初始化项目,由ModelHub XC社区提供模型

Model: W-61/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-05-04 21:47:48 +08:00
commit ba2948d9f0
498 changed files with 17185 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: W-61/llama-3-8b-base-sft-ultrachat-8xh200
tags:
- alignment-handbook
- new-dpo
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551
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. -->
# llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551
This model is a fine-tuned version of [W-61/llama-3-8b-base-sft-ultrachat-8xh200](https://huggingface.co/W-61/llama-3-8b-base-sft-ultrachat-8xh200) on the HuggingFaceH4/ultrafeedback_binarized dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5654
- Fcm Dpo/beta: 0.0053
- Margin Dpo/margin Mean: 76.3970
- Margin Dpo/margin Std: 117.3571
- Logps/chosen: -452.7190
- Logps/rejected: -508.2192
- Logps/ref Chosen: -287.8268
- Logps/ref Rejected: -266.9300
- Kl/chosen Kl Mean: -164.8922
- Kl/rejected Kl Mean: -241.2892
- Kl/mean: -203.0907
- Kl/std: 105.0539
- Logits/chosen: -0.8540
- Logits/rejected: -0.8401
## 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 |
|:-------------:|:------:|:----:|:---------------:|:------------:|:----------------------:|:---------------------:|:------------:|:--------------:|:----------------:|:------------------:|:-----------------:|:-------------------:|:---------:|:--------:|:-------------:|:---------------:|
| 4.7729 | 0.4188 | 200 | 0.5649 | 0.0093 | 45.6673 | 71.5906 | -366.5140 | -391.2846 | -287.8268 | -266.9300 | -78.6872 | -124.3545 | -101.5209 | 64.7014 | -0.8783 | -0.8629 |
| 4.5768 | 0.8377 | 400 | 0.5654 | 0.0053 | 76.3970 | 117.3571 | -452.7190 | -508.2192 | -287.8268 | -266.9300 | -164.8922 | -241.2892 | -203.0907 | 105.0539 | -0.8540 | -0.8401 |
### Framework versions
- Transformers 4.51.0
- Pytorch 2.3.1+cu121
- Datasets 2.21.0
- Tokenizers 0.21.4

27
all_results.json Normal file
View File

@@ -0,0 +1,27 @@
{
"epoch": 0.9989528795811519,
"eval_KL/chosen_KL_mean": -174.78041076660156,
"eval_KL/mean": -214.0789337158203,
"eval_KL/rejected_KL_mean": -253.3774871826172,
"eval_KL/std": 108.87296295166016,
"eval_fcm_dpo/beta": 0.004879707470536232,
"eval_logits/chosen": -0.8583829998970032,
"eval_logits/rejected": -0.8428274393081665,
"eval_logps/chosen": -462.607177734375,
"eval_logps/ref_chosen": -287.8268127441406,
"eval_logps/ref_rejected": -266.9300231933594,
"eval_logps/rejected": -520.3075561523438,
"eval_loss": 0.5694514513015747,
"eval_margin_dpo/margin_mean": 78.59707641601562,
"eval_margin_dpo/margin_std": 121.47869110107422,
"eval_runtime": 78.2536,
"eval_samples": 2000,
"eval_samples_per_second": 25.558,
"eval_steps_per_second": 3.195,
"total_flos": 0.0,
"train_loss": 4.705121593155451,
"train_runtime": 6559.8554,
"train_samples": 61135,
"train_samples_per_second": 9.32,
"train_steps_per_second": 0.073
}

29
config.json Normal file
View File

@@ -0,0 +1,29 @@
{
"architectures": [
"LlamaForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"bos_token_id": 128000,
"eos_token_id": 128001,
"head_dim": 128,
"hidden_act": "silu",
"hidden_size": 4096,
"initializer_range": 0.02,
"intermediate_size": 14336,
"max_position_embeddings": 8192,
"mlp_bias": false,
"model_type": "llama",
"num_attention_heads": 32,
"num_hidden_layers": 32,
"num_key_value_heads": 8,
"pretraining_tp": 1,
"rms_norm_eps": 1e-05,
"rope_scaling": null,
"rope_theta": 500000.0,
"tie_word_embeddings": false,
"torch_dtype": "float32",
"transformers_version": "4.51.0",
"use_cache": true,
"vocab_size": 128256
}

21
eval_results.json Normal file
View File

@@ -0,0 +1,21 @@
{
"epoch": 0.9989528795811519,
"eval_KL/chosen_KL_mean": -174.78041076660156,
"eval_KL/mean": -214.0789337158203,
"eval_KL/rejected_KL_mean": -253.3774871826172,
"eval_KL/std": 108.87296295166016,
"eval_fcm_dpo/beta": 0.004879707470536232,
"eval_logits/chosen": -0.8583829998970032,
"eval_logits/rejected": -0.8428274393081665,
"eval_logps/chosen": -462.607177734375,
"eval_logps/ref_chosen": -287.8268127441406,
"eval_logps/ref_rejected": -266.9300231933594,
"eval_logps/rejected": -520.3075561523438,
"eval_loss": 0.5694514513015747,
"eval_margin_dpo/margin_mean": 78.59707641601562,
"eval_margin_dpo/margin_std": 121.47869110107422,
"eval_runtime": 78.2536,
"eval_samples": 2000,
"eval_samples_per_second": 25.558,
"eval_steps_per_second": 3.195
}

9
generation_config.json Normal file
View File

@@ -0,0 +1,9 @@
{
"bos_token_id": 128000,
"do_sample": true,
"eos_token_id": 128001,
"max_length": 4096,
"temperature": 0.6,
"top_p": 0.9,
"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": -0.02793477475643158, "std": 0.6237562894821167, "min": -2.3379974365234375, "p10": -0.7331054687499999, "median": -0.01795196533203125, "p90": 0.5892669677734375, "max": 1.66827392578125, "pos_frac": 0.4921875, "sample": [0.1674652099609375, 0.0925140380859375, 0.15960693359375, -0.04441070556640625, 1.4956512451171875, 0.2601776123046875, -0.199493408203125, 0.20355224609375, -0.645050048828125, 0.5536956787109375, 1.0389938354492188, -0.570037841796875, 0.210723876953125, 0.2867774963378906, 0.47210693359375, -0.19388961791992188, 0.446380615234375, -1.2149658203125, 0.22918701171875, 1.66827392578125, -0.710540771484375, -0.70892333984375, 0.5078010559082031, 0.137725830078125, -1.50103759765625, -0.093780517578125, -0.19947052001953125, 0.131744384765625, 0.47847747802734375, 0.55963134765625, -0.65667724609375, 0.0279388427734375, -0.722320556640625, -0.1856689453125, -0.7859954833984375, -0.54620361328125, -0.29620361328125, 0.31866455078125, -0.099853515625, -0.62738037109375, 0.5283050537109375, 0.406494140625, -0.2256622314453125, 0.79986572265625, 0.8321533203125, -0.26869964599609375, -0.19608306884765625, -0.14408111572265625, -0.6990966796875, -1.551727294921875, -0.18653106689453125, 0.19101333618164062, -0.385894775390625, -0.2215576171875, 0.587554931640625, 0.01473236083984375, 0.3006134033203125, 0.037418365478515625, -0.1775360107421875, -0.0821533203125, 0.2209625244140625, -0.175567626953125, -1.319793701171875, -0.29390716552734375, -0.7856903076171875, -1.133575439453125, -2.3379974365234375, -0.582763671875, -0.407562255859375, 0.3812713623046875, -0.08927154541015625, 0.34177398681640625, 0.3060302734375, 0.201385498046875, 1.053680419921875, 0.5845870971679688, -1.07904052734375, 0.0, -0.0983428955078125, 0.143341064453125, -0.0359039306640625, 1.14263916015625, -0.6820068359375, -0.6137542724609375, 0.59326171875, 0.4530029296875, 0.039764404296875, -0.214813232421875, 0.38768768310546875, 0.952484130859375, 0.9595947265625, 0.2894439697265625, 0.4634552001953125, 0.46490478515625, -0.1591033935546875, -0.5848388671875, -0.758270263671875, 0.43994140625, -0.9514617919921875, -0.097900390625, -0.6257400512695312, -0.23140716552734375, 0.3023681640625, 1.04931640625, -0.2509117126464844, 0.964508056640625, -0.245361328125, 0.3499603271484375, 0.4720001220703125, 0.113616943359375, 0.24009323120117188, -0.6871795654296875, -0.339508056640625, -0.3977317810058594, -1.02557373046875, 0.550262451171875, -0.042972564697265625, 0.3683624267578125, -0.143951416015625, 0.9142074584960938, -0.1939697265625, 0.131011962890625, -1.03314208984375, 0.1519756317138672, -0.14737701416015625, 0.253753662109375, 0.50762939453125, -0.6018524169921875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000001.npy"}
{"epoch": 0.0020942408376963353, "step": 2, "batch_size": 128, "mean": 0.014312252402305603, "std": 0.6668511629104614, "min": -1.97247314453125, "p10": -0.8530120849609375, "median": 0.042736053466796875, "p90": 0.8162986755371093, "max": 2.1260986328125, "pos_frac": 0.515625, "sample": [-0.946807861328125, 0.58673095703125, -0.07851600646972656, 0.3317985534667969, -0.2125396728515625, -0.79388427734375, -0.061969757080078125, 0.03466033935546875, -1.2871551513671875, 0.349517822265625, -0.5836944580078125, -0.7088851928710938, 0.17462158203125, -0.19950103759765625, -0.4486236572265625, 0.1865081787109375, 0.31939697265625, 0.51416015625, 0.603607177734375, -0.423065185546875, 0.29950523376464844, 0.19829559326171875, -0.7109375, 0.547821044921875, -1.3863983154296875, -0.48331451416015625, -0.283416748046875, 0.917938232421875, 0.440765380859375, -0.3415374755859375, -0.322021484375, -0.1216583251953125, -0.14400482177734375, -1.139923095703125, -0.21550750732421875, -0.4138031005859375, 1.39569091796875, -0.739013671875, -1.128082275390625, 0.9674072265625, 0.99932861328125, 0.067962646484375, 0.8251266479492188, 0.1144866943359375, -0.85760498046875, -0.9853515625, 0.0034122467041015625, -0.28753662109375, -0.39898681640625, 0.06365966796875, 1.42681884765625, -1.2214508056640625, 2.1260986328125, 0.19042205810546875, 0.0, -0.1408672332763672, 0.07848358154296875, -0.04154014587402344, -1.97247314453125, -0.28961181640625, 0.4712066650390625, 0.6232757568359375, -0.58184814453125, 0.06168365478515625, 0.069488525390625, 0.7487945556640625, -1.0627593994140625, 0.75299072265625, -0.244537353515625, 0.9249114990234375, -0.5446319580078125, -0.7351150512695312, 0.09808349609375, 0.936187744140625, -0.0081024169921875, 0.050811767578125, 0.619873046875, 0.8125152587890625, -0.04132080078125, 0.397979736328125, 0.41588592529296875, -0.06989288330078125, -0.0491485595703125, 0.23805618286132812, 0.6832046508789062, 0.138214111328125, 1.1375579833984375, 0.3428459167480469, 0.33001708984375, 0.06795883178710938, -0.851043701171875, 0.13287353515625, -0.1810455322265625, 1.65087890625, -0.05999755859375, 0.15135574340820312, -0.0111846923828125, 0.565032958984375, -0.2957305908203125, -0.1913604736328125, 0.564849853515625, -0.92486572265625, 0.498077392578125, -0.666595458984375, 0.482147216796875, 0.06561279296875, 0.35321044921875, 0.1884613037109375, -0.12225723266601562, 1.531219482421875, 0.1341400146484375, 0.465179443359375, 0.7451171875, 1.089202880859375, 0.103302001953125, -0.5128173828125, 0.10821533203125, -0.3601531982421875, -0.32855224609375, -0.611328125, 0.48419189453125, -1.094207763671875, -0.7486114501953125, -0.090087890625, -0.925384521484375, -0.3560943603515625, -0.7297210693359375, 0.6012115478515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000002.npy"}
{"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.031267568469047546, "std": 0.8380438685417175, "min": -2.111053466796875, "p10": -0.9439788818359375, "median": -0.021651268005371094, "p90": 1.1467407226562496, "max": 2.515533447265625, "pos_frac": 0.453125, "sample": [-1.28814697265625, -0.07103538513183594, -0.05497169494628906, -0.12131500244140625, 0.13126373291015625, 0.2555694580078125, -0.668975830078125, 2.01055908203125, 0.0, -0.3183441162109375, -1.412109375, 0.003631591796875, 0.347503662109375, -0.98388671875, 0.106353759765625, -0.3690071105957031, -0.006103515625, 1.0726318359375, -0.5364532470703125, -0.49603271484375, 0.118621826171875, -1.077972412109375, 0.9483642578125, 0.2100677490234375, 1.560699462890625, 0.5394668579101562, 1.82513427734375, -0.7479248046875, 0.120880126953125, -0.412628173828125, -0.00499725341796875, -0.2504901885986328, -0.047119140625, 1.113067626953125, -1.26611328125, 0.7889251708984375, 0.8341827392578125, 1.58978271484375, -0.2430267333984375, 1.10076904296875, -0.2168426513671875, 1.225311279296875, 0.640960693359375, 0.0060882568359375, -0.0556640625, 0.32293701171875, 0.7973785400390625, 1.5943145751953125, 0.0726318359375, 0.619476318359375, 1.936279296875, 0.0882415771484375, 0.14508056640625, -0.16872215270996094, -1.796539306640625, 0.752838134765625, -0.95147705078125, 1.9586181640625, -0.14643096923828125, -0.06540679931640625, 0.0, 1.587890625, 2.515533447265625, -0.9515228271484375, -0.561920166015625, -0.3231201171875, -0.343505859375, -0.87841796875, -0.059295654296875, 0.14874649047851562, -1.67828369140625, -0.5377578735351562, -0.812042236328125, -0.440887451171875, 0.1028289794921875, -1.2353515625, 0.165618896484375, -0.8890380859375, -0.61175537109375, -0.1220703125, -1.8140106201171875, 0.3423309326171875, 0.9155960083007812, -0.016530990600585938, -0.087615966796875, -2.111053466796875, -1.171234130859375, 0.6122589111328125, 1.227203369140625, 0.2709197998046875, 0.04858970642089844, -0.39849853515625, 0.043701171875, 0.839111328125, -0.000240325927734375, -0.230072021484375, -0.52569580078125, 0.29656219482421875, -0.345306396484375, -0.034450531005859375, -0.2574920654296875, -0.8118438720703125, 0.06622123718261719, 0.2177886962890625, -0.0840606689453125, 1.005401611328125, 0.0370941162109375, -0.221923828125, 1.3839111328125, 0.4352378845214844, -0.842254638671875, -0.8394317626953125, 1.979827880859375, 0.4345703125, -0.3636627197265625, -0.40975189208984375, -0.427490234375, -0.5615692138671875, -0.320465087890625, -0.07671356201171875, -0.940765380859375, -0.3389091491699219, 0.12660980224609375, -0.48377227783203125, -0.02677154541015625, 0.854766845703125, 0.01410675048828125, 0.45648193359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000003.npy"}
{"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": 0.014057323336601257, "std": 0.7843911647796631, "min": -2.80841064453125, "p10": -0.896697998046875, "median": 0.05974006652832031, "p90": 0.9421966552734374, "max": 2.08056640625, "pos_frac": 0.546875, "sample": [0.5777435302734375, 0.1022796630859375, -0.06212425231933594, -0.56561279296875, 1.00152587890625, 0.07903289794921875, 0.060260772705078125, -0.11956787109375, -0.629119873046875, -1.027923583984375, 0.8026123046875, 0.7906494140625, 0.2710113525390625, 0.917633056640625, 0.16495513916015625, -0.5012054443359375, -0.18646240234375, 0.343780517578125, 0.24261474609375, 1.95355224609375, -0.698028564453125, -1.07440185546875, 0.1237030029296875, 0.568023681640625, -0.8204345703125, -0.1162261962890625, 0.683807373046875, 0.01313018798828125, 0.18963623046875, 0.13036346435546875, 0.946319580078125, 0.2847003936767578, -0.36212158203125, -0.92352294921875, -0.173583984375, -0.23052978515625, 0.8025360107421875, 0.4698486328125, 0.255584716796875, 0.226226806640625, 0.0008068084716796875, -0.18502044677734375, 0.08452606201171875, -0.086761474609375, -0.915283203125, 0.050506591796875, 0.5007476806640625, 0.32561492919921875, -1.554473876953125, 0.0062255859375, -0.74285888671875, 0.4923095703125, -0.7332763671875, 0.1400146484375, -0.6456756591796875, 0.0592193603515625, 0.104278564453125, -0.2489013671875, -0.3109130859375, -0.307647705078125, 0.255950927734375, -0.430145263671875, 0.5121612548828125, 0.18499755859375, 1.0537109375, 0.743255615234375, -0.071624755859375, 0.31781005859375, -0.07161712646484375, -0.073974609375, -0.88873291015625, 0.60430908203125, 1.830322265625, 0.60601806640625, -0.295013427734375, 0.566864013671875, 1.05462646484375, 0.166656494140625, 1.16424560546875, 0.5223617553710938, -0.242889404296875, -0.06635284423828125, 0.16748046875, -1.559326171875, 1.2594451904296875, 0.37310791015625, 0.85894775390625, -0.646087646484375, 1.1846923828125, 0.56781005859375, -0.383544921875, -1.7186279296875, 0.1512298583984375, -0.6855621337890625, -1.480010986328125, 0.69586181640625, -0.11541748046875, 0.12482452392578125, -0.886962890625, 0.58123779296875, 1.263763427734375, -0.025482177734375, -0.204864501953125, -0.229644775390625, -0.09563446044921875, 1.074249267578125, -0.45147705078125, 0.9404296875, -0.6405029296875, -2.80841064453125, 0.0042572021484375, -0.345611572265625, -0.8049468994140625, -2.1396484375, -0.41864013671875, -0.99560546875, -0.403564453125, 0.620452880859375, 2.08056640625, 0.5211410522460938, 0.3208160400390625, -1.00933837890625, -2.09039306640625, -0.64764404296875, 0.424560546875, -0.577972412109375, 0.82098388671875, 1.141357421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000004.npy"}
{"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": -0.09676751494407654, "std": 0.8048688173294067, "min": -2.527008056640625, "p10": -1.0744827270507813, "median": -0.13527679443359375, "p90": 0.8201629638671875, "max": 2.2528076171875, "pos_frac": 0.4609375, "sample": [0.9181289672851562, 0.7486572265625, -0.4413909912109375, 1.52789306640625, -1.1087799072265625, 0.551483154296875, -0.93011474609375, -0.5218658447265625, 0.3164520263671875, -0.243377685546875, -0.846893310546875, -0.064697265625, 0.17481231689453125, 0.37621307373046875, -0.1549530029296875, -1.92303466796875, 0.21524810791015625, 0.80975341796875, 0.018798828125, -0.59320068359375, 0.5089111328125, -0.1220703125, 0.034099578857421875, -0.59356689453125, 1.341583251953125, -0.72412109375, -1.406890869140625, 0.05438995361328125, 0.4376068115234375, -0.16168212890625, -0.718475341796875, -0.3076324462890625, 0.10137939453125, -1.22216796875, -0.1484832763671875, 0.775665283203125, -0.3787841796875, -0.1975555419921875, -0.725738525390625, 0.60760498046875, 0.766387939453125, -1.2200927734375, 0.19757080078125, 0.305206298828125, -0.942718505859375, 0.2131195068359375, 1.405242919921875, 0.4385986328125, -0.08254432678222656, 1.58056640625, 0.937469482421875, -0.249298095703125, -0.81158447265625, 0.429168701171875, 0.163909912109375, 1.8443603515625, -0.607666015625, -0.4388580322265625, -0.518951416015625, 0.1850738525390625, 0.2864990234375, 0.227569580078125, 0.0004730224609375, 0.7201156616210938, -1.1615486145019531, 0.12421417236328125, -0.7943572998046875, 0.759613037109375, -0.63885498046875, 0.511993408203125, -0.4091796875, -1.3127059936523438, -0.9027023315429688, -0.21480560302734375, 0.034664154052734375, 0.7526397705078125, -1.059783935546875, -0.8564453125, 0.50262451171875, -0.3701057434082031, -0.187591552734375, -1.6954345703125, -1.405364990234375, 0.268463134765625, -0.71673583984375, -2.02642822265625, 0.88958740234375, 0.611297607421875, 1.60919189453125, 0.36871337890625, -1.624542236328125, 0.07500267028808594, 1.668975830078125, 0.008392333984375, 0.844451904296875, 0.242340087890625, 0.8642578125, 0.6764602661132812, -0.85125732421875, -0.913787841796875, 0.52166748046875, -0.89794921875, -0.36893463134765625, -0.075439453125, -0.254608154296875, -2.527008056640625, -0.1004638671875, -0.288848876953125, -0.1626434326171875, 0.199188232421875, -0.2298431396484375, 0.36370849609375, -0.4189300537109375, -0.425201416015625, 0.2063446044921875, -0.3517608642578125, -0.17319488525390625, -0.19012451171875, -1.300567626953125, 0.0589599609375, -0.433135986328125, -0.43851470947265625, -0.8970947265625, -0.45862579345703125, 2.2528076171875, 0.0478668212890625, -0.1655120849609375, -0.362457275390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000005.npy"}
{"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": 0.15789739787578583, "std": 0.7085481882095337, "min": -1.73492431640625, "p10": -0.6399253845214843, "median": 0.1323986053466797, "p90": 1.1030853271484375, "max": 2.843017578125, "pos_frac": 0.5859375, "sample": [-0.6997833251953125, -0.828338623046875, -1.3643875122070312, 1.8658447265625, -0.40704345703125, 0.259918212890625, 0.59625244140625, -0.12753677368164062, -0.36907958984375, 0.2119598388671875, -0.2258148193359375, 0.2214508056640625, 1.12884521484375, 0.059967041015625, -0.2137908935546875, 0.17736053466796875, -1.0443115234375, -0.12081146240234375, -0.08068656921386719, 1.102874755859375, -0.19354248046875, 0.13538360595703125, -0.68035888671875, 1.0816192626953125, -0.4855194091796875, 0.05792236328125, -0.233673095703125, 0.521942138671875, 1.1549072265625, 0.0033416748046875, -0.5340194702148438, 0.61737060546875, 0.0219573974609375, -0.022186279296875, 0.36981964111328125, -0.874267578125, 1.1963653564453125, 0.5468292236328125, 0.22760009765625, 0.161376953125, 0.40167236328125, -0.59124755859375, -0.519683837890625, 1.094696044921875, -0.18087196350097656, 0.4409637451171875, 0.12941360473632812, 0.628387451171875, 0.454620361328125, 1.1990966796875, 0.3653602600097656, -1.6201324462890625, -0.490264892578125, 0.4878082275390625, -1.444671630859375, -0.225128173828125, -0.7119140625, 0.8034820556640625, 0.44091796875, 2.843017578125, -0.21212005615234375, 1.433990478515625, 0.0650634765625, -0.6225967407226562, 0.660888671875, 0.3805694580078125, -0.019412994384765625, 0.25555419921875, 0.16253662109375, 0.6353607177734375, 1.10357666015625, -0.7830657958984375, 0.665191650390625, 0.4334716796875, 0.300079345703125, -0.198333740234375, -0.3330230712890625, 0.0762939453125, 0.0, 0.702667236328125, 0.23760986328125, -0.8583984375, 0.22928619384765625, -0.26784515380859375, 0.7787322998046875, 0.0762939453125, -0.4934844970703125, 1.800018310546875, 0.3086090087890625, 0.19066619873046875, -0.5574951171875, -0.51678466796875, 0.07152938842773438, 1.255462646484375, 1.058990478515625, 0.6465911865234375, 0.516754150390625, 0.25018310546875, -0.1027069091796875, 1.148712158203125, -0.03436279296875, 0.34857177734375, -0.1194000244140625, 1.599365234375, 0.47356414794921875, -0.0157470703125, -0.56488037109375, 0.59088134765625, -0.44225311279296875, 0.00707244873046875, -0.36883544921875, -0.1352252960205078, 1.23602294921875, 0.8073348999023438, 0.43753814697265625, 0.49603271484375, -1.73492431640625, -1.17828369140625, 0.124847412109375, 0.9148712158203125, -0.2836132049560547, -0.275634765625, -0.0491485595703125, -0.327880859375, 0.535491943359375, 0.532470703125, -0.18187522888183594, 0.6181716918945312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000006.npy"}
