -171.976318359375, 16.388870239257812, 38.315765380859375, 10.393463134765625, -48.95953369140625, 14.886260986328125, 0.239349365234375, 107.5604248046875, 218.52532958984375, 7.318693161010742, 108.3800048828125, 143.27273559570312, 0.0, 89.58871459960938, -155.37155151367188, -59.72662353515625, 12.457206726074219, -160.5742645263672, 163.10540771484375, -144.78021240234375, 187.61447143554688, 83.16644287109375, -4.7972564697265625, 28.467864990234375, 139.34381103515625, -0.0804595947265625, -212.4197998046875, 189.4566650390625, -8.452117919921875, 106.13638305664062, 60.312957763671875, 216.2518310546875, -22.25140380859375, 131.2239990234375, 36.81614685058594, 35.586395263671875, 82.398193359375, -6.6699676513671875, 216.11526489257812, 151.32833862304688, 178.5795440673828, -52.18682861328125, -147.9783935546875, -51.74444580078125, -39.455772399902344, 19.598541259765625, 71.02606964111328, 17.86383056640625, 132.5322265625, 17.34942626953125, 145.4684295654297, 141.0142364501953, 148.7715606689453, 135.55088806152344, 161.89776611328125, 192.07562255859375, 183.06719970703125, 165.63934326171875, 154.6943359375, 59.10528564453125, 257.6688232421875, 12.0625, -49.20171356201172, -52.44938659667969, 214.06759643554688, 169.28341674804688, 146.39561462402344, 90.43159484863281, 145.76869201660156, 8.470260620117188, 23.155807495117188, 49.636497497558594, 75.02392578125, 152.51016235351562, -58.0836181640625, 67.08915710449219, 152.13864135742188, 167.27706909179688, 23.32476806640625, -143.48924255371094, 59.98712158203125, 146.25173950195312, 52.965362548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000442.npy"}
{"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 54.126277923583984, "std": 92.55221557617188, "min": -183.62619018554688, "p10": -46.89767379760742, "median": 31.02823257446289, "p90": 176.6766815185547, "max": 236.302001953125, "pos_frac": 0.6875, "sample": [57.318450927734375, 9.379974365234375, 111.79364013671875, 20.451950073242188, -55.66864013671875, 57.60577392578125, 191.0713653564453, 15.757843017578125, 189.8472900390625, -14.451881408691406, 32.67987823486328, 71.21197509765625, -9.872001647949219, 112.4013671875, -12.699249267578125, -129.89442443847656, -17.805267333984375, 124.68984985351562, 175.85427856445312, -5.770763397216797, 44.03404235839844, 82.132568359375, 77.99024963378906, 46.761962890625, -5.73565673828125, 22.812286376953125, 122.35171508789062, 161.5621795654297, 29.3765869140625, -57.90746307373047, -33.46875, 79.47100830078125, -160.365478515625, 6.527332305908203, -8.966827392578125, 34.821380615234375, 170.92926025390625, 17.421672821044922, 167.56338500976562, 11.020374298095703, 177.305908203125, -20.4359130859375, -10.13763427734375, -0.794891357421875, 18.28570556640625, 73.00480651855469, 236.302001953125, 5.4105072021484375, 107.79856872558594, -36.80885314941406, 68.6300048828125, -176.03826904296875, -50.120941162109375, -76.865234375, 98.92001342773438, 150.29087829589844, 100.85005187988281, 229.47903442382812, 87.75927734375, 195.48922729492188, 127.3831787109375, 28.463470458984375, -8.145637512207031, 150.1500244140625, 176.40701293945312, 9.390953063964844, -4.7098388671875, 93.37061309814453, 154.20050048828125, 152.5716552734375, -9.779617309570312, -4.31207275390625, 192.31591796875, 16.869647979736328, 179.27638244628906, -4.3770751953125, -1.0909194946289062, 145.30345153808594, -81.36907958984375, 10.805450439453125, -24.592304229736328, 177.32086181640625, 174.27099609375, 120.38866424560547, 130.0247802734375, -45.516273498535156, 15.350357055664062, -101.03086853027344, 136.9499969482422, 208.8167724609375, 192.1566162109375, 140.3472900390625, 51.99876403808594, 21.006820678710938, -85.79756164550781, 135.05841064453125, 13.094070434570312, 201.32821655273438, 133.65179443359375, 155.68936157226562, 87.452392578125, -39.17431640625, -162.44195556640625, 169.40794372558594, -5.547454833984375, 1.6259765625, 117.34867858886719, 139.933837890625, 61.274322509765625, 8.902297973632812, 139.4756622314453, 12.23394775390625, -37.06646728515625, 14.835556030273438, -2.5758514404296875, -11.205543518066406, 82.90628051757812, 179.8032684326172, -1.4061279296875, -183.62619018554688, 159.7880859375, 155.16122436523438, 158.2984161376953, 4.576744079589844, -3.31378173828125, -121.3997802734375, 11.4395751953125, 7.658233642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000443.npy"}
{"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 56.777381896972656, "std": 95.95818328857422, "min": -206.815673828125, "p10": -41.84227752685547, "median": 40.43410873413086, "p90": 183.94414215087892, "max": 314.99578857421875, "pos_frac": 0.6953125, "sample": [3.5597763061523438, -1.83123779296875, 183.18280029296875, 20.65155029296875, 76.02816772460938, 64.76116943359375, -19.0469970703125, 2.984619140625, 20.67932891845703, 59.22540283203125, -9.190399169921875, 35.92926025390625, -35.545684814453125, 40.41162109375, 151.76156616210938, 30.3973388671875, 140.58139038085938, 59.457618713378906, -27.394195556640625, 40.45659637451172, 105.14616394042969, 29.615570068359375, 133.13824462890625, -41.12078857421875, 19.085205078125, -35.603179931640625, 160.26010131835938, 22.332557678222656, 5.7432708740234375, 48.234710693359375, 17.230972290039062, 158.5274658203125, 110.93946838378906, -17.762130737304688, 183.8529510498047, -29.625396728515625, -10.451370239257812, 53.968536376953125, 169.79632568359375, 150.59442138671875, 164.69528198242188, 210.5880126953125, -15.35369873046875, -4.011030197143555, 40.87579345703125, 151.9034423828125, 47.181884765625, -146.13861083984375, -129.51290893554688, -42.43011474609375, 226.44357299804688, 136.03842163085938, -71.07821655273438, 210.1986083984375, -5.803997039794922, 69.833984375, 234.60787963867188, 183.42578125, -3.133544921875, -13.88555908203125, 1.7126388549804688, 314.99578857421875, 144.22125244140625, -34.74647521972656, 44.56878662109375, -5.275421142578125, 193.80404663085938, 132.80751037597656, -81.09651184082031, 216.66201782226562, 150.6807098388672, 41.51814270019531, -61.84765625, 187.87591552734375, -15.76580810546875, 61.26324462890625, 193.708251953125, 169.66339111328125, 155.80197143554688, 87.39117431640625, 177.23623657226562, -5.439855575561523, 1.2849960327148438, 198.12448120117188, -16.797954559326172, 165.20065307617188, -8.443008422851562, 151.76309204101562, 130.89614868164062, -91.55475616455078, 3.622039794921875, 197.744873046875, -19.025054931640625, 136.69395446777344, 42.61481475830078, 30.042190551757812, 62.079986572265625, -39.169921875, 11.7691650390625, -72.978271484375, -23.678131103515625, 7.01776123046875, 184.53073120117188, 5.726062774658203, -122.40765380859375, 44.065185546875, -65.63139343261719, 159.45468139648438, -41.59034729003906, 141.67359924316406, 6.365419387817383, 77.80303955078125, 9.435302734375, 124.13778686523438, 105.42007446289062, 32.57085418701172, 184.15692138671875, 143.93064880371094, 152.15121459960938, 140.60366821289062, -3.1546993255615234, 12.92806625366211, -57.1754150390625, 1.1922492980957031, -178.62371826171875, 25.82305908203125, 138.5810089111328, -206.815673828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000444.npy"}
{"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 45.23573303222656, "std": 94.24959564208984, "min": -172.30262756347656, "p10": -60.04111328124999, "median": 20.990142822265625, "p90": 177.54124755859374, "max": 228.98348999023438, "pos_frac": 0.65625, "sample": [9.994354248046875, -172.30262756347656, 194.12286376953125, 82.22439575195312, 45.095977783203125, 196.3885498046875, -5.6262664794921875, 39.500389099121094, 177.01727294921875, -105.2745361328125, 10.129806518554688, 98.09214782714844, 160.2143096923828, -65.25723266601562, -35.15350341796875, 3.983856201171875, 25.993682861328125, 13.443511962890625, 208.13961791992188, 17.39476776123047, 182.33438110351562, -9.4534912109375, 5.08099365234375, -123.94534301757812, 20.878021240234375, 139.69903564453125, 34.26060485839844, -2.9395103454589844, 166.45074462890625, 25.249343872070312, 205.64312744140625, 85.68350219726562, 42.092041015625, -1.4570541381835938, -25.479644775390625, -12.838760375976562, 7.55339241027832, 47.09626388549805, 124.53507232666016, 90.0106201171875, 9.7684326171875, 179.59234619140625, 201.85003662109375, -1.79571533203125, 138.08338928222656, 3.4603443145751953, -1.4208831787109375, 29.0281982421875, 65.33158874511719, -136.13397216796875, -37.942665100097656, 92.29193115234375, 169.2662353515625, -65.810791015625, 146.29238891601562, 0.445556640625, 48.74041748046875, 183.30029296875, 20.25482177734375, 173.6603240966797, 116.211181640625, 62.94281005859375, 228.98348999023438, 3.335418701171875, 171.08526611328125, 136.396240234375, 126.48260498046875, -160.0589141845703, 52.39447021484375, 202.8455810546875, 28.6988525390625, 58.420257568359375, -24.531219482421875, 141.76512145996094, 145.44073486328125, 5.953826904296875, -18.045318603515625, -151.58209228515625, 167.70550537109375, 139.98095703125, -1.06243896484375, -63.382049560546875, -45.85551452636719, -135.9302978515625, 4.990699768066406, 26.533203125, 17.367294311523438, -33.238677978515625, 114.27156829833984, -0.6504135131835938, 87.675537109375, 152.01522827148438, 195.6785888671875, 156.02520751953125, 41.66424560546875, 145.12353515625, -3.176614761352539, -148.27871704101562, 178.76385498046875, -2.826171875, 3.7066650390625, 67.07696533203125, 173.69110107421875, 21.102264404296875, -7.595703125, 21.658493041992188, 8.709541320800781, 19.17132568359375, 179.47698974609375, -1.22052001953125, -45.75518798828125, -12.722610473632812, 0.0, -11.667484283447266, 14.083686828613281, -58.609283447265625, -5.64111328125, -6.27789306640625, -5.0304718017578125, 172.81442260742188, -58.01544189453125, -50.30908203125, -167.61944580078125, -1.7819538116455078, 59.1976318359375, 167.83267211914062, -80.62023162841797, 157.5491180419922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000445.npy"}
{"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 64.60104370117188, "std": 95.87442016601562, "min": -213.35491943359375, "p10": -37.68828582763672, "median": 48.32743835449219, "p90": 182.75007019042968, "max": 244.83969116210938, "pos_frac": 0.7578125, "sample": [134.68020629882812, 225.82403564453125, 10.677192687988281, 4.4630126953125, 1.039703369140625, 42.300498962402344, 146.16354370117188, 36.12938690185547, 164.2154998779297, -16.991561889648438, 8.8798828125, -66.67146301269531, 123.98062133789062, 64.90325164794922, -19.12432861328125, 31.81671905517578, 213.36422729492188, 22.299781799316406, -18.796554565429688, 46.478759765625, 13.550704956054688, 135.03314208984375, 179.14788818359375, 157.40121459960938, -8.593597412109375, -49.085601806640625, 242.81106567382812, 4.4854583740234375, 10.934455871582031, -83.015869140625, 11.523040771484375, 12.491928100585938, -26.56793212890625, 38.820152282714844, -59.714202880859375, 37.19407653808594, 172.94915771484375, 3.3024215698242188, 16.441497802734375, 52.52947235107422, 183.14999389648438, 3.8065719604492188, 106.09131622314453, 18.850326538085938, 48.603179931640625, 149.79934692382812, 182.57867431640625, 145.5118408203125, 174.464599609375, -213.35491943359375, 127.88198852539062, 130.33505249023438, 209.76995849609375, 3.5928077697753906, 141.7433319091797, 162.73265075683594, 11.703369140625, 48.05169677734375, -202.07440185546875, 110.68783569335938, 175.03671264648438, -1.3686027526855469, -39.48661804199219, 0.0, 54.84796142578125, 169.9056396484375, 37.411376953125, 114.16936492919922, 152.560302734375, 152.7321319580078, -14.25518798828125, 161.8757781982422, 129.27203369140625, 168.083251953125, 158.83074951171875, 12.028106689453125, 170.65350341796875, 141.7110595703125, -53.642547607421875, 0.0, 78.94659423828125, -89.2142333984375, 209.06893920898438, -10.30792236328125, 28.513015747070312, -92.83251953125, 111.74127197265625, 213.02593994140625, 133.7919921875, 85.889892578125, -31.978897094726562, 244.83969116210938, 205.67431640625, 170.47201538085938, 133.54879760742188, 71.39459228515625, 109.30368041992188, 147.07382202148438, -23.30673599243164, 90.66168212890625, 20.166263580322266, -11.363128662109375, 197.72479248046875, 203.58050537109375, -199.02438354492188, -146.4193572998047, -13.595108032226562, 16.34515380859375, -12.091064453125, 5.95245361328125, 50.0036506652832, 0.0, -36.917572021484375, 1.63189697265625, 147.1160888671875, -47.68963623046875, 196.68658447265625, 83.01309204101562, 217.93014526367188, 15.439895629882812, 5.2244873046875, 57.34703063964844, 19.365127563476562, -12.384651184082031, 167.77606201171875, 80.66627502441406, 143.80686950683594, 80.77450561523438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000446.npy"}