{"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": -0.04685226082801819, "std": 0.7290181517601013, "min": -2.1793212890625, "p10": -1.057093048095703, "median": 0.001293182373046875, "p90": 0.8444915771484374, "max": 2.2413330078125, "pos_frac": 0.5, "sample": [-0.512939453125, -0.3822021484375, -1.86102294921875, -1.517822265625, 0.4340667724609375, -1.04095458984375, -0.4211273193359375, 1.1981201171875, -0.4830322265625, -0.00601959228515625, -0.4135894775390625, 1.0240859985351562, -0.56463623046875, 0.1683807373046875, 0.2978057861328125, 0.14208984375, 0.0304107666015625, 0.73565673828125, 0.0, 0.1358509063720703, 0.0889739990234375, -1.1334991455078125, 0.24188995361328125, 0.5165863037109375, -0.061492919921875, -0.074005126953125, -0.6263427734375, 0.15819549560546875, 1.025634765625, 0.130889892578125, -0.4346923828125, -1.14422607421875, -0.14131927490234375, 0.2091217041015625, -0.130950927734375, 1.25128173828125, 0.06964874267578125, 0.423675537109375, -0.3153076171875, 0.40641021728515625, -0.4366455078125, 0.02197265625, -0.4303131103515625, -1.2186279296875, 0.498138427734375, 1.572418212890625, 0.2967529296875, 0.00962066650390625, -0.3398704528808594, -1.941192626953125, -0.788360595703125, -0.2528076171875, -1.098724365234375, -0.8795166015625, -1.10955810546875, 0.46419525146484375, -0.307098388671875, -1.119110107421875, 0.21161651611328125, -0.1584014892578125, 0.17049026489257812, 0.863525390625, -0.08649063110351562, 0.345794677734375, 0.549774169921875, -0.792816162109375, -0.3385009765625, -0.061443328857421875, -0.840667724609375, -0.899810791015625, 0.0417327880859375, 0.582183837890625, 0.09797286987304688, -0.217926025390625, 0.00258636474609375, -0.028045654296875, 0.4697685241699219, 0.268951416015625, -0.6781005859375, 0.870635986328125, -1.1446533203125, -0.22791290283203125, -0.9514617919921875, 0.1794891357421875, 0.42572021484375, -1.0947494506835938, 0.61578369140625, 1.328033447265625, -0.139892578125, -0.010059356689453125, -0.5308837890625, 0.19586181640625, -0.31914520263671875, 0.019989013671875, -0.61981201171875, 1.098907470703125, -0.035633087158203125, 0.4322509765625, 0.41053009033203125, -0.117095947265625, 0.4585723876953125, 0.443634033203125, 0.80859375, -0.69317626953125, 0.1433563232421875, 0.01806640625, 1.38494873046875, -0.36346435546875, -0.65948486328125, 1.382568359375, -0.9922332763671875, 0.2853355407714844, 2.2413330078125, 1.1507415771484375, 0.46521759033203125, 0.836334228515625, 0.48199462890625, -0.2577972412109375, 0.655242919921875, -2.1793212890625, -0.50732421875, -0.6044921875, 0.3651885986328125, -0.06974983215332031, -0.5322723388671875, 0.2535400390625, -1.1625213623046875, 0.39711761474609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000007.npy"}
{"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": 0.06536369025707245, "std": 0.7604699730873108, "min": -2.32281494140625, "p10": -0.7010711669921875, "median": 0.009276390075683594, "p90": 1.1825836181640625, "max": 2.370635986328125, "pos_frac": 0.5, "sample": [0.4959716796875, -0.21425628662109375, 0.1622161865234375, -0.55096435546875, -0.74310302734375, 0.33551025390625, -0.912841796875, -0.439300537109375, -0.06709098815917969, 0.545867919921875, -0.04981803894042969, -0.066497802734375, 0.480224609375, 0.820526123046875, 0.17926025390625, 0.0594940185546875, -0.29779052734375, 0.139678955078125, -0.085052490234375, 2.00531005859375, 1.4771728515625, -0.7679443359375, 0.236358642578125, -1.53125, 0.690887451171875, 0.217071533203125, 0.276641845703125, 1.180084228515625, 0.556610107421875, 2.370635986328125, -0.001220703125, -0.76580810546875, 0.40478515625, 0.42645263671875, -0.4090728759765625, -1.510345458984375, 0.18509674072265625, -0.3624267578125, -0.585784912109375, 0.0260009765625, -2.32281494140625, -0.811767578125, -0.6610107421875, -0.1352386474609375, 0.2779541015625, -0.6892242431640625, -0.3006725311279297, -0.00350189208984375, 1.618438720703125, -0.5628662109375, -1.423736572265625, -0.644287109375, 1.971435546875, -0.48166465759277344, 0.9233551025390625, -0.04290771484375, 0.53521728515625, -0.087310791015625, -0.609130859375, -0.54620361328125, 0.22540283203125, -0.3990478515625, -0.62640380859375, -0.90850830078125, -0.655059814453125, 0.5997314453125, 0.1851348876953125, -0.5430679321289062, 0.23309326171875, -0.6934814453125, 0.475921630859375, 1.2393798828125, -0.328369140625, -0.03849029541015625, 0.05908203125, -0.27301025390625, 0.1458740234375, -0.2873954772949219, -0.31261634826660156, 0.99176025390625, 0.3014545440673828, 0.20001220703125, -0.0215301513671875, -0.051422119140625, -0.449859619140625, 0.019773483276367188, 0.604461669921875, 0.0476837158203125, -0.35253143310546875, 1.570281982421875, 1.347076416015625, 1.543853759765625, 1.032135009765625, 0.32091522216796875, -0.004405975341796875, -0.405853271484375, 0.79150390625, 1.18841552734375, -0.4019775390625, 1.08587646484375, 0.23843765258789062, 0.37786865234375, 0.52374267578125, -1.059906005859375, 0.08937835693359375, 0.34255409240722656, 0.19757080078125, 0.22045135498046875, -0.29400634765625, -0.718780517578125, 0.0838775634765625, -0.520172119140625, 1.2064208984375, -0.49176025390625, -1.568634033203125, -0.15474700927734375, 0.284271240234375, 1.423492431640625, -0.512451171875, 0.395416259765625, -0.6635894775390625, 1.6997833251953125, -0.144775390625, 0.19763946533203125, -0.09282684326171875, -0.3215789794921875, 0.308807373046875, -0.05107688903808594], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000008.npy"}
{"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": 0.038522377610206604, "std": 0.7108509540557861, "min": -2.277435302734375, "p10": -0.9022430419921874, "median": 0.07098388671875, "p90": 0.7992996215820312, "max": 2.477996826171875, "pos_frac": 0.5234375, "sample": [0.6158065795898438, 0.4061279296875, 0.8234710693359375, -0.09960556030273438, 0.6761474609375, -1.2474365234375, -0.9561767578125, -0.35784149169921875, 0.48388671875, -0.188232421875, 0.3715667724609375, 0.9644775390625, 1.20513916015625, -1.1633758544921875, -0.117156982421875, 0.5235595703125, 0.384307861328125, -2.277435302734375, -1.00738525390625, 0.5623931884765625, -0.08954620361328125, 2.477996826171875, 0.18555831909179688, 0.31842041015625, -0.49123382568359375, 0.034576416015625, -1.0582275390625, 0.7506866455078125, 0.8843231201171875, 0.26409912109375, 0.7385101318359375, -1.42364501953125, 0.161773681640625, -0.28627777099609375, 0.20623779296875, 0.3849945068359375, 0.5484619140625, -0.3119354248046875, -0.2969512939453125, -0.2093353271484375, 0.9416351318359375, -0.0390472412109375, 1.8173828125, 0.3452301025390625, -0.462066650390625, 0.435638427734375, -0.4283447265625, -0.29327392578125, 0.1617431640625, 0.06085205078125, 1.087158203125, 0.001861572265625, 0.08111572265625, 1.364105224609375, -2.259796142578125, -0.5080986022949219, -0.3365478515625, -1.22015380859375, -0.941741943359375, -0.651519775390625, 1.6179962158203125, 0.16172027587890625, 0.4237060546875, 0.11087417602539062, -0.44623565673828125, -1.2725830078125, 0.3884124755859375, -0.2442626953125, 0.7556610107421875, 0.1743011474609375, 0.52813720703125, -0.07076644897460938, 0.7889404296875, -0.3575019836425781, -0.4266357421875, -0.22972869873046875, 0.6014404296875, 0.37796783447265625, -0.22564697265625, -0.2216033935546875, -0.079620361328125, -1.6187744140625, -0.02545166015625, 0.310699462890625, 0.7202606201171875, 0.36041259765625, 0.0, -0.165283203125, -0.45721435546875, 0.3748779296875, 0.1466064453125, -0.1956310272216797, -0.040252685546875, -0.1801605224609375, -0.0846405029296875, 0.15636062622070312, 0.107025146484375, -0.419952392578125, 0.10076141357421875, -0.5043792724609375, 0.7502975463867188, 0.0, -0.465911865234375, 0.55126953125, -0.348052978515625, 0.14640045166015625, 0.0, 0.76641845703125, 0.5621185302734375, 0.12177658081054688, -0.425750732421875, -0.23138427734375, 0.32049560546875, -0.341400146484375, 0.43304443359375, 0.12641143798828125, -1.071685791015625, -0.33740234375, 1.29400634765625, -0.0767974853515625, -0.705169677734375, -0.88531494140625, 0.12506103515625, 0.39764404296875, -0.347137451171875, 0.9193115234375, 0.292510986328125, 0.8734130859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000009.npy"}
{"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": 0.024518877267837524, "std": 0.6897330284118652, "min": -2.066192626953125, "p10": -0.7695381164550781, "median": 0.0019989013671875, "p90": 1.017115783691406, "max": 1.515869140625, "pos_frac": 0.5, "sample": [1.146087646484375, 0.0618438720703125, 0.052490234375, 0.6706008911132812, -0.0954742431640625, -0.607513427734375, 0.18732070922851562, 1.0900726318359375, 0.003997802734375, 1.3246994018554688, 0.295166015625, 0.22359466552734375, -0.4818267822265625, -1.38909912109375, 0.6209716796875, 0.22324752807617188, 0.218017578125, -0.542327880859375, 1.0641021728515625, 0.0852203369140625, 0.39276123046875, 0.5743865966796875, 0.94586181640625, 1.39190673828125, 1.3975372314453125, -0.2615375518798828, 0.493438720703125, -0.2855949401855469, 0.635711669921875, -0.2780914306640625, -0.810516357421875, -0.61761474609375, -0.14056015014648438, 1.44219970703125, -0.033100128173828125, -0.515228271484375, 0.229766845703125, -0.138275146484375, 0.981781005859375, 1.4600372314453125, 1.515869140625, -0.013916015625, -0.3234405517578125, 0.534423828125, 0.62664794921875, 0.2018890380859375, -0.24239349365234375, -0.5211105346679688, -0.21148681640625, -0.40350341796875, 0.13094329833984375, 0.31272125244140625, -0.05816650390625, -0.101959228515625, -0.1809539794921875, -0.163818359375, 0.285400390625, -1.0488128662109375, 0.07068634033203125, 0.086029052734375, -0.7150421142578125, 0.27758026123046875, -0.239044189453125, 0.2714424133300781, 0.03206634521484375, 0.05499267578125, -0.424530029296875, -0.44244384765625, 0.0968017578125, -0.27569580078125, -0.7659835815429688, 0.796356201171875, -1.6470794677734375, 0.3076019287109375, -0.32659912109375, -0.03282928466796875, -0.46174049377441406, 1.32861328125, 1.08306884765625, 0.32379150390625, 0.6851348876953125, -0.03216552734375, -0.0118408203125, 0.29248046875, 0.08642959594726562, 0.474151611328125, -0.599853515625, -0.16237258911132812, 1.4053955078125, -0.446868896484375, -0.1535797119140625, -0.23221969604492188, 0.5245132446289062, -0.091796875, 0.793548583984375, 0.60302734375, 0.1625823974609375, -0.77783203125, -0.0943603515625, -0.9351959228515625, -0.44989013671875, 0.10429000854492188, 0.03204345703125, -1.0600738525390625, 1.091064453125, -0.001983642578125, -0.5459747314453125, -0.632049560546875, -0.1932220458984375, -1.016693115234375, 0.996978759765625, 0.17170143127441406, -0.2494049072265625, 0.0, -0.337799072265625, -0.56610107421875, 0.017360687255859375, -1.5029296875, 0.08796119689941406, -2.066192626953125, -0.7895355224609375, -0.2246551513671875, -1.75775146484375, 0.39825439453125, 0.60430908203125, -1.5067901611328125, 0.392608642578125, -0.098724365234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000010.npy"}
{"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": -0.07755695283412933, "std": 0.7567636966705322, "min": -2.94720458984375, "p10": -0.931756591796875, "median": -0.01511383056640625, "p90": 0.8551338195800781, "max": 1.7593536376953125, "pos_frac": 0.484375, "sample": [1.1083984375, -0.52227783203125, 0.299652099609375, 0.371917724609375, 0.5069427490234375, -0.359527587890625, 0.13336181640625, -0.894744873046875, 1.62762451171875, 0.298370361328125, 0.8483200073242188, 0.0680999755859375, 0.05914306640625, -1.068267822265625, -0.374481201171875, 0.2596588134765625, -0.356231689453125, -0.052959442138671875, -0.83782958984375, 0.4793548583984375, 0.06060600280761719, 0.0659942626953125, -0.5799560546875, -0.3809814453125, -0.87445068359375, -0.049468994140625, -0.141937255859375, -0.0122528076171875, 0.31097412109375, 0.7536239624023438, -0.22637939453125, 0.279296875, 0.07067108154296875, -2.94720458984375, -0.66387939453125, 0.155426025390625, -0.225799560546875, -0.274871826171875, -0.2395782470703125, 0.060516357421875, 0.416961669921875, -0.017974853515625, -0.333648681640625, -0.806610107421875, -1.482879638671875, 0.2703857421875, -0.07700347900390625, 0.2313079833984375, 0.026611328125, -1.9356689453125, -0.786865234375, 0.87103271484375, -0.3904266357421875, 0.091888427734375, 0.015899658203125, 0.4134521484375, -0.1555023193359375, 0.3551025390625, -0.6497802734375, -0.2655487060546875, -0.1154937744140625, 0.2225341796875, 1.7593536376953125, -2.50146484375, 0.9776611328125, -0.2428436279296875, 0.100830078125, 0.09375, -0.5674362182617188, -0.23796463012695312, -1.52154541015625, -0.966278076171875, 0.651885986328125, -0.71282958984375, -0.0017910003662109375, 1.7180938720703125, 0.19792938232421875, 0.6796722412109375, 0.003875732421875, 0.142333984375, 0.08461761474609375, -0.2061767578125, -0.4196128845214844, 0.9606094360351562, -1.5228729248046875, 0.22530746459960938, 0.064910888671875, 1.491607666015625, 0.355010986328125, -1.0123291015625, -0.5213623046875, 0.17903518676757812, -0.499725341796875, 0.6192665100097656, 0.8082427978515625, -0.4869384765625, 0.419921875, 0.97906494140625, -0.2993507385253906, -0.09440803527832031, -1.2771949768066406, 1.70184326171875, -0.916961669921875, -1.2582550048828125, 0.8204345703125, -0.5811767578125, 1.03375244140625, -1.0365447998046875, -0.55242919921875, -0.6615982055664062, -0.380462646484375, 0.1410064697265625, -0.051727294921875, 0.88824462890625, -0.310272216796875, -0.148956298828125, 0.4489593505859375, -1.096771240234375, 0.019561767578125, 1.01898193359375, 0.50445556640625, -0.3531494140625, -0.8017425537109375, -0.6112823486328125, 0.366302490234375, 0.24042129516601562, -0.2473602294921875, -0.15606689453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000011.npy"}
{"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": 0.007990524172782898, "std": 0.8173657059669495, "min": -2.60650634765625, "p10": -0.9114288330078125, "median": 0.035919189453125, "p90": 0.9665359497070312, "max": 2.189697265625, "pos_frac": 0.5234375, "sample": [1.35516357421875, -0.150177001953125, -0.048004150390625, -1.08660888671875, 0.47264671325683594, -0.35053253173828125, -0.5423431396484375, -0.55841064453125, 1.0142822265625, 0.1753082275390625, 0.12201690673828125, 0.8516845703125, -2.1033935546875, 0.12783050537109375, -0.495147705078125, 0.6344680786132812, 0.09329986572265625, 0.103851318359375, -0.19036865234375, -0.01153564453125, 0.2943115234375, 0.94659423828125, -0.05877685546875, -0.5453033447265625, -0.540313720703125, 0.1572723388671875, -0.24874496459960938, -0.22475814819335938, 2.189697265625, -0.98236083984375, -2.25103759765625, -0.2437896728515625, -1.9047698974609375, -1.32623291015625, 0.36285400390625, 0.55499267578125, -0.21570968627929688, 0.316253662109375, 1.349517822265625, 0.30339813232421875, -0.01969146728515625, 0.05560302734375, -0.212738037109375, -0.19598388671875, 0.03046417236328125, -0.131683349609375, 1.633453369140625, -0.718505859375, 0.086456298828125, -0.2555694580078125, 0.8637237548828125, 0.41064453125, -0.1033172607421875, -0.562530517578125, 0.2029876708984375, -0.21923828125, -0.172515869140625, 0.386474609375, 0.046600341796875, -0.2543487548828125, 0.2878570556640625, -1.33392333984375, 0.821502685546875, 0.11602783203125, -0.0088653564453125, 0.6516571044921875, 0.4866790771484375, 1.11785888671875, 0.6680526733398438, 0.9613494873046875, 0.3936004638671875, -0.350341796875, 0.45652008056640625, -1.111419677734375, 0.9786376953125, 0.09234619140625, 0.22759628295898438, 0.1303863525390625, -0.968658447265625, 0.5071258544921875, -0.32110595703125, 0.0811309814453125, 0.94903564453125, 0.12804412841796875, -0.111419677734375, -0.255523681640625, -0.07061767578125, -0.4879150390625, 1.995849609375, 1.646270751953125, 2.04364013671875, 0.16455078125, -0.6492156982421875, 0.05731201171875, -0.49261474609375, 1.257080078125, 0.11773681640625, 0.09041595458984375, 0.69195556640625, 0.46258544921875, -0.7489013671875, -0.7205352783203125, -0.5143890380859375, -0.1102294921875, -1.58197021484375, 0.935394287109375, -0.8473663330078125, -0.907135009765625, -0.2099151611328125, 0.0169677734375, -2.17352294921875, -2.60650634765625, -0.3612518310546875, -0.3156585693359375, 1.6290130615234375, -0.5894775390625, 0.12421798706054688, 0.132080078125, -0.92144775390625, -0.133514404296875, 0.04071044921875, 0.0311279296875, 0.4931640625, -0.1270294189453125, 1.014892578125, 0.3275871276855469, -0.8289337158203125, 0.336822509765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000012.npy"}
{"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": 0.07708097994327545, "std": 0.7010467648506165, "min": -1.9060211181640625, "p10": -0.7886245727539062, "median": 0.017599105834960938, "p90": 0.8777374267578124, "max": 2.23712158203125, "pos_frac": 0.5390625, "sample": [-1.0448989868164062, 0.3188362121582031, -0.8307075500488281, 2.23712158203125, 1.3733367919921875, 0.1849365234375, 0.7174072265625, -1.0235595703125, 0.1231842041015625, -0.0503692626953125, 0.95904541015625, -0.65753173828125, 0.3523216247558594, 0.00777435302734375, 0.25898170471191406, -0.20507431030273438, 0.515380859375, -0.6690673828125, -0.2138519287109375, -0.26833343505859375, 0.44669342041015625, -0.02789306640625, 0.645751953125, 0.665069580078125, 0.25775146484375, -0.013780593872070312, 0.1707000732421875, 0.436492919921875, -0.206390380859375, 0.454833984375, -0.073455810546875, 0.04193115234375, 0.001129150390625, -0.7771759033203125, 0.9586257934570312, -0.4607086181640625, -0.09823417663574219, -0.191131591796875, 0.10660552978515625, -0.3729248046875, -0.09792327880859375, 0.2306976318359375, 0.17108154296875, -0.12445068359375, 0.39501953125, 0.0189361572265625, -0.6192626953125, -0.27337646484375, 0.736297607421875, 2.0421142578125, -0.5751190185546875, -0.091339111328125, -0.42667388916015625, -1.9060211181640625, -0.2662353515625, 0.0, -0.7353515625, -1.19049072265625, 0.00940704345703125, 0.4159088134765625, 0.6392059326171875, -0.8770294189453125, 0.49402618408203125, 0.12021636962890625, 0.353118896484375, 1.5157470703125, -0.46514892578125, 0.46173095703125, 0.550445556640625, -0.22698020935058594, -0.14013671875, -0.55029296875, 0.47263336181640625, -0.815338134765625, 0.19903564453125, -1.4752197265625, 0.20210838317871094, 0.4437255859375, 0.6783409118652344, 0.8125228881835938, -0.016002655029296875, -0.01131439208984375, 0.05901336669921875, -0.369964599609375, -0.2288360595703125, -0.57830810546875, -0.0189971923828125, 0.10107421875, -0.30181884765625, 0.59649658203125, 1.70416259765625, 0.17911529541015625, 0.452423095703125, 0.5056304931640625, 1.112518310546875, 0.190460205078125, -0.6761703491210938, 1.77850341796875, 0.2998809814453125, -0.306854248046875, 1.158172607421875, 1.004852294921875, -0.21111297607421875, -0.4867401123046875, 0.6649322509765625, -0.824462890625, -0.610443115234375, -0.9385986328125, -0.289886474609375, 0.686859130859375, -0.07910537719726562, 1.7423095703125, -0.524993896484375, 0.016262054443359375, 0.9027099609375, -1.36639404296875, 0.00907135009765625, -1.06365966796875, 0.2882537841796875, 0.732635498046875, 0.6993408203125, 0.867034912109375, 0.42913818359375, 0.2905769348144531, -1.126861572265625, 0.07591629028320312, -0.739593505859375, -0.06561279296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000013.npy"}
{"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": 0.08273807168006897, "std": 0.7090755105018616, "min": -1.55999755859375, "p10": -0.7181819915771483, "median": 0.01276397705078125, "p90": 0.9265304565429687, "max": 2.33282470703125, "pos_frac": 0.5234375, "sample": [1.30706787109375, 0.44802093505859375, -0.3735198974609375, 0.031005859375, 0.6586380004882812, -0.5030670166015625, -0.07122802734375, 1.1796875, 0.07386970520019531, 0.205108642578125, 0.4107513427734375, 0.43210601806640625, 0.1119384765625, -0.34619140625, 1.359161376953125, 0.288177490234375, 1.26397705078125, -0.143463134765625, -0.22109222412109375, 1.5304718017578125, -0.09791183471679688, -0.12671852111816406, -0.02099609375, 0.4188232421875, 0.751708984375, 0.01287841796875, 0.0084381103515625, -0.14391326904296875, -1.55999755859375, -0.6080322265625, 0.1324462890625, -0.78948974609375, 0.58685302734375, 0.12811279296875, 1.0620651245117188, 0.3685302734375, 0.575958251953125, -0.3223552703857422, -0.06927490234375, 0.09375, 0.15250396728515625, -1.473480224609375, 0.06119537353515625, 2.33282470703125, 0.0, 0.13975143432617188, -0.704681396484375, -1.0545654296875, 0.4649238586425781, -0.09259033203125, -0.37713623046875, -0.204010009765625, 0.3148956298828125, 0.63531494140625, 0.7102203369140625, -0.887786865234375, 0.594482421875, -1.14434814453125, -0.431640625, -0.205963134765625, 1.96807861328125, -1.16607666015625, 0.922882080078125, -0.653839111328125, -1.302490234375, 0.7883148193359375, 1.8548736572265625, -0.38775634765625, 0.381256103515625, 0.69390869140625, 0.135162353515625, -0.017547607421875, -0.7496833801269531, 0.263824462890625, 0.260833740234375, -0.30667877197265625, -0.1858673095703125, -0.924468994140625, -0.067596435546875, 0.1226959228515625, -0.13478851318359375, -0.19512939453125, 0.2773895263671875, 0.5295028686523438, -0.1123046875, -0.092620849609375, -0.4196014404296875, -0.02065277099609375, -0.46692657470703125, 0.11839485168457031, -0.5254058837890625, -0.31111907958984375, -1.471954345703125, 0.01971435546875, -0.53656005859375, -1.553070068359375, -0.014926910400390625, -0.31053924560546875, -0.639404296875, -0.7517547607421875, -0.3011474609375, 0.09172821044921875, -0.01422119140625, 0.452392578125, 0.196746826171875, 0.697479248046875, -0.19087982177734375, -0.5620269775390625, 0.9350433349609375, 0.3277435302734375, 0.1596832275390625, 0.68768310546875, 1.885650634765625, 0.3461151123046875, -0.19344329833984375, 0.54705810546875, -0.4887237548828125, 1.0738983154296875, -0.4713287353515625, -0.24813461303710938, 0.0076904296875, 0.541778564453125, 1.84619140625, 0.901824951171875, 0.23235321044921875, 0.0126495361328125, 0.3226318359375, -0.0962371826171875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000014.npy"}
{"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": -0.022674620151519775, "std": 0.7878087162971497, "min": -2.36004638671875, "p10": -0.8428226470947265, "median": -0.06363582611083984, "p90": 0.7831573486328123, "max": 2.896148681640625, "pos_frac": 0.4609375, "sample": [0.4971923828125, 0.0605621337890625, -0.180755615234375, 2.1986083984375, 0.75396728515625, 0.729095458984375, -0.599334716796875, -0.68121337890625, -0.7398834228515625, -0.07224273681640625, -0.41057586669921875, -0.794342041015625, -1.7591705322265625, 0.614898681640625, 0.1930999755859375, -0.930023193359375, -0.445648193359375, 0.03129768371582031, 0.325042724609375, 0.53314208984375, -2.36004638671875, 0.09606170654296875, 0.378143310546875, 0.749603271484375, 1.25347900390625, -0.3118896484375, -1.051422119140625, -0.7059326171875, 0.839019775390625, -0.0799560546875, -1.262359619140625, -0.24176025390625, 0.75921630859375, 0.62261962890625, 0.551177978515625, -0.09076690673828125, -0.17828369140625, -1.552978515625, -0.80438232421875, -0.32584381103515625, 0.59814453125, -0.56549072265625, 0.238494873046875, -0.2460784912109375, -0.2050933837890625, -0.7506256103515625, -0.12345123291015625, 0.6569976806640625, 0.27508544921875, 0.212432861328125, -0.14358139038085938, -0.1248779296875, -0.68927001953125, -0.776123046875, -1.46978759765625, -1.694671630859375, 0.08514404296875, -0.06752777099609375, -0.2980060577392578, -0.608428955078125, -0.1013031005859375, -0.13840866088867188, 1.35406494140625, 0.14178848266601562, 0.15484619140625, -0.08174896240234375, 1.5846405029296875, 0.291595458984375, 0.3282470703125, 0.50054931640625, 0.34725189208984375, -0.3450431823730469, -0.5468292236328125, 0.240203857421875, 0.7034912109375, -0.1628265380859375, 0.3263092041015625, 0.72930908203125, -0.053863525390625, 0.2745208740234375, 0.89178466796875, -0.0044002532958984375, -0.18182373046875, -0.382110595703125, 1.414886474609375, -0.694091796875, 1.8244781494140625, 0.99334716796875, -0.06457328796386719, -1.6390380859375, -0.4048194885253906, -0.31226348876953125, -0.8380699157714844, -0.0626983642578125, 0.587860107421875, -0.73089599609375, -0.4068946838378906, 2.896148681640625, -0.15478515625, -1.0653076171875, -1.8204803466796875, -1.9267578125, 0.638763427734375, -0.5042266845703125, 0.2654876708984375, -0.198211669921875, -0.23089599609375, 0.21812820434570312, -0.0026092529296875, -0.21654510498046875, 0.20624923706054688, 0.17726898193359375, -0.09116744995117188, 0.6808319091796875, 0.8715362548828125, 0.19522476196289062, -0.42462158203125, 0.33303070068359375, 0.038299560546875, 0.112640380859375, 0.225982666015625, 0.22296142578125, 1.030914306640625, 0.876129150390625, 0.379852294921875, -0.2196807861328125, -0.010772705078125, -0.853912353515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000015.npy"}
{"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": 0.12514957785606384, "std": 0.8233881592750549, "min": -3.1328125, "p10": -0.741607666015625, "median": 0.06545257568359375, "p90": 1.1740386962890623, "max": 2.621490478515625, "pos_frac": 0.5546875, "sample": [0.34825897216796875, -0.4013023376464844, -0.0426788330078125, 1.23138427734375, -0.7164459228515625, -0.2118072509765625, 1.20892333984375, 0.76361083984375, -1.78863525390625, -0.7742919921875, -0.1000518798828125, -0.017604827880859375, 0.1479034423828125, 0.469146728515625, -0.9598388671875, 0.18381500244140625, -0.0155029296875, -0.149993896484375, -0.238555908203125, 0.3748016357421875, 0.1327972412109375, 0.4717864990234375, 0.874603271484375, 0.15972900390625, -0.335418701171875, 0.14912033081054688, -0.45526123046875, 2.425323486328125, -0.15350341796875, 0.901031494140625, -0.41131591796875, -0.49566650390625, 0.1524658203125, -0.32373046875, -0.18511962890625, -0.33721923828125, -0.72760009765625, 0.220794677734375, -0.6326675415039062, 0.4139823913574219, 0.31915283203125, 0.0688018798828125, 0.185516357421875, 0.52459716796875, 0.971710205078125, -0.015146255493164062, -0.1387939453125, 0.515625, 0.69952392578125, -0.14957427978515625, 0.5116615295410156, 1.30010986328125, -0.4322052001953125, -1.082611083984375, 0.971160888671875, 1.379547119140625, -0.09621429443359375, -0.0764923095703125, -0.27147674560546875, 0.7734375, 1.42926025390625, 0.32355499267578125, -0.2143096923828125, -1.46026611328125, 0.062103271484375, 1.37237548828125, 0.1067962646484375, 2.621490478515625, -0.5115966796875, -0.291229248046875, -0.1451263427734375, 1.100677490234375, -0.3347015380859375, -0.176361083984375, -0.789093017578125, 0.9419708251953125, -2.526214599609375, 0.9810028076171875, 0.33087158203125, 1.159088134765625, -0.12206268310546875, 0.2693443298339844, -0.0452423095703125, 1.3607025146484375, -0.971405029296875, 0.724365234375, 0.0, 0.33941650390625, 1.822998046875, 1.437408447265625, 0.3898773193359375, 0.76007080078125, -0.7140960693359375, -0.15106201171875, 0.02471923828125, 0.5140380859375, -0.4097747802734375, 0.5792388916015625, -0.3227386474609375, 0.13982200622558594, 1.025238037109375, 0.471588134765625, -0.54730224609375, 0.4958953857421875, 0.10009002685546875, 0.03859710693359375, -0.461456298828125, 0.03359222412109375, -1.00140380859375, -0.828369140625, 0.3245086669921875, 0.548248291015625, 1.7159881591796875, 0.391693115234375, -0.19537353515625, 0.012237548828125, 0.12353515625, 0.05175018310546875, 0.155181884765625, 0.0310516357421875, -0.5056304931640625, -0.0078125, 0.423095703125, -1.32110595703125, -3.1328125, -0.967620849609375, 1.8955078125, 0.43072509765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000016.npy"}
{"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": 0.07308512926101685, "std": 0.7401096224784851, "min": -2.0937652587890625, "p10": -0.7711944580078125, "median": 0.05328941345214844, "p90": 0.9602111816406249, "max": 2.783172607421875, "pos_frac": 0.53125, "sample": [0.17833709716796875, 0.3831024169921875, 1.034912109375, -0.691986083984375, 0.2466888427734375, -0.621002197265625, 0.90228271484375, 1.23272705078125, 1.63861083984375, -0.2069091796875, -0.4587249755859375, -0.3228302001953125, 0.97076416015625, 0.386810302734375, -0.770965576171875, -0.45510101318359375, -0.951904296875, 0.192626953125, -1.13519287109375, 0.944091796875, -0.6173095703125, 0.0269012451171875, 0.601470947265625, -0.125640869140625, 0.10845947265625, -0.5670928955078125, 0.0811767578125, 0.69464111328125, 0.31085205078125, 0.17681884765625, 0.5191192626953125, -0.2266845703125, -0.2950897216796875, 0.576934814453125, -0.2176055908203125, -0.20287704467773438, -0.7886962890625, 0.784881591796875, -0.21978759765625, 0.0952606201171875, 1.2115478515625, -1.002166748046875, 0.846405029296875, -0.5826873779296875, -0.9344711303710938, -1.1362457275390625, -0.248077392578125, 0.08103561401367188, -2.0937652587890625, 0.1060638427734375, 1.3144378662109375, -0.28582000732421875, -0.690643310546875, 1.3616180419921875, -0.28000640869140625, 0.026336669921875, 0.162567138671875, 0.6970367431640625, 0.5748062133789062, -0.286468505859375, 0.13117218017578125, 0.36891937255859375, -0.138458251953125, -0.03365325927734375, -0.3602294921875, -0.02362060546875, 0.40838623046875, 0.42156982421875, -0.5754165649414062, 0.35009765625, 0.144805908203125, -0.06915283203125, 0.9556884765625, 0.4521026611328125, 0.8892974853515625, -0.10113525390625, -0.985809326171875, 1.0007705688476562, -0.15399169921875, -0.2440185546875, 2.61810302734375, -0.1213531494140625, -0.771728515625, -0.431060791015625, 0.5417861938476562, 0.235748291015625, -0.27947998046875, -0.362213134765625, 0.81573486328125, 0.663970947265625, 0.07596206665039062, 1.2478179931640625, -0.21269989013671875, 1.07513427734375, -0.3552093505859375, 0.08526611328125, 0.37403106689453125, -0.271453857421875, 0.049358367919921875, 1.17022705078125, -0.20666885375976562, -0.1571044921875, -0.081329345703125, 0.423065185546875, -0.000652313232421875, -0.131134033203125, 0.19546890258789062, -0.634307861328125, -1.8159942626953125, 0.783050537109375, -0.6365852355957031, 0.3207855224609375, -1.306243896484375, 0.31488800048828125, 0.057220458984375, -0.16748046875, 0.2069244384765625, 0.787353515625, -1.6710968017578125, 0.49981689453125, 0.08063507080078125, 0.00136566162109375, -0.91766357421875, -0.1645050048828125, 2.783172607421875, -0.715301513671875, 0.50958251953125, 0.358795166015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000017.npy"}
{"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": 0.08329518139362335, "std": 0.7457202672958374, "min": -2.16644287109375, "p10": -0.8602066040039062, "median": 0.07223129272460938, "p90": 0.97064208984375, "max": 2.49310302734375, "pos_frac": 0.546875, "sample": [-0.734130859375, -0.248291015625, -0.19499969482421875, -0.32635498046875, 0.2801322937011719, -0.39813232421875, -0.09390640258789062, -0.7427101135253906, 0.20451736450195312, -0.00246429443359375, 1.30633544921875, 0.12225341796875, 0.0796051025390625, 0.15673828125, 0.1629180908203125, 0.261810302734375, -0.89208984375, 1.074676513671875, 0.3868865966796875, 0.96954345703125, -0.3725128173828125, 0.5880661010742188, 1.62457275390625, -1.0032882690429688, 1.328033447265625, -0.019184112548828125, 0.2644805908203125, 0.9699554443359375, 0.254913330078125, -0.3449211120605469, -0.03863525390625, 0.648956298828125, -0.2439727783203125, -0.527801513671875, 0.60369873046875, 0.14406585693359375, -0.292877197265625, 0.20755577087402344, -0.219696044921875, 0.2811279296875, 0.09127044677734375, 1.223541259765625, 2.04541015625, -1.113800048828125, -0.11679840087890625, 0.5983047485351562, -1.2213134765625, -0.8398590087890625, 0.4142913818359375, -0.1889801025390625, -0.150543212890625, -0.1673583984375, 0.709259033203125, 0.44280242919921875, 0.76116943359375, -0.22637939453125, -0.329193115234375, -0.55291748046875, -0.8997802734375, 0.5919342041015625, 0.025299072265625, -0.8465423583984375, -0.120361328125, 0.017791748046875, -0.49468994140625, -0.50970458984375, 0.8251953125, 1.06536865234375, 2.22869873046875, 1.640625, 0.3588905334472656, 0.11907196044921875, 0.385223388671875, -1.99749755859375, 0.5396270751953125, 0.256317138671875, 0.849334716796875, 0.23729705810546875, 0.666473388671875, -0.9175262451171875, 0.33038330078125, 0.38523101806640625, 0.294097900390625, -0.1023406982421875, 0.199432373046875, -0.26654815673828125, 0.33676910400390625, -1.02392578125, 0.2198486328125, 0.27492523193359375, 1.39532470703125, 0.06777191162109375, 0.551971435546875, -0.26563262939453125, -2.16644287109375, -0.3822021484375, -0.26898956298828125, -0.34528350830078125, 0.1155242919921875, 0.00958251953125, 0.076690673828125, 0.755859375, -1.016510009765625, 2.49310302734375, 0.1888885498046875, -0.286773681640625, 0.03594970703125, -0.23944091796875, 0.379058837890625, 0.9722442626953125, -0.21331787109375, 0.9839935302734375, -0.39825439453125, 0.482696533203125, -0.5012664794921875, -0.670989990234375, 0.040863037109375, -0.320892333984375, 0.678680419921875, -1.283782958984375, -0.1824951171875, -0.0398101806640625, -0.0015106201171875, -1.340911865234375, 0.78631591796875, -1.0301513671875, 0.846893310546875, -0.517669677734375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000018.npy"}
{"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": 0.05183883011341095, "std": 0.6757497191429138, "min": -2.7303466796875, "p10": -0.7616226196289062, "median": 0.06371593475341797, "p90": 0.8451553344726561, "max": 2.03155517578125, "pos_frac": 0.5859375, "sample": [0.1214141845703125, 0.587188720703125, 0.280670166015625, 0.040771484375, 0.32183837890625, 0.3543701171875, 1.227752685546875, 0.203338623046875, -0.334075927734375, -2.7303466796875, -0.3370208740234375, -0.3280353546142578, 0.3773651123046875, -1.3072509765625, 0.4508247375488281, 0.35939788818359375, 0.331390380859375, 0.35137939453125, -0.16143798828125, 0.0838165283203125, 0.43633270263671875, 0.52166748046875, -0.25356483459472656, -0.1063232421875, -0.1039581298828125, -0.0558319091796875, 0.06937217712402344, 0.01239013671875, 1.556640625, -0.34467315673828125, -1.2137451171875, -0.309234619140625, -0.225738525390625, -1.089111328125, 0.3819732666015625, 0.1110076904296875, 1.51080322265625, 0.25482177734375, 0.208282470703125, -0.1354656219482422, -0.127410888671875, 0.20131492614746094, -1.3712158203125, 0.3370819091796875, -0.7847900390625, 0.94366455078125, 0.2288970947265625, 0.8819732666015625, -0.02484130859375, 0.3447723388671875, -0.062713623046875, -0.551971435546875, 0.829376220703125, 1.9601287841796875, 0.591094970703125, -1.26800537109375, 0.489105224609375, -1.24151611328125, 0.294586181640625, 0.10474395751953125, 0.251312255859375, -0.4443817138671875, 0.1832427978515625, 0.5273056030273438, -0.6339569091796875, 0.5665130615234375, 0.0110931396484375, -0.044769287109375, 0.63311767578125, -0.873779296875, 1.132080078125, 0.353668212890625, -0.2054595947265625, -0.4097900390625, -0.31668853759765625, -1.3878021240234375, -0.975616455078125, 0.0981597900390625, -0.425201416015625, 0.2559814453125, 0.20745849609375, 0.30780029296875, 0.2061138153076172, -0.0873870849609375, -0.362548828125, -0.24964141845703125, 0.4532947540283203, 2.03155517578125, 0.11634063720703125, 0.974761962890625, 0.4144287109375, -0.2626953125, 0.6014556884765625, -0.28961181640625, 0.0118408203125, 0.55072021484375, -0.502349853515625, 0.035614013671875, -0.3645210266113281, 0.0352935791015625, 0.0580596923828125, 0.209564208984375, 1.0581207275390625, -0.02056884765625, -1.36468505859375, 0.001220703125, 0.036407470703125, 0.89361572265625, 0.0273284912109375, -0.7516937255859375, -0.0827789306640625, 0.02513885498046875, -0.93182373046875, -0.153076171875, -0.460601806640625, 1.001220703125, 0.731964111328125, -0.4281005859375, -0.288818359375, 0.2458343505859375, 0.526947021484375, 0.13067626953125, 1.5521240234375, -0.3854522705078125, -0.1606597900390625, -0.141387939453125, 0.1916046142578125, 0.098968505859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000019.npy"}
{"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": 0.13456162810325623, "std": 0.807576060295105, "min": -2.681640625, "p10": -0.8323196411132813, "median": 0.07622146606445312, "p90": 1.2231781005859372, "max": 2.426055908203125, "pos_frac": 0.578125, "sample": [0.1411895751953125, 0.846954345703125, 0.268768310546875, -0.4376220703125, 0.3137664794921875, 0.8060302734375, 1.4393310546875, 0.284759521484375, -0.260986328125, -0.9382553100585938, 0.548370361328125, 0.1310882568359375, -0.1028900146484375, 1.13970947265625, 0.69873046875, -0.0060882568359375, 0.01824951171875, 1.5909423828125, -0.11574554443359375, -0.035003662109375, -0.796051025390625, -0.905426025390625, -0.5245361328125, 0.3528861999511719, -0.28114891052246094, -0.5561752319335938, 0.6452484130859375, -1.268890380859375, -0.2262115478515625, -0.85211181640625, 1.579071044921875, 0.031215667724609375, 0.03556060791015625, -1.828033447265625, -0.2722587585449219, -0.8238372802734375, -0.36383056640625, 0.079925537109375, 0.136627197265625, 1.33233642578125, 1.49493408203125, 0.0653076171875, 0.8400421142578125, -0.26868438720703125, 0.5540618896484375, 0.6973876953125, 0.3921966552734375, 0.177459716796875, -0.812835693359375, 0.26239013671875, 0.06732177734375, -0.13137054443359375, 0.0, -1.1202392578125, 1.81903076171875, -0.07448577880859375, -0.3178138732910156, -1.638427734375, 0.4792938232421875, -0.2416839599609375, -0.324310302734375, 0.7984619140625, -0.641845703125, -2.681640625, 1.20452880859375, -0.050506591796875, 0.0832366943359375, 0.00861358642578125, 0.8794708251953125, 0.19384765625, 0.18817138671875, -0.06170654296875, 1.4508819580078125, -0.2039031982421875, -0.269561767578125, 0.4263916015625, 0.07251739501953125, 0.706817626953125, -0.52001953125, 0.41015625, -0.2447509765625, 1.446075439453125, 0.18389892578125, 1.0287437438964844, 0.3858642578125, -0.189910888671875, 0.508056640625, 1.4700927734375, -0.13909149169921875, -1.81207275390625, 0.2824592590332031, 0.919189453125, 0.89337158203125, 0.341156005859375, 0.305389404296875, 1.695831298828125, -1.460906982421875, 0.5166244506835938, 0.29189300537109375, 0.15335655212402344, -0.9891357421875, -0.43792724609375, -0.1086273193359375, -0.010784149169921875, 2.426055908203125, -0.960296630859375, 0.2139892578125, 0.2613525390625, -0.15274810791015625, 0.054718017578125, -0.7452392578125, 0.0343017578125, 0.25060272216796875, 0.22576904296875, -0.219207763671875, -0.19130325317382812, 2.33343505859375, -0.6062774658203125, -1.025360107421875, 0.03921318054199219, 1.266693115234375, 0.549591064453125, 0.42938232421875, 1.110931396484375, -0.06395149230957031, 0.7029953002929688, -0.010284423828125, 0.531585693359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000020.npy"}
{"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": 0.2021702378988266, "std": 0.7579217553138733, "min": -3.73876953125, "p10": -0.6106536865234374, "median": 0.22327804565429688, "p90": 1.0775634765625, "max": 1.962493896484375, "pos_frac": 0.6484375, "sample": [0.330352783203125, 0.0654754638671875, 0.17352294921875, 0.087249755859375, 0.804534912109375, 0.5150146484375, -0.32275390625, 1.17938232421875, -0.50750732421875, -0.8224868774414062, 0.22784423828125, -2.401458740234375, 0.249114990234375, 0.4215240478515625, -0.26116943359375, 0.5981216430664062, -0.732940673828125, -0.069305419921875, 0.8043212890625, -0.42939186096191406, 0.277587890625, -0.2006969451904297, -1.2412109375, -0.587371826171875, 0.5413360595703125, 0.0169677734375, -0.76727294921875, -0.12493896484375, -0.2301025390625, 0.21985626220703125, -0.41063690185546875, -0.836883544921875, 0.594146728515625, 1.0267333984375, 0.0424652099609375, -0.2997589111328125, -0.0162200927734375, 0.405120849609375, -1.670745849609375, 1.962493896484375, 0.49664306640625, -0.406890869140625, 0.0330810546875, 0.7851791381835938, 0.03484344482421875, 1.0919189453125, -0.04459571838378906, -0.22210693359375, 0.8766555786132812, 1.2423095703125, 0.5712890625, 0.9412689208984375, 0.0, 0.4705657958984375, 0.007568359375, -0.22690963745117188, -0.00750732421875, 0.5438232421875, -0.1279582977294922, -0.12891387939453125, 1.136474609375, 0.512939453125, -0.05199241638183594, -0.30352783203125, 0.8564453125, -0.66497802734375, -0.89031982421875, -0.172210693359375, 0.149322509765625, -0.138946533203125, 0.81219482421875, 0.3314056396484375, 0.619964599609375, 0.57080078125, -0.453948974609375, 1.0714111328125, 1.64691162109375, -0.49859619140625, 0.814056396484375, 0.124053955078125, -0.21184539794921875, 0.2266998291015625, 0.51806640625, 0.5742034912109375, 0.708587646484375, 1.6687774658203125, 0.09047317504882812, 0.14777374267578125, 0.1557464599609375, 0.17871856689453125, 0.783721923828125, 0.524810791015625, -0.15252685546875, 0.3514080047607422, 0.414031982421875, 0.39151763916015625, 0.16564178466796875, -0.8623046875, 0.87152099609375, 1.1884918212890625, 0.51507568359375, 0.417633056640625, 1.25408935546875, 1.4710693359375, 1.93548583984375, 0.01837158203125, 0.663543701171875, 0.51690673828125, 0.10872650146484375, 0.499053955078125, -0.02587890625, -0.2994384765625, 0.238739013671875, 1.203765869140625, 0.5224533081054688, -0.729644775390625, 0.9512252807617188, 0.634735107421875, -0.050567626953125, -0.47283935546875, -3.73876953125, 1.4307861328125, 0.3478736877441406, 0.636962890625, 0.2962188720703125, 0.0493011474609375, 0.4992218017578125, -0.7618618011474609], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000021.npy"}
{"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": 0.02922128140926361, "std": 0.7573904991149902, "min": -1.93780517578125, "p10": -0.7646926879882812, "median": 0.024512290954589844, "p90": 0.9037590026855468, "max": 2.946014404296875, "pos_frac": 0.5234375, "sample": [-0.5732879638671875, 0.3726959228515625, -0.20117950439453125, 0.6068115234375, 2.651519775390625, -0.46558380126953125, 0.27862548828125, 0.06258010864257812, 0.364898681640625, -0.31927490234375, 1.21148681640625, -0.6317825317382812, 0.08575439453125, 0.10466575622558594, -1.322967529296875, 2.946014404296875, -1.64263916015625, -0.627655029296875, 1.0374908447265625, 0.241668701171875, -0.1685943603515625, -0.43106842041015625, -0.098236083984375, 0.41170501708984375, 0.8812255859375, 0.15667724609375, -0.17431640625, 0.1470947265625, 0.44890594482421875, 0.5267868041992188, -0.7353515625, -0.7009429931640625, 0.0, -0.5244140625, -0.240264892578125, 0.5576171875, -0.13818359375, 1.0868377685546875, -0.4529380798339844, 0.3055267333984375, -0.290863037109375, -0.680755615234375, -0.09124755859375, 1.8883056640625, 0.047882080078125, -0.4026641845703125, 0.9931640625, -0.49798583984375, -1.93780517578125, -0.749847412109375, 0.028532028198242188, -0.12072372436523438, 0.029144287109375, 0.6051483154296875, 0.09842681884765625, 0.9335861206054688, 0.334686279296875, 0.554412841796875, 0.953948974609375, 1.98712158203125, 0.163482666015625, -0.263458251953125, 0.165130615234375, 0.507659912109375, 0.18526458740234375, 0.551971435546875, 0.19842529296875, -0.661346435546875, 0.8505172729492188, 0.603057861328125, -0.5455322265625, -1.14111328125, 0.8909759521484375, -0.1602783203125, 0.50848388671875, -0.2116851806640625, -0.085296630859375, 0.412750244140625, -0.38360595703125, -0.87152099609375, 0.086273193359375, -0.070526123046875, -0.4551849365234375, 0.24848365783691406, 0.587615966796875, -0.060882568359375, -0.67926025390625, 0.216705322265625, 1.67108154296875, -0.1356201171875, 1.0271759033203125, -0.044921875, 0.1498870849609375, -0.7248382568359375, 0.872711181640625, -0.9622955322265625, -1.299652099609375, -0.2480010986328125, -0.68634033203125, 0.5269775390625, -1.52520751953125, 0.003780364990234375, -0.32269287109375, 0.001495361328125, 0.12259674072265625, -0.29864501953125, -0.3219146728515625, 0.687042236328125, -1.320709228515625, -0.61395263671875, 0.1362152099609375, -0.131256103515625, 0.31473541259765625, 0.2928009033203125, 0.2630157470703125, 0.26495361328125, 0.0204925537109375, -1.15814208984375, 1.02801513671875, 0.660125732421875, -0.134979248046875, -0.82061767578125, -0.7993316650390625, -0.9185943603515625, 0.10736465454101562, -0.70904541015625, 0.029903411865234375, -0.570770263671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000022.npy"}