{"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 68.24501037597656, "std": 86.7730712890625, "min": -189.84716796875, "p10": -26.289457702636717, "median": 54.00117492675781, "p90": 176.22299499511718, "max": 227.52081298828125, "pos_frac": 0.7734375, "sample": [-56.79020690917969, 10.946208953857422, 148.1617431640625, 149.13059997558594, 33.67912292480469, 9.160724639892578, 195.72958374023438, 11.048797607421875, -34.60693359375, 162.3765869140625, 52.5469970703125, 145.68882751464844, 40.555084228515625, 181.1514892578125, 175.84710693359375, 81.50477600097656, 0.0, -5.66473388671875, 188.32138061523438, 10.958354949951172, -5.5841827392578125, 36.0977783203125, 157.3182373046875, 24.146575927734375, 110.4207763671875, -11.05743408203125, 2.663482666015625, 147.06068420410156, -141.31256103515625, 143.29971313476562, -125.00550842285156, 24.1572265625, 140.76695251464844, 118.40931701660156, -18.54405975341797, 29.71082305908203, 170.48033142089844, 156.4340057373047, 27.615028381347656, 143.0993194580078, 3.2359848022460938, 161.44070434570312, 181.8943328857422, 179.2476806640625, 44.780853271484375, 0.4041900634765625, 157.31649780273438, 158.0435791015625, 174.6171112060547, 0.0, 170.36285400390625, 20.300384521484375, 63.60980224609375, 137.98318481445312, -34.21833801269531, -39.037139892578125, -8.10174560546875, 11.3428955078125, -117.97698211669922, 170.25868225097656, 11.1385498046875, 54.78363800048828, 0.520172119140625, 69.14877319335938, -19.946365356445312, 209.589599609375, 135.70086669921875, 2.855224609375, -40.81404113769531, 16.096145629882812, 70.64250183105469, -32.38043212890625, -26.174728393554688, 7.6578369140625, 83.156005859375, 0.9510650634765625, 104.96562194824219, -35.532196044921875, 81.20599365234375, 177.43661499023438, 227.52081298828125, -2.33087158203125, -4.12019157409668, 104.84246826171875, -189.84716796875, 0.0, 7.483512878417969, 144.95755004882812, 54.6309814453125, 177.10006713867188, 53.371368408203125, 39.077117919921875, 29.97686767578125, 138.79342651367188, 212.89382934570312, 145.29257202148438, 162.56597900390625, 227.27996826171875, 0.0, 172.32728576660156, 64.67254638671875, 141.68753051757812, 27.0361328125, 21.5927734375, 35.67755126953125, 79.18984985351562, 30.4810791015625, -129.46957397460938, 167.4936981201172, 76.25997924804688, -11.70355224609375, 143.41500854492188, 37.220611572265625, 183.88262939453125, 158.6064453125, 141.35610961914062, 15.689971923828125, 140.78443908691406, -8.367507934570312, -8.497782707214355, 61.253173828125, 175.4339599609375, 152.88510131835938, 91.42274475097656, 20.143157958984375, 189.0941162109375, -26.557159423828125, 126.4356460571289], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000447.npy"}
{"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 59.0895881652832, "std": 91.56965637207031, "min": -165.05474853515625, "p10": -30.521873474121094, "median": 39.63543701171875, "p90": 176.8189437866211, "max": 248.64320373535156, "pos_frac": 0.75, "sample": [18.90478515625, 9.69293212890625, -117.89053344726562, 14.934417724609375, 80.10202026367188, 185.35562133789062, 7.118927001953125, -165.05474853515625, 116.96817016601562, 154.24310302734375, -4.28948974609375, -139.69403076171875, 113.85055541992188, 39.535919189453125, 1.39599609375, 102.57626342773438, 143.33432006835938, 166.28347778320312, 85.09555053710938, -7.075090408325195, 71.86209869384766, 181.80459594726562, 248.64320373535156, 5.937103271484375, 91.01361083984375, 29.48168182373047, 32.61216735839844, -34.24392318725586, 17.72216796875, 8.86273193359375, -13.614387512207031, 39.734954833984375, 191.5303955078125, -107.1041259765625, 145.65664672851562, -43.58917236328125, 160.19284057617188, 177.08502197265625, 46.0782470703125, 89.736572265625, 99.25845336914062, 25.6278076171875, 187.40545654296875, -0.0186920166015625, 80.03385925292969, 5.997926712036133, 92.31948852539062, 137.33135986328125, -4.4490814208984375, 212.2509765625, -11.845962524414062, -3.25677490234375, -9.996917724609375, 99.22520446777344, 2.4732742309570312, -15.265411376953125, 43.94306945800781, 129.60772705078125, -89.64468383789062, 186.55078125, 144.7162322998047, 67.05548095703125, 13.8233642578125, 0.19327926635742188, 171.81106567382812, 212.46261596679688, 17.26803970336914, 13.453208923339844, 162.51150512695312, 204.666748046875, 165.06906127929688, 128.4916534423828, 126.0165786743164, 32.53466796875, 12.566886901855469, 127.7418212890625, 180.56732177734375, 132.40476989746094, 156.7978973388672, 93.83853149414062, -26.44732666015625, 184.45993041992188, -0.9968643188476562, 163.9880828857422, 19.597335815429688, -154.3326873779297, -145.08383178710938, -156.568115234375, 162.96295166015625, 14.7430419921875, -12.009490966796875, -1.7630462646484375, 164.37966918945312, 167.2075958251953, 21.01300048828125, -140.01461791992188, 105.46881103515625, -1.7766284942626953, 29.45208740234375, 153.87265014648438, 156.95419311523438, 15.909257888793945, 176.7049102783203, 8.314952850341797, 137.54342651367188, -27.86566162109375, -48.1146240234375, 88.22987365722656, 51.227203369140625, -30.400131225585938, 168.25733947753906, -30.805938720703125, -17.544769287109375, 5.0283203125, -28.050981521606445, 22.977294921875, 8.4033203125, 17.39324951171875, 55.197601318359375, 1.3680610656738281, 130.58303833007812, -11.2186279296875, 8.903827667236328, 168.68984985351562, 227.3155517578125, 145.10096740722656, 49.4228515625, 89.46128845214844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000448.npy"}
{"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 67.56897735595703, "std": 92.10833740234375, "min": -191.53350830078125, "p10": -26.79997062683105, "median": 60.672203063964844, "p90": 185.17783203125, "max": 221.21044921875, "pos_frac": 0.75, "sample": [170.00009155273438, 101.84515380859375, -37.75312805175781, 129.36146545410156, 58.084869384765625, -15.6273193359375, 10.542938232421875, 199.53143310546875, -5.397773742675781, 29.046875, 15.184295654296875, 36.0133056640625, 127.9344482421875, 38.658485412597656, -9.126129150390625, 145.05148315429688, 202.17672729492188, 180.01748657226562, 16.7415771484375, -2.5243616104125977, -127.40336608886719, 139.2512969970703, 196.1947021484375, 112.88619232177734, 58.1412353515625, 116.01600646972656, -47.66664123535156, 60.82965087890625, 6.3758649826049805, 185.9371337890625, 203.7874755859375, 181.71505737304688, 162.07672119140625, 192.42803955078125, 25.8809814453125, 168.97894287109375, 17.789794921875, 91.42153930664062, 156.14468383789062, 124.69107055664062, 179.17721557617188, 35.8428955078125, 138.5140380859375, 15.079124450683594, 221.21044921875, 212.12936401367188, 0.494781494140625, -3.759286880493164, 194.92431640625, 159.92721557617188, -43.28722381591797, 80.56784057617188, 13.500839233398438, 30.218475341796875, 6.540510177612305, 151.77891540527344, 18.828643798828125, 137.7820587158203, 4.845458984375, 91.59434509277344, 22.245254516601562, 65.18319702148438, 166.4983673095703, 19.919464111328125, 114.1518783569336, -4.3193206787109375, 44.500579833984375, 62.533416748046875, -8.36236572265625, 164.59970092773438, -80.3245849609375, 85.40888977050781, 168.20010375976562, 156.54412841796875, -191.53350830078125, -1.2269096374511719, 1.7676467895507812, 153.43994140625, 92.526611328125, 166.1932373046875, 152.35650634765625, 104.97325134277344, 216.07565307617188, 60.51475524902344, 124.69348907470703, 118.23086547851562, 166.03482055664062, 38.88323974609375, 135.12545776367188, -29.036991119384766, -2.8818740844726562, 36.96221923828125, -5.845245361328125, 195.6304931640625, 180.58245849609375, 184.8524169921875, -128.0900115966797, -9.376337051391602, 127.69424438476562, 61.0115966796875, -37.954437255859375, 146.11846923828125, -20.11939239501953, -22.98223876953125, 89.0498046875, -25.84124755859375, -127.31631469726562, -10.604927062988281, 155.9833984375, 22.03857421875, 6.399559020996094, 146.32666015625, -18.408172607421875, -3.2205467224121094, 41.06549072265625, -4.3423004150390625, 156.60067749023438, 67.59581756591797, 1.8154754638671875, 189.4248046875, 37.840972900390625, -166.38124084472656, 191.285400390625, 16.56745147705078, 104.0511474609375, -167.2069549560547, -99.3997802734375, -7.042022705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000449.npy"}
{"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 48.72777557373047, "std": 91.16922760009766, "min": -174.68331909179688, "p10": -37.52312316894529, "median": 26.412479400634766, "p90": 176.03271484375, "max": 217.99484252929688, "pos_frac": 0.6875, "sample": [0.0, -128.594970703125, 155.70628356933594, -12.076454162597656, 73.07550048828125, -7.2059478759765625, 156.5182647705078, -15.66314697265625, 180.2935791015625, 48.78460693359375, 125.7589111328125, 192.79193115234375, -1.46929931640625, 60.843902587890625, 161.1131591796875, 11.031784057617188, -11.150894165039062, 129.89422607421875, 16.786514282226562, 20.149808883666992, -86.29940795898438, 102.19108581542969, 193.48614501953125, 170.22508239746094, 134.02743530273438, 25.872314453125, -15.426872253417969, 16.570037841796875, 86.57051086425781, 32.021881103515625, 13.614067077636719, -6.3943328857421875, -10.28912353515625, 2.052459716796875, 59.19316101074219, 14.65966796875, 211.03143310546875, -147.91476440429688, 120.51583862304688, -93.16766357421875, -1.9853477478027344, 182.7454833984375, 19.224727630615234, 110.35357666015625, 16.313522338867188, 24.25151824951172, -25.445188522338867, 79.802001953125, 194.92193603515625, 50.32585144042969, -1.52069091796875, 168.96096801757812, 17.71361541748047, 10.049522399902344, 51.971435546875, -3.924774169921875, 38.67436218261719, 193.52171325683594, 5.969879150390625, -143.90252685546875, 60.55645751953125, -174.68331909179688, 157.82257080078125, 42.5264892578125, -51.10205078125, 187.415771484375, 155.4481658935547, 3.472585678100586, 117.45938110351562, 109.88955688476562, -2.0340576171875, 130.41099548339844, -1.192413330078125, 175.03515625, 6.23394775390625, 26.95264434814453, 188.65423583984375, 155.22021484375, -129.94093322753906, 16.748565673828125, 112.49295043945312, 44.91668701171875, -20.045608520507812, -0.449127197265625, 187.78271484375, 33.88507080078125, -1.0062217712402344, 17.8912353515625, 22.70172882080078, -23.3822021484375, 10.8822021484375, 107.26304626464844, 178.3603515625, 217.99484252929688, 29.463958740234375, 166.84872436523438, -31.574493408203125, -123.4605712890625, -31.703582763671875, -138.3924560546875, 15.643375396728516, 105.047607421875, 123.88037109375, 154.7499237060547, 200.19061279296875, 48.253936767578125, -121.2303466796875, 66.66934204101562, 7.7352294921875, 164.2302703857422, 171.55455017089844, 10.0435791015625, 38.5291748046875, -18.73065185546875, 169.62490844726562, -124.16947174072266, 0.1836700439453125, -14.6324462890625, 43.152557373046875, 127.15676879882812, 0.0, -14.615470886230469, 112.27920532226562, 53.963783264160156, -10.059539794921875, 159.4340362548828, -22.47979736328125, -111.7615966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000450.npy"}
{"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 58.11521911621094, "std": 92.8734359741211, "min": -194.8868408203125, "p10": -38.43915328979491, "median": 39.380245208740234, "p90": 173.68260803222657, "max": 243.57357788085938, "pos_frac": 0.75, "sample": [58.581451416015625, 39.820106506347656, -24.143646240234375, 8.686199188232422, 204.2728729248047, 158.75363159179688, 38.94038391113281, 2.2910003662109375, -194.8868408203125, -27.54034423828125, 136.18585205078125, 7.741739273071289, 194.74404907226562, 21.6978759765625, 36.07763671875, 117.4024658203125, 94.18048095703125, 9.696563720703125, 187.36538696289062, -67.49066162109375, 164.583984375, -7.0760498046875, 15.12957763671875, 66.9521484375, 210.53744506835938, -0.5609130859375, 243.57357788085938, 13.9990234375, -12.972640991210938, 35.066070556640625, -65.31044006347656, 6.338996887207031, -175.52146911621094, 188.05218505859375, 97.62496948242188, 42.29798889160156, 24.16181182861328, 116.61825561523438, 36.29248046875, 145.14749145507812, 16.926288604736328, 157.37225341796875, 134.67694091796875, 147.59466552734375, 143.12527465820312, 171.23394775390625, 15.150497436523438, -0.0945281982421875, 199.27020263671875, -1.38067626953125, 5.2381591796875, 201.3134765625, 22.881576538085938, 213.61911010742188, 153.84548950195312, 158.71047973632812, -35.25147247314453, 59.70892333984375, -178.64859008789062, 150.57994079589844, 146.4593505859375, 142.76339721679688, 27.331039428710938, 168.0635986328125, 3.0633392333984375, 40.9100341796875, -125.44551849365234, 1.9725189208984375, 87.7583236694336, 124.7333984375, 1.19122314453125, 10.07745361328125, -34.769134521484375, 165.1114501953125, 9.853240966796875, -48.976104736328125, 77.13255310058594, 10.367828369140625, 164.1165771484375, 94.51522827148438, -0.74444580078125, -3.3666229248046875, -26.69573974609375, 58.32427215576172, 175.83700561523438, -23.890350341796875, 29.791610717773438, 102.593017578125, 17.269760131835938, 16.52887725830078, 134.30905151367188, 130.06326293945312, -25.7515869140625, 170.55731201171875, 33.20750427246094, -170.27639770507812, 65.06363677978516, -45.8770751953125, -31.277496337890625, 173.654052734375, 49.236572265625, 165.96450805664062, 74.61386108398438, 173.74923706054688, -10.274993896484375, -100.24800109863281, -109.3394775390625, -66.867431640625, 111.84063720703125, 52.809234619140625, 155.98165893554688, 31.105087280273438, 178.20590209960938, 168.9937744140625, -3.7112655639648438, 17.061161041259766, 105.8985595703125, 3.423675537109375, 101.73605346679688, 243.0548095703125, 49.374935150146484, -3.1038360595703125, -23.325729370117188, 2.6540756225585938, 140.95335388183594, 106.49600219726562, -54.333221435546875, 148.06698608398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000451.npy"}