{"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": 0.16596047580242157, "std": 0.8672677874565125, "min": -2.57501220703125, "p10": -0.7601032257080078, "median": 0.16407489776611328, "p90": 1.3259002685546875, "max": 2.399017333984375, "pos_frac": 0.578125, "sample": [-0.2332916259765625, 0.9008331298828125, -0.2158203125, 0.6430816650390625, 0.1190185546875, 1.826904296875, -0.339996337890625, 0.39141845703125, 0.154632568359375, -1.033203125, 1.325103759765625, -1.683837890625, -0.17205810546875, 1.63287353515625, 1.3277587890625, -0.05826568603515625, 1.4246826171875, -0.3553466796875, 0.15964317321777344, -0.33221435546875, 0.4403533935546875, -0.876953125, 0.0777587890625, 1.42498779296875, 0.9282684326171875, 0.42322540283203125, -0.4222564697265625, 0.604461669921875, -0.59765625, -0.17845916748046875, 0.0, -0.48900604248046875, 0.018218994140625, -0.6578292846679688, 1.00750732421875, 0.9924468994140625, -0.42368316650390625, 0.34954833984375, 0.536590576171875, -0.6951904296875, 1.20977783203125, 1.581451416015625, 0.887786865234375, -0.09128189086914062, -0.7958984375, 1.0121612548828125, 0.2504730224609375, 0.16850662231445312, 1.3824462890625, 0.16992950439453125, -0.7447624206542969, -0.016332626342773438, -0.5066604614257812, 0.7100677490234375, 1.2027587890625, -2.57501220703125, 0.10870361328125, -0.223236083984375, 0.3419647216796875, 0.2946319580078125, -0.543701171875, 0.6662406921386719, 0.2664794921875, -0.20397186279296875, -0.6387176513671875, -0.47298431396484375, 0.9483642578125, 1.36773681640625, -0.014923095703125, 0.372894287109375, 1.4950332641601562, -0.2662506103515625, -0.2193450927734375, 0.025390625, 0.414093017578125, 0.59619140625, 1.714599609375, 0.013097763061523438, 0.2542724609375, -0.9069976806640625, 0.59381103515625, -0.11787223815917969, -0.21788787841796875, -0.1003570556640625, 1.06170654296875, -0.279022216796875, 0.943206787109375, 0.587066650390625, 0.4126739501953125, -1.534332275390625, -1.1240921020507812, -0.7135677337646484, -0.52178955078125, -0.26454925537109375, 0.06421661376953125, 0.6409912109375, 0.635955810546875, 1.524322509765625, -0.46048736572265625, 0.3583984375, -2.056884765625, 0.6475982666015625, 0.9473800659179688, -0.1983489990234375, -0.28265380859375, 1.81256103515625, -0.40167236328125, 0.18145751953125, 0.5107421875, 0.485809326171875, 0.5171012878417969, 1.048492431640625, 0.082916259765625, -1.152862548828125, -0.03513336181640625, 0.8402099609375, 2.399017333984375, 0.4948577880859375, -1.2496490478515625, -2.526123046875, 0.999969482421875, -0.1664581298828125, -0.2178802490234375, 0.2572784423828125, -2.08428955078125, 0.4748077392578125, 0.7165908813476562, 0.530487060546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000023.npy"}
{"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": 0.30209383368492126, "std": 0.8906621336936951, "min": -2.3345947265625, "p10": -0.64970703125, "median": 0.2223663330078125, "p90": 1.3180511474609375, "max": 3.998870849609375, "pos_frac": 0.640625, "sample": [0.48974609375, 1.1321258544921875, 0.30727386474609375, 0.2435455322265625, 0.153656005859375, 1.31353759765625, 0.06866455078125, -0.133758544921875, 0.1299285888671875, 1.562835693359375, 0.1304473876953125, 0.459381103515625, -0.58642578125, 0.257781982421875, 0.34491729736328125, 0.43218994140625, -0.19049072265625, 3.998870849609375, 0.109161376953125, -0.2331523895263672, 1.9501953125, -0.0894775390625, -0.41607666015625, 0.0, 0.0046844482421875, -0.07082366943359375, -0.352386474609375, 0.059722900390625, -0.36029052734375, 0.3438720703125, -2.3345947265625, 0.13634490966796875, 0.57666015625, -0.80889892578125, 0.68450927734375, 0.4410400390625, 1.5418701171875, 0.048553466796875, -0.6672210693359375, 0.62908935546875, -0.666839599609375, 0.093353271484375, -0.0955963134765625, -0.642364501953125, -0.349853515625, 0.286834716796875, 0.0, 0.7935791015625, 0.694793701171875, 0.48641204833984375, 0.70623779296875, 0.135467529296875, 0.6439056396484375, 0.29247283935546875, 1.8472900390625, 0.6888961791992188, -0.462158203125, -0.12742996215820312, 1.13128662109375, 0.736541748046875, -2.3056640625, -0.2138671875, 1.9388427734375, 0.4954204559326172, 2.4365234375, 0.8757400512695312, -0.0775146484375, -0.345001220703125, -0.55499267578125, -0.877471923828125, 0.42132568359375, -0.7783203125, -0.7462844848632812, 0.220428466796875, 0.3954315185546875, -0.38604736328125, -0.442108154296875, 1.328582763671875, 1.125274658203125, 1.2577362060546875, -1.6411590576171875, -0.691314697265625, 1.198455810546875, 0.64239501953125, 0.0215606689453125, 0.6610107421875, 1.030029296875, -1.072906494140625, 1.999481201171875, 0.02577972412109375, 1.48187255859375, 0.81732177734375, -0.1627197265625, -1.4139633178710938, 0.33636474609375, 0.11559295654296875, -0.041046142578125, -0.141845703125, 0.6492156982421875, 1.207550048828125, -0.4356689453125, 0.22430419921875, -0.6027679443359375, 1.740203857421875, -0.536895751953125, 0.9632568359375, 0.685150146484375, 2.85162353515625, 0.3581390380859375, 1.04730224609375, -0.18707275390625, -0.05694580078125, 0.5621185302734375, 1.2640380859375, 0.3513336181640625, 0.63330078125, -0.29193115234375, 0.379547119140625, 1.2321319580078125, -0.06103515625, 1.1099853515625, 1.53607177734375, 0.1636962890625, 0.63677978515625, -0.8822021484375, 0.01514434814453125, -0.3466796875, 0.02553558349609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000024.npy"}
{"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.09393027424812317, "std": 0.74391770362854, "min": -1.98236083984375, "p10": -0.7755744934082031, "median": 0.07217788696289062, "p90": 0.8964721679687496, "max": 3.3544921875, "pos_frac": 0.546875, "sample": [-0.145599365234375, 0.368743896484375, 0.858551025390625, -0.3092498779296875, -0.17816162109375, 1.91864013671875, 1.6959228515625, -0.1282501220703125, 3.3544921875, -1.1124267578125, 0.654296875, -0.1862335205078125, 0.1373138427734375, -0.32916259765625, 0.16576385498046875, -1.4742431640625, -1.0170364379882812, 0.07488250732421875, 0.55792236328125, -0.109405517578125, 0.04929351806640625, -0.26465415954589844, -0.2920379638671875, 0.602020263671875, -0.3258056640625, -0.84521484375, -0.2447357177734375, -0.024932861328125, -0.4301795959472656, -0.25091552734375, 0.3800678253173828, 0.411834716796875, 0.0, -0.2587471008300781, -0.399658203125, 0.1827106475830078, 0.6955413818359375, 0.1281585693359375, -0.134307861328125, 0.2447509765625, -0.76727294921875, 0.0, -0.471435546875, -0.7237548828125, -0.4573516845703125, 0.3074188232421875, -0.9068145751953125, 0.302490234375, -0.7949447631835938, 0.51751708984375, 0.4038963317871094, 0.6739501953125, 0.2427520751953125, -0.5145454406738281, 0.60284423828125, -0.8607940673828125, -0.43651580810546875, 0.533447265625, 0.5452117919921875, -0.913970947265625, 0.5597915649414062, 0.1591033935546875, 0.2958984375, 1.8216552734375, 0.000179290771484375, -0.1768646240234375, 0.6132965087890625, -0.4837493896484375, -1.242767333984375, -0.169891357421875, 0.984954833984375, 1.1905288696289062, 1.1238327026367188, -1.98236083984375, 0.699859619140625, 0.222503662109375, -0.16220855712890625, 0.16951942443847656, -0.7268218994140625, 0.63525390625, -0.8140106201171875, 0.1089935302734375, -0.718414306640625, 0.130645751953125, -0.4156494140625, -0.5821533203125, 0.12320327758789062, 0.49204444885253906, -0.1353302001953125, 1.031005859375, 0.04913330078125, 0.49393463134765625, 0.673065185546875, 0.358428955078125, 1.778045654296875, -0.48553466796875, 0.307037353515625, 1.2725677490234375, -0.3201904296875, 0.03302001953125, -0.23236846923828125, -0.7660751342773438, 0.454742431640625, -0.0063800811767578125, 1.53515625, 2.057525634765625, 0.2659912109375, -0.3526153564453125, 0.5941925048828125, -0.8497314453125, 0.0694732666015625, 0.52386474609375, 0.3192138671875, 0.0589599609375, -1.193450927734375, 0.1246185302734375, 1.303863525390625, 0.38631439208984375, 0.259185791015625, 0.1697845458984375, -0.134429931640625, -0.07120132446289062, -0.760894775390625, -0.4986839294433594, 0.529388427734375, -0.164703369140625, 0.104217529296875, 0.08348464965820312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000025.npy"}
{"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": 0.11699323356151581, "std": 0.7820940613746643, "min": -1.880157470703125, "p10": -0.7319900512695312, "median": 0.041594505310058594, "p90": 1.0985809326171876, "max": 2.72607421875, "pos_frac": 0.546875, "sample": [2.69024658203125, -0.036678314208984375, 0.0628662109375, -0.5890350341796875, 1.855194091796875, -0.09760284423828125, 0.02162933349609375, -0.418701171875, 0.144134521484375, -0.24734115600585938, -0.21799468994140625, 0.810760498046875, 0.0, -0.37221527099609375, -0.579193115234375, -1.1531295776367188, 1.351959228515625, -0.6485328674316406, -0.108306884765625, -0.2954254150390625, 0.111175537109375, 0.1143798828125, -0.2622528076171875, 0.255096435546875, 0.153411865234375, 0.4729766845703125, -0.4236297607421875, 0.630645751953125, -0.24224853515625, 0.22359466552734375, 0.050140380859375, -1.138946533203125, 0.11418914794921875, -0.54180908203125, -1.642333984375, 2.72607421875, -0.441253662109375, -1.5420074462890625, 1.031463623046875, -0.3197784423828125, -0.269927978515625, 0.5809574127197266, -0.00136566162109375, -0.117279052734375, 0.27252197265625, -1.261474609375, 0.6829299926757812, 0.27899169921875, -0.1206207275390625, -0.02616119384765625, 0.602203369140625, 0.019439697265625, 0.34128570556640625, 1.10009765625, 0.687957763671875, 0.025554656982421875, 0.18280029296875, -0.728240966796875, -0.4547119140625, -0.4302978515625, 0.4258861541748047, 0.56640625, 0.314361572265625, 0.4053192138671875, 0.865478515625, 0.72039794921875, 0.1475830078125, -0.8118896484375, 1.93115234375, 0.21700286865234375, 1.097930908203125, -0.36276817321777344, 0.550567626953125, -0.667999267578125, 0.03934478759765625, -0.310455322265625, -0.44629669189453125, -0.23987388610839844, 1.43792724609375, -0.7576255798339844, 0.526947021484375, 0.429534912109375, 0.4513092041015625, -0.61529541015625, 1.165802001953125, -0.2130107879638672, 0.88299560546875, 0.6467132568359375, 0.483551025390625, 0.279327392578125, 1.27764892578125, -0.5867691040039062, 0.7880287170410156, 0.548095703125, 1.8017578125, 0.0715179443359375, -0.4328460693359375, -0.38562965393066406, 0.09381103515625, 0.8597869873046875, -0.5220565795898438, -0.4588508605957031, -0.324462890625, 0.9377593994140625, 0.439910888671875, 0.00390625, 0.6411590576171875, 0.0, -0.78656005859375, 0.04384422302246094, 0.2491455078125, 0.7432861328125, -0.6451568603515625, 1.4316253662109375, 0.993560791015625, 0.2565460205078125, -0.94256591796875, -0.620208740234375, 1.33526611328125, -0.95330810546875, 1.401611328125, -1.383056640625, -0.10530853271484375, 0.017059326171875, -0.150787353515625, -0.7407379150390625, -1.880157470703125, -0.0682373046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000026.npy"}
{"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": 0.1983897089958191, "std": 0.9162200093269348, "min": -1.98486328125, "p10": -0.781134033203125, "median": 0.14107513427734375, "p90": 1.411931610107422, "max": 3.3624267578125, "pos_frac": 0.59375, "sample": [-0.0594482421875, -0.381988525390625, -0.30393218994140625, -0.4259071350097656, -0.212921142578125, -0.39727783203125, -0.114013671875, 0.05560302734375, 0.0376434326171875, 0.25617218017578125, 0.134368896484375, -0.064544677734375, -0.26025390625, 0.472503662109375, -0.5374755859375, 0.0760498046875, -0.3921051025390625, -0.5557937622070312, 1.37847900390625, 1.59259033203125, 0.15892791748046875, 0.632843017578125, 2.087799072265625, -0.99468994140625, -0.36993408203125, 1.63409423828125, -0.3621559143066406, -0.351898193359375, 0.20370864868164062, -0.1218719482421875, 2.22369384765625, 0.3861083984375, 0.9517364501953125, 1.53240966796875, 0.6868438720703125, 0.159698486328125, 0.1886749267578125, 0.7705078125, 0.95050048828125, -0.798309326171875, -0.82220458984375, 0.00555419921875, -0.518035888671875, 0.5953750610351562, -0.2374725341796875, 2.4959716796875, 0.2836456298828125, -0.279754638671875, 0.399566650390625, 0.41790008544921875, 0.49249267578125, -0.242462158203125, -0.4028472900390625, 0.4275054931640625, 0.354705810546875, 0.914886474609375, 0.3047370910644531, 0.5047607421875, -0.0092315673828125, 3.3624267578125, -1.6397705078125, -1.8633575439453125, -1.7265472412109375, 0.513214111328125, 0.39495086669921875, 0.4145965576171875, 0.97100830078125, 0.23072433471679688, 0.68292236328125, 0.101318359375, -1.866729736328125, -1.174468994140625, -1.877655029296875, -0.43145751953125, 0.230499267578125, 0.4078369140625, 1.648834228515625, 1.117828369140625, -1.98486328125, 0.052490234375, -0.03487396240234375, 0.93017578125, 0.775665283203125, -0.773773193359375, -1.179351806640625, 1.170196533203125, 0.506378173828125, -0.13866424560546875, -0.056522369384765625, -0.0183258056640625, 2.1931610107421875, 0.073699951171875, -1.51776123046875, 0.1477813720703125, 0.1956329345703125, -1.25946044921875, 0.86224365234375, 0.08286285400390625, -0.6391143798828125, 1.756378173828125, -0.340789794921875, -0.46672630310058594, -0.4365997314453125, 1.2178955078125, 0.2660999298095703, 1.7410888671875, 1.163116455078125, 0.2245025634765625, -0.5382537841796875, -0.03359794616699219, 1.252960205078125, 0.19155502319335938, 0.06497383117675781, -0.60791015625, 0.23876953125, -0.2227783203125, -0.5508193969726562, 0.16769027709960938, -0.5091094970703125, 0.552703857421875, 1.021575927734375, 1.4104690551757812, 1.1754150390625, 1.41534423828125, 1.66595458984375, -0.0264434814453125, 0.107574462890625, 0.05756378173828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000027.npy"}
{"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.12946686148643494, "std": 0.8944807648658752, "min": -3.92034912109375, "p10": -0.9535995483398437, "median": 0.04969024658203125, "p90": 1.1806701660156247, "max": 3.215972900390625, "pos_frac": 0.5390625, "sample": [0.123870849609375, -3.92034912109375, 0.026763916015625, 2.47119140625, 0.2997589111328125, -0.020261764526367188, 0.13652801513671875, -0.22995376586914062, 0.3066234588623047, 1.11627197265625, 0.027563095092773438, 0.81927490234375, 0.446014404296875, -0.1057281494140625, 0.31097412109375, 1.5355987548828125, -0.2540435791015625, 0.3320465087890625, 1.0167388916015625, 1.0763397216796875, -0.1657562255859375, -0.3871002197265625, 0.7600631713867188, 0.149932861328125, -0.0242462158203125, 0.1260986328125, -0.8668212890625, 0.03092193603515625, -0.007171630859375, -0.23211669921875, 0.0182952880859375, 1.0161895751953125, -1.279296875, -0.2471923828125, -0.32314300537109375, 0.4527740478515625, -0.5265045166015625, 1.5356903076171875, 0.314056396484375, -0.9625396728515625, -0.597625732421875, -0.09869384765625, 0.505035400390625, -0.396484375, -0.22088623046875, 1.51617431640625, -0.38330078125, 0.3145751953125, 1.31304931640625, 0.31659698486328125, -0.792388916015625, 0.0315399169921875, -1.019866943359375, 2.465576171875, -0.964324951171875, 0.38916015625, -0.6269340515136719, 0.067840576171875, 0.349853515625, 0.734832763671875, 0.7815704345703125, -0.07733917236328125, 1.5164031982421875, -0.265625, -1.1455078125, 0.310943603515625, -0.1383209228515625, -0.98291015625, 1.2574462890625, -0.042327880859375, 0.373260498046875, 0.170318603515625, 0.5611953735351562, -0.21045684814453125, 0.22991943359375, -0.2996826171875, -0.54107666015625, -0.229400634765625, 0.28832244873046875, 2.25799560546875, 0.3629913330078125, -0.2143096923828125, -0.97589111328125, -0.50115966796875, 0.563079833984375, -0.5610237121582031, 0.9637451171875, -0.3572540283203125, -0.41280364990234375, 0.42560386657714844, 0.229278564453125, 3.215972900390625, 0.8057708740234375, -1.362457275390625, -0.0492401123046875, 0.10975265502929688, 2.127166748046875, -0.3446388244628906, -0.555145263671875, 0.6485137939453125, -0.995269775390625, 0.9850883483886719, -0.9969024658203125, 0.6947784423828125, 0.261474609375, -0.611328125, 0.43550872802734375, -0.5916824340820312, 1.624664306640625, 0.5240936279296875, -0.94976806640625, -0.1677703857421875, 1.4197998046875, -0.15433120727539062, -0.885040283203125, 0.34381103515625, 0.2921142578125, 0.27190399169921875, 0.78204345703125, -0.204742431640625, -1.22845458984375, 0.87188720703125, -0.3333740234375, 1.14776611328125, -0.977935791015625, 0.21931076049804688, 0.0, -0.9396514892578125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000028.npy"}
{"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.22749751806259155, "std": 0.864125669002533, "min": -1.97357177734375, "p10": -0.7692523956298828, "median": 0.1534290313720703, "p90": 1.3392272949218749, "max": 2.81060791015625, "pos_frac": 0.5859375, "sample": [-0.19775772094726562, 1.327545166015625, 0.36492919921875, 1.041473388671875, 0.5747222900390625, 0.171875, -0.69964599609375, 0.1502685546875, 0.17864990234375, 1.41790771484375, 0.09771728515625, 0.69061279296875, 1.80889892578125, 1.03753662109375, -0.16559410095214844, 0.0, 0.173126220703125, 0.3168182373046875, 0.451690673828125, 0.012355804443359375, -0.948394775390625, 0.235992431640625, -0.686859130859375, 0.70361328125, 0.06689453125, 0.213134765625, 0.184356689453125, -0.219757080078125, -0.19171905517578125, 0.35256195068359375, 1.3172607421875, 0.453582763671875, 0.769927978515625, -0.327484130859375, -0.470001220703125, -0.67950439453125, 0.716064453125, -0.01904296875, 1.8431549072265625, -0.20539093017578125, -1.040008544921875, 0.059673309326171875, 0.734161376953125, -0.36406707763671875, 0.033721923828125, -0.7525749206542969, -0.127227783203125, 1.015655517578125, 0.7245941162109375, -0.0011138916015625, -1.69866943359375, -0.1781005859375, -1.1396484375, -0.253204345703125, -0.2238006591796875, -1.325469970703125, -1.03009033203125, 0.6118927001953125, 1.64508056640625, 0.24072265625, -1.97357177734375, 0.3009033203125, 0.01629638671875, -0.85675048828125, 0.22765350341796875, 1.1963043212890625, 1.29443359375, -0.2816162109375, 0.0, 0.6475906372070312, 1.060760498046875, -0.202545166015625, -0.9800567626953125, -0.7292022705078125, 0.6868896484375, -0.1799774169921875, -0.047515869140625, 0.03211784362792969, 2.140228271484375, 1.025604248046875, 1.366485595703125, -0.5676116943359375, 0.3297119140625, -0.13677215576171875, -0.369354248046875, 0.17887115478515625, 0.2999114990234375, 2.81060791015625, -0.258453369140625, 0.278289794921875, -0.6541824340820312, 0.5159912109375, -0.31591796875, 1.522125244140625, 1.060150146484375, -0.6368560791015625, -0.4549560546875, 0.03936767578125, -0.6419601440429688, 2.0516357421875, 1.2417755126953125, -0.12170791625976562, 0.5479888916015625, 0.7054443359375, -0.976043701171875, -0.5350341796875, -1.598876953125, -0.09332275390625, 0.6827392578125, 1.11102294921875, -0.4815673828125, 0.4417572021484375, -0.12835693359375, 0.0133819580078125, -0.80816650390625, 1.186126708984375, 0.0028839111328125, 1.0763702392578125, 1.95452880859375, 0.7647552490234375, 0.15658950805664062, 0.8898086547851562, 1.5687255859375, 0.5598907470703125, 1.48114013671875, -1.3930206298828125, 2.533599853515625, -0.25042724609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000029.npy"}
{"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": 0.29195964336395264, "std": 0.9811336994171143, "min": -3.5572509765625, "p10": -0.8699172973632813, "median": 0.28391265869140625, "p90": 1.4794342041015622, "max": 3.22747802734375, "pos_frac": 0.65625, "sample": [0.48785400390625, -0.533721923828125, 1.454833984375, 0.471832275390625, -0.10218429565429688, 0.652557373046875, 1.145233154296875, -1.492767333984375, 0.802825927734375, 0.4835662841796875, 0.04846954345703125, 0.7067718505859375, 1.41131591796875, -1.091796875, 0.6809234619140625, -0.640533447265625, 0.299957275390625, -1.70306396484375, 0.2893524169921875, 0.2416839599609375, 0.465728759765625, 1.06951904296875, -0.222808837890625, -0.2784404754638672, 0.3752899169921875, 1.272491455078125, 1.99749755859375, 0.22615623474121094, 0.5209465026855469, 1.124664306640625, 0.2962646484375, -0.18688583374023438, -0.0812835693359375, -1.57366943359375, -0.5459403991699219, 1.1182861328125, -0.9193115234375, -0.20188140869140625, 0.161163330078125, 2.20611572265625, 0.34954833984375, -0.41269683837890625, 0.328155517578125, 0.4095458984375, 1.359619140625, -1.631500244140625, 0.029094696044921875, 0.7571868896484375, 0.11858367919921875, -1.0284042358398438, 0.7079315185546875, -0.1335906982421875, 1.061431884765625, 0.86370849609375, 2.0887451171875, 0.05585479736328125, -1.162078857421875, 0.1384735107421875, 0.2470703125, 1.5947418212890625, 1.1790771484375, -0.03668212890625, 0.278472900390625, -0.7198257446289062, -0.32872772216796875, -0.18576812744140625, 0.2783966064453125, 0.3065185546875, 0.35225677490234375, 1.588104248046875, -0.11589622497558594, 2.426513671875, 0.2948150634765625, 0.23413467407226562, 0.639404296875, 3.22747802734375, -1.2650833129882812, 1.5499114990234375, 0.00720977783203125, 0.7021484375, -0.5029983520507812, 1.7679443359375, 0.6317138671875, -0.261749267578125, 0.4652748107910156, -0.513916015625, 2.703582763671875, 1.277557373046875, -0.3169708251953125, -0.303131103515625, 0.14255523681640625, -1.1623077392578125, 0.5183486938476562, 1.3631591796875, -1.0111083984375, 1.87213134765625, 0.4810791015625, 0.004364013671875, -0.5091552734375, 0.13330078125, 0.9300537109375, -0.500518798828125, 0.6013031005859375, 1.3371047973632812, 0.136474609375, 0.05467987060546875, 1.08819580078125, 0.590240478515625, -0.5870208740234375, -0.86578369140625, 2.766265869140625, -0.5621337890625, 1.13751220703125, 0.88958740234375, 0.26032447814941406, -0.8795623779296875, 0.5856056213378906, 0.3250579833984375, -0.17240142822265625, 1.536834716796875, -0.859375, -0.45104217529296875, -0.3835296630859375, -3.5572509765625, 0.1412181854248047, 0.4565887451171875, 0.528167724609375, -0.54632568359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000030.npy"}