{"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 55.82073974609375, "std": 88.11543273925781, "min": -165.18515014648438, "p10": -34.19954948425293, "median": 31.84935760498047, "p90": 180.9626037597656, "max": 235.0968017578125, "pos_frac": 0.75, "sample": [-72.4466552734375, 147.7152099609375, 43.72979736328125, 198.0146484375, -137.76580810546875, 9.666099548339844, 12.340644836425781, 188.4412841796875, -11.619293212890625, -25.352561950683594, 184.18878173828125, 0.9866943359375, 5.7532196044921875, 8.958305358886719, -2.300518035888672, 10.5494384765625, 65.36029052734375, -5.655059814453125, -5.595703125, 15.907470703125, -0.12291908264160156, -88.09669494628906, 114.67840576171875, 158.24484252929688, 145.28274536132812, 55.431854248046875, 40.94390869140625, 37.66682434082031, 0.71234130859375, -67.00077819824219, -114.13272094726562, 201.15670776367188, -35.86850357055664, 204.02688598632812, 125.08238220214844, -16.36865234375, 184.22418212890625, 175.27264404296875, -16.55841064453125, 146.18374633789062, 96.5813980102539, 14.599365234375, 7.114723205566406, 42.47003173828125, 146.82598876953125, 13.570993423461914, 32.62499237060547, 135.1498565673828, 105.22415161132812, 24.886314392089844, 172.3741455078125, 36.568115234375, 134.474609375, -10.7615966796875, 11.569168090820312, -50.60858154296875, 13.569656372070312, 17.796905517578125, 16.077499389648438, 21.1611328125, 11.062782287597656, 42.31304931640625, -61.06536865234375, 102.81997680664062, -65.72412109375, 152.82591247558594, 132.6224365234375, 174.80325317382812, 17.819580078125, 14.214111328125, 10.7847900390625, 180.59042358398438, 31.07372283935547, 12.9229736328125, 174.98953247070312, 235.0968017578125, 160.23316955566406, -17.53900146484375, 46.4039306640625, 129.93551635742188, -150.88275146484375, 111.84466552734375, 187.4171905517578, -3.59423828125, 181.83102416992188, 155.54029846191406, -5.688591003417969, 150.8937530517578, 190.5662841796875, 28.780731201171875, 43.01936340332031, 49.9490966796875, 20.169940948486328, -1.3474578857421875, 29.615684509277344, -44.00862121582031, 156.25778198242188, 77.92646789550781, 60.537139892578125, -19.72894287109375, 218.91140747070312, 55.08966064453125, 24.154327392578125, 152.15325927734375, 58.077659606933594, 1.5765705108642578, -27.6441650390625, 20.270431518554688, 175.46115112304688, 191.75621032714844, 220.63873291015625, 179.5245361328125, 24.898529052734375, 38.44268798828125, 170.24737548828125, 0.0, 59.38859558105469, -1.9841079711914062, -136.75625610351562, -33.484283447265625, 41.61328125, 71.597900390625, -165.18515014648438, 53.72686767578125, 3.142486572265625, 174.0344696044922, 7.405792236328125, -0.1895008087158203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000452.npy"}
{"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 63.530242919921875, "std": 87.26377868652344, "min": -182.3031005859375, "p10": -25.84448547363281, "median": 49.808624267578125, "p90": 177.20116577148437, "max": 244.75921630859375, "pos_frac": 0.7890625, "sample": [35.317047119140625, 151.2789306640625, 10.804656982421875, 123.95635986328125, 148.09820556640625, 5.6279754638671875, 20.633056640625, -3.5009918212890625, 156.6371307373047, 102.50993347167969, 76.5113296508789, 59.46958923339844, 40.8765869140625, 55.97467041015625, -19.58514404296875, -52.032318115234375, 42.96038818359375, -182.3031005859375, 76.98330688476562, 183.2794189453125, -23.710418701171875, 173.1876220703125, 172.04428100585938, 153.4554901123047, 57.7628173828125, 7.73504638671875, -44.532989501953125, 167.3084716796875, 47.621307373046875, 166.1200714111328, 24.531036376953125, 142.07852172851562, 40.51347351074219, 142.4890594482422, 34.676483154296875, 29.8465576171875, 38.0244140625, -4.23797607421875, 188.48353576660156, 190.26959228515625, 7.164520263671875, 176.10101318359375, 35.03692626953125, 81.27850341796875, 137.57601928710938, -134.12399291992188, 15.037353515625, 17.331085205078125, 13.61260986328125, -51.45916748046875, 3.345916748046875, 157.07943725585938, 5.6813201904296875, 20.543563842773438, -169.33929443359375, 127.33126831054688, -5.306243896484375, -47.799163818359375, 51.407470703125, 7.512237548828125, 135.5009307861328, -8.198692321777344, 0.0, 137.93772888183594, 90.76411437988281, 21.09869384765625, 125.92733001708984, 13.14459228515625, 165.419677734375, 27.26239013671875, 90.36740112304688, 78.23455810546875, 53.59942626953125, -159.98471069335938, 22.110626220703125, 179.7681884765625, 195.82861328125, 169.82252502441406, -2.892780303955078, 158.70130920410156, -11.271591186523438, 196.28012084960938, 93.78768920898438, -10.40533447265625, 9.158233642578125, 19.8416748046875, 189.86062622070312, -17.715576171875, -29.4644775390625, 190.82568359375, 46.35394287109375, 96.0546875, -50.304107666015625, -36.15167236328125, 171.51422119140625, 21.148651123046875, -155.84030151367188, 163.63002014160156, -8.937530517578125, 244.75921630859375, 77.26343536376953, 50.6429443359375, 76.02839660644531, 78.22232055664062, 27.410064697265625, 65.58392333984375, 0.7679595947265625, 3.4865341186523438, 180.24557495117188, 68.75407409667969, 165.51348876953125, 135.6129150390625, 26.0289306640625, 158.08151245117188, 48.97430419921875, 113.28956604003906, 185.54452514648438, 188.67437744140625, -8.6710205078125, 8.76861572265625, 161.9833984375, 7.144744873046875, 133.75918579101562, -24.293060302734375, 174.47207641601562, -48.39208984375, 58.20928192138672, 216.0543212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000453.npy"}
{"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 58.397159576416016, "std": 90.00226593017578, "min": -234.87823486328125, "p10": -23.874639892578124, "median": 26.47570037841797, "p90": 185.29647216796874, "max": 232.79953002929688, "pos_frac": 0.7578125, "sample": [55.2279052734375, 174.0263671875, -154.65396118164062, -23.7467041015625, -5.471527099609375, 9.3709716796875, 185.39678955078125, 154.77740478515625, 8.198699951171875, 38.51249694824219, 27.329452514648438, 33.268470764160156, -10.551332473754883, 135.1917724609375, 23.66887664794922, 117.74793243408203, -146.9798583984375, 6.655733108520508, 88.85177612304688, -3.1011714935302734, 136.10202026367188, 27.75267791748047, 22.786376953125, 4.577178955078125, 217.68377685546875, 232.79953002929688, -6.50213623046875, 191.32431030273438, -234.87823486328125, 5.21099853515625, 20.85454559326172, 155.4134063720703, 105.62728118896484, 134.89723205566406, 170.6780242919922, -14.72890853881836, 202.61660766601562, -30.245864868164062, 17.62811279296875, 27.774917602539062, 166.4168701171875, 17.614059448242188, 186.27764892578125, 138.08087158203125, 63.52827453613281, 154.4395751953125, 113.39581298828125, 2.3797607421875, 187.53985595703125, -10.93377685546875, 19.34417724609375, 25.6219482421875, 44.5931396484375, -0.07403564453125, 152.79312133789062, -16.69281005859375, 129.533935546875, 3.9228668212890625, 169.744140625, 185.46484375, 127.13470458984375, 156.21849060058594, 192.29937744140625, 107.0665283203125, 199.0242919921875, 174.80584716796875, 164.72511291503906, 25.1217041015625, 17.72125244140625, -0.927490234375, 192.0865478515625, 164.0402374267578, -2.64166259765625, -61.33778762817383, 12.355865478515625, 34.496795654296875, -22.645362854003906, 73.83206176757812, -0.3316650390625, -107.65826416015625, -21.510055541992188, 2.5569076538085938, 182.32656860351562, -41.7542724609375, -49.788726806640625, 41.72332763671875, 151.6936492919922, 63.00286865234375, -24.17315673828125, 166.13992309570312, 24.092575073242188, 13.874053955078125, 19.3487548828125, 185.25347900390625, 216.34774780273438, 2.4381103515625, 22.167625427246094, 7.663116455078125, 179.9567108154297, 3.453563690185547, -32.08677673339844, 9.599334716796875, 176.22671508789062, 22.644590377807617, 37.58319091796875, 31.69537353515625, 198.50982666015625, 56.817344665527344, -79.21395874023438, 151.90036010742188, 4.190399169921875, 2.82647705078125, 74.35906982421875, 1.7397613525390625, 19.67596435546875, 142.0850067138672, 144.051025390625, -4.3187103271484375, -76.44107055664062, 42.25751495361328, 15.567146301269531, -10.770889282226562, 32.885009765625, 179.13137817382812, -18.35821533203125, -2.70098876953125, 18.151123046875, -87.45118713378906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000454.npy"}
{"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 47.93603515625, "std": 88.79603576660156, "min": -198.65896606445312, "p10": -41.11850204467773, "median": 32.74117851257324, "p90": 166.56744384765625, "max": 230.85751342773438, "pos_frac": 0.703125, "sample": [148.391845703125, 88.10140228271484, 26.171920776367188, -14.504867553710938, 136.82693481445312, 81.27423095703125, 117.2375717163086, 171.65367126464844, 19.892868041992188, -25.42980194091797, -6.815277099609375, 86.69247436523438, -4.63690185546875, 138.7948760986328, 62.48517990112305, 164.4243621826172, 178.10092163085938, 141.38978576660156, 224.45559692382812, -85.981689453125, 8.253631591796875, 39.470672607421875, 10.557907104492188, 156.220947265625, 230.85751342773438, 167.58946228027344, 163.4476318359375, 16.33612060546875, 31.70882797241211, 8.557281494140625, -18.88201904296875, -175.83633422851562, -0.6137676239013672, 27.674423217773438, 34.284271240234375, 7.792694091796875, -11.903213500976562, -24.7607421875, -42.71973419189453, 21.651824951171875, 15.802047729492188, -6.9718017578125, 136.8272705078125, 215.02359008789062, -15.75634765625, 50.108642578125, 154.49537658691406, -4.848175048828125, 167.1737060546875, 28.99810791015625, 4.7751007080078125, 13.18927001953125, 31.3819580078125, -7.945587158203125, -4.6685638427734375, -2.574951171875, 14.260162353515625, -157.22714233398438, -41.63568878173828, -0.7508068084716797, 14.6265869140625, -40.25276184082031, 5.771425247192383, -34.14741516113281, -32.539459228515625, -74.97012329101562, 93.577880859375, 38.762359619140625, 145.9601287841797, 118.3128662109375, -165.18914794921875, 129.912841796875, 171.1514892578125, 166.3076171875, 45.535400390625, -15.820098876953125, -18.85833740234375, 172.86651611328125, 165.880859375, 85.30747985839844, 68.53567504882812, 5.368110656738281, 8.338119506835938, -120.40565490722656, 160.32986450195312, -1.3131942749023438, -167.1614532470703, 71.8764419555664, -4.086578369140625, 38.159358978271484, -19.425140380859375, 35.34857177734375, 1.84661865234375, 40.89726257324219, 15.882705688476562, 133.62142944335938, 215.41546630859375, 216.07647705078125, 118.30825805664062, -40.8968505859375, 4.53936767578125, 6.090187072753906, 125.96826171875, 62.14141845703125, 5.159759521484375, -3.724519729614258, 16.439376831054688, 165.83706665039062, 35.43121337890625, 46.402259826660156, -51.04136657714844, 111.60455322265625, 84.91104125976562, 69.56932067871094, 161.08966064453125, -68.89996337890625, -198.65896606445312, 48.59637451171875, -68.2705078125, 163.139892578125, 147.298828125, 63.50103759765625, 169.96896362304688, 55.4207763671875, 87.691650390625, 49.10650634765625, 175.9460906982422, 33.773529052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000455.npy"}
{"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 70.70343017578125, "std": 101.13316345214844, "min": -188.63711547851562, "p10": -50.25546112060547, "median": 44.93635559082031, "p90": 192.0644790649414, "max": 337.4088134765625, "pos_frac": 0.7265625, "sample": [44.2999267578125, -8.229530334472656, -57.01251220703125, 151.1400909423828, 41.72039794921875, -6.989990234375, 159.1106414794922, -83.55551147460938, 188.35943603515625, 120.39846801757812, 175.31097412109375, -188.63711547851562, 20.02520751953125, -0.72637939453125, -32.885009765625, 200.9249267578125, 137.06588745117188, -50.37950134277344, 7.2001800537109375, 18.56512451171875, 181.1690673828125, 118.5015869140625, -20.8507080078125, 168.86656188964844, 1.1916160583496094, -19.092227935791016, 107.7022705078125, 52.060089111328125, 163.43130493164062, -83.34378051757812, 151.04339599609375, -143.7017822265625, 147.35169982910156, -50.202301025390625, 119.75398254394531, 182.32696533203125, 102.28741455078125, 151.47998046875, 127.46942138671875, 133.41433715820312, 220.85699462890625, -20.407135009765625, 188.5501708984375, 23.766647338867188, 191.23748779296875, 162.39306640625, 201.74288940429688, -5.463623046875, 10.66436767578125, 160.767578125, 40.31927490234375, 121.58602905273438, -8.228363037109375, 33.273155212402344, -53.053436279296875, 17.729278564453125, 4.573638916015625, 22.72199249267578, 23.57830810546875, 209.71905517578125, 16.28704833984375, -6.649345397949219, 17.86078643798828, -122.9824447631836, -25.49871826171875, 187.36068725585938, 152.32508850097656, -147.9196014404297, 28.283782958984375, 20.51416015625, 14.120674133300781, 136.80307006835938, -5.091136932373047, 139.45675659179688, 48.76789093017578, 142.47158813476562, 214.13214111328125, 143.188720703125, -57.378387451171875, 68.21084594726562, 150.5342559814453, 195.2257080078125, -149.56768798828125, 189.456298828125, 7.810028076171875, -7.3123779296875, 21.772705078125, 206.88861083984375, 185.37042236328125, 196.6313018798828, 132.876220703125, 204.19866943359375, -3.286895751953125, 182.12625122070312, 0.0, 3.891529083251953, 127.33499145507812, 180.30502319335938, 141.2108154296875, 129.2451629638672, 15.22393798828125, 12.995147705078125, 77.6749267578125, -0.853179931640625, 107.188720703125, -82.19464111328125, 156.17276000976562, 5.255096435546875, 12.42559814453125, -17.3243408203125, -1.222137451171875, 125.17778015136719, 45.572784423828125, 132.22528076171875, 193.99412536621094, 0.003437042236328125, 270.889892578125, 15.131744384765625, 337.4088134765625, 223.69537353515625, 176.8294677734375, 31.597747802734375, -13.93402099609375, -4.672393798828125, 172.85206604003906, -29.343399047851562, -127.42306518554688, 182.82464599609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000456.npy"}