{"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.25791555643081665, "std": 0.9926722049713135, "min": -1.765899658203125, "p10": -0.8736709594726562, "median": 0.1938934326171875, "p90": 1.4479736328124995, "max": 4.930572509765625, "pos_frac": 0.609375, "sample": [0.394012451171875, 1.651885986328125, 0.5006103515625, -0.213531494140625, -0.43982696533203125, 1.201324462890625, -0.93463134765625, 1.21923828125, -0.1156158447265625, 1.265380859375, 1.9107666015625, -0.6523666381835938, 1.20965576171875, -1.242767333984375, 0.493255615234375, 0.1003875732421875, 0.16872406005859375, 0.3196907043457031, 1.17608642578125, 0.8575592041015625, 0.5372314453125, -0.737335205078125, 0.34259033203125, -0.785064697265625, -0.124725341796875, -1.70855712890625, 1.3603515625, 0.1221771240234375, -0.3773040771484375, 0.1708526611328125, 0.2702178955078125, 1.09271240234375, 0.9082870483398438, -0.0919036865234375, 0.656341552734375, 0.0806732177734375, 0.2582550048828125, 0.501373291015625, 0.30499267578125, 0.0, 0.3744964599609375, 0.6653976440429688, 0.0, -0.5572509765625, 0.2923698425292969, 0.3236541748046875, -0.673095703125, 0.5829544067382812, 0.4262542724609375, -0.9324951171875, -1.4639892578125, -1.4250335693359375, 0.6922149658203125, 1.369293212890625, 0.3292121887207031, 0.14962005615234375, 2.27972412109375, -0.817779541015625, -0.038909912109375, 0.241180419921875, 0.0, 0.76019287109375, 0.6300048828125, -0.6357421875, 0.4727935791015625, 2.179107666015625, 0.2156982421875, -0.776397705078125, 1.83746337890625, 0.10735511779785156, 0.7696685791015625, 2.19720458984375, -0.7344894409179688, 0.23595428466796875, -1.1400299072265625, 0.788421630859375, -0.0269317626953125, 0.1345195770263672, 2.374664306640625, -1.140106201171875, 0.6129112243652344, 0.7907791137695312, 1.8634033203125, -0.09807395935058594, -0.724822998046875, 0.172088623046875, -1.6700439453125, 2.66552734375, 0.8021087646484375, 1.58392333984375, -0.62396240234375, 0.45101356506347656, 2.362335205078125, 0.015472412109375, 0.2860107421875, -1.1254196166992188, -0.507720947265625, -1.356658935546875, 0.9812774658203125, 0.0, 0.322113037109375, 4.930572509765625, 0.3252105712890625, -0.4068641662597656, -0.7500762939453125, -0.860015869140625, -0.9055328369140625, -0.15639686584472656, 1.39208984375, 0.07483673095703125, 0.16754150390625, 0.15426254272460938, 0.0097503662109375, -0.69451904296875, -0.01983642578125, 0.363433837890625, -0.1895885467529297, -0.0848846435546875, -0.1512451171875, 1.578369140625, 0.6746826171875, 0.27646636962890625, -0.16523170471191406, -0.128204345703125, -0.6803932189941406, 0.87579345703125, -1.765899658203125, 0.6284332275390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000031.npy"}
{"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": 0.3075369596481323, "std": 1.0586572885513306, "min": -2.978271484375, "p10": -0.825103759765625, "median": 0.21083831787109375, "p90": 1.66893310546875, "max": 3.86505126953125, "pos_frac": 0.640625, "sample": [1.71502685546875, 0.1270294189453125, -1.19415283203125, -0.08203125, -0.214447021484375, -0.46966552734375, 0.036617279052734375, -0.081024169921875, -0.0399627685546875, 0.19140625, -0.54693603515625, 0.6373443603515625, -1.2335739135742188, 0.24920654296875, 1.0608978271484375, -0.19302749633789062, 1.40106201171875, 0.711212158203125, -0.24945068359375, 1.227813720703125, -0.3949737548828125, 0.6341552734375, 0.5079345703125, 0.468780517578125, 0.0662689208984375, 1.66021728515625, 0.4145050048828125, -0.05580329895019531, 0.8563232421875, 0.9962997436523438, 0.11590576171875, 0.06357574462890625, -0.10582733154296875, -1.261993408203125, -0.6361083984375, 0.6435546875, -0.654388427734375, 1.050872802734375, 2.788970947265625, -0.1013641357421875, 0.137725830078125, -1.076934814453125, 0.6942138671875, -0.45542144775390625, 0.2093505859375, 0.775970458984375, 0.2123260498046875, 0.46610260009765625, -0.64935302734375, -0.6865692138671875, 1.630859375, -1.15911865234375, -0.000213623046875, -0.8466796875, 1.9698486328125, -0.4935455322265625, 0.432220458984375, 3.86505126953125, 2.54620361328125, 1.20269775390625, 1.1199951171875, 0.015472412109375, 0.7885284423828125, 0.59796142578125, 0.110137939453125, -0.81585693359375, 0.148223876953125, -0.2901611328125, 0.5426025390625, 0.648956298828125, 0.083160400390625, -0.007781982421875, 0.08706283569335938, 0.36602783203125, 0.7402496337890625, -0.0132598876953125, -0.1298980712890625, -2.978271484375, 0.8172836303710938, 1.785797119140625, 0.20196533203125, 0.9224853515625, 0.13564109802246094, -0.0582122802734375, -1.383331298828125, 1.3652801513671875, 0.541290283203125, -0.2910003662109375, -0.10609054565429688, 0.05706787109375, -0.6437149047851562, -0.6842575073242188, 1.8831024169921875, 2.03326416015625, -2.456298828125, 0.386749267578125, 0.4695281982421875, -0.46771240234375, 0.7004623413085938, -0.8960723876953125, 0.29355621337890625, 2.3907470703125, 1.2059783935546875, 0.7702789306640625, -2.0987548828125, 0.790191650390625, -0.18743324279785156, -0.6298294067382812, 0.2570343017578125, 0.5391044616699219, 0.90350341796875, 0.16854095458984375, 0.41656494140625, 1.59698486328125, 2.561798095703125, 0.411041259765625, 0.31005859375, 2.21954345703125, 0.96771240234375, -1.32843017578125, 0.12799072265625, 0.5141830444335938, -2.76934814453125, 1.68927001953125, -0.24945068359375, 0.819976806640625, 0.639984130859375, 2.82257080078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000032.npy"}
{"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.4025673568248749, "std": 1.1038322448730469, "min": -2.568206787109375, "p10": -0.8117660522460936, "median": 0.341888427734375, "p90": 1.5911224365234375, "max": 4.952239990234375, "pos_frac": 0.671875, "sample": [0.23919677734375, 1.09088134765625, 0.210479736328125, -1.21868896484375, 0.25537872314453125, 0.432464599609375, -1.1789321899414062, 0.520263671875, 0.921905517578125, 0.46675872802734375, 0.823089599609375, 0.3404541015625, 4.952239990234375, 0.359619140625, 0.0, 0.5460739135742188, -0.4650726318359375, 0.493743896484375, -0.2992706298828125, -2.456817626953125, -0.21087646484375, 1.5872955322265625, 0.4358634948730469, -0.5882568359375, -0.7606658935546875, 0.08612060546875, 0.1539306640625, 0.560272216796875, 2.02716064453125, -0.35693359375, -0.039154052734375, -0.205230712890625, 0.67144775390625, -1.6739501953125, 0.9413700103759766, 0.3739013671875, 0.5902252197265625, 1.601409912109375, -2.3753662109375, 0.803863525390625, 0.34332275390625, 0.017955780029296875, -0.161376953125, 0.7974853515625, -2.1793212890625, -1.0218505859375, 3.61968994140625, -0.42440032958984375, 0.41717529296875, -0.325958251953125, 1.459808349609375, -1.5395355224609375, -0.141143798828125, 0.0, 1.6446075439453125, 0.3297119140625, -0.06862068176269531, 0.293701171875, 1.5861968994140625, 0.83831787109375, 2.664398193359375, -1.514373779296875, 0.13316726684570312, 0.8952255249023438, 0.07613372802734375, -0.4586181640625, 0.2725982666015625, -0.03554534912109375, 0.65142822265625, 1.56329345703125, 1.345458984375, -0.3062286376953125, -1.3206787109375, 1.358856201171875, 0.48150634765625, -0.231475830078125, -1.0907135009765625, 1.505035400390625, 0.28208160400390625, 0.5966796875, 0.6053466796875, -0.4098358154296875, 3.217529296875, 0.66094970703125, 1.483123779296875, 0.1387939453125, 0.8198699951171875, 1.677581787109375, 1.4163055419921875, 0.4371337890625, 0.6898956298828125, 0.468963623046875, 1.3341445922851562, -0.15678787231445312, 0.23188018798828125, 1.81585693359375, 0.315277099609375, 1.2281494140625, -2.568206787109375, 1.64306640625, 0.0371246337890625, 0.033985137939453125, 1.6000518798828125, 0.6383056640625, -0.266082763671875, 0.581268310546875, 0.1124420166015625, 1.360809326171875, 0.0711517333984375, 1.466644287109375, 0.18817138671875, -0.401275634765625, 1.97869873046875, -0.471160888671875, 1.1597213745117188, 0.8801422119140625, -0.930999755859375, 1.2630424499511719, -0.216796875, -0.13836669921875, 1.5732421875, 0.564300537109375, 1.55145263671875, -0.24407958984375, -0.00616455078125, 0.10216522216796875, -0.3156394958496094, 2.2972412109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000033.npy"}
{"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.7076154947280884, "std": 1.1651321649551392, "min": -1.8955078125, "p10": -0.4405685424804687, "median": 0.5039710998535156, "p90": 2.5209045410156246, "max": 4.37408447265625, "pos_frac": 0.75, "sample": [-0.0379638671875, -0.3769073486328125, 0.9371337890625, 0.78009033203125, -0.081146240234375, 1.306060791015625, -1.7076873779296875, 0.9054679870605469, 0.290802001953125, 1.62493896484375, 1.114166259765625, 0.765594482421875, 4.08111572265625, 0.53851318359375, 0.8441162109375, 2.42095947265625, -0.3287353515625, -0.2506866455078125, 0.6247272491455078, 0.1166839599609375, 0.3619384765625, -0.2542724609375, 1.3106689453125, 0.89361572265625, -0.415771484375, 2.476776123046875, 2.8836669921875, -0.8440208435058594, -0.04302787780761719, -1.2755126953125, -0.4198455810546875, -0.0644683837890625, -0.19757080078125, -0.19708633422851562, 0.2737884521484375, 0.636566162109375, -0.49627685546875, 0.091796875, -0.22515869140625, 0.344696044921875, -0.9473876953125, -0.12310791015625, 0.7906951904296875, 2.314208984375, 2.90521240234375, 0.30626678466796875, -0.9298858642578125, 0.182830810546875, 0.21856689453125, 0.02154541015625, 0.786590576171875, 0.4890289306640625, 0.2041149139404297, 2.13604736328125, 1.9534912109375, 0.604736328125, 0.3712005615234375, 0.80194091796875, 3.15826416015625, 3.14080810546875, 1.370635986328125, 0.4184989929199219, 0.752685546875, 0.0577850341796875, 4.37408447265625, 3.1414794921875, 0.71990966796875, 0.12385177612304688, 0.8720169067382812, 0.093597412109375, 0.8721466064453125, 0.356292724609375, 0.246612548828125, -0.0240325927734375, 3.169586181640625, -1.0356674194335938, 0.2725410461425781, 0.181854248046875, 0.8045425415039062, -0.22188186645507812, 2.095703125, 2.01446533203125, 2.623870849609375, -0.3508453369140625, 0.91351318359375, 2.77734375, 0.0460662841796875, 1.0701904296875, 0.4196014404296875, 1.1541748046875, 1.8102569580078125, -0.603363037109375, 3.167236328125, -0.62579345703125, 1.75537109375, 1.16033935546875, 0.0341033935546875, 0.006256103515625, 0.583221435546875, 0.005035400390625, 0.204010009765625, 0.62457275390625, 0.81268310546875, 1.793701171875, 2.8974609375, -0.3165569305419922, 1.159576416015625, 0.5189132690429688, 0.4394874572753906, 2.640380859375, 1.782196044921875, -0.05706787109375, -1.8955078125, -0.488922119140625, 0.296661376953125, 0.421173095703125, 0.468780517578125, 1.688690185546875, -1.431304931640625, 0.784912109375, 0.058197021484375, 1.149261474609375, 1.1151962280273438, 0.655181884765625, -1.44232177734375, 0.9321441650390625, 1.27130126953125, 1.0957794189453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000034.npy"}
{"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": 0.35799336433410645, "std": 1.3265807628631592, "min": -5.3641357421875, "p10": -0.9346420288085937, "median": 0.4135322570800781, "p90": 1.8745010375976556, "max": 3.50433349609375, "pos_frac": 0.65625, "sample": [0.23506927490234375, 0.68267822265625, -1.096343994140625, 0.8162841796875, -0.7832489013671875, 3.50433349609375, 0.1436767578125, 1.568084716796875, -2.484954833984375, 0.22650146484375, 0.45184326171875, 0.24144744873046875, -1.510986328125, -0.601898193359375, 0.5522003173828125, 0.37522125244140625, -0.715484619140625, -1.5738525390625, -0.373321533203125, -0.18956947326660156, -0.3362884521484375, 0.96539306640625, -0.913360595703125, 0.9614410400390625, -0.42522430419921875, -0.154693603515625, 0.07720947265625, -0.5492782592773438, -0.59405517578125, -0.087890625, 0.13665771484375, -0.14126968383789062, -0.8198699951171875, 0.3209686279296875, 1.561370849609375, 1.445159912109375, 0.632354736328125, 0.0418701171875, 0.9834709167480469, 0.0, -0.1296539306640625, 1.058624267578125, 0.4601593017578125, 1.095947265625, -0.09722900390625, -0.140838623046875, 2.362030029296875, -1.4100799560546875, 0.644287109375, -0.59710693359375, 0.48394775390625, -0.33077239990234375, 2.838470458984375, 0.6132659912109375, 3.4248046875, -0.3862152099609375, 2.683258056640625, 0.594085693359375, 1.032379150390625, -0.10748672485351562, 2.57464599609375, 1.27825927734375, 3.061187744140625, 2.0937271118164062, -0.58648681640625, 0.87615966796875, 0.0589447021484375, 0.03997039794921875, -1.515380859375, 0.8595790863037109, 0.968780517578125, -3.08624267578125, -1.019989013671875, 0.6256103515625, 0.5582504272460938, 0.9459228515625, 0.06386566162109375, 1.319366455078125, 2.03076171875, 1.12213134765625, 0.5243911743164062, 0.2837104797363281, 2.156951904296875, 0.7668304443359375, 1.0643997192382812, 0.91290283203125, -0.09454536437988281, 0.8291015625, -4.741302490234375, 0.062084197998046875, 0.336090087890625, 0.9002609252929688, 0.2191448211669922, 0.5670547485351562, 0.2854938507080078, 1.021484375, -0.3477783203125, -5.3641357421875, 1.717437744140625, 0.816558837890625, -2.5262451171875, 2.198883056640625, 0.0, -0.9842987060546875, 2.1546478271484375, 1.2127685546875, 0.509674072265625, 0.4608154296875, 0.640594482421875, -0.4116973876953125, 1.620880126953125, 0.317169189453125, -2.0887451171875, 0.0748748779296875, 1.4486312866210938, 1.7076416015625, 0.7516822814941406, 1.8208465576171875, -0.86932373046875, 1.1776771545410156, -0.1263427734375, 0.156036376953125, -0.380615234375, 0.901031494140625, 1.745025634765625, 1.70550537109375, -0.2423858642578125, 1.99969482421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000035.npy"}
{"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.6478171348571777, "std": 1.347158670425415, "min": -3.26531982421875, "p10": -1.0355697631835936, "median": 0.5635337829589844, "p90": 2.4320068359374996, "max": 6.1717529296875, "pos_frac": 0.7109375, "sample": [0.11541748046875, 1.192230224609375, 0.800079345703125, 1.1237640380859375, -0.416961669921875, 1.63275146484375, -1.028076171875, 0.33155059814453125, 0.985748291015625, -0.16632461547851562, 0.929718017578125, -0.788848876953125, 1.90631103515625, 0.14404296875, -0.37677001953125, 1.1528053283691406, -1.0530548095703125, 1.059600830078125, 1.84014892578125, 2.906768798828125, 0.8694000244140625, 1.6153564453125, -1.3035392761230469, 0.0, 1.2481002807617188, 0.688323974609375, -0.4534912109375, 1.046417236328125, 0.31280517578125, 0.4451484680175781, 4.04022216796875, 2.3949737548828125, 0.697296142578125, 0.299224853515625, -1.3660888671875, 2.6665191650390625, 0.5664291381835938, 0.717010498046875, -0.04430389404296875, 2.5184173583984375, 6.1717529296875, -0.058685302734375, 0.06109619140625, 1.0579071044921875, 0.560638427734375, 1.7147216796875, 0.6783447265625, 2.1754913330078125, 1.1831817626953125, 1.220703125, 0.0069580078125, 1.3950653076171875, -0.02581787109375, 0.644317626953125, -0.6152801513671875, -0.23766326904296875, -1.46417236328125, 0.702239990234375, 3.136688232421875, 3.03021240234375, 0.0, 0.797027587890625, 1.863861083984375, 0.7791290283203125, 1.343292236328125, 1.853179931640625, 1.2501564025878906, 0.13376617431640625, 0.5408782958984375, 0.223968505859375, 3.213775634765625, -0.49383544921875, -2.219512939453125, -1.52716064453125, -0.8163604736328125, 0.5466041564941406, 0.4135589599609375, 3.059844970703125, 1.00244140625, 0.07708740234375, -1.2901611328125, 1.2767333984375, -0.571380615234375, -0.02056884765625, -1.614501953125, 0.02828216552734375, -1.71746826171875, -0.475433349609375, -0.2833709716796875, -0.13396453857421875, 1.023162841796875, 2.21600341796875, 1.2798004150390625, 0.6136627197265625, -1.22607421875, 1.549072265625, 0.02635955810546875, 0.053436279296875, -0.2879180908203125, 0.2551116943359375, -3.26531982421875, 0.975982666015625, 0.441558837890625, 0.320404052734375, 0.73468017578125, 0.7117385864257812, 1.42108154296875, 2.57684326171875, -1.503326416015625, 1.5763397216796875, 0.036174774169921875, 0.526031494140625, 0.430023193359375, 1.5543212890625, -0.4418182373046875, -0.7048988342285156, 3.26953125, 2.853363037109375, 2.02850341796875, 0.069122314453125, 0.98565673828125, 3.7601318359375, 0.9758148193359375, 0.39863014221191406, -1.138824462890625, -0.3600196838378906, 0.2958221435546875, 1.0637359619140625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000036.npy"}
{"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.517514169216156, "std": 1.699217438697815, "min": -3.240509033203125, "p10": -1.7440246582031247, "median": 0.5259323120117188, "p90": 2.5485458374023438, "max": 5.255523681640625, "pos_frac": 0.65625, "sample": [0.294921875, 2.1383056640625, 1.95306396484375, 0.269805908203125, 1.157958984375, -0.5359477996826172, 5.12664794921875, 0.02181243896484375, 1.413970947265625, -1.555877685546875, 3.16632080078125, -0.09488868713378906, 0.83551025390625, 2.3699493408203125, 3.080780029296875, 2.995849609375, 1.525787353515625, -0.8009452819824219, 2.2149658203125, 0.0272064208984375, 0.8290863037109375, 1.284423828125, -0.4155292510986328, 0.27130126953125, -2.951812744140625, -0.39886474609375, 0.5214385986328125, 1.057891845703125, 0.419921875, -3.1549072265625, 1.108795166015625, -0.4475555419921875, 0.6150970458984375, -1.54888916015625, 1.7608642578125, 0.24151611328125, -0.52166748046875, -2.44873046875, 1.757598876953125, 0.568511962890625, 0.438507080078125, -2.08905029296875, -0.7108001708984375, 2.722015380859375, 2.5298919677734375, 0.817718505859375, 0.07600975036621094, -0.48504638671875, -1.683197021484375, -2.7242431640625, -0.0926513671875, -1.4496917724609375, 1.14886474609375, 2.592071533203125, -0.462890625, -0.30078125, -0.61700439453125, 1.45745849609375, 0.97515869140625, -1.371124267578125, 0.518829345703125, 0.65533447265625, 1.466522216796875, 0.673187255859375, -3.240509033203125, 0.460052490234375, 3.2076416015625, 0.842315673828125, 2.46734619140625, -1.9442138671875, 1.8214111328125, 0.530426025390625, 1.746490478515625, 1.096466064453125, 1.3840484619140625, 0.406646728515625, 0.591094970703125, 4.9144287109375, 0.812744140625, -1.885955810546875, -0.7708663940429688, 5.255523681640625, 0.50347900390625, 0.246490478515625, 2.249359130859375, 1.141693115234375, -0.62738037109375, -0.2763824462890625, 0.09527587890625, 4.491943359375, -0.947509765625, 2.88507080078125, -0.367218017578125, 0.73468017578125, 0.223388671875, 0.6415290832519531, 0.353271484375, 0.8919486999511719, 0.14556884765625, -0.6964111328125, 1.71514892578125, 0.16497421264648438, -2.385650634765625, -0.4053802490234375, -0.8587684631347656, 2.38739013671875, -0.37664794921875, 1.760894775390625, 1.847503662109375, -0.2521533966064453, 0.8227996826171875, -2.7255859375, 3.267181396484375, -2.38580322265625, 2.18768310546875, 1.327911376953125, 2.45367431640625, -2.791748046875, -0.594146728515625, -1.3651885986328125, 0.5340576171875, -2.47412109375, 1.229095458984375, 1.7673492431640625, 1.688140869140625, 0.94427490234375, -1.566802978515625, 2.70306396484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000037.npy"}
{"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.5638105273246765, "std": 1.5764820575714111, "min": -5.3186492919921875, "p10": -1.2328718185424805, "median": 0.4490928649902344, "p90": 2.453620147705077, "max": 4.1275634765625, "pos_frac": 0.6640625, "sample": [2.17828369140625, 2.3337020874023438, 0.12239646911621094, -0.5903091430664062, -0.20775604248046875, 1.0621566772460938, -0.459381103515625, 3.84197998046875, 3.129852294921875, 0.7481002807617188, 0.199737548828125, -0.7667465209960938, 0.3329315185546875, 1.405914306640625, 0.290283203125, 0.249053955078125, 0.809295654296875, -0.6314010620117188, 1.8116607666015625, -1.1605224609375, 3.906524658203125, -1.757476806640625, 0.11203384399414062, 0.2033538818359375, 0.6120834350585938, -0.9814605712890625, -1.3816680908203125, -0.27276611328125, 1.0642852783203125, -0.11987113952636719, 0.1941680908203125, 1.0570068359375, 0.028076171875, 1.3671875, 0.7793426513671875, 1.890167236328125, 2.2320404052734375, -0.018306732177734375, -0.54583740234375, -0.2432861328125, 0.9035720825195312, 0.5587196350097656, 4.007720947265625, -1.1630020141601562, -2.0482177734375, 0.2881927490234375, 3.7493896484375, 1.659149169921875, 0.3995361328125, -0.350433349609375, -0.58221435546875, 0.44925689697265625, 0.039703369140625, 0.04776763916015625, -0.2797355651855469, -2.1454849243164062, -1.5689697265625, -0.38675689697265625, 0.9217529296875, 0.528167724609375, 0.2261962890625, 0.7828369140625, 2.733428955078125, 1.3399200439453125, 2.0541229248046875, 0.11728286743164062, 0.7304840087890625, -5.3186492919921875, 1.127960205078125, -0.61712646484375, 3.079833984375, 2.135833740234375, 0.2703857421875, -1.27301025390625, -2.8971099853515625, 0.53857421875, 3.3532562255859375, 0.7551345825195312, 2.1824798583984375, -2.7845001220703125, -0.15319061279296875, -1.1272964477539062, 1.018402099609375, 0.8472900390625, 0.3918304443359375, 2.7647705078125, 3.9847412109375, 1.9219207763671875, 1.327850341796875, -0.09665679931640625, 0.65948486328125, -0.191925048828125, 1.81378173828125, 0.8491668701171875, 0.4489288330078125, 1.2820205688476562, 1.8966827392578125, -1.3824615478515625, 1.832275390625, -2.787750244140625, -1.2975234985351562, 0.0, -1.0498809814453125, -2.3634033203125, -0.0319671630859375, 1.713623046875, 3.581207275390625, -0.08351898193359375, 1.99884033203125, 1.807159423828125, 4.1275634765625, -1.19189453125, 2.084197998046875, 1.28338623046875, 0.6575927734375, 0.321685791015625, 0.241943359375, -0.358795166015625, 0.4304046630859375, 3.1573486328125, 0.733642578125, 1.5421905517578125, -0.998077392578125, 1.0076408386230469, 2.19830322265625, 1.1797943115234375, -1.2156696319580078, 1.0018119812011719], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000038.npy"}
{"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 1.2989695072174072, "std": 1.9712737798690796, "min": -3.0497589111328125, "p10": -0.705713653564453, "median": 1.0809516906738281, "p90": 3.966165161132812, "max": 7.58636474609375, "pos_frac": 0.7421875, "sample": [1.1461334228515625, 0.758087158203125, 2.35919189453125, 1.3494110107421875, 0.60467529296875, 1.44525146484375, -0.4686431884765625, -0.5409317016601562, -0.99920654296875, 2.509033203125, 1.34759521484375, 3.1433258056640625, 4.475372314453125, 1.843963623046875, 3.117095947265625, 0.66448974609375, -0.223297119140625, 2.560272216796875, 1.611968994140625, 1.6180419921875, -0.163848876953125, 1.42803955078125, -0.7679901123046875, 2.47991943359375, 6.4005126953125, 1.4956588745117188, 1.7297019958496094, -1.5471038818359375, 0.7793426513671875, 2.890380859375, 0.49932861328125, -0.243072509765625, -0.4215087890625, -0.2599639892578125, 5.678466796875, 0.901702880859375, 0.14313507080078125, 1.112762451171875, 1.18817138671875, -0.538970947265625, 1.9505615234375, 1.7270965576171875, 2.9386444091796875, 1.3825511932373047, 1.0686874389648438, 5.156402587890625, 3.922027587890625, -0.01807403564453125, 2.1480865478515625, 5.323974609375, 2.3719635009765625, -1.0088424682617188, 2.9544830322265625, 0.28375244140625, 0.7355804443359375, 0.45184326171875, -2.071868896484375, 0.477142333984375, -1.08197021484375, 1.12078857421875, -0.14794921875, 0.0040435791015625, -3.0497589111328125, 2.76708984375, 0.0282745361328125, 2.6994476318359375, -0.38970947265625, -0.21901512145996094, 1.215606689453125, -1.244171142578125, 1.30615234375, 0.488525390625, -0.0926361083984375, 4.06915283203125, 0.8063201904296875, -0.35076904296875, 1.94927978515625, -2.0625, 2.3389053344726562, 1.586212158203125, 2.07220458984375, -0.5528373718261719, 3.21075439453125, 1.0932159423828125, 0.314788818359375, 1.6017303466796875, 6.284423828125, 5.52569580078125, 1.062835693359375, 0.14897537231445312, 0.6572990417480469, -0.3337860107421875, -0.6790237426757812, -0.5475616455078125, 0.0, 0.022979736328125, 0.957305908203125, -1.03173828125, 2.620849609375, 1.71380615234375, 0.5765228271484375, 1.2986679077148438, 7.58636474609375, -2.5112457275390625, 5.291015625, 3.0434188842773438, 1.51800537109375, 1.40625, 5.995574951171875, -2.246734619140625, 0.9922943115234375, 0.8607177734375, 0.3084087371826172, 1.8244400024414062, 2.997039794921875, 0.1863727569580078, 4.3327789306640625, 3.7918701171875, 3.4423675537109375, -0.22387313842773438, 2.6981430053710938, 0.7054557800292969, 0.9512939453125, 5.159912109375, -2.496917724609375, 0.6161727905273438, 0.02191925048828125, 1.3581161499023438], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000039.npy"}
{"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.7541956901550293, "std": 1.9831246137619019, "min": -3.4212646484375, "p10": -1.182305908203125, "median": 0.424560546875, "p90": 3.2183593749999995, "max": 9.6217041015625, "pos_frac": 0.640625, "sample": [0.08135414123535156, -1.975555419921875, 0.30417633056640625, 0.901031494140625, -0.9898529052734375, 1.4245147705078125, -2.01617431640625, 0.421722412109375, 0.07646942138671875, 1.91015625, 2.416046142578125, -0.7684707641601562, 4.160888671875, 1.93707275390625, 0.964141845703125, 3.1125106811523438, -0.6464424133300781, 1.38763427734375, -0.13750839233398438, 1.2583541870117188, -1.0421295166015625, 3.17034912109375, -0.028165817260742188, 0.7925834655761719, 1.8031196594238281, 6.0887451171875, -1.1861724853515625, -1.0860824584960938, 0.040374755859375, 4.0589599609375, -0.13534927368164062, 3.0547332763671875, -0.3122100830078125, 1.9428863525390625, 0.28509521484375, -0.131103515625, 0.38250732421875, -1.1806488037109375, 6.13555908203125, -0.1728515625, 9.6217041015625, 2.84844970703125, 1.06463623046875, 5.5084228515625, -1.3377914428710938, -0.298126220703125, -3.2933502197265625, 1.6166534423828125, -0.0606689453125, 1.4879913330078125, -1.0135345458984375, -2.197021484375, 0.897918701171875, 0.24676513671875, 0.658905029296875, 3.56378173828125, 1.717926025390625, 1.8351364135742188, 0.145751953125, -0.01220703125, 0.307342529296875, 0.0050601959228515625, 0.7810211181640625, -3.4212646484375, 1.47027587890625, 1.294952392578125, -0.483489990234375, 2.45513916015625, 1.044342041015625, 0.41300010681152344, -0.39647674560546875, 0.16863250732421875, 0.48140716552734375, 2.677398681640625, 0.361175537109375, 1.7142753601074219, 0.581390380859375, 0.037616729736328125, -2.081390380859375, 3.43353271484375, -0.6903533935546875, 0.427398681640625, -0.56195068359375, -0.4933433532714844, -0.9405517578125, 0.5151519775390625, -0.9372177124023438, 3.83203125, 0.26619720458984375, -0.29791259765625, 1.8818359375, 1.145355224609375, 3.168487548828125, 3.7889404296875, 0.985809326171875, 2.611114501953125, 4.430816650390625, -1.123016357421875, 1.161102294921875, 0.89111328125, 1.260650634765625, 0.549224853515625, -2.043731689453125, -1.6241340637207031, -1.11102294921875, -0.307891845703125, -0.4881134033203125, -0.5086536407470703, -1.0382766723632812, 1.115570068359375, -2.62646484375, 0.90863037109375, 0.780731201171875, 0.6881370544433594, -0.348663330078125, -3.2479248046875, -1.80633544921875, 0.6128692626953125, 3.33038330078125, 5.14019775390625, 2.085845947265625, -0.0008373260498046875, 2.0709991455078125, -1.026641845703125, 0.3881378173828125, 1.511505126953125, 0.026031494140625, 2.03826904296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000040.npy"}
{"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.7986251711845398, "std": 1.8719266653060913, "min": -6.334716796875, "p10": -1.0891143798828125, "median": 0.6809883117675781, "p90": 3.198910522460937, "max": 5.9659423828125, "pos_frac": 0.640625, "sample": [2.971435546875, -0.9154052734375, 3.3365478515625, -0.5117568969726562, 1.691802978515625, 0.1300048828125, 2.4547119140625, 0.2845306396484375, 0.601715087890625, -0.871795654296875, 0.5979499816894531, 3.101776123046875, -0.0337066650390625, 0.2649993896484375, -2.3432769775390625, 2.681671142578125, 0.0203094482421875, -1.582061767578125, 5.205780029296875, 3.62261962890625, 0.16278076171875, 4.653106689453125, 2.597900390625, 0.692138671875, -1.50244140625, 1.353546142578125, 0.694244384765625, 2.280120849609375, -1.083099365234375, 1.5473213195800781, 1.4810791015625, -6.334716796875, 0.6916275024414062, 5.439239501953125, -2.7725830078125, -0.5693607330322266, 4.51336669921875, 4.452728271484375, 0.3628387451171875, 2.0688629150390625, 3.048583984375, 0.4234466552734375, 1.4298095703125, 0.697418212890625, 0.09625244140625, 1.139129638671875, 4.108642578125, -3.2275238037109375, 0.634002685546875, 3.93695068359375, 0.8565673828125, -0.7697601318359375, 1.16375732421875, -1.2378997802734375, 3.3636474609375, -2.82427978515625, -1.21514892578125, 1.4837646484375, 1.530731201171875, 0.3664703369140625, 1.26727294921875, -0.820159912109375, -0.1132965087890625, 0.41900634765625, -0.80682373046875, -0.3847503662109375, -0.6065216064453125, -0.038177490234375, 0.0, 0.709197998046875, -0.30487060546875, 0.4149169921875, 1.2907257080078125, -0.234405517578125, -0.384185791015625, -2.48992919921875, 2.7844161987304688, -1.4535751342773438, -0.98760986328125, 3.70263671875, 1.5240478515625, 0.9927902221679688, -0.02320098876953125, 0.75677490234375, 1.865234375, 0.107177734375, 2.5380706787109375, 1.8868408203125, 0.836944580078125, 0.7825927734375, -0.063140869140625, 2.2870635986328125, 1.8445510864257812, -0.846527099609375, 0.85394287109375, 2.185455322265625, -0.73236083984375, 3.036468505859375, -0.22574234008789062, 2.028594970703125, 5.9659423828125, -1.1031494140625, 1.673004150390625, 1.20172119140625, -0.4084625244140625, -0.400299072265625, -0.20355224609375, -1.2910003662109375, 0.906005859375, 0.13953399658203125, 2.021394729614258, 2.30267333984375, -0.58074951171875, -0.964599609375, 3.546356201171875, -0.51849365234375, 3.139923095703125, -0.408203125, -0.3367729187011719, 0.0104827880859375, -0.90435791015625, 1.044769287109375, 2.87310791015625, 1.032562255859375, 0.9834976196289062, 0.67034912109375, 2.275054931640625, -0.4832763671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000041.npy"}
{"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 1.014228105545044, "std": 2.421198844909668, "min": -6.979034423828125, "p10": -1.1556396484375, "median": 0.6179161071777344, "p90": 4.249966430664062, "max": 9.26141357421875, "pos_frac": 0.703125, "sample": [0.2535419464111328, 1.3414993286132812, 4.217620849609375, 1.0489501953125, 0.948883056640625, 5.7491455078125, 2.657623291015625, 1.8829345703125, 1.9075927734375, 0.4194755554199219, -0.4732666015625, 4.406494140625, -0.530029296875, 2.1269912719726562, -1.872039794921875, 0.17532730102539062, 3.7109375, -0.1290283203125, -1.6427154541015625, 7.573089599609375, -0.21021270751953125, -0.4620819091796875, -1.01934814453125, 1.87335205078125, 5.5780029296875, 0.46746826171875, 0.219451904296875, -0.22432899475097656, 0.035980224609375, -0.62640380859375, -0.95928955078125, 0.9282245635986328, 3.6318359375, 0.72802734375, 4.95379638671875, 1.9746932983398438, 0.7994384765625, 1.5945968627929688, 0.5596160888671875, 0.696990966796875, -0.3801116943359375, -0.41555023193359375, 1.677032470703125, -0.8226165771484375, 4.13519287109375, 0.8427886962890625, 0.6177749633789062, 0.164093017578125, 0.899169921875, 0.33929443359375, 1.6732330322265625, 1.8555355072021484, 2.6204071044921875, 0.9464225769042969, 0.374725341796875, -0.2053375244140625, 0.668487548828125, -0.6998291015625, -1.139984130859375, 2.0061264038085938, -4.50347900390625, -0.279754638671875, -1.0702362060546875, -1.192169189453125, -0.19696044921875, 5.077606201171875, 0.12639617919921875, 0.2346038818359375, -1.2397918701171875, 1.302032470703125, 0.7972412109375, 0.14297103881835938, 6.117767333984375, 5.93060302734375, -2.779052734375, 0.033050537109375, -6.62994384765625, 0.451690673828125, 5.2750244140625, 0.5034294128417969, 0.46901702880859375, 0.05364990234375, 0.358673095703125, 1.9807586669921875, 3.51177978515625, 1.3369903564453125, 4.8916015625, 1.238311767578125, -0.44376564025878906, -1.2903594970703125, -1.3095703125, 2.070556640625, 6.33807373046875, -0.43503570556640625, 3.199493408203125, 0.5313568115234375, -3.066497802734375, 9.26141357421875, -1.112701416015625, 0.6973724365234375, 3.0853347778320312, 0.5190048217773438, 0.0, 3.50518798828125, 1.8699951171875, 1.183258056640625, 4.325439453125, -0.8444747924804688, -0.11597824096679688, 2.0768890380859375, 2.87457275390625, 2.046630859375, 0.100799560546875, -0.38245201110839844, 0.25164794921875, 3.36566162109375, 1.819183349609375, 0.3544158935546875, 1.6672515869140625, 0.6180572509765625, 1.1139678955078125, 2.780181884765625, -4.06817626953125, 3.0555419921875, -6.979034423828125, 2.4057464599609375, -2.7711944580078125, 0.11190032958984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000042.npy"}
{"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 1.1711583137512207, "std": 2.3596653938293457, "min": -9.3397216796875, "p10": -1.1896347045898437, "median": 0.9083251953125, "p90": 4.779904174804687, "max": 5.8336181640625, "pos_frac": 0.6875, "sample": [2.363250732421875, 3.86553955078125, 1.70721435546875, -1.232177734375, 2.750946044921875, 3.911285400390625, 5.64068603515625, 0.77703857421875, 0.2507896423339844, 0.1475067138671875, -0.5667152404785156, 2.2542877197265625, -0.4943084716796875, 4.810516357421875, 0.5306549072265625, 1.82757568359375, 0.31191253662109375, 1.532806396484375, 3.54754638671875, -4.76507568359375, 5.40399169921875, 1.0354461669921875, 2.15740966796875, 4.166778564453125, -0.85528564453125, -0.3696746826171875, -2.25445556640625, 2.7733917236328125, -0.2589550018310547, 5.081451416015625, 0.65301513671875, 1.9528121948242188, -1.8294677734375, -9.3397216796875, 3.30126953125, 1.0184307098388672, 0.7102279663085938, 2.430938720703125, 1.0142345428466797, -0.44559478759765625, -1.1667556762695312, 1.91162109375, -1.0966758728027344, 5.263702392578125, 0.8651275634765625, 0.18530845642089844, -0.61297607421875, -1.0565567016601562, -1.765411376953125, 1.4614715576171875, -0.81402587890625, 0.7391510009765625, 0.2867088317871094, -1.1714019775390625, 1.23651123046875, -0.68359375, 5.7139892578125, 3.775970458984375, 1.3814926147460938, 5.1668701171875, 0.871429443359375, 0.912445068359375, 0.754547119140625, -2.1844635009765625, 5.74078369140625, 3.468109130859375, 5.625762939453125, -1.7999267578125, 1.759613037109375, 3.813568115234375, 5.07220458984375, -0.14849853515625, 0.4230804443359375, -0.120025634765625, -1.641357421875, -0.25920867919921875, -0.00701904296875, -3.521514892578125, 0.8432769775390625, 1.692779541015625, 0.6508712768554688, 2.5098876953125, 0.1136474609375, 0.6782684326171875, 1.416107177734375, -0.8058624267578125, 1.039642333984375, -2.4184951782226562, 0.0454559326171875, -0.9301300048828125, 2.595123291015625, 1.2607421875, -0.03216552734375, 1.6374053955078125, 4.76678466796875, -0.5154457092285156, 1.0304412841796875, 1.65423583984375, -0.96942138671875, 0.6120223999023438, 3.8756103515625, 1.30731201171875, 1.6492919921875, -1.032867431640625, 3.929046630859375, 4.97802734375, 4.13140869140625, 2.402099609375, 2.969757080078125, 2.1719970703125, 4.927764892578125, 4.69842529296875, 0.38482666015625, -1.3738670349121094, 0.47967529296875, 1.774078369140625, 2.188262939453125, 1.7670783996582031, 5.8336181640625, 2.68695068359375, -0.307037353515625, -1.53070068359375, 0.8362846374511719, 1.02880859375, -0.21356201171875, -0.4623298645019531, 0.904205322265625, -0.8686485290527344], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000043.npy"}
{"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 1.6126174926757812, "std": 3.082651376724243, "min": -9.148681640625, "p10": -1.3003250122070313, "median": 1.5271987915039062, "p90": 5.095428466796874, "max": 9.6348876953125, "pos_frac": 0.765625, "sample": [3.725189208984375, 1.485137939453125, 9.6348876953125, 1.573760986328125, 1.3369293212890625, 2.3592529296875, 0.502197265625, 0.9106559753417969, 1.5604095458984375, -0.9148235321044922, 3.8358001708984375, 4.116363525390625, 1.3332061767578125, -0.6149578094482422, 8.31787109375, 0.9810791015625, 2.420806884765625, 0.599273681640625, -2.659271240234375, 1.06671142578125, 2.701202392578125, -0.5531005859375, 0.215911865234375, 0.1153411865234375, 0.5393733978271484, 1.7476730346679688, 3.1490478515625, 0.36089134216308594, -7.785247802734375, -0.9251060485839844, 5.557769775390625, 2.54248046875, -0.54248046875, -0.49737548828125, 0.6839008331298828, 3.50177001953125, 1.3385009765625, -1.3136444091796875, 1.94354248046875, 2.156036376953125, 9.35015869140625, -0.10661125183105469, 1.3787078857421875, 4.488494873046875, 3.35479736328125, 0.529510498046875, 5.458099365234375, 3.335418701171875, 2.3085784912109375, 0.25482177734375, 0.6045894622802734, 1.57403564453125, -0.777679443359375, 3.6265869140625, 0.7865829467773438, 6.42132568359375, -4.268798828125, 2.9354248046875, -0.052120208740234375, 2.71173095703125, 4.3795623779296875, 4.83099365234375, 3.0504150390625, 7.82635498046875, 4.209564208984375, 0.4851837158203125, 2.557159423828125, -9.148681640625, 6.23748779296875, 4.377593994140625, -7.50115966796875, 3.0596580505371094, -0.141754150390625, 6.76971435546875, 0.5640869140625, 5.7073516845703125, 3.34197998046875, -4.9635009765625, -0.5931854248046875, -4.59381103515625, 1.914794921875, -0.33563232421875, 2.89263916015625, 0.8778057098388672, 3.8621826171875, -0.49372100830078125, 1.5218353271484375, 5.25115966796875, 5.95751953125, 1.5357742309570312, 2.650360107421875, 4.630126953125, 1.130340576171875, 2.592315673828125, -1.7068328857421875, 4.240692138671875, -1.29461669921875, 1.176422119140625, -1.2911376953125, -2.0549163818359375, 4.59423828125, -0.19777297973632812, -1.401824951171875, 2.71746826171875, 1.0072669982910156, 3.308837890625, 2.1258544921875, 1.532562255859375, 0.6051101684570312, 0.530120849609375, 4.832122802734375, 0.792388916015625, 0.34885406494140625, 7.185516357421875, 2.2022628784179688, 4.5546875, 2.410247802734375, 1.0256195068359375, -3.547607421875, 0.3560523986816406, -6.256500244140625, 1.7467613220214844, 0.4025421142578125, 0.7099456787109375, 5.0286865234375, -0.5019912719726562, 3.150848388671875, 3.220001220703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000044.npy"}
{"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 1.3019399642944336, "std": 2.854621648788452, "min": -6.177703857421875, "p10": -1.8806610107421873, "median": 0.8657302856445312, "p90": 5.267817687988281, "max": 11.86224365234375, "pos_frac": 0.6953125, "sample": [2.467987060546875, -2.6693878173828125, 5.3011322021484375, -1.4840621948242188, 0.39821434020996094, 0.5640869140625, -0.21433639526367188, 0.381256103515625, 0.9236106872558594, -1.0701904296875, 4.145538330078125, 4.996368408203125, -1.029876708984375, 0.546966552734375, -2.12066650390625, -1.0941162109375, 0.08378410339355469, 2.6968994140625, 0.0, -3.4576416015625, -0.71856689453125, 2.4463653564453125, 1.1016693115234375, 0.9458694458007812, -1.3348388671875, 6.571807861328125, 1.4068756103515625, 1.3524169921875, 0.878173828125, 2.8469619750976562, 1.3092193603515625, 0.5765762329101562, 0.17523193359375, 3.647003173828125, -0.6318321228027344, 1.999267578125, 1.78936767578125, 0.041942596435546875, 5.393890380859375, 5.4090576171875, 4.08038330078125, 2.438232421875, 2.0966949462890625, -0.9315948486328125, 11.86224365234375, 1.1959304809570312, 4.96630859375, -1.4893112182617188, 5.8367919921875, 2.041351318359375, -1.4713592529296875, 0.5789871215820312, -1.214752197265625, -2.2230072021484375, 1.1362495422363281, 2.862060546875, 0.8532867431640625, 1.9093170166015625, -0.3244132995605469, -6.177703857421875, -4.80426025390625, 3.040130615234375, -0.8397293090820312, 2.6770782470703125, -0.9949951171875, 2.0623779296875, -1.096771240234375, 1.8252754211425781, 2.512939453125, 1.6272201538085938, 0.394287109375, 3.4885711669921875, 0.0804443359375, 0.9977035522460938, 8.52960205078125, 1.7313232421875, 5.076751708984375, -1.829437255859375, 7.69232177734375, 0.05328369140625, 1.9255142211914062, 6.64642333984375, 0.30088233947753906, 9.163665771484375, 0.5555229187011719, 0.4625701904296875, -5.000579833984375, 0.0478973388671875, 2.2403945922851562, -2.181671142578125, 3.9227294921875, 5.33074951171875, 4.4609375, -2.2933502197265625, 4.1346435546875, 0.44272613525390625, 3.3079833984375, 4.138702392578125, 1.56890869140625, -2.150115966796875, -1.07470703125, -1.154754638671875, 2.0231552124023438, -0.4257965087890625, 0.7966156005859375, -0.4910888671875, -0.34039306640625, 0.5786895751953125, 0.1573638916015625, 0.9082756042480469, 0.8168144226074219, -1.32244873046875, 2.1280975341796875, 3.1923828125, 1.8486976623535156, -2.00018310546875, 3.648590087890625, -2.96563720703125, 6.856903076171875, 2.73614501953125, 0.57489013671875, 5.659393310546875, 0.1018829345703125, -0.0040130615234375, 5.2535400390625, -0.5905532836914062, -2.471435546875, 0.3615264892578125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000045.npy"}
{"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 1.5269722938537598, "std": 2.5979697704315186, "min": -6.100555419921875, "p10": -1.1179203033447265, "median": 1.2063064575195312, "p90": 4.972218322753905, "max": 10.398193359375, "pos_frac": 0.75, "sample": [3.1607666015625, 4.15618896484375, 2.194580078125, 0.3736724853515625, 0.9866790771484375, -0.36370849609375, 5.17559814453125, 4.7320556640625, 1.20745849609375, -0.30213165283203125, -1.321502685546875, 0.3464202880859375, -1.5662841796875, 1.399383544921875, 3.011688232421875, -0.3340606689453125, 0.05206298828125, 0.16521263122558594, 1.9788818359375, -0.360260009765625, -1.028961181640625, 1.2051544189453125, 1.3154296875, -2.805267333984375, 3.06964111328125, 1.300537109375, -2.935333251953125, 0.3291015625, 5.31280517578125, -0.5555419921875, -0.1952075958251953, 0.7825851440429688, 4.580902099609375, 6.424774169921875, 5.9169921875, 3.6717529296875, 5.825103759765625, 0.5257492065429688, 0.359771728515625, 3.175445556640625, 3.381622314453125, 2.718292236328125, 0.8802566528320312, 5.670745849609375, 2.21014404296875, 1.4224853515625, 2.22369384765625, -0.6669578552246094, -0.7461051940917969, 1.433929443359375, 4.613861083984375, 3.4676513671875, 1.786285400390625, 1.215972900390625, 0.944183349609375, 3.742401123046875, 2.11859130859375, -1.074951171875, 1.5195846557617188, 0.575469970703125, 0.9216728210449219, 0.9586563110351562, 0.4972190856933594, -0.6652984619140625, 0.01995849609375, 0.5288772583007812, 0.19183349609375, -4.04595947265625, -0.960113525390625, 3.02362060546875, -4.2435302734375, 0.52166748046875, 6.088714599609375, -1.337493896484375, 0.8155670166015625, 3.777252197265625, -1.9869499206542969, -0.1680145263671875, 1.7129364013671875, 3.878021240234375, 1.3201751708984375, 2.93719482421875, 6.378173828125, 0.51080322265625, 1.575225830078125, 1.2009201049804688, -6.100555419921875, -1.077392578125, 1.4543304443359375, 6.22021484375, 0.0649261474609375, 4.8850555419921875, 4.3037109375, 0.8086318969726562, 0.484100341796875, 3.53009033203125, 0.6794509887695312, -1.103515625, -2.889129638671875, 3.7150115966796875, -0.3234539031982422, 3.20880126953125, -1.3526611328125, 0.4034271240234375, 10.398193359375, 2.9185409545898438, 0.1442413330078125, 1.272705078125, 6.81011962890625, 4.59393310546875, 3.163238525390625, -4.1483154296875, -0.7150382995605469, -0.09961700439453125, 0.19331932067871094, 2.6370849609375, 6.016357421875, 1.8257675170898438, 0.2779541015625, 3.552490234375, 0.0, 2.5413818359375, 2.584747314453125, -1.1515312194824219, 0.38748931884765625, 7.71990966796875, 2.01348876953125, 3.74853515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000046.npy"}