{"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 63.70589065551758, "std": 94.12210845947266, "min": -202.6751708984375, "p10": -32.47643508911133, "median": 42.84027099609375, "p90": 184.00229492187498, "max": 224.0930633544922, "pos_frac": 0.71875, "sample": [190.41690063476562, 16.329376220703125, -30.053482055664062, -7.00836181640625, 126.00775146484375, 34.50714111328125, 167.3257293701172, 174.94241333007812, 198.8544921875, 15.321990966796875, 166.38082885742188, -0.5306148529052734, 200.72998046875, 124.3687744140625, 5.39599609375, 207.88201904296875, -7.95086669921875, 35.582733154296875, -131.48817443847656, 181.26669311523438, 130.77505493164062, 158.49789428710938, 157.86767578125, 224.0930633544922, 144.61227416992188, 46.41937255859375, 178.43533325195312, 13.580673217773438, 178.42251586914062, 19.47698974609375, 44.19000244140625, -18.4073486328125, -3.469482421875, -33.458038330078125, 151.6153564453125, 174.9346923828125, -32.055747985839844, 41.49053955078125, -64.34353637695312, 116.4742431640625, -1.96063232421875, 77.8944320678711, -31.769485473632812, -31.97479248046875, 5.116939544677734, -202.6751708984375, 194.26309204101562, 23.15057373046875, 188.12380981445312, -11.149932861328125, -46.777099609375, 73.912841796875, 126.58558654785156, -55.21875, 117.27302551269531, 8.602117538452148, -162.6619873046875, 138.65011596679688, 50.78936767578125, 25.62542724609375, -27.017822265625, -20.38494110107422, 24.25933837890625, -26.76988983154297, 40.32200622558594, 68.1231689453125, 137.42237854003906, 27.89171600341797, 221.84999084472656, -12.7703857421875, 7.1049957275390625, 152.89523315429688, 148.03421020507812, 39.831512451171875, 22.383056640625, -153.17555236816406, -48.560699462890625, 39.84918212890625, 55.905738830566406, 99.03938293457031, 24.0484619140625, 177.39706420898438, 145.3562774658203, 101.7828369140625, 118.44790649414062, -27.78375244140625, 131.44924926757812, -1.651123046875, 165.81790161132812, 211.1357421875, 7.869390487670898, -30.973907470703125, 171.884033203125, -121.65130615234375, -49.61968994140625, 173.57470703125, -12.28778076171875, 122.50654602050781, -18.39990234375, 182.23593139648438, 149.40354919433594, 48.49639892578125, 36.79725646972656, 50.61543273925781, -46.67535400390625, 201.440185546875, 30.988311767578125, 127.4817123413086, 13.341354370117188, 109.88810729980469, 189.63656616210938, 120.39122772216797, 3.09674072265625, 188.2232208251953, 175.329833984375, 145.08993530273438, 5.782341003417969, 28.260759353637695, -145.9571533203125, 105.42424011230469, 191.50299072265625, 1.1254940032958984, 0.0, 180.30294799804688, -2.42108154296875, 155.24476623535156, 51.878448486328125, -10.937362670898438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000457.npy"}
{"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 56.09027862548828, "std": 93.1148681640625, "min": -183.668701171875, "p10": -39.49065246582031, "median": 24.006179809570312, "p90": 188.73720397949216, "max": 238.29405212402344, "pos_frac": 0.7578125, "sample": [12.078216552734375, -51.062408447265625, 12.520172119140625, -162.29879760742188, 201.40089416503906, 17.31268310546875, -49.538238525390625, 181.17523193359375, -9.125198364257812, -41.529022216796875, 24.492340087890625, -82.69108581542969, -46.5965576171875, 0.7630157470703125, 2.7865428924560547, 82.30508422851562, 34.384490966796875, 31.4754638671875, -24.64117431640625, 57.319854736328125, 190.93258666992188, -1.5170097351074219, 130.260986328125, -0.9241561889648438, 27.367706298828125, 4.706901550292969, 173.22564697265625, 187.79632568359375, -7.6734619140625, 154.76644897460938, 21.508255004882812, -10.597885131835938, 142.4962615966797, 215.72216796875, 38.059356689453125, 20.935089111328125, 20.590606689453125, 75.70956420898438, 176.91366577148438, -30.126449584960938, 191.159912109375, -3.335315704345703, -148.1405029296875, 23.52001953125, 7.33367919921875, 5.06365966796875, 11.87603759765625, 11.164813995361328, 9.172290802001953, 32.15599822998047, 5.4128570556640625, -66.56700134277344, -17.472213745117188, 171.63128662109375, 150.80340576171875, -46.75286865234375, 3.10028076171875, -1.2316665649414062, 135.25958251953125, 205.60711669921875, 37.1474609375, 2.5939016342163086, 227.50238037109375, 69.70574951171875, 131.0331573486328, 4.007354736328125, 22.264678955078125, -118.26360321044922, -38.6170654296875, 203.56884765625, 169.18243408203125, 57.220947265625, -21.512725830078125, 30.1063232421875, -72.00384521484375, 159.9611053466797, 119.72433471679688, -18.867767333984375, 7.302482604980469, 146.95216369628906, 164.01666259765625, -28.08746337890625, -9.910758972167969, 49.67372131347656, 203.10446166992188, 28.02764892578125, 238.29405212402344, 20.426025390625, 60.23773193359375, 159.95803833007812, 34.329620361328125, 201.21905517578125, 150.37257385253906, 139.1085662841797, 215.88507080078125, -3.9807357788085938, 146.55886840820312, -9.35382080078125, 5.385980606079102, 136.2186279296875, 42.646881103515625, 12.717041015625, 8.42669677734375, 171.09552001953125, 98.13333129882812, 14.320610046386719, 8.916389465332031, 16.81976318359375, 164.78892517089844, 178.5706329345703, 39.338958740234375, 203.05137634277344, 169.6519317626953, 108.41363525390625, 173.15386962890625, 17.4249267578125, 175.7296142578125, 1.0047149658203125, -177.726318359375, -183.668701171875, 141.20123291015625, 192.05963134765625, 20.892242431640625, 5.093475341796875, 141.334716796875, 33.56683349609375, 5.716617584228516, -24.056556701660156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000458.npy"}
{"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 42.02314758300781, "std": 97.81919860839844, "min": -190.47576904296875, "p10": -56.00793380737304, "median": 19.36656951904297, "p90": 181.75391693115233, "max": 236.77850341796875, "pos_frac": 0.6328125, "sample": [3.5175933837890625, -50.51249694824219, 4.870328903198242, 158.00689697265625, -0.9971923828125, 157.0654296875, -148.61013793945312, 43.59051513671875, 30.787307739257812, 30.547134399414062, 63.654571533203125, 125.18891906738281, 11.38357162475586, -17.63916015625, 13.573661804199219, 6.189540863037109, 182.3549346923828, 107.58990478515625, -17.72743034362793, 30.5980224609375, 59.35968017578125, 20.7265625, -83.90899658203125, 31.905242919921875, 29.801666259765625, -13.75146484375, 152.74435424804688, 1.5582046508789062, 162.066162109375, -55.06214141845703, -17.473682403564453, 0.0, 3.774932861328125, -38.631195068359375, 166.499755859375, -132.67019653320312, 92.00418853759766, -44.264339447021484, -12.476089477539062, 88.19509887695312, 174.44073486328125, -20.304046630859375, 19.0579833984375, 65.10650634765625, -155.5384521484375, 70.6673812866211, -44.103759765625, 27.27386474609375, -15.564189910888672, 188.68048095703125, 5.059074401855469, -1.6659488677978516, -3.408489227294922, 222.04385375976562, 44.75581359863281, -2.4300537109375, 112.56898498535156, -157.3065643310547, 174.8057403564453, 236.77850341796875, 136.40420532226562, -26.115982055664062, 181.496337890625, 149.8090362548828, 160.15737915039062, 200.71658325195312, 41.60711669921875, 145.18927001953125, -152.2418212890625, -139.560546875, 16.3392333984375, 41.566650390625, 163.85365295410156, 163.74603271484375, -17.677719116210938, -172.20867919921875, 198.35968017578125, 64.73194885253906, 7.640106201171875, -58.21478271484375, -190.47576904296875, -6.983642578125, -3.202301025390625, -160.66433715820312, 182.39727783203125, 50.771240234375, -18.64453125, 174.85955810546875, 0.0, 7.5447845458984375, 151.0472869873047, -23.34619140625, 24.86944580078125, 217.32220458984375, -114.636474609375, 205.7645263671875, -6.423736572265625, 149.70111083984375, 17.267044067382812, 32.31159210205078, -76.24162292480469, 143.0831298828125, 18.270263671875, -18.144577026367188, -10.053955078125, 99.87247467041016, 186.46279907226562, 14.386260986328125, 8.120285034179688, -8.51080322265625, -43.509490966796875, -22.22967529296875, -0.71722412109375, -40.20416259765625, 166.1363983154297, 195.88262939453125, -1.54541015625, 19.675155639648438, 175.06076049804688, 31.63262939453125, -4.822654724121094, 192.16744995117188, 205.48373413085938, 19.946090698242188, 40.14453125, 7.6824951171875, 63.563720703125, 137.5480499267578], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000459.npy"}
{"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 42.88510513305664, "std": 101.46121215820312, "min": -198.57568359375, "p10": -100.17281341552734, "median": 26.034446716308594, "p90": 170.75658416748047, "max": 234.8712158203125, "pos_frac": 0.71875, "sample": [14.650344848632812, 70.98306274414062, -135.0450439453125, 159.92626953125, 60.750091552734375, 166.2899169921875, -164.14248657226562, 134.53475952148438, -43.3480224609375, 191.6373291015625, -135.60403442382812, 17.104583740234375, 17.59503173828125, 12.098846435546875, 30.413314819335938, 7.82037353515625, -76.64419555664062, 16.13214111328125, 169.72421264648438, 192.22503662109375, -45.878143310546875, 28.9688720703125, 70.36688232421875, 117.72532653808594, 61.05531311035156, 11.137939453125, 163.82073974609375, -8.535110473632812, 205.822021484375, -9.438568115234375, -14.45068359375, 102.98019409179688, 89.37869262695312, 234.29681396484375, 90.60678100585938, -47.217315673828125, 136.060546875, -147.83065795898438, 230.00918579101562, -19.345733642578125, -178.49314880371094, 182.94064331054688, -100.94244384765625, 0.5961341857910156, 95.06500244140625, 135.80621337890625, -37.14654541015625, 145.6898956298828, 5.55975341796875, -99.84297180175781, 86.08444213867188, 2.044445037841797, 0.5897884368896484, -1.0944747924804688, -45.54273986816406, 17.438796997070312, -10.487808227539062, -2.510986328125, -2.423175811767578, -16.82861328125, 174.98440551757812, 56.11822509765625, 151.83297729492188, 10.298171997070312, 27.49200439453125, -125.44357299804688, 0.04692840576171875, 44.574981689453125, 165.96188354492188, 0.19031524658203125, 19.264678955078125, 120.28317260742188, 23.951324462890625, 17.8558349609375, 128.00799560546875, 130.91912841796875, 164.96578979492188, 154.84716796875, 22.3487548828125, 30.562236785888672, 196.72177124023438, 27.636276245117188, -10.2171630859375, 228.9058837890625, 54.7801513671875, 234.8712158203125, 181.88934326171875, 169.34738159179688, -136.652587890625, 169.04681396484375, 124.10830688476562, 115.59481811523438, -143.50473022460938, 7.0986328125, -44.68034362792969, 8.64801025390625, 159.392578125, -198.57568359375, 131.15838623046875, -96.1043701171875, 67.53429412841797, -147.94268798828125, 150.15478515625, 28.32232666015625, 16.412757873535156, -117.10307312011719, -10.824516296386719, -71.7130126953125, 168.19210815429688, 37.977935791015625, 27.277435302734375, 155.667236328125, 33.046234130859375, 164.25018310546875, -122.7362060546875, 5.051977157592773, 5.605525970458984, 149.50807189941406, 15.56231689453125, 184.5327911376953, 173.1654510498047, 13.00909423828125, 28.42425537109375, -89.83807373046875, -80.69686889648438, 5.794525146484375, 24.791458129882812, 44.20332336425781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000460.npy"}
{"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 55.16642761230469, "std": 100.26421356201172, "min": -190.27838134765625, "p10": -79.07128601074218, "median": 34.62944793701172, "p90": 187.42049407958984, "max": 228.96841430664062, "pos_frac": 0.703125, "sample": [17.141082763671875, -134.1241455078125, 188.32586669921875, 23.581058502197266, 77.50135803222656, 123.00408935546875, 161.5043487548828, 28.87884521484375, -81.513916015625, 192.26095581054688, 161.69146728515625, 0.0, 15.28179931640625, 228.96841430664062, 124.15412902832031, 146.302978515625, 143.99008178710938, 73.490966796875, -4.903961181640625, 10.259368896484375, 64.9146957397461, 187.14581298828125, 127.10511016845703, -86.32211303710938, 78.45977783203125, -8.969696044921875, 11.8031005859375, 203.34976196289062, -39.46040344238281, 158.67117309570312, 0.8589363098144531, 202.92169189453125, 58.458984375, 154.94960021972656, -37.609710693359375, 159.93304443359375, 198.88232421875, 20.849197387695312, -168.22930908203125, -78.02444458007812, 188.82620239257812, -151.208984375, 3.1525421142578125, 19.30010986328125, 137.09542846679688, 2.8242950439453125, 130.32858276367188, 144.88720703125, 188.06141662597656, -5.832244873046875, 144.82974243164062, -110.59883117675781, 24.711212158203125, -1.3353271484375, 11.631969451904297, -11.657150268554688, 155.6922607421875, -187.52264404296875, -43.022491455078125, 89.90618896484375, 135.17076110839844, 2.1821823120117188, -14.615921020507812, 146.29287719726562, 193.02053833007812, 23.079025268554688, -190.27838134765625, 4.7198638916015625, 178.46481323242188, 7.10675048828125, -11.557682037353516, 168.33627319335938, -0.95391845703125, 178.1368408203125, -13.57525634765625, 202.03814697265625, 85.15533447265625, 159.34197998046875, -59.082275390625, -43.234710693359375, 154.29092407226562, 6.317169189453125, 142.3658447265625, 85.11434936523438, 163.13064575195312, 67.59488677978516, -24.768280029296875, -118.85931396484375, -8.483802795410156, 48.177764892578125, 32.03977966308594, 128.6326141357422, 15.082061767578125, 143.6631317138672, 43.42266845703125, 199.51466369628906, 10.877944946289062, 44.67138671875, 160.7532958984375, -130.85028076171875, 107.8874740600586, 1.4521522521972656, 191.10350036621094, 20.901275634765625, -88.019287109375, -21.14971923828125, 20.84130859375, 157.8990478515625, 160.97125244140625, -11.69073486328125, -61.83613967895508, 126.73883056640625, -0.5660057067871094, 115.46161651611328, 18.8326416015625, 1.4870834350585938, 126.65620422363281, 126.20771789550781, -160.30841064453125, -20.96057891845703, 37.2191162109375, -5.340980529785156, -123.42401123046875, 192.85174560546875, -19.7513484954834, 92.57168579101562, 147.47251892089844, 179.80813598632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000461.npy"}