{"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 1.0877865552902222, "std": 3.561202049255371, "min": -11.3851318359375, "p10": -2.5177749633789057, "median": 0.962315559387207, "p90": 5.202832031249999, "max": 11.583099365234375, "pos_frac": 0.6640625, "sample": [-1.7659912109375, 0.9816379547119141, 1.0021209716796875, -0.40731048583984375, 6.13897705078125, -4.7452392578125, -4.679229736328125, 2.8140869140625, 6.2197265625, -1.81329345703125, 2.3084716796875, 1.6322860717773438, -1.3507843017578125, -3.387836456298828, -7.13153076171875, -0.436981201171875, 0.229400634765625, 0.06427001953125, 7.217620849609375, -6.588775634765625, 3.26165771484375, -6.3487548828125, 1.2385597229003906, -1.6065826416015625, 3.27178955078125, 0.680419921875, -0.686981201171875, -2.2544708251953125, -6.73345947265625, -6.1063079833984375, 4.1047821044921875, 0.0, 3.951904296875, -2.8692169189453125, -2.056427001953125, 3.539398193359375, 4.1922607421875, 3.579833984375, 0.346771240234375, 1.4085617065429688, 0.9429931640625, -0.5102252960205078, 11.583099365234375, 5.107635498046875, 3.965576171875, 0.257293701171875, 1.4208984375, 7.6300048828125, 0.759002685546875, 0.34307861328125, -0.2672119140625, -1.0410194396972656, 7.955169677734375, 2.1059722900390625, 1.232696533203125, 5.424957275390625, -7.8310546875, 7.792236328125, 3.7711029052734375, 2.7394866943359375, -2.367156982421875, 3.167877197265625, -1.538543701171875, 3.08489990234375, 2.6017684936523438, 0.2140655517578125, -1.2619781494140625, 0.5849075317382812, -0.01174163818359375, 6.95257568359375, 2.2032699584960938, 0.04737091064453125, -0.697174072265625, 0.3636932373046875, 0.4129486083984375, 4.546669006347656, 0.0, 6.3838043212890625, 8.396026611328125, 9.669036865234375, 2.885528564453125, -1.33990478515625, 0.0811614990234375, -3.13360595703125, 3.03302001953125, -1.3163986206054688, 0.37503623962402344, -1.92572021484375, 1.1389694213867188, 4.8807373046875, 2.9503822326660156, 2.61602783203125, -0.5454368591308594, 4.55548095703125, 1.2094955444335938, 0.86859130859375, 3.2471923828125, 1.8125457763671875, 5.9359130859375, 1.2227020263671875, -0.803192138671875, 2.2372894287109375, 0.0903778076171875, 1.493408203125, 4.560577392578125, -2.35601806640625, -0.4448680877685547, 1.250396728515625, 2.0975341796875, -0.40468597412109375, 2.969970703125, -0.6641464233398438, 0.396392822265625, 1.1654644012451172, 0.915863037109375, 3.346466064453125, -1.8189010620117188, 3.109560012817383, -0.41367340087890625, 1.45306396484375, 0.3523101806640625, 4.140960693359375, 1.525726318359375, 4.05438232421875, 0.49652099609375, -11.3851318359375, -4.14483642578125, 4.114768981933594], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000047.npy"}
{"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 1.4889018535614014, "std": 3.1710166931152344, "min": -4.91607666015625, "p10": -1.72886962890625, "median": 0.888702392578125, "p90": 5.851847839355469, "max": 18.12615966796875, "pos_frac": 0.6953125, "sample": [2.1102142333984375, 0.115447998046875, 1.607208251953125, 6.499847412109375, 3.5440673828125, 0.386871337890625, 3.35382080078125, 8.55218505859375, 3.93988037109375, -0.31695556640625, -1.5337753295898438, -0.3052215576171875, 0.2585868835449219, -4.91607666015625, -2.016529083251953, 9.338134765625, -2.601348876953125, -3.53546142578125, -0.525177001953125, -0.22767257690429688, 2.274383544921875, 0.6035804748535156, 3.110870361328125, 1.75213623046875, 1.6886367797851562, 0.43325042724609375, 0.04846954345703125, 0.6290130615234375, -0.700927734375, 0.122894287109375, 2.0064220428466797, 8.05511474609375, 10.18035888671875, 2.2548828125, 2.902862548828125, 1.7316017150878906, 0.07381057739257812, 2.894775390625, 1.341552734375, 4.0924072265625, 0.3029327392578125, 3.313232421875, 1.6827239990234375, -0.15942764282226562, 2.3305435180664062, 0.5782699584960938, -3.7471923828125, 2.4069976806640625, -1.820343017578125, -0.67681884765625, 9.4241943359375, 1.9837360382080078, 0.514739990234375, 0.95184326171875, 2.65185546875, 3.6923370361328125, 0.88580322265625, -0.00992584228515625, 0.27443885803222656, 1.901702880859375, -1.439666748046875, 3.249298095703125, 0.7696533203125, 0.5230484008789062, 1.809844970703125, -4.596282958984375, 3.11627197265625, -1.2913818359375, 4.5911865234375, 2.060699462890625, -1.689666748046875, 1.610595703125, 0.0, 0.3674163818359375, 3.501434326171875, 1.3125457763671875, 4.28985595703125, 0.10884475708007812, 0.2528076171875, 1.67669677734375, 0.7279205322265625, 4.841644287109375, -0.5099334716796875, 7.16680908203125, -0.539764404296875, 5.84759521484375, 1.74951171875, 0.6757583618164062, -0.4902362823486328, 2.1829986572265625, -0.945404052734375, -0.0794677734375, -0.6886444091796875, 0.9779052734375, 2.8770751953125, 6.6138916015625, 2.058074951171875, 1.25872802734375, 1.5167388916015625, 1.4709701538085938, -1.6582794189453125, 5.8617706298828125, 0.0, 7.7779541015625, -0.31072998046875, 3.9135284423828125, 18.12615966796875, 0.9593048095703125, -2.61798095703125, 2.6371383666992188, -0.5025787353515625, -1.3611259460449219, -2.841888427734375, 0.394775390625, 0.3577613830566406, -1.98699951171875, -2.0548248291015625, -2.7753677368164062, 0.3285064697265625, 2.06744384765625, 1.48919677734375, 0.4034423828125, -2.53875732421875, -0.521514892578125, -0.343292236328125, 0.8916015625, 6.03948974609375, 6.20355224609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000048.npy"}
{"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 1.7377345561981201, "std": 3.4948861598968506, "min": -8.925811767578125, "p10": -2.544602584838867, "median": 1.2350263595581055, "p90": 6.092053222656249, "max": 10.2752685546875, "pos_frac": 0.6875, "sample": [0.43280029296875, 4.952484130859375, 2.0712432861328125, 1.5931777954101562, -2.6429595947265625, 4.41650390625, 5.19537353515625, 5.966064453125, 5.6060791015625, 0.7917137145996094, -0.4694976806640625, 0.937591552734375, -4.459808349609375, 2.136932373046875, -0.396575927734375, 9.61895751953125, 1.4388275146484375, 2.0846900939941406, -1.236175537109375, 5.9833984375, -1.6680908203125, -0.8310165405273438, 3.8153076171875, 3.240142822265625, 5.803192138671875, -0.647216796875, -2.634796142578125, 9.374908447265625, 0.193145751953125, 8.844451904296875, 1.4692916870117188, 5.1137847900390625, 5.00726318359375, 1.69580078125, 0.4373779296875, 1.239786148071289, -0.14862060546875, 1.2596435546875, 0.22407150268554688, 0.201873779296875, 1.1302127838134766, 5.268310546875, -1.9628143310546875, 1.2266387939453125, 4.456809997558594, 2.4462127685546875, 2.6529083251953125, 5.330230712890625, 0.003509521484375, -0.417205810546875, 0.551910400390625, -5.976066589355469, 0.3042182922363281, 6.3455810546875, -2.6017494201660156, 2.03466796875, -0.9766693115234375, 0.6294517517089844, -0.1580657958984375, -0.970916748046875, -0.10207366943359375, 6.62841796875, 3.072021484375, 2.21917724609375, -1.1838836669921875, 2.7047576904296875, 0.1534423828125, -2.870820999145508, 5.230865478515625, 7.18487548828125, 3.46319580078125, 4.685577392578125, -0.391998291015625, -1.6308822631835938, 1.2302665710449219, 6.64617919921875, -8.925811767578125, 7.16015625, 7.8702545166015625, -0.76702880859375, 1.0665626525878906, 5.42071533203125, 5.95037841796875, 0.1767120361328125, 0.82781982421875, 3.000335693359375, -2.6763916015625, 4.884674072265625, 8.14373779296875, 5.13995361328125, 1.267669677734375, -0.3752555847167969, 4.267326354980469, 10.2752685546875, 2.92083740234375, -2.520111083984375, -0.0657958984375, 0.51861572265625, 1.3464279174804688, 2.334716796875, -2.7655029296875, 0.943328857421875, -4.0584716796875, -1.9467315673828125, 8.453277587890625, -0.2946014404296875, -4.287750244140625, 5.60595703125, 5.138427734375, 0.14544105529785156, 4.193061828613281, -0.3071136474609375, 2.4277267456054688, 1.1815261840820312, 3.50177001953125, 2.5064697265625, -0.5312232971191406, 7.7745361328125, 1.8095779418945312, 1.1805496215820312, -7.903656005859375, 0.7150344848632812, 1.896026611328125, -3.896453857421875, 2.7863006591796875, -1.362213134765625, -0.060638427734375, -0.0238189697265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000049.npy"}
{"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 1.0769885778427124, "std": 4.051243782043457, "min": -10.503265380859375, "p10": -3.8183746337890625, "median": 0.7360382080078125, "p90": 5.851010131835937, "max": 12.685791015625, "pos_frac": 0.6171875, "sample": [2.84783935546875, -2.6560821533203125, 0.7581787109375, -0.7508182525634766, -1.265899658203125, 1.1827812194824219, -0.2755126953125, -0.14319229125976562, 1.4534912109375, 11.97540283203125, -3.1953125, -0.03871345520019531, 2.83673095703125, 3.86395263671875, 11.48333740234375, 2.15435791015625, 3.42816162109375, -8.464324951171875, 5.6588134765625, -0.0426483154296875, 0.71185302734375, -3.27093505859375, -5.637542724609375, 4.5257568359375, 2.87109375, 3.500579833984375, 2.7518463134765625, 0.10700225830078125, 3.41595458984375, 0.858551025390625, -0.53363037109375, 11.149658203125, 4.4538726806640625, -1.159912109375, 2.312286376953125, 1.0468292236328125, 0.000732421875, 2.797821044921875, -0.9605712890625, 1.666351318359375, -1.4637985229492188, -4.6749114990234375, 3.9441375732421875, 4.73321533203125, 0.18025970458984375, 5.32562255859375, -0.402099609375, 4.636322021484375, -1.2278480529785156, -0.38006591796875, 8.7357177734375, -0.5635223388671875, 3.1244125366210938, 0.8476543426513672, 1.2644119262695312, -1.783447265625, 7.37615966796875, -2.1212310791015625, 2.0095367431640625, 4.0299072265625, 1.2510223388671875, 1.0087432861328125, -3.934967041015625, 0.3958702087402344, -4.09228515625, -3.842620849609375, -1.731292724609375, -0.8298797607421875, 2.8170547485351562, -0.46637916564941406, -1.470703125, -1.1471576690673828, -0.7552928924560547, 1.7893829345703125, 0.36785888671875, 0.19256591796875, 6.159332275390625, -1.14794921875, 9.133407592773438, 0.011384963989257812, 1.315643310546875, -1.396209716796875, -10.503265380859375, -3.894500732421875, 0.06824493408203125, 5.2945556640625, -3.533538818359375, -4.346771240234375, 3.6962051391601562, -3.8079833984375, 5.843109130859375, -3.2813720703125, 10.48370361328125, 0.6626434326171875, 0.070648193359375, -4.511962890625, -0.5283966064453125, -0.4162445068359375, 1.056549072265625, 3.0437164306640625, 3.820648193359375, 5.801055908203125, 2.620086669921875, 4.76092529296875, -1.707000732421875, 0.900909423828125, 0.32010650634765625, 5.86944580078125, 0.5015792846679688, -10.307464599609375, -7.05499267578125, 2.9598922729492188, 12.685791015625, 1.577880859375, -1.5762939453125, -6.254425048828125, 6.095947265625, 2.421051025390625, -2.239471435546875, 0.713897705078125, 2.39727783203125, 1.24432373046875, 7.448516845703125, 3.52740478515625, 1.41552734375, 0.13531494140625, -0.5564460754394531, 6.301605224609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000050.npy"}
{"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 1.8410539627075195, "std": 4.9682440757751465, "min": -13.018218994140625, "p10": -2.9789962768554688, "median": 1.408905029296875, "p90": 8.135543823242188, "max": 16.4383544921875, "pos_frac": 0.65625, "sample": [2.73223876953125, 0.00830078125, 4.63636589050293, 2.019317626953125, 1.6278228759765625, -1.3385238647460938, 4.42755126953125, 6.905670166015625, -4.375762939453125, -0.024169921875, 5.4207000732421875, 7.0566558837890625, 6.02691650390625, 3.595184326171875, -1.2570877075195312, -1.1067352294921875, 8.32232666015625, 0.73516845703125, 1.675872802734375, 6.07568359375, -9.10394287109375, -2.955718994140625, 2.094757080078125, -2.01983642578125, 6.0652923583984375, 16.4383544921875, 0.1233978271484375, 2.84796142578125, 10.02691650390625, 1.1739730834960938, -0.34783935546875, 10.13897705078125, 10.67169189453125, 6.9023895263671875, 2.0658531188964844, 7.757568359375, 7.4583740234375, 4.5310516357421875, 1.28961181640625, -0.97369384765625, 2.7892913818359375, 5.486907958984375, 8.095855712890625, 4.0557861328125, -0.36028289794921875, 6.77862548828125, 4.95648193359375, 0.9320259094238281, -6.02362060546875, -8.8829345703125, 8.872421264648438, -2.4112091064453125, -13.018218994140625, 4.00567626953125, -1.122772216796875, 5.6535797119140625, 1.6848373413085938, 1.39007568359375, -8.29168701171875, 5.597442626953125, -0.51025390625, 2.44464111328125, -2.0465545654296875, 1.1688308715820312, -12.7972412109375, -2.80462646484375, 0.30361175537109375, -1.3819580078125, 3.594879150390625, 6.517608642578125, 1.6806449890136719, 1.249114990234375, -0.8154296875, 4.2974853515625, -2.468353271484375, 0.0, 8.2281494140625, -5.28472900390625, 1.427734375, 6.79827880859375, 1.1939697265625, -0.7506942749023438, -0.24782943725585938, 2.7490081787109375, -11.472991943359375, -1.05169677734375, -0.344451904296875, 1.99639892578125, -3.41607666015625, 13.24664306640625, -0.3118743896484375, -2.2491912841796875, 0.2962188720703125, 0.2412567138671875, 0.7585372924804688, 2.951324462890625, -2.493438720703125, 2.3484039306640625, 8.499496459960938, -1.145416259765625, -1.56890869140625, -3.507843017578125, -1.0205841064453125, 6.849151611328125, 14.9451904296875, 11.8385009765625, 6.755302429199219, 9.5787353515625, 1.15679931640625, 0.957550048828125, 2.814422607421875, 0.7369403839111328, 2.3470458984375, 2.5316619873046875, -1.8306732177734375, -1.408782958984375, -3.0333099365234375, -3.935821533203125, 0.53076171875, 1.92108154296875, 9.107315063476562, 1.4927978515625, 2.544342041015625, -0.3071441650390625, 1.03375244140625, 5.014190673828125, 0.4637451171875, 1.71234130859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000051.npy"}
{"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 3.134784698486328, "std": 5.010632514953613, "min": -8.86920166015625, "p10": -1.689622497558593, "median": 2.2425689697265625, "p90": 10.310006713867187, "max": 19.779022216796875, "pos_frac": 0.7734375, "sample": [4.92498779296875, 1.89239501953125, -4.6318359375, 1.3013916015625, 0.6520843505859375, 2.2231903076171875, 11.40985107421875, 1.7095108032226562, 1.1309795379638672, 8.525848388671875, 2.0199127197265625, 2.778717041015625, -2.1578521728515625, -7.48211669921875, 3.6728515625, 12.99951171875, 4.934539794921875, 4.1327362060546875, -0.3137969970703125, 0.5847549438476562, 8.246734619140625, 1.639862060546875, 2.1641616821289062, 0.6855545043945312, 3.78302001953125, -1.1118621826171875, 1.763702392578125, 8.468475341796875, 6.183563232421875, -2.1782989501953125, -8.86920166015625, -0.401123046875, 1.1987495422363281, 2.4884796142578125, 3.668701171875, 0.11295318603515625, 0.9419479370117188, -3.81060791015625, 2.6882476806640625, 2.306640625, 13.98199462890625, 0.5806922912597656, -1.147796630859375, -0.397369384765625, 9.896682739257812, 1.5807571411132812, 6.0350494384765625, 4.09521484375, 5.7498779296875, 8.96502685546875, 3.3953933715820312, 2.7155303955078125, 3.468170166015625, 2.361083984375, -0.013647079467773438, 4.88702392578125, -0.8244342803955078, 2.584014892578125, 0.8952713012695312, 4.42431640625, 0.9586639404296875, 1.433929443359375, 7.314239501953125, -3.7861328125, 9.02197265625, 2.5335693359375, -3.01641845703125, -1.4258956909179688, 1.1052398681640625, -0.41412353515625, 2.75408935546875, 2.9077186584472656, 10.466400146484375, 5.1822967529296875, -5.9215087890625, -6.020294189453125, 1.5141754150390625, 6.51751708984375, 5.879058837890625, 13.038726806640625, 2.084442138671875, 11.112060546875, 1.0486297607421875, 2.5020370483398438, 10.094757080078125, -5.59698486328125, 1.3586387634277344, 6.26165771484375, 19.779022216796875, 3.98883056640625, 3.249237060546875, -0.005401611328125, 5.416290283203125, 14.15313720703125, 4.609405517578125, 0.8865203857421875, 0.02532196044921875, 12.6024169921875, 1.9369964599609375, 1.05645751953125, -1.48895263671875, 10.24298095703125, 3.18084716796875, 1.61083984375, 15.07818603515625, 7.275733947753906, -6.547210693359375, 3.3082427978515625, 14.8831787109375, 11.36187744140625, -0.8779296875, 11.386810302734375, -0.3157196044921875, -1.3848953247070312, 7.861114501953125, 0.05766105651855469, 1.0606689453125, 2.2619476318359375, 1.9839324951171875, 6.759796142578125, 5.02813720703125, 4.061431884765625, -5.964813232421875, -1.4383506774902344, 2.019073486328125, 0.11418914794921875, 9.83251953125, -0.215789794921875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000052.npy"}
{"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 2.2812438011169434, "std": 5.517298698425293, "min": -12.26483154296875, "p10": -3.808552551269531, "median": 1.5682868957519531, "p90": 10.357786560058592, "max": 15.48968505859375, "pos_frac": 0.65625, "sample": [6.735076904296875, -2.6312713623046875, -7.77984619140625, -4.711578369140625, 2.6993408203125, -1.46990966796875, 10.524017333984375, -7.6962738037109375, 0.327239990234375, -2.0407180786132812, -1.006134033203125, -2.6741943359375, -5.364776611328125, -9.30462646484375, 3.0826873779296875, 6.9078369140625, 5.3909912109375, -2.57403564453125, 4.476165771484375, -0.149444580078125, -1.63140869140625, 0.569122314453125, -2.349945068359375, 10.01776123046875, 9.97796630859375, 12.903533935546875, 5.580543518066406, 9.41656494140625, 6.8687744140625, 8.609283447265625, 0.7275238037109375, 11.42535400390625, 0.24188995361328125, 3.3330078125, 2.8128662109375, -1.01556396484375, 7.197021484375, -12.26483154296875, 0.4767646789550781, 3.7927017211914062, -0.218292236328125, 1.0800819396972656, -0.9779396057128906, -2.093109130859375, 0.7779312133789062, 2.345947265625, -3.631805419921875, -1.6628341674804688, 6.34039306640625, -4.6409912109375, 6.098175048828125, 11.57525634765625, 0.6590576171875, -9.51605224609375, 0.20776748657226562, -1.20928955078125, 7.575592041015625, 1.587564468383789, -0.4902496337890625, 9.787750244140625, 10.286544799804688, 2.9886436462402344, -2.8558502197265625, -0.3453369140625, 4.6765594482421875, -2.6756248474121094, 10.91015625, 0.6081085205078125, 4.236610412597656, -1.6010169982910156, 1.3888168334960938, 4.763031005859375, 15.412933349609375, 5.6829833984375, 11.9573974609375, 10.964874267578125, -3.9268035888671875, 2.616546630859375, 5.73724365234375, 0.63519287109375, 3.38885498046875, 1.9925537109375, 6.572357177734375, 1.2469482421875, -0.152923583984375, 10.7916259765625, -9.28057861328125, 10.572998046875, 1.588277816772461, -3.75787353515625, 2.9809722900390625, 1.9196929931640625, 11.33612060546875, 1.3548202514648438, 1.57373046875, 15.48968505859375, -1.0363960266113281, -6.68670654296875, 1.6318740844726562, 0.7327880859375, -0.6477527618408203, 7.472129821777344, 1.7717056274414062, -7.265869140625, 0.360015869140625, -1.6415481567382812, 0.66534423828125, 6.0599212646484375, -0.41553497314453125, -9.61663818359375, 0.03730010986328125, -0.20218658447265625, 6.00640869140625, 6.092376708984375, 2.2624359130859375, 1.301513671875, 10.70556640625, 0.0, 8.59600830078125, -0.55938720703125, 8.3919677734375, -2.452056884765625, 1.5628433227539062, 4.801055908203125, 1.8947601318359375, 2.4923782348632812, 7.669891357421875, 9.912277221679688], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000053.npy"}
{"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 1.9015836715698242, "std": 5.819508075714111, "min": -13.0858154296875, "p10": -5.057424926757812, "median": 1.6472930908203125, "p90": 9.10979309082031, "max": 19.09716796875, "pos_frac": 0.6015625, "sample": [8.694061279296875, 5.16094970703125, 8.188812255859375, 5.07745361328125, 3.068115234375, 1.9945602416992188, -2.397857666015625, 12.09649658203125, 0.1908416748046875, -4.365936279296875, 0.949798583984375, -6.87200927734375, 0.039859771728515625, -1.315399169921875, 10.079833984375, -0.2485485076904297, 13.47967529296875, 5.131614685058594, -7.12127685546875, -1.99237060546875, 2.834014892578125, 7.777313232421875, 5.888519287109375, 0.4031982421875, -4.3885650634765625, 3.3134841918945312, 6.69793701171875, 1.07855224609375, 2.69158935546875, -0.12543487548828125, 0.8669281005859375, 7.593719482421875, -3.0487213134765625, 1.763275146484375, -0.095458984375, 8.255859375, 15.08197021484375, -6.6849365234375, 7.4183349609375, 2.9422607421875, 4.6484832763671875, 2.4811134338378906, 17.56121826171875, -1.41302490234375, 13.430206298828125, 1.000152587890625, -5.699005126953125, -4.6932830810546875, -0.70892333984375, -6.88189697265625, 6.145599365234375, 13.0692138671875, 0.63934326171875, 1.617340087890625, -0.23722076416015625, 2.8541793823242188, -0.2330322265625, -0.6549072265625, 5.41668701171875, -1.2237625122070312, 0.5789604187011719, 3.629974365234375, -5.336181640625, -5.499542236328125, 3.6130218505859375, 3.2245407104492188, -3.2327880859375, 3.4652557373046875, 7.933197021484375, 2.03277587890625, -1.0506057739257812, 1.948577880859375, 5.180145263671875, 3.717376708984375, -2.19427490234375, 3.0454559326171875, 4.34521484375, -3.4954833984375, 3.6845016479492188, 0.1327800750732422, -4.642730712890625, -4.860382080078125, 1.67724609375, -2.3127365112304688, 6.9259033203125, 3.7666015625, -4.37603759765625, 3.3861541748046875, 3.423675537109375, 4.69464111328125, -1.15374755859375, 10.78326416015625, -6.848968505859375, 5.300567626953125, 4.353179931640625, -0.3628692626953125, -6.53289794921875, 19.09716796875, 18.15234375, -0.264007568359375, 8.334808349609375, 4.112762451171875, 7.67462158203125, 1.146209716796875, 10.832611083984375, -4.196941375732422, -8.698486328125, -1.7997589111328125, -13.0858154296875, 11.15789794921875, -0.710205078125, -5.4618988037109375, -1.0426025390625, -2.7781982421875, 12.1651611328125, -0.237701416015625, -2.6014938354492188, -0.824432373046875, 0.13623046875, 1.991058349609375, -2.832305908203125, -1.47393798828125, 3.215555191040039, -10.5382080078125, -4.937957763671875, 2.409149169921875, 3.9112472534179688, 4.38507080078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000054.npy"}