{"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 71.19913482666016, "std": 100.18830108642578, "min": -184.68463134765625, "p10": -40.58117904663086, "median": 79.2378044128418, "p90": 193.40660400390624, "max": 244.286865234375, "pos_frac": 0.78125, "sample": [149.2210693359375, 62.887176513671875, 6.1152191162109375, 119.5311279296875, 183.2686767578125, 43.389739990234375, 154.30874633789062, 122.64056396484375, 96.70160675048828, 193.14089965820312, -152.3807373046875, 132.14410400390625, 197.50631713867188, 31.095169067382812, 195.91867065429688, 6.42254638671875, 111.57405090332031, 155.95965576171875, 81.19021606445312, 5.32904052734375, 39.65997314453125, 147.24151611328125, 143.0880889892578, 47.30462646484375, 26.98770523071289, -1.714385986328125, 167.06385803222656, -1.1503334045410156, 166.3133544921875, -124.18009948730469, -146.31607055664062, 96.43557739257812, -3.469696044921875, 18.292205810546875, -1.26153564453125, 159.84927368164062, 194.02658081054688, -80.71231079101562, -14.39434814453125, 14.607757568359375, 159.55316162109375, 160.70904541015625, 176.216552734375, 6.718505859375, 231.5452880859375, 114.23218536376953, 95.71467590332031, 182.68801879882812, 133.41943359375, 3.4945220947265625, 196.25613403320312, 7.793800354003906, 148.88938903808594, -116.02536010742188, 13.063011169433594, -15.662948608398438, 185.208984375, 37.138641357421875, 8.714942932128906, 18.8568115234375, 159.02914428710938, 213.75872802734375, 30.110034942626953, 169.22532653808594, -20.47283935546875, 96.96411895751953, -143.85888671875, 127.2691650390625, 243.37994384765625, -29.237136840820312, 168.19293212890625, 174.44326782226562, 158.1844940185547, 138.7769775390625, 113.51730346679688, 104.85076904296875, -29.00677490234375, 25.14752197265625, 10.481128692626953, -24.49993896484375, 14.546550750732422, -30.310546875, 1.23309326171875, 2.253387451171875, -162.62322998046875, 72.33940124511719, 152.83102416992188, 231.21279907226562, -107.73484802246094, 126.30775451660156, -89.60769653320312, 204.02999877929688, 77.28539276123047, 45.72938537597656, 105.09858703613281, 244.286865234375, -40.18981170654297, 117.1552734375, 140.55484008789062, 169.83694458007812, 126.88748931884766, -51.846405029296875, 238.14990234375, 17.919708251953125, -19.843902587890625, 20.2998046875, 146.42752075195312, 54.950164794921875, 28.91546630859375, 13.9986572265625, 158.7197265625, 6.7627105712890625, 62.21331787109375, 184.67095947265625, -159.01083374023438, 212.36859130859375, -31.243118286132812, 121.57975006103516, 138.0164031982422, -20.056076049804688, 196.80499267578125, 21.818557739257812, 0.2125244140625, -41.49436950683594, 130.40594482421875, 131.61453247070312, 156.2810516357422, -184.68463134765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000462.npy"}
{"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 51.63710021972656, "std": 88.98045349121094, "min": -171.2911376953125, "p10": -31.284088134765625, "median": 31.871049880981445, "p90": 177.57069702148436, "max": 221.43707275390625, "pos_frac": 0.6953125, "sample": [94.6634521484375, 61.80421447753906, 55.0048828125, 7.522499084472656, 18.9635009765625, -2.364686965942383, 196.2218017578125, 12.7587890625, -171.2911376953125, 85.48971557617188, 167.12652587890625, -1.70501708984375, -155.04441833496094, -40.7655029296875, -1.96630859375, 189.18692016601562, 70.0884780883789, 95.11335754394531, 19.12127685546875, -1.8889141082763672, 18.44647979736328, -164.29385375976562, -22.651336669921875, 7.1389312744140625, 94.09150695800781, -2.4982147216796875, -97.95748901367188, 35.55198669433594, 180.949951171875, 183.77935791015625, -160.2574005126953, 140.48043823242188, 14.61614990234375, -31.243804931640625, -46.37152099609375, 193.14508056640625, 185.61358642578125, 162.50718688964844, 221.43707275390625, -52.559722900390625, 76.8154296875, -33.43548583984375, 116.49601745605469, 210.05194091796875, 82.5679931640625, 5.67572021484375, 200.6259765625, -1.9743499755859375, 27.55267333984375, 2.117156982421875, 15.956703186035156, 31.116920471191406, 176.8133544921875, 62.528411865234375, 8.295074462890625, 19.69903564453125, 185.17758178710938, -25.073577880859375, 166.915283203125, 141.69314575195312, 90.09130859375, 168.031005859375, 18.713592529296875, -10.239791870117188, -28.745513916015625, 19.431228637695312, 179.33782958984375, -0.2116851806640625, 120.43414306640625, 57.390869140625, 165.42930603027344, -31.378082275390625, -6.97772216796875, 66.501953125, 57.547760009765625, 9.30499267578125, 131.63247680664062, -29.954757690429688, -25.19213104248047, 45.00175476074219, 160.9346160888672, 14.0982666015625, 155.02520751953125, 137.83572387695312, 76.80758666992188, 6.669961929321289, 23.632408142089844, 149.65542602539062, 96.42471313476562, -22.016077041625977, -15.42266845703125, 119.19512939453125, -27.66827392578125, 0.0, -11.7703857421875, 5.080055236816406, 8.270450592041016, 192.12020874023438, 36.6123046875, -13.854408264160156, 96.67306518554688, 130.83705139160156, -28.839263916015625, 2.8237171173095703, 30.28986358642578, -116.15176391601562, 37.94140625, 32.625179290771484, 124.74801635742188, -0.41904640197753906, 10.428634643554688, -70.53042602539062, -21.9759521484375, 108.47293853759766, -1.67230224609375, 161.74288940429688, 165.31588745117188, 38.63438034057617, 167.90460205078125, 87.67132568359375, 212.7723388671875, 38.711944580078125, 51.002685546875, 119.81597900390625, 140.85546875, 166.4534912109375, -166.1573486328125, -25.783233642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000463.npy"}
{"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 54.353965759277344, "std": 91.09596252441406, "min": -208.1600341796875, "p10": -39.79985961914062, "median": 31.549667358398438, "p90": 173.71971435546874, "max": 210.3790283203125, "pos_frac": 0.75, "sample": [0.0, 99.73350524902344, 50.641151428222656, 154.93173217773438, -10.861740112304688, 143.1937255859375, 170.09796142578125, -33.11643981933594, 174.5941162109375, 0.10890388488769531, 58.81163024902344, 93.9764404296875, 150.13612365722656, 33.798980712890625, 146.1426544189453, 26.472881317138672, 167.64834594726562, -4.33349609375, 158.91497802734375, 117.26875305175781, 47.52288818359375, 161.8389129638672, 11.112276077270508, 34.9139404296875, 55.101348876953125, 149.64105224609375, 17.583839416503906, -160.27613830566406, 210.3790283203125, 13.450973510742188, -2.077056884765625, -14.91497802734375, 17.932357788085938, 196.78244018554688, 185.32769775390625, 176.63629150390625, 87.7076416015625, 4.603416442871094, -10.349945068359375, 125.49626159667969, -2.951019287109375, 23.0811767578125, 179.16259765625, 87.88055419921875, 140.1763153076172, 4.55108642578125, -13.370269775390625, 15.026397705078125, 1.1515960693359375, 110.2684326171875, 16.466567993164062, 25.74713134765625, 3.8212432861328125, 144.66358947753906, 13.667613983154297, 123.1221923828125, 155.18231201171875, 128.2689208984375, -160.71917724609375, 118.05477905273438, -4.4622802734375, 156.11166381835938, -7.732696533203125, 163.82508850097656, 191.41976928710938, -10.7889404296875, 114.3128662109375, 27.021163940429688, 31.3056640625, 117.34326171875, -105.05975341796875, -208.1600341796875, -150.96212768554688, 3.4335174560546875, 11.726715087890625, 11.415435791015625, -16.745208740234375, -49.187255859375, 117.61715698242188, 19.82630157470703, 171.04933166503906, 203.3265380859375, 162.81109619140625, 196.32052612304688, -43.34271240234375, 173.344970703125, 2.170328140258789, -38.281494140625, 13.4715576171875, 3.706570625305176, 42.95109558105469, -113.3565673828125, 167.39492797851562, -12.396514892578125, -70.970703125, 5.503593444824219, -1.7683792114257812, 146.49227905273438, 7.141845703125, 10.64554214477539, 74.6783218383789, 181.54249572753906, 43.41993713378906, 176.71197509765625, 182.82675170898438, -57.54779052734375, 86.06015014648438, -0.511566162109375, 148.16064453125, -123.57174682617188, 13.657318115234375, 9.538475036621094, -16.06247329711914, 54.29644775390625, 182.369384765625, 48.11984634399414, 137.5376739501953, 153.91848754882812, 146.47116088867188, -186.9046630859375, 18.4158935546875, 30.29998779296875, -47.77024459838867, 71.29487609863281, 23.402698516845703, -24.936767578125, 41.797515869140625, 31.793670654296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000464.npy"}
{"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 70.5584487915039, "std": 96.23728942871094, "min": -190.62271118164062, "p10": -32.49734878540038, "median": 46.29881286621094, "p90": 191.46531524658204, "max": 228.50613403320312, "pos_frac": 0.7421875, "sample": [-2.9503631591796875, -7.7647705078125, 13.583465576171875, 28.879348754882812, 175.3585205078125, -6.874521255493164, -37.302398681640625, 192.65908813476562, 9.438812255859375, 26.030776977539062, 1.5506439208984375, 8.222900390625, -2.9623260498046875, 12.25604248046875, -133.06881713867188, -36.73638916015625, 190.20260620117188, 165.88775634765625, 205.9259033203125, 184.24798583984375, 164.55828857421875, -45.3665771484375, 43.84889221191406, 168.70639038085938, 35.545867919921875, -2.387847900390625, 3.578277587890625, 174.06858825683594, 191.92453002929688, 148.86599731445312, 2.612762451171875, 153.10987854003906, 37.419830322265625, 138.81521606445312, -1.2204437255859375, -12.978424072265625, -36.35028076171875, 188.89077758789062, 146.17752075195312, 223.24453735351562, 0.13134765625, 174.14508056640625, 1.2813720703125, -29.966583251953125, -44.084716796875, 36.17718505859375, -8.112541198730469, 49.16637420654297, -25.41412353515625, 145.1562042236328, 142.5503692626953, 144.85140991210938, 132.613525390625, 144.1638946533203, 59.21498107910156, -98.96944427490234, 217.56619262695312, 41.55851745605469, -10.977081298828125, 14.394927978515625, 25.75042724609375, 52.124542236328125, 46.35858154296875, 117.03312683105469, 179.87619018554688, 23.330490112304688, 197.87429809570312, 220.45220947265625, -13.3482666015625, 27.796844482421875, 101.48418426513672, 164.79693603515625, 157.9104461669922, 148.5697479248047, 126.50497436523438, -98.658447265625, -5.68353271484375, 53.043975830078125, 28.182174682617188, 100.2497329711914, 46.239044189453125, 9.81463623046875, 3.70458984375, 12.0848388671875, 7.279205322265625, 144.00592041015625, 195.63272094726562, -61.82572937011719, 75.32494354248047, -15.0946044921875, 186.68988037109375, 208.26934814453125, 228.50613403320312, 47.441383361816406, 144.270751953125, 26.12835693359375, 200.50341796875, 176.06280517578125, 194.80035400390625, 5.656364440917969, 175.1644287109375, 175.19944763183594, 200.17355346679688, 177.09646606445312, 9.574462890625, 153.74075317382812, -19.080322265625, 174.8311767578125, -156.31802368164062, 38.51617431640625, -94.19207763671875, -0.198211669921875, 180.9844970703125, 177.8572998046875, 163.94461059570312, 191.2685089111328, -145.25128173828125, 157.93405151367188, 58.104644775390625, 10.379066467285156, -30.846092224121094, 120.87823486328125, -16.89202880859375, -17.43109893798828, 169.49267578125, -190.62271118164062, -8.71319580078125, 91.64825439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000465.npy"}
{"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 73.34831237792969, "std": 88.70015716552734, "min": -171.140869140625, "p10": -12.968212127685543, "median": 56.04029846191406, "p90": 177.6454833984375, "max": 229.6402587890625, "pos_frac": 0.8125, "sample": [7.51933479309082, 132.20501708984375, 148.2097930908203, 163.05422973632812, 132.29115295410156, 178.9063720703125, 37.048126220703125, 0.0, 103.75799560546875, 58.283782958984375, 167.21800231933594, -27.463115692138672, 136.16793823242188, 111.60372924804688, 91.65093994140625, 222.591796875, 125.5737075805664, 4.479656219482422, 152.07247924804688, 82.94378662109375, 136.9913787841797, 25.467994689941406, 83.29405212402344, -136.07794189453125, 17.485076904296875, 50.585968017578125, 116.13508605957031, 27.179840087890625, 142.9267578125, 129.5812225341797, -1.7630996704101562, 167.0047607421875, 4.477973937988281, 139.39529418945312, 155.47232055664062, 2.239330291748047, 166.02615356445312, -8.558792114257812, 164.74212646484375, 66.85556030273438, 226.8621063232422, 10.381683349609375, 219.87918090820312, 53.79681396484375, 154.13775634765625, 186.03582763671875, 27.92163848876953, -87.28788757324219, -133.39691162109375, 6.033966064453125, 11.034027099609375, 148.76998901367188, 6.683807373046875, 155.2242889404297, 158.89559936523438, 169.7886962890625, -11.972747802734375, -171.140869140625, 29.99169921875, 169.37823486328125, 138.73472595214844, 155.1842803955078, 124.20176696777344, 70.29354858398438, 18.843063354492188, 143.9134979248047, 177.1051025390625, 140.2117462158203, 200.44961547851562, 153.72921752929688, 0.0, 159.16461181640625, 220.82183837890625, -30.284820556640625, -15.290962219238281, 27.136009216308594, 139.87030029296875, -27.700302124023438, 53.381317138671875, 229.6402587890625, 6.322669982910156, 10.953399658203125, 213.11053466796875, 103.91238403320312, 31.052520751953125, -168.029541015625, 22.3365478515625, 49.19818115234375, 20.9935302734375, 120.48538970947266, 159.32400512695312, 21.756729125976562, 20.033782958984375, -22.655715942382812, 213.8753662109375, 2.9841718673706055, 39.41059875488281, 158.806640625, -76.79852294921875, -147.4908447265625, 51.25438690185547, 26.16259765625, -4.3755340576171875, 179.41102600097656, 23.358131408691406, 0.0, 170.43801879882812, 21.67657470703125, -0.09482955932617188, 94.47933959960938, 195.8404541015625, 109.68663024902344, -4.196136474609375, 152.19973754882812, 102.17374420166016, -6.4552764892578125, 24.981231689453125, 40.327484130859375, 25.172821044921875, 2.07568359375, 26.359893798828125, 15.641700744628906, 202.1165771484375, 153.72866821289062, -22.854736328125, -6.474086761474609, 118.50762939453125, 29.863418579101562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000466.npy"}
{"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 53.36249923706055, "std": 98.25377655029297, "min": -166.10977172851562, "p10": -73.90275192260742, "median": 24.43262481689453, "p90": 182.34429626464842, "max": 220.75762939453125, "pos_frac": 0.65625, "sample": [175.02166748046875, 157.45730590820312, 116.67286682128906, 105.40229797363281, 39.03760528564453, -9.413589477539062, 83.9354248046875, 129.9842529296875, 124.1131591796875, -1.136993408203125, -144.6851806640625, -166.10977172851562, -10.58404541015625, -10.145858764648438, 17.937408447265625, 154.9659423828125, -114.17620849609375, 205.8741455078125, -5.1142578125, 98.33287048339844, -20.82537841796875, 142.1907958984375, -19.082290649414062, 126.54801940917969, 172.34573364257812, 60.52581787109375, 24.959548950195312, 142.68760681152344, -0.39177703857421875, 13.423309326171875, 23.90570068359375, 161.95651245117188, 87.14996337890625, 161.55645751953125, 188.38134765625, 97.0418701171875, 29.792465209960938, 190.43728637695312, 87.51686096191406, -93.62493896484375, 193.00778198242188, -1.3974609375, 0.20099449157714844, 185.88232421875, 17.914276123046875, 153.97747802734375, -15.50828742980957, -42.099517822265625, 181.4896240234375, -127.96178436279297, 12.673072814941406, -130.72152709960938, -76.23201751708984, 163.22056579589844, -52.71044921875, 149.43096923828125, -62.68474197387695, -47.774505615234375, 159.74185180664062, 203.43536376953125, 19.8023681640625, 14.007356643676758, -1.1262149810791016, 10.851806640625, 141.24151611328125, 17.741943359375, 0.05438995361328125, 90.07655334472656, -8.460723876953125, -72.90449523925781, 5.179962158203125, -150.3299102783203, 159.5233154296875, 0.0, 0.0956573486328125, -6.4958343505859375, 90.30096435546875, 194.9903106689453, 4.274749755859375, -1.8455734252929688, 0.0, 52.56732177734375, 165.93801879882812, -11.879241943359375, 92.15557861328125, 106.95913696289062, 146.58987426757812, -37.31108093261719, 213.14364624023438, -118.88436889648438, -72.65190124511719, 201.38958740234375, 155.21380615234375, 141.84461975097656, 181.61825561523438, -0.2089691162109375, -20.727279663085938, 22.188385009765625, -7.21099853515625, 87.13945007324219, -36.38763427734375, 170.57708740234375, 166.68408203125, 40.534027099609375, -23.548171997070312, 17.49852752685547, -134.16893005371094, -85.31903076171875, 14.3679780960083, 92.7947998046875, 184.03839111328125, 131.13182067871094, -123.40960693359375, 156.33319091796875, 8.426227569580078, 8.13299560546875, -12.163970947265625, 0.8125457763671875, 220.75762939453125, 77.95675659179688, 170.4306640625, 118.64057922363281, 43.15435791015625, 171.7079315185547, 193.13726806640625, 195.68374633789062, -103.0484619140625, -26.954345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000467.npy"}
{"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 47.74378967285156, "std": 86.20987701416016, "min": -146.6684112548828, "p10": -46.49939880371094, "median": 24.58973217010498, "p90": 168.46994018554688, "max": 249.279541015625, "pos_frac": 0.6953125, "sample": [35.358123779296875, 200.6881103515625, 150.1929168701172, -21.30438232421875, 50.48255920410156, 21.1741943359375, -21.43707275390625, 166.96145629882812, 58.13999938964844, 33.25567626953125, 32.69903564453125, -44.025054931640625, 20.91949462890625, 191.44607543945312, -79.51620483398438, -15.245857238769531, 0.433868408203125, 177.20166015625, -116.42558288574219, -24.5181884765625, 63.85639953613281, 25.130544662475586, -120.02899169921875, 110.8365478515625, -9.183502197265625, 221.935546875, 54.05987548828125, 67.30197143554688, -10.1072998046875, -31.103744506835938, 0.88970947265625, 30.60638427734375, 190.94827270507812, -23.080415725708008, 34.605224609375, -16.08177947998047, -19.393447875976562, 167.90826416015625, -115.557861328125, 24.048919677734375, 9.628715515136719, 26.36968994140625, 11.5404052734375, 156.9576416015625, -6.097908020019531, 5.5732421875, 71.24200439453125, -41.549896240234375, 37.27928161621094, 4.7857208251953125, -10.462203979492188, 3.822376251220703, 66.01129150390625, 0.63983154296875, 14.540008544921875, 46.88848876953125, 11.04132080078125, 160.11932373046875, 137.935791015625, 161.0847930908203, 172.56198120117188, -28.0489501953125, -146.6684112548828, 10.876808166503906, 249.279541015625, 7.01849365234375, 18.70380401611328, 176.15618896484375, 1.92095947265625, 90.44862365722656, 55.4234619140625, 126.9505615234375, -26.67181396484375, 75.0804443359375, 153.61798095703125, 162.96630859375, -49.999755859375, 18.244125366210938, 8.242301940917969, 129.74819946289062, -79.54554748535156, 204.67169189453125, 31.723094940185547, 179.08087158203125, -16.9591064453125, 133.5196533203125, 144.011962890625, -21.21065330505371, 145.67849731445312, 119.07083892822266, -13.3546142578125, 5.440107345581055, -0.172393798828125, 200.861328125, 1.1851806640625, 34.65980529785156, 130.5245361328125, 2.2375640869140625, 66.14437866210938, -0.28436279296875, 159.3551483154297, 167.02108764648438, 77.96965789794922, 8.69171142578125, -46.05242919921875, -32.934814453125, -54.40997314453125, 63.564788818359375, 55.14398193359375, 121.21774291992188, -8.307952880859375, -5.07476806640625, 133.55804443359375, -98.43592834472656, -56.611602783203125, 15.452392578125, 155.4647979736328, 12.637161254882812, -3.6341094970703125, -122.57935333251953, 153.90232849121094, 157.1263427734375, 48.195404052734375, 180.02044677734375, -47.542327880859375, 169.780517578125, -53.377716064453125, 156.50897216796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000468.npy"}
{"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 62.263973236083984, "std": 98.77171325683594, "min": -246.3162841796875, "p10": -30.97147521972656, "median": 36.74424743652344, "p90": 181.24105987548828, "max": 232.63800048828125, "pos_frac": 0.765625, "sample": [21.4447021484375, 179.00399780273438, 0.0, 14.157524108886719, 31.7509765625, 60.78350830078125, -25.17493438720703, 19.613037109375, 9.779190063476562, 79.64451599121094, 130.2997589111328, 34.81915283203125, 103.72747802734375, -246.3162841796875, 23.980224609375, -5.9354248046875, 204.92112731933594, 162.52613830566406, 23.93896484375, 142.10647583007812, -17.034133911132812, 87.33157348632812, 6.617473602294922, 4.9886016845703125, -162.4049072265625, 176.57821655273438, 21.32391357421875, 166.76495361328125, 182.64675903320312, 18.49139404296875, 147.98297119140625, 59.386962890625, 222.2652587890625, 173.16793823242188, 1.8783721923828125, 114.12547302246094, 51.127685546875, 104.41607666015625, 175.27008056640625, 24.98291015625, -2.3876571655273438, -27.2608642578125, 31.0865478515625, 1.28924560546875, 148.69520568847656, -0.748260498046875, 18.72588348388672, 20.602020263671875, 49.654541015625, 36.87060546875, 142.47933959960938, 182.13311767578125, 170.4104461669922, 188.78814697265625, -150.42486572265625, 122.29312896728516, 169.5828857421875, 204.59854125976562, 169.3778076171875, -60.044158935546875, 5.081167221069336, -10.41583251953125, 196.6796875, -167.966796875, -88.59552001953125, 66.1663818359375, 24.060882568359375, -11.42523193359375, -32.094085693359375, 180.85874938964844, 16.601669311523438, 148.3293914794922, 82.0211181640625, 168.3389129638672, -219.21041870117188, 10.558429718017578, 40.24163818359375, 157.73947143554688, -7.5810699462890625, 232.63800048828125, 171.20823669433594, 92.61666870117188, 36.617889404296875, 2.027099609375, 174.6319580078125, -16.918930053710938, -213.31451416015625, 5.176887512207031, 99.26727294921875, 21.923561096191406, 58.37553405761719, 197.78936767578125, -5.5374603271484375, 174.80210876464844, -37.22723388671875, -20.966888427734375, 0.43839263916015625, 3.4504776000976562, -14.551761627197266, 150.73292541503906, -39.66206359863281, 0.34583282470703125, -30.4903564453125, 220.60842895507812, 5.075567245483398, -61.5281982421875, 180.829833984375, 112.16805267333984, 175.91262817382812, 193.59417724609375, 161.09335327148438, 7.226287841796875, 175.91656494140625, 47.5960693359375, -2.5294265747070312, 71.7452392578125, 22.858028411865234, 162.34182739257812, 137.57907104492188, 32.550559997558594, 186.866455078125, 169.470703125, 150.40313720703125, -16.814239501953125, 84.782470703125, 35.45574951171875, 208.76275634765625, -35.637115478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000469.npy"}
{"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 71.37315368652344, "std": 97.80847930908203, "min": -182.68182373046875, "p10": -51.383074951171864, "median": 59.47267150878906, "p90": 194.6878204345703, "max": 238.26173400878906, "pos_frac": 0.8046875, "sample": [-17.1260986328125, 20.54193115234375, 181.80221557617188, 44.053192138671875, -2.2502899169921875, 63.71380615234375, 5.94573974609375, 159.8166046142578, 175.2454833984375, 23.8792724609375, -32.51324462890625, 4.7391357421875, -0.7211647033691406, 175.12313842773438, 115.60223388671875, 155.01724243164062, 179.76123046875, 1.4408721923828125, -13.77099609375, 50.40533447265625, 130.38397216796875, -88.20635986328125, -117.76788330078125, -138.55349731445312, 216.9588623046875, 146.836669921875, 92.48052215576172, 23.4727783203125, 169.2197723388672, 170.70547485351562, 227.46466064453125, 140.98184204101562, -33.99463653564453, 0.3885498046875, 146.66366577148438, -72.26852416992188, 194.58642578125, 142.74417114257812, 27.6422119140625, -106.06694030761719, 11.75244140625, 212.57473754882812, 138.38699340820312, 19.74114990234375, 7.826019287109375, 163.17413330078125, 198.847900390625, 18.765121459960938, 42.60612869262695, 141.33737182617188, 12.165786743164062, 166.01828002929688, 26.9779052734375, 19.430431365966797, -17.4205322265625, 157.53790283203125, 145.1078338623047, 2.36029052734375, 155.19264221191406, 17.740753173828125, 70.52230072021484, 3.8115177154541016, 165.77748107910156, 87.37397766113281, 23.234466552734375, 44.92491912841797, 7.348991394042969, -182.68182373046875, 19.787994384765625, 172.8780517578125, 49.850555419921875, 136.15884399414062, 146.95126342773438, 1.2642593383789062, -127.20611572265625, -65.6961441040039, -81.32830810546875, 137.25799560546875, 182.1156463623047, 158.524658203125, 34.485809326171875, 55.501983642578125, 141.86782836914062, 213.73910522460938, 126.18989562988281, -20.842613220214844, 5.7929534912109375, 197.896484375, 194.92440795898438, 195.31253051757812, -25.06926727294922, 159.8016357421875, 138.007080078125, 156.47439575195312, -2.72540283203125, 62.180511474609375, 8.538124084472656, 138.4956817626953, 115.5118408203125, 11.97137451171875, 177.91824340820312, 186.85336303710938, 125.71762084960938, 238.26173400878906, 15.73272705078125, 180.26251220703125, 0.9228458404541016, 6.455474853515625, -16.372039794921875, -156.91445922851562, 20.6724853515625, -105.02494812011719, 204.7647705078125, -48.370513916015625, 125.65038299560547, 214.865478515625, 4.74884033203125, 56.76483154296875, 75.439697265625, 76.818603515625, 200.46728515625, -122.177734375, 70.03369140625, 39.311004638671875, 205.60397338867188, 174.487060546875, 177.85760498046875, -58.412384033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000470.npy"}
{"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 50.581974029541016, "std": 90.76143646240234, "min": -167.3115997314453, "p10": -33.534861373901364, "median": 28.177406311035156, "p90": 180.35684204101562, "max": 236.809326171875, "pos_frac": 0.6875, "sample": [-87.00999450683594, 165.50473022460938, 30.578765869140625, 90.43692016601562, 145.33396911621094, 215.47265625, 158.6685333251953, -128.78656005859375, -12.905508041381836, 159.33541870117188, -54.70390319824219, -21.206146240234375, 0.669830322265625, 109.09345245361328, -1.6747207641601562, 65.19940185546875, 19.242401123046875, 49.19873046875, -24.8160400390625, -39.108978271484375, 27.414520263671875, 30.088653564453125, 19.696502685546875, 54.84832763671875, -31.829498291015625, -11.293235778808594, 198.66094970703125, -0.03265380859375, -28.12249755859375, 35.24461364746094, 187.13552856445312, 59.6121826171875, 61.46826171875, 10.270660400390625, -4.1445770263671875, 54.24922180175781, 149.95689392089844, 125.1129150390625, 25.94823455810547, 15.826690673828125, 15.301315307617188, -6.84698486328125, 168.83181762695312, 179.8980712890625, -37.514041900634766, 108.62918090820312, 29.494430541992188, 208.6533660888672, 154.44970703125, -10.886993408203125, 5.4411163330078125, 8.5189208984375, 172.591552734375, 121.8074951171875, -125.90472412109375, 139.4921875, 132.95838928222656, 3.928129196166992, -3.3736114501953125, 31.058982849121094, 219.30679321289062, 236.809326171875, 14.392724990844727, -2.68243408203125, -30.374286651611328, -3.0351810455322266, -8.033622741699219, 4.7204437255859375, -167.3115997314453, 24.66485595703125, 130.85951232910156, 28.940292358398438, 156.41184997558594, 42.54542541503906, -146.33946228027344, 33.99334716796875, 18.0205078125, 81.34331512451172, 57.204429626464844, -26.7874755859375, 103.638916015625, -17.172760009765625, -1.425537109375, -16.937820434570312, 53.259033203125, 140.35726928710938, 8.739883422851562, 19.01329803466797, 137.39654541015625, -4.668121337890625, 124.02951049804688, 162.59039306640625, -11.836002349853516, 56.56494140625, 15.287887573242188, 179.6095733642578, 182.85406494140625, -67.21646881103516, -9.926620483398438, -28.2945556640625, 31.462909698486328, 18.163108825683594, 3.330108642578125, 61.9102783203125, 151.48536682128906, 182.5895538330078, 49.332916259765625, 235.33859252929688, 21.397750854492188, 8.555389404296875, 153.07199096679688, 181.42730712890625, 192.41531372070312, -132.82920837402344, 174.495849609375, 6.517364501953125, 131.32723999023438, 0.0, 46.01129150390625, -96.09089660644531, 191.94232177734375, 155.30178833007812, 188.6607666015625, -17.7359619140625, 16.82275390625, -19.5504150390625, -118.95542907714844, -147.58035278320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000471.npy"}
{"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 67.26830291748047, "std": 96.3628158569336, "min": -169.08126831054688, "p10": -35.70909729003906, "median": 48.5146484375, "p90": 181.36097717285156, "max": 344.33056640625, "pos_frac": 0.7421875, "sample": [344.33056640625, -7.4663238525390625, 3.317647933959961, -135.0176544189453, 4.548046112060547, 2.332672119140625, -14.17835807800293, 151.45831298828125, 4.540435791015625, 37.955322265625, 100.90200805664062, 95.86970520019531, 179.27496337890625, 37.05828094482422, -58.82341766357422, 155.62890625, -19.857200622558594, 9.836143493652344, -36.63897705078125, 85.51811218261719, 21.33160400390625, 31.98907470703125, 129.50323486328125, 48.435638427734375, -13.490554809570312, 198.99288940429688, 142.98876953125, -131.0001220703125, 3.875469207763672, 154.5263671875, 57.841827392578125, -83.68074035644531, 19.048004150390625, 215.98583984375, -4.23614501953125, -29.127944946289062, 149.83450317382812, 170.94918823242188, -38.4593505859375, 6.4593505859375, 25.004974365234375, 140.71270751953125, 118.37646484375, 160.1544189453125, -28.336746215820312, 20.76007080078125, 8.40264892578125, -3.5052490234375, 11.83270263671875, 16.872161865234375, 145.19256591796875, 58.20158386230469, 180.88641357421875, -115.87081909179688, 154.97329711914062, 153.33050537109375, 173.5167236328125, 147.4818115234375, 66.29415893554688, 31.401168823242188, -2.347381591796875, 100.5853271484375, 6.901115417480469, -1.7235908508300781, 150.57501220703125, 180.831298828125, 146.6297607421875, 93.75401306152344, 128.01107788085938, -2.8651580810546875, 57.77323532104492, 39.207191467285156, 42.078857421875, 171.70469665527344, 152.31793212890625, 214.6707763671875, 180.38531494140625, -8.816368103027344, 173.97056579589844, 24.721343994140625, -20.24420166015625, 238.03732299804688, 167.8224334716797, 212.79412841796875, 37.14617919921875, 212.84619140625, 206.31292724609375, 177.68402099609375, 153.09361267089844, -169.08126831054688, -7.0435943603515625, -148.13775634765625, 133.47154235839844, 76.54623413085938, 3.1072998046875, 138.5501708984375, 147.34136962890625, 8.885940551757812, -35.310577392578125, 109.08502197265625, 172.29373168945312, -13.064117431640625, 182.46829223632812, 138.66290283203125, 21.99950408935547, 80.85773468017578, -3.58123779296875, 12.579437255859375, -3.334310531616211, 1.95050048828125, 195.28616333007812, 184.8441925048828, 4.538187026977539, 17.402801513671875, 48.593658447265625, -125.85530090332031, -38.16719055175781, 245.5587158203125, -16.32562255859375, 53.42181396484375, 192.23837280273438, 103.69488525390625, -119.42352294921875, 112.4510498046875, 137.0087432861328, 142.24063110351562, -50.846580505371094, -16.460609436035156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000472.npy"}
{"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 67.0908203125, "std": 94.13557434082031, "min": -190.6127471923828, "p10": -36.9123306274414, "median": 47.86979293823242, "p90": 188.3549835205078, "max": 225.67584228515625, "pos_frac": 0.75, "sample": [11.37359619140625, 189.40444946289062, 47.62367248535156, -4.056970596313477, 133.58758544921875, 5.3707275390625, 55.77203369140625, 17.71856689453125, 53.15069580078125, -15.19952392578125, 65.78544616699219, 35.71894836425781, -68.3502197265625, 36.32734298706055, -129.18035888671875, 65.0751953125, 32.35333251953125, 7.6320037841796875, 96.2762451171875, 158.7564697265625, 127.29328918457031, 151.19532775878906, 199.85101318359375, 225.67584228515625, 83.78753662109375, -127.42340087890625, -15.700637817382812, 132.31153869628906, 119.21763610839844, 31.5440673828125, 190.27066040039062, 180.77392578125, 167.01846313476562, 37.195648193359375, -16.62488555908203, -48.89796447753906, 16.433319091796875, -112.25182342529297, 215.787353515625, -8.124427795410156, 48.092796325683594, 147.73175048828125, 39.11676025390625, -15.308082580566406, -14.268539428710938, -24.269073486328125, -42.226806640625, 197.076416015625, 56.74029541015625, 24.2423095703125, 26.765106201171875, -6.3205108642578125, -0.12948226928710938, 197.8424072265625, -18.28277587890625, 185.3616943359375, -23.699737548828125, 158.1880645751953, 207.4910888671875, -170.49057006835938, 71.53372192382812, 192.33050537109375, 192.5535888671875, 4.195899963378906, 31.361419677734375, -88.54034423828125, 220.69454956054688, 176.71539306640625, 141.20809936523438, 198.216064453125, 132.5205078125, 144.81387329101562, 121.69989013671875, 26.672271728515625, 4.398406982421875, 155.7313232421875, 152.13998413085938, 18.29694366455078, 2.80712890625, 174.42327880859375, -2.9660415649414062, -35.2559814453125, 177.34112548828125, 146.74940490722656, 150.20416259765625, 183.34930419921875, -190.6127471923828, -119.13080596923828, 163.46023559570312, -18.564849853515625, 43.8253173828125, 123.88656616210938, 3.198577880859375, 48.41290283203125, -0.70068359375, -80.64985656738281, 27.08230972290039, 29.24456787109375, 42.518463134765625, -40.77714538574219, 187.90521240234375, 144.41502380371094, 145.6559600830078, 0.0, 158.17169189453125, 115.60382080078125, -97.62088012695312, 148.4615478515625, 25.089447021484375, 145.08973693847656, 154.72488403320312, 17.641220092773438, 1.7589035034179688, 181.09030151367188, 24.635009765625, -4.571533203125, 4.3134765625, -20.359786987304688, 28.604293823242188, 47.64678955078125, 163.72650146484375, 160.34402465820312, 168.23675537109375, 66.42871856689453, 214.67791748046875, 149.8348846435547, 153.5508575439453, 60.05870056152344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000473.npy"}
{"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 49.61687469482422, "std": 90.84080505371094, "min": -188.11920166015625, "p10": -38.14109878540039, "median": 32.91844940185547, "p90": 169.47246398925782, "max": 301.0908203125, "pos_frac": 0.7265625, "sample": [-1.2395477294921875, 13.03509521484375, 301.0908203125, -4.2396392822265625, -12.385101318359375, -29.21429443359375, 155.42774963378906, 197.04122924804688, 78.41453552246094, 160.4417724609375, 7.242401123046875, 28.860595703125, -8.5150146484375, -140.5533447265625, 7.246818542480469, -142.66859436035156, 6.31036376953125, -2.1556396484375, 72.90807342529297, 37.12012481689453, 120.02466583251953, 177.60958862304688, 4.7282257080078125, -70.682373046875, 163.19683837890625, -49.412532806396484, -95.93182373046875, 88.2373046875, -15.529411315917969, 85.6986083984375, 146.67233276367188, 53.812744140625, 0.6566009521484375, 145.31028747558594, 49.7652587890625, 137.8758544921875, 233.154052734375, 21.46186065673828, 1.7048873901367188, 189.43450927734375, 146.09295654296875, 169.75015258789062, 190.0384521484375, 180.0938262939453, 70.91522216796875, -2.483287811279297, 0.0, 1.5035552978515625, 25.43927001953125, 10.280593872070312, -0.9982681274414062, -9.713775634765625, -126.61213684082031, 22.367141723632812, 63.0958366394043, 127.31909942626953, 30.39276885986328, -40.24767303466797, 169.35345458984375, 155.15518188476562, 1.3550567626953125, -7.60699462890625, -7.03302001953125, 153.5274658203125, 138.2570343017578, 22.702926635742188, 159.35015869140625, 70.60842895507812, -37.23828125, 10.38568115234375, 113.78421020507812, 71.52668762207031, -147.08343505859375, 12.934650421142578, 39.595306396484375, -28.39373779296875, 194.961181640625, 134.45204162597656, 114.83094787597656, -7.7550811767578125, -35.1766357421875, 25.05810546875, -0.9952812194824219, -0.017572402954101562, -0.7938671112060547, -188.11920166015625, -88.6521987915039, 133.7520294189453, -162.9834747314453, 152.6130828857422, 90.83322143554688, -184.98025512695312, -31.99713134765625, 2.99530029296875, 144.758056640625, 8.620189666748047, 40.95597839355469, -14.0982666015625, 33.793670654296875, 122.81576538085938, 80.09698486328125, 172.3516845703125, 176.60601806640625, 64.367919921875, 212.92245483398438, 88.9254150390625, 124.63259887695312, 16.870635986328125, 51.48968505859375, 28.224533081054688, 35.17901611328125, 33.01829528808594, 109.48394775390625, 89.37673950195312, 38.0616455078125, 59.9210205078125, 153.81004333496094, 137.57183837890625, 30.561019897460938, 39.0740966796875, 31.158584594726562, 173.00218200683594, 32.818603515625, -134.80010986328125, 27.42242431640625, 19.239471435546875, 30.3095703125, 84.0224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000474.npy"}
{"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 52.17831802368164, "std": 90.4847640991211, "min": -198.77328491210938, "p10": -55.222452545166, "median": 35.564849853515625, "p90": 166.9704376220703, "max": 253.37875366210938, "pos_frac": 0.7109375, "sample": [197.05014038085938, 164.94540405273438, -88.19881439208984, 161.4484100341797, 97.75658416748047, 21.32098388671875, -198.77328491210938, 214.91024780273438, -65.10118865966797, -49.104034423828125, -11.273574829101562, -82.92608642578125, -50.98870849609375, -48.208717346191406, 13.052978515625, 136.85067749023438, 135.22206115722656, 158.46571350097656, 167.78671264648438, -10.448074340820312, 195.01785278320312, -14.682525634765625, 74.29991912841797, 64.19696807861328, 178.5034942626953, 161.6325225830078, 151.62342834472656, -9.436309814453125, 229.24951171875, -2.047098159790039, 69.20801544189453, 41.72119140625, -71.53509521484375, 162.8049774169922, 45.50215148925781, 20.16242218017578, -108.42235565185547, 29.822494506835938, 1.4639720916748047, 166.62060546875, 142.40866088867188, 50.78968811035156, -139.94764709472656, 72.15724182128906, 91.41557312011719, -14.379119873046875, 126.42217254638672, -10.41986083984375, 157.0595703125, -28.067596435546875, 14.893634796142578, -94.48242950439453, 160.368408203125, 149.26211547851562, -117.06857299804688, 165.81402587890625, 189.35824584960938, -2.001554489135742, 20.860761642456055, 99.64654541015625, 158.33326721191406, 27.170181274414062, 7.442329406738281, -23.83148956298828, 29.375732421875, 50.22064208984375, 20.766754150390625, 48.229339599609375, 8.60614013671875, 30.66009521484375, 16.942047119140625, 172.85006713867188, 77.80076599121094, -10.792724609375, 140.696533203125, -40.896240234375, 1.2709674835205078, 253.37875366210938, 36.15191650390625, -1.7731704711914062, 46.93879699707031, 203.49658203125, 160.07305908203125, -23.565399169921875, 155.19613647460938, -45.61326217651367, 64.42575073242188, 7.59478759765625, 55.568603515625, 24.01123809814453, 0.20449447631835938, 48.87099075317383, 128.17941284179688, 139.18344116210938, -2.240997314453125, 161.6361541748047, 147.5250244140625, 169.37173461914062, 26.049362182617188, 119.16212463378906, 12.152763366699219, -114.74163818359375, 37.02593994140625, 0.37359619140625, -8.878662109375, -32.427001953125, -112.42886352539062, 37.08502197265625, 71.08351135253906, 79.90917205810547, 95.82926940917969, 112.02101135253906, -2.9688148498535156, 13.8426513671875, 159.6165771484375, -34.36181640625, 93.95225524902344, -67.50996398925781, -95.6243896484375, 20.31219482421875, 180.6103515625, 34.977783203125, 19.476097106933594, 203.41943359375, 88.55282592773438, 10.681976318359375, -5.638500213623047, 10.22845458984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000475.npy"}
{"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 61.14632797241211, "std": 94.19873046875, "min": -177.97396850585938, "p10": -30.47313537597656, "median": 35.833984375, "p90": 182.21788635253907, "max": 245.55169677734375, "pos_frac": 0.7109375, "sample": [-126.09014892578125, -126.19009399414062, -4.2986602783203125, -38.593505859375, 19.7236328125, 152.31472778320312, -1.6144561767578125, -9.127567291259766, 18.113800048828125, 111.25599670410156, 139.9999237060547, 88.74456787109375, 172.5168914794922, 204.90997314453125, 124.80388641357422, 212.1192626953125, -31.672393798828125, 189.2992706298828, 149.13307189941406, -6.002239227294922, -8.90582275390625, 37.6243896484375, 21.8218994140625, 26.278350830078125, 138.07968139648438, -16.718551635742188, 170.89816284179688, -3.6953182220458984, -177.97396850585938, 193.9152374267578, 15.89825439453125, 24.518035888671875, -40.20133972167969, 193.80487060546875, 171.71902465820312, 13.44036865234375, 181.33578491210938, -79.19314575195312, 28.154998779296875, -29.95916748046875, -0.504638671875, -176.8988037109375, 134.8443603515625, 22.243209838867188, 101.71757507324219, 36.06610107421875, 143.28463745117188, 193.51907348632812, 12.588478088378906, -74.3284912109375, 119.5343017578125, 173.9713134765625, 10.53759765625, 10.465476989746094, -1.2186717987060547, 203.53561401367188, 11.946624755859375, -15.47296142578125, 176.4472198486328, 30.042022705078125, 159.378662109375, 118.55461883544922, -24.557098388671875, 177.8243408203125, 12.97589111328125, 194.05563354492188, 191.68212890625, 12.486083984375, 144.20834350585938, 68.2847900390625, 31.3094482421875, -29.805633544921875, 225.4207305908203, 37.68243408203125, 53.61992645263672, -1.7325286865234375, 146.01263427734375, -140.62969970703125, 162.5618896484375, 61.87115478515625, 153.4340057373047, 145.171875, 144.97079467773438, 2.67742919921875, 9.497085571289062, 2.8881378173828125, -4.558143615722656, 142.59873962402344, -173.80360412597656, 127.54088592529297, -11.677093505859375, 118.181884765625, 184.276123046875, 3.5007553100585938, 44.16386413574219, -15.1341552734375, -15.2918701171875, 83.25467681884766, 3.32843017578125, 145.8075714111328, -21.834686279296875, 35.60186767578125, -69.50018310546875, 245.55169677734375, 20.13787841796875, 28.430618286132812, -8.10595703125, 57.06501770019531, 115.93968200683594, 108.78314208984375, 82.63232421875, 156.74945068359375, 151.53802490234375, 30.407363891601562, 5.587921142578125, -21.022369384765625, 127.9912109375, -24.169357299804688, -4.6045989990234375, 126.66494750976562, 172.11764526367188, 139.95462036132812, -95.46429443359375, 178.7672576904297, -29.835784912109375, 186.19415283203125, 147.6105499267578, 79.00254821777344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000476.npy"}
{"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 66.19078063964844, "std": 99.39743041992188, "min": -195.85678100585938, "p10": -45.87492218017578, "median": 52.34887504577637, "p90": 186.89508056640625, "max": 238.49456787109375, "pos_frac": 0.7109375, "sample": [-113.111328125, 56.002685546875, 157.50709533691406, -113.03851318359375, 187.66070556640625, 36.91668701171875, -87.79916381835938, 138.95623779296875, -27.666702270507812, 112.50970458984375, -8.719696044921875, 183.94082641601562, -1.1752777099609375, 165.318359375, 59.092437744140625, -15.683929443359375, 214.29522705078125, 98.73953247070312, -195.85678100585938, -45.71684265136719, -4.6142578125, -41.694305419921875, 57.58378601074219, 182.7088623046875, 188.7056121826172, 191.62890625, 146.16253662109375, -10.621234893798828, 15.088821411132812, 30.21160888671875, 156.0514678955078, 179.74319458007812, 26.551666259765625, 222.36578369140625, 137.37042236328125, 42.600555419921875, 221.5594482421875, 33.399436950683594, -12.637481689453125, 0.0, 17.67327880859375, 166.25096130371094, 11.851104736328125, -30.470428466796875, 138.0936279296875, -12.495086669921875, 9.97149658203125, 151.97637939453125, 161.64715576171875, -68.44100952148438, -28.29846954345703, 13.098297119140625, -28.268692016601562, 174.89495849609375, 32.711669921875, 14.51123046875, 157.74539184570312, 28.391265869140625, 160.82830810546875, 38.62178039550781, 147.07611083984375, 173.86245727539062, 238.49456787109375, -12.476348876953125, 122.43887329101562, 132.14691162109375, 196.71963500976562, -119.64459228515625, 78.67664337158203, -42.830657958984375, 129.945556640625, 171.66604614257812, 128.2680206298828, 152.50369262695312, 96.18841552734375, 186.56695556640625, -43.98152160644531, 8.534866333007812, 1.719024658203125, 161.45074462890625, -24.80841064453125, 169.42503356933594, 117.61004638671875, 1.9260025024414062, -47.60783386230469, -164.50228881835938, 166.76260375976562, 61.99821853637695, 48.695064544677734, 133.26361083984375, -3.2140378952026367, 14.608963012695312, -14.917949676513672, 206.06307983398438, -7.95611572265625, 163.62721252441406, -126.67988586425781, 144.89263916015625, 36.824127197265625, 162.80804443359375, -77.377685546875, 230.0621337890625, -22.214736938476562, -156.45315551757812, 17.21087646484375, -21.636199951171875, -46.2437744140625, -2.7010498046875, 202.58828735351562, 92.41891479492188, 42.44873046875, 226.18609619140625, 179.44989013671875, 189.63873291015625, 1.6786346435546875, 167.74832153320312, 176.3553466796875, -71.02442932128906, 174.3015899658203, 13.624755859375, 60.4659423828125, 143.2899169921875, 149.88853454589844, 6.1993408203125, 4.299278259277344, 108.35054779052734, 56.621673583984375, 6.4706878662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.4-s_star-0.35-20260430-140517/margin_logs/step_0000477.npy"}

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2b4e5bed25bd4783993e7d8bed8c0dcb604850687b6133068c9ebbcfe9593fce
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:923d8d34ebce2d992a81fde80eff4a6f422a21f64bc55e9cf132991c17436c41
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:eb403180791aa8a9a910919e998e58cdf97cbac5d5d44510196f2576fbb9cc0f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:02658446b1fa14a7f2804d9588e8252b3fbd8fa3fc043c68bb939a6e25392af0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9488c49904b6678373cc3c7758a552c97db7f8a84804b34aa86ad46440721502
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:962f864bc9125f352ec23a928c916258ad3ca6726b7f1dddf7a4ce72814073ad
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1d917220e34c0d8f080a3e5ba767cf5528a909758a02bf05a153c0da54cee0b5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:90384d8c0c4644bb7d507447e4488f4de16ecb692701e6db25f1e6854968cd52
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f651d46bb7f2443b1506101fcf7563106ae44343d0ab3cea4c8090db753211f3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f7ba361c3b87d03a9240c6c3e3acad7b0276491ff7999a600561c465b46bb1e7
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:06c8466e7362cd1abf3e6b20d05478b7e95cc2d861814fea4b32b342e49d7436
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e7375f559b07d27f8ff529eccac5c136437cc561ca3b6b655a18ca930c5e0ab2
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:fc7afd10fa703085ef9ff2ed3fc7c7a872b6424f65ec9e80ec64fb5792cd5e9e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:82774e3ddfd7091b7d7e95e6af25b4f3b474098e867a5f24848f057ff6bf32f7
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:56e365cd9e65afdd13c375c94b1433ce234302b403ae3d22ee3f6d6230cec913
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a6b240f0d8ae143c084cc89532e20369e1d1066565d96f7cc48ed6c9c6593338
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6c788b4ea14ddb0ed0575d53b063e11a939d38c98837fb02effc063714906e1e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7a137737b3f4306558dc079b21be1565067185f7460c2095c1b4568379c72849
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ae35fde66670d7e211fb761621fc576e55a8a6e9ceb42fa54fe597912eddea97
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0c8d1c2cd92f2256460eb0383d6961cad275cf347c63759380fe0510e1bf39b6
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f8fbd66b6d71a254fd5ac727c7ac41ac3dcd4e9c200db7f3f5af10336ff2565f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3879842a9bb379c673b4ba928689595229c0f22c7fe9ae74d4da0dff2a8000aa
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f83db63a030596050a6f6a7b0c3d54c2f4da79bc4abc091cecf9b6064a5d86ff
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:538d12318cd8ef62333994dc0fe1c829c835a5b133ca49e6a5d1459ec502329b
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c52c0274b0b6a381adbb22cdc50ff53ff92d8962369582ccb7bf36eb5fbcb3a1
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e52decf9582096ff744043b149ab870f2654443a84c65c4402185a8e774fb206
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c97e5611abb141522e7850deddc9a61e741d89e6da5e40957d53850045b37c27
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3c4aba6662cbdfea2ade11e771a2bc8165255d9e2e14b4a9d125a3e50990dba0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4c0dd2894f7092ff2fa56577ab462a02a6041c277bb35f9ab70bf7fc797d55ea
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0daf93979342d034714780c7249a36778ecc6bfc3f1386d1de031c25a847ee18
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d32475d864778de4cd0751957f929a082d49a6a285c7e0c1cd7fdb1292d9f1ee
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3dc03ac28da1e65c9cbc200d046fbeeea6c7330a1802b7770cd8b8ca25fbbb6b
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:75db5f0df1d5154f9dc6e56e87625af7139db2d822596840f786e708d966c7ab
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6986d0dc57aafb5421fd00c1e7255b8624e085ccfac2e0ac95aad12476b02931
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bfb51930af3e600156c23bdfde096ab58bbe710a1fe0dea3406f17054677261d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7cd44212c06aaadc94e53c116969a2d0c74bbc259ab926eaa59b17ee2799695f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:485643ad89399fbc981ef141cf7ff7310bea99427cf1dd1eb7bb8e15d4381dd3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2860dd5f9d6def68e7e2d4f6fe7f8dfd8985f9fbab18fe34f7d058dc6300e0f5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:00c86c6d5f0f318a1e1b5818903fdfa302599d63e3dc62f4db910fd93a2ab4af
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a7b61be5ca7747266e5b12df656de8f387e51693bc35cacd56e4c8f619ec748c
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3bec56921169be8a97ff79ccc322e5492a0063731456b2d99ba2f73a38c07b70
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2576333868bfbdb8d555629dc6dec73688baaa960cb22c86bb41761bb7f177b3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:24655c243d79f09d818df1e464aee9d69c5780b24b8db5c88adaf978ef3220f5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a696619030f430c0c604046eae8b4be599f7b989a560137fc2fff2ec1a24be32
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1ed90285ed3117f2a5d0e96dacd79be675a1a993a1c1c529c1dbbbe89dd22c86
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c2fd220177428585ebc75d4bf0179722f419613ff8f1c0074a67f9847eca362e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:496ddca5b60c65b9a0634d127fcd4aa9630edbf0f4cba67b056d15c8a9b87638
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d64cc2f19cff8750f4a13fe9634dd49f213ec730fb195b51f3a1478020131b8a
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2fe3968019b1a4edc55c4e5b8106b99a3df2d0905a2c00babbd757f0e52ea02f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c8f2f5ccd7a51b8a97fbc455512924d21bb2f44efc854856b6f96046d6013ee9
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:fc2d9f0676de5644ac5cdb093ab46ffa2f5090943d42fc3da004eaf5bb5fda40
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2bac66266cd0e80091dccd1ea247440a66e1037ff3714deea06666a9f84445b7
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1347856b82f3d05d090bfa754babcf8171f33a64661bb2aae5eb80b0e75c76ff
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c0b7916d8dbe7883232cd0d1f18a515b7314c873c2c523ac3a3b10377dec590b
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bbf0170b7c14c6d38d1996bbb8a34ac674f1ac627c659a56bb75b2ba07c8b134
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3f2b87955927fa16e86aee807b005dcde9e5ed751103f7d9b23b440bbab74f32
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b735a00a678e9820c84a128606424d5570bad7340229e3e4b04835443d578c96
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d4b08edf1323f8bc7e26c8e792b0f6191606f70efe03b74fff7e73c29c22aa07
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0ef1699ba77c0556b778fd04451d90b05a1d8e4aab33c16a7d10c8dc55e18678
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0adae8f87427685c0b41f00695496bc1f707a523431e2423e76fc41a6db72b0f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:490dae20adb21369fcce5e2c7a24145a730e1b8f3ff7040d2ee94a19f7af19cc
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:584c7b0970b0d24900ae9393a61cb55c1c1f1755e9b232883b57c23b148cae37
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a4c204d46f37a791a7277cdb321a6557aef39fa0ab6d9ea6b961423297c6e745
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:331853c5ecf9f42d18ec04aadd47768794802dbad13a1512f9c0fe1c4569771e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e71be419541bc5f3c83dd81423255111f717fd138e39b6c4075819d92d0dcb08
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:fb5111d800401e2ee428de0ea2ff4ca46064a269c7306c96b35d02f5715a40ad
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a33be3e17b30a7c2be10c146ceb5dfa779638e13ead6e5a6fb3955ee047e81fa
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:538b0692e9e1faf83fd9914b2fe5f0cd5eff574ae84f7e154aa1c65eb875d490
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:21f31671f056e34be70c4027623321900d63b5319bd0656165e0a672f0f4739e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0ab1d7910c691f709a7c4b272e840e18379f261c87847be89a49d6ad0ff8dd78
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:398047ceb043bd76957662b002bc69ff8d5b59b911de5cb02b6f474ac6678312
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:dd9b9469d5319cdac9025782e23c3141cf400928acb96c7f650e7af84959700f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cc8fb2b43d2871ce026eb7d0792d5be784d6991f6c9d80dd2180ca586c4b043a
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:43ca01908a10433fc49a9c957d311545ce08309fd4be5a11ad1a36f46e337593
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a5a100b783f6b49439d7daaf60ff6e5a1f2e3651eadef7c0050647feba4bd0c8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bb6bc80864a8987dca378e5c55f6e94e63a58f7e61b6f8610cf0070c75240b08
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:265e53421c41f735d58383ffbe23c7c12cc46c820744ea8acb08989bc2024d0d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7bfd68f32ba1994d07f50a56b4178ef60873cf0616aa332f012ba6683430501f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:32d1c9e63e8fb9229816f7021d59cf4bb8a0f33a99356ce417f91c40ace4de06
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c359e3864bf41d590af9d06b8e072b4399dfa113ca748238f7bb4271f20f4a2e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ffed058bbd805ee71976e59a1167b1ef431ee6caa134fcc0d81ac62ca3e4181d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9ef2db929f237a98d764309ab634e3f5052589155b916967874cd2a5e22fe4ca
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bdd73e6908010f70c503b29d175f6ec8ac965226a99b5650292f953a46060d89
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d6d39b5a34ec550033690b028dabcf282d8bab6f3648328b53279607b116534c
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d1ad90d9e83d4a20361ff6ef20fb302eb349cd80dd2e2800d405b80714ea9720
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8c8569e02f87a68ce4f1e2abdfe75da4ecf25e7635f5dd2574e0e99e6fc93be5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:53842dfdbc82f14bf87ba6b583ca4b470ae33481a260374e0451e3a72ba72ced
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ef47693993fd3fa4af2d1693fd7b1c65dd2a8e5799b1bdcc0880e0177f1ff653
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5c46c9a1ddfd8de8e9402f170f40b252d68c8a3a3384e462dcf29c99cb9c81b0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bf61b09035a6061b3335136672fc8d41b77b9c69592da160dbf5c68c346458b8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5887c2df119973b707ce5161e9b6cf0bdc6f20775b8e1173fbd101434a1c0613
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cb1486bf237f6d47832040d315168c20550642c36a1b2bb194a5f33d5340de39
size 640

Some files were not shown because too many files have changed in this diff Show More