{"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 2.304323673248291, "std": 4.944877624511719, "min": -14.5762939453125, "p10": -2.6727951049804686, "median": 1.6263885498046875, "p90": 9.442633056640624, "max": 19.6861572265625, "pos_frac": 0.640625, "sample": [-4.24102783203125, -1.749755859375, -0.051971435546875, -1.9656982421875, 1.6199951171875, -2.7015380859375, 5.100170135498047, 0.76947021484375, 6.820770263671875, -0.5361480712890625, 4.4854583740234375, 3.7509765625, 5.50006103515625, 4.5430450439453125, 0.474273681640625, 1.0686092376708984, 2.209016799926758, -3.280975341796875, 2.1513633728027344, 3.2948646545410156, 4.6959991455078125, 0.5686187744140625, 3.57000732421875, 9.32275390625, 10.445465087890625, -0.17481040954589844, 0.8142318725585938, 10.568450927734375, 2.0922393798828125, -3.933837890625, -0.21795654296875, 0.15082550048828125, -1.99334716796875, 2.20306396484375, 2.426666259765625, 3.108417510986328, -0.251556396484375, 4.5137939453125, -0.2971038818359375, 12.644012451171875, -0.58795166015625, -1.5321426391601562, -0.42291259765625, 0.8031005859375, 1.892303466796875, 2.9779815673828125, 2.0193634033203125, 1.014678955078125, 3.684967041015625, 11.28289794921875, 2.0244522094726562, -3.2591705322265625, 2.26507568359375, 2.5089111328125, -2.535003662109375, 3.000244140625, -7.5834197998046875, 7.71075439453125, -0.71990966796875, 1.16961669921875, -1.2664947509765625, 1.632781982421875, 12.94195556640625, 0.9168701171875, 15.740203857421875, -0.106201171875, -2.6604766845703125, 1.762420654296875, 4.913566589355469, -1.6184139251708984, -2.6378021240234375, 2.525249481201172, -3.2384033203125, -0.54974365234375, 0.025218963623046875, -0.06654930114746094, -0.858917236328125, 3.77325439453125, 3.28076171875, 5.244453430175781, 1.3295364379882812, 13.8829345703125, 3.8155517578125, -0.0604095458984375, -0.836090087890625, -3.276416778564453, -1.7280864715576172, 10.468490600585938, 9.127227783203125, 0.4695281982421875, -0.0275115966796875, 6.077484130859375, 3.76116943359375, 3.702117919921875, 13.629241943359375, -1.579660415649414, 7.576934814453125, -1.8024406433105469, 1.6806182861328125, 5.9495391845703125, -4.267242431640625, 4.997802734375, 3.0240821838378906, 10.59625244140625, 0.7468338012695312, -5.0415191650390625, 19.6861572265625, 2.37957763671875, 5.0146636962890625, -1.6176910400390625, -14.5762939453125, 1.1513023376464844, 8.005584716796875, -0.7419967651367188, -0.849578857421875, 0.0445404052734375, 3.817718505859375, -3.451019287109375, 6.166721343994141, -0.806121826171875, 0.570098876953125, 3.2508392333984375, -4.421966552734375, 9.72235107421875, -1.682647705078125, 14.453155517578125, 8.283370971679688, 5.354236602783203], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000055.npy"}
{"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 3.135500907897949, "std": 6.564188003540039, "min": -14.288055419921875, "p10": -3.2719635009765615, "median": 2.0939483642578125, "p90": 11.519609069824218, "max": 26.132354736328125, "pos_frac": 0.6328125, "sample": [3.22198486328125, -11.259719848632812, 6.8441925048828125, 12.9232177734375, 3.1547698974609375, 0.7704544067382812, -1.782501220703125, 5.81585693359375, -1.1070632934570312, 9.729217529296875, 8.136322021484375, 7.130615234375, -0.75616455078125, -1.9861907958984375, -0.4741363525390625, 0.395477294921875, -2.0690078735351562, -2.6763153076171875, 11.8035888671875, -4.523406982421875, 2.272552490234375, 0.7356109619140625, 8.32073974609375, -6.16943359375, -0.1533660888671875, 26.132354736328125, 9.340179443359375, 9.86749267578125, 8.79998779296875, -1.6643829345703125, 11.97540283203125, 6.225250244140625, 2.7094039916992188, -2.3345947265625, 13.88702392578125, 4.2375640869140625, -6.26544189453125, 4.153800964355469, 5.668891906738281, 1.0431785583496094, 15.829635620117188, 0.37689208984375, -0.00135040283203125, -0.08897209167480469, 1.1304397583007812, 23.0557861328125, 4.931915283203125, -0.6732692718505859, 1.3863677978515625, 0.941375732421875, -0.24468994140625, -2.173736572265625, 8.178482055664062, -9.12322998046875, 6.9723052978515625, -4.66290283203125, 0.1866321563720703, -1.515035629272461, 4.444183349609375, 0.55364990234375, 10.584144592285156, 15.14764404296875, 9.548065185546875, 8.405349731445312, -4.93841552734375, -2.96435546875, 3.81683349609375, 3.326263427734375, 5.33648681640625, 2.5693016052246094, -0.7133712768554688, -1.0516281127929688, -11.55975341796875, 1.3326091766357422, -0.08182525634765625, 4.8434600830078125, -5.383392333984375, 4.76611328125, -0.1825103759765625, -0.490325927734375, -14.288055419921875, 3.478851318359375, 6.2215423583984375, -3.989715576171875, 4.773567199707031, 17.7615966796875, 0.86529541015625, 10.022857666015625, -4.106475830078125, 7.431938171386719, 0.404205322265625, 17.6668701171875, 3.67822265625, 8.200157165527344, 0.600799560546875, 18.8543701171875, 4.14910888671875, -1.1042404174804688, -1.7363624572753906, -0.5408782958984375, 5.216529846191406, -5.3942108154296875, 2.3863754272460938, -1.72540283203125, -2.81927490234375, 12.58343505859375, -0.7267074584960938, 8.9383544921875, -1.9153289794921875, -1.571319580078125, 8.878387451171875, 3.364959716796875, 5.015899658203125, -2.6817474365234375, 0.5874443054199219, 4.004974365234375, -0.6619186401367188, 1.567098617553711, 3.2744903564453125, 1.91534423828125, -2.842681884765625, -0.90570068359375, 6.1638641357421875, 8.86993408203125, 6.6540985107421875, 11.397903442382812, 14.446044921875, 5.091033935546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000056.npy"}
{"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 4.166539669036865, "std": 5.891612529754639, "min": -8.799026489257812, "p10": -2.571661376953125, "median": 3.1914596557617188, "p90": 12.354284667968749, "max": 20.1947021484375, "pos_frac": 0.765625, "sample": [-2.391143798828125, 7.223114013671875, -7.12310791015625, 1.853240966796875, -2.5480728149414062, -4.9229736328125, 14.2816162109375, 1.2161903381347656, 3.50396728515625, 2.1957244873046875, 0.324005126953125, -0.979522705078125, 10.928207397460938, 1.7963180541992188, 9.449073791503906, 2.3361129760742188, 4.96240234375, 8.193603515625, 4.38543701171875, 6.797210693359375, 5.63331413269043, 12.669403076171875, -0.48310279846191406, 1.8654632568359375, 1.20465087890625, 10.43133544921875, -8.78863525390625, 4.22576904296875, -0.6872978210449219, -2.6267013549804688, 4.536376953125, 6.3093109130859375, -1.3552703857421875, 1.5894012451171875, 6.37762451171875, 20.1947021484375, 6.305389404296875, 4.122528076171875, -7.497123718261719, 7.8671875, 10.190383911132812, 0.12701797485351562, 4.797332763671875, 17.928970336914062, -1.7426910400390625, 4.485252380371094, -0.6445999145507812, -0.3033447265625, 3.15325927734375, 8.914093017578125, -1.2680816650390625, 2.807332992553711, 5.40667724609375, 16.46722412109375, 4.38970947265625, 6.65203857421875, -0.9202079772949219, 1.743316650390625, 5.420074462890625, 12.2337646484375, 2.322052001953125, 1.114471435546875, 0.8955497741699219, 6.18841552734375, 3.9192657470703125, 11.22967529296875, 14.45953369140625, 12.0877685546875, -8.799026489257812, 4.9467926025390625, 15.982452392578125, 5.944793701171875, -2.5187835693359375, -1.758941650390625, 14.32305908203125, 2.04925537109375, 4.179168701171875, 7.43902587890625, 3.092437744140625, 2.87451171875, 1.654388427734375, -0.8290252685546875, 17.57958984375, 8.73614501953125, -7.349761962890625, -3.38385009765625, -0.7766494750976562, 12.635498046875, -3.091827392578125, -4.045654296875, 7.368804931640625, 9.27239990234375, 0.285308837890625, 2.9934959411621094, 2.60040283203125, -4.1434326171875, 9.306396484375, 1.3643951416015625, 5.3114013671875, -3.679718017578125, 2.743408203125, 0.35540771484375, 10.7144775390625, 3.2296600341796875, 1.4500350952148438, 15.676300048828125, -0.00360870361328125, 14.220733642578125, 7.7371368408203125, 2.855743408203125, 0.1640625, 1.0438232421875, 8.063491821289062, 1.8623123168945312, 7.704196929931641, 10.6807861328125, 8.647369384765625, -0.18769454956054688, 0.9269733428955078, 0.36617279052734375, 10.591705322265625, -3.266754150390625, 14.84942626953125, 3.6420745849609375, 6.8438568115234375, 0.41895294189453125, 9.788909912109375, 7.205108642578125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000057.npy"}
{"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 2.3927717208862305, "std": 6.001771926879883, "min": -20.890838623046875, "p10": -4.524801635742187, "median": 2.4249649047851562, "p90": 9.737667846679688, "max": 18.853668212890625, "pos_frac": 0.6953125, "sample": [6.1597900390625, 10.493377685546875, 5.0509796142578125, 4.49932861328125, 3.5159530639648438, -1.3066139221191406, 2.5824432373046875, -15.91571044921875, 3.68597412109375, 6.66015625, 3.519237518310547, 9.751617431640625, 3.8676986694335938, 6.94927978515625, 1.912384033203125, -4.32623291015625, 1.5109634399414062, 9.7293701171875, 9.731689453125, -1.6788330078125, 10.496978759765625, 10.981964111328125, 6.63262939453125, -7.8641204833984375, -7.619293212890625, 1.646209716796875, 5.6915283203125, 5.460845947265625, 12.158065795898438, -0.732177734375, 0.7699623107910156, 5.7923431396484375, -2.4832763671875, -0.721435546875, 7.21478271484375, -5.541412353515625, -3.68603515625, 7.4839935302734375, 0.08465957641601562, 3.17120361328125, -12.2620849609375, -0.473175048828125, 0.2828369140625, 0.8230438232421875, 4.89056396484375, -5.5559234619140625, 0.7988052368164062, 13.42401123046875, 0.106903076171875, -20.890838623046875, 0.22054290771484375, 15.598968505859375, -0.7257843017578125, -2.4402618408203125, -0.32611083984375, -0.341552734375, -2.7942352294921875, 5.344831466674805, 2.66485595703125, 7.563720703125, 5.765419006347656, -1.0311431884765625, 6.90753173828125, -3.1819305419921875, 3.301177978515625, 5.32586669921875, -4.988128662109375, 9.359695434570312, 1.589263916015625, 2.267486572265625, 2.584930419921875, -5.403232574462891, -1.7190093994140625, 2.9940185546875, 5.047451019287109, 5.749053955078125, 17.65423583984375, 2.8217620849609375, -2.44610595703125, 3.3974533081054688, 5.80145263671875, -0.9249267578125, 18.853668212890625, -7.710296630859375, 4.2933197021484375, 4.7751617431640625, -0.5610504150390625, 0.8991012573242188, 3.0394515991210938, 8.70428466796875, 0.897857666015625, -0.0843505859375, 1.78466796875, 2.7873382568359375, 9.8592529296875, 5.32806396484375, 1.179931640625, -8.844512939453125, 1.44976806640625, 4.019744873046875, 1.19512939453125, 0.8691387176513672, -2.5377044677734375, -2.243927001953125, 8.820892333984375, 2.9630775451660156, 0.5669517517089844, 8.368270874023438, 5.1493682861328125, 2.256500244140625, -9.508880615234375, 2.80926513671875, -2.1669158935546875, 10.0693359375, 1.976165771484375, 1.74462890625, -1.5478057861328125, 10.64031982421875, 8.448722839355469, 11.085311889648438, 6.6950225830078125, 1.9839401245117188, -5.78472900390625, 8.79248046875, -1.5617828369140625, -3.243408203125, 6.3310546875, 1.321258544921875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000058.npy"}
{"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 1.6042370796203613, "std": 7.531189918518066, "min": -23.535430908203125, "p10": -5.928500366210937, "median": 0.73529052734375, "p90": 10.893254089355468, "max": 22.7860107421875, "pos_frac": 0.609375, "sample": [0.673980712890625, 9.8692626953125, 5.4454345703125, -20.271209716796875, 1.2344284057617188, 6.8971405029296875, 1.143798828125, -3.530120849609375, 4.0225372314453125, 3.96697998046875, 1.2860107421875, 5.0447540283203125, -6.533416748046875, 8.962921142578125, -9.888092041015625, 0.43402862548828125, 2.5724639892578125, 0.3828544616699219, -6.806549072265625, 3.8915863037109375, 0.5322265625, 0.18642807006835938, 0.1614055633544922, 0.1698455810546875, -6.148681640625, 0.4243927001953125, 0.796600341796875, 11.786224365234375, -12.4200439453125, -2.12445068359375, -3.438079833984375, 3.3176727294921875, -5.0297393798828125, 12.035400390625, 0.0, 12.19415283203125, -0.2092742919921875, 4.9276580810546875, 0.10612106323242188, -7.7659912109375, 0.9937095642089844, 16.6358642578125, 7.705230712890625, 16.51513671875, -1.0374298095703125, -14.291839599609375, 3.8579788208007812, 22.7860107421875, 5.56903076171875, 9.2137451171875, 7.565040588378906, 0.37530517578125, 13.982177734375, 20.046844482421875, -2.235015869140625, 2.9940261840820312, -0.29694366455078125, -5.834136962890625, -4.264923095703125, 1.0985641479492188, -4.924072265625, 3.897613525390625, 6.058258056640625, -1.669708251953125, -5.4120025634765625, 16.198272705078125, 20.191375732421875, -1.330718994140625, 5.9831695556640625, 4.0350494384765625, -3.2474212646484375, 10.691055297851562, -4.13775634765625, 21.5419921875, -1.6471443176269531, -14.20989990234375, 1.5754508972167969, -23.535430908203125, -4.616302490234375, 0.12549400329589844, 5.445953369140625, -0.413360595703125, 2.8522491455078125, 1.797088623046875, 3.7193374633789062, 7.7538299560546875, 9.46099853515625, -1.107940673828125, -5.4713287353515625, -9.839157104492188, 6.677947998046875, 4.904541015625, 0.511627197265625, -0.6797466278076172, -4.193748474121094, 10.45855712890625, -1.5240478515625, 1.95806884765625, -4.835418701171875, 5.24517822265625, -0.610198974609375, 2.600128173828125, 2.7815628051757812, 4.1737060546875, 0.16564178466796875, -9.56793212890625, -4.45306396484375, -5.538055419921875, -0.769622802734375, 4.1187744140625, 0.8687744140625, 7.257232666015625, 17.4776611328125, -0.177337646484375, -0.4732246398925781, -3.8036575317382812, 11.36505126953125, 5.792427062988281, 4.103271484375, -2.15411376953125, -9.501708984375, 4.283042907714844, 3.218963623046875, -0.26007080078125, 0.9510002136230469, 0.4495086669921875, -3.138763427734375, -1.783599853515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000059.npy"}
{"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 3.361804246902466, "std": 7.240179538726807, "min": -20.16302490234375, "p10": -4.359419822692871, "median": 2.5728492736816406, "p90": 12.459918212890624, "max": 22.0103759765625, "pos_frac": 0.7265625, "sample": [-17.65692138671875, 10.93902587890625, 3.975027084350586, 0.91302490234375, 5.776275634765625, 8.117919921875, 0.6341018676757812, 5.8875579833984375, 4.798126220703125, -12.40093994140625, 1.0587158203125, 0.10322952270507812, 0.3333396911621094, 5.2371368408203125, 0.0797119140625, 5.37353515625, 1.95758056640625, -0.89312744140625, 6.10113525390625, 17.580078125, 6.212890625, -5.82049560546875, 16.362823486328125, -4.503387451171875, -3.512186050415039, 2.7361984252929688, 1.893463134765625, 0.0, -3.5796642303466797, 16.0994873046875, -3.751495361328125, 9.8936767578125, 2.03106689453125, 8.1431884765625, 2.84869384765625, -6.373565673828125, 8.422576904296875, -9.78173828125, 3.7151565551757812, -0.31482696533203125, 0.23113250732421875, 5.539802551269531, 3.68707275390625, 3.7930450439453125, -0.5837631225585938, -6.2214202880859375, -4.325433731079102, -1.81024169921875, 9.981887817382812, 1.4382514953613281, -2.4559288024902344, 1.48077392578125, 2.382110595703125, 12.42401123046875, 0.11742591857910156, -4.438720703125, 8.107933044433594, -7.5714111328125, 0.7461166381835938, 7.215576171875, 2.2532806396484375, 9.458953857421875, 13.307548522949219, 0.06304931640625, -20.16302490234375, 12.093048095703125, -0.877716064453125, -4.1304779052734375, 1.9782943725585938, -2.528656005859375, 2.4095001220703125, 10.33209228515625, 8.479110717773438, 12.543701171875, 10.912689208984375, 4.388427734375, 1.9215927124023438, 5.0648040771484375, 1.8059616088867188, 6.709197998046875, 15.921783447265625, 4.6421356201171875, -1.2485504150390625, 3.864471435546875, 21.928314208984375, 0.417083740234375, 9.546539306640625, -17.439178466796875, 4.62116813659668, 9.07843017578125, 3.436981201171875, 2.366046905517578, -3.57183837890625, -0.12280464172363281, 7.6842041015625, 12.600311279296875, 1.1956558227539062, 13.311279296875, -0.8030548095703125, 4.611083984375, 4.808868408203125, 15.919189453125, -1.9730224609375, -6.00445556640625, 17.838851928710938, 9.935211181640625, 1.5766525268554688, 7.5623626708984375, 10.634307861328125, 0.0, 22.0103759765625, -9.9443359375, 6.4834442138671875, -0.683837890625, 0.8521041870117188, 0.7216339111328125, 4.245086669921875, 1.9173774719238281, 7.9748382568359375, -3.71514892578125, 6.270027160644531, 12.007537841796875, -0.6394996643066406, 15.7088623046875, 5.450897216796875, 1.6688385009765625, 3.2875823974609375, 9.961135864257812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000060.npy"}
{"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 3.241795301437378, "std": 8.2119140625, "min": -21.18560791015625, "p10": -6.889498901367188, "median": 2.589496612548828, "p90": 14.551977539062499, "max": 22.756622314453125, "pos_frac": 0.65625, "sample": [-7.446380615234375, -11.222152709960938, 0.0307159423828125, -3.4595489501953125, 2.45343017578125, 5.865234375, -2.097412109375, 5.60186767578125, 1.8210468292236328, 5.881919860839844, 9.509292602539062, -16.3607177734375, -2.80438232421875, 0.2541961669921875, -2.2467498779296875, 3.0608177185058594, -21.18560791015625, -1.1053619384765625, -7.88134765625, 1.610107421875, 2.9620361328125, 2.7255630493164062, 14.422088623046875, 10.641845703125, 10.789306640625, -2.229095458984375, 9.180816650390625, -7.052001953125, 18.085113525390625, 15.929885864257812, 1.3423995971679688, 0.36707305908203125, -0.24167442321777344, 0.05826568603515625, 18.988983154296875, 2.3447418212890625, -0.9070510864257812, 14.4287109375, 6.4267578125, 5.800939559936523, -2.72540283203125, -0.03826904296875, 11.964370727539062, -4.084724426269531, -1.02166748046875, 0.8638458251953125, 11.927520751953125, 6.257476806640625, 1.338165283203125, 3.6019439697265625, -1.389312744140625, 14.932388305664062, 16.40399169921875, 20.67755126953125, 21.654953002929688, 12.610137939453125, 6.7376708984375, 5.85772705078125, 14.839599609375, 1.26318359375, 8.428085327148438, 4.851749420166016, 5.07415771484375, 0.75494384765625, 6.306098937988281, 14.27325439453125, 15.30963134765625, 22.253997802734375, 12.68182373046875, 7.43463134765625, 0.44219207763671875, 3.341480255126953, 7.686382293701172, -7.264984130859375, -4.794338226318359, 4.3309326171875, -0.06988716125488281, -4.947296142578125, 5.29913330078125, -6.819854736328125, 11.931610107421875, 0.4793701171875, 4.31890869140625, -0.591705322265625, 1.0185699462890625, 6.338584899902344, 0.2590179443359375, -2.5140609741210938, -2.88470458984375, 5.308135986328125, -1.7169647216796875, -7.6790008544921875, -9.43255615234375, 6.8659820556640625, 0.29885101318359375, 3.843505859375, 2.0207595825195312, 19.478607177734375, -2.9886322021484375, 14.04205322265625, 15.25811767578125, 22.756622314453125, 8.7154541015625, 0.0, -0.8297271728515625, 4.670783996582031, 4.214591979980469, 2.9467620849609375, 2.8134613037109375, -12.88177490234375, 9.031723022460938, -2.317626953125, -1.4342803955078125, 1.8144378662109375, -15.589813232421875, 6.7887420654296875, -15.80084228515625, -2.3387451171875, -2.211700439453125, 9.089385986328125, -2.90057373046875, -4.044807434082031, -1.6829071044921875, 5.141563415527344, 4.5541839599609375, 6.30438232421875, 12.300643920898438, -8.40155029296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000061.npy"}
{"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 3.471477746963501, "std": 7.654454708099365, "min": -15.12786865234375, "p10": -5.151655960083008, "median": 2.0964183807373047, "p90": 15.503648376464843, "max": 31.2110595703125, "pos_frac": 0.6796875, "sample": [17.01385498046875, 3.644866943359375, 17.50994873046875, 9.244163513183594, 6.997894287109375, 3.2690353393554688, 9.32696533203125, 5.28338623046875, -11.1309814453125, -3.0682220458984375, -1.23150634765625, -3.2094459533691406, -4.5874176025390625, 9.111907958984375, 0.62152099609375, 1.1197509765625, 1.2433929443359375, 7.0322113037109375, -0.6433563232421875, -3.56121826171875, 5.353080749511719, 3.8796119689941406, 7.040679931640625, 1.7361984252929688, -5.4823455810546875, 19.6630859375, -6.4764556884765625, 0.09283447265625, 3.9691505432128906, 1.939666748046875, 1.205841064453125, -0.84576416015625, 7.884040832519531, 5.8027801513671875, 1.7329235076904297, 0.6377773284912109, 2.8123779296875, -4.3048553466796875, 5.530864715576172, -2.7765045166015625, 3.7543792724609375, 11.006500244140625, 8.477096557617188, -2.01776123046875, -5.7119293212890625, -2.725250244140625, -0.48169708251953125, 0.04803466796875, 7.437957763671875, 19.564239501953125, -15.12786865234375, 15.760284423828125, -5.18975830078125, 15.0958251953125, 5.57012939453125, -7.1790771484375, 16.15802001953125, 0.0, 8.546859741210938, -3.7733078002929688, -13.709136962890625, -0.941802978515625, 8.5706787109375, -0.04154205322265625, 3.084136962890625, 0.6160812377929688, -1.3199539184570312, 0.7894287109375, 0.722991943359375, 7.51812744140625, 17.76776123046875, 3.843902587890625, 1.7627944946289062, -5.135326385498047, 8.27374267578125, 16.059783935546875, 5.919464111328125, -3.8671112060546875, 1.963287353515625, 4.560466766357422, 3.3174896240234375, -4.541481018066406, 5.038238525390625, 3.710296630859375, 1.2183837890625, 13.7803955078125, 16.938507080078125, 1.0400390625, -1.462066650390625, 8.8406982421875, 0.7183341979980469, -3.504119873046875, 27.591156005859375, 17.643356323242188, 9.50860595703125, 11.182281494140625, -8.197906494140625, 31.2110595703125, -6.812530517578125, -6.572418212890625, 9.824874877929688, 1.2479400634765625, -3.1180496215820312, 4.82080078125, 15.393661499023438, 8.054458618164062, 1.68212890625, 7.892448425292969, 2.2295494079589844, 16.72479248046875, 3.652587890625, -3.5839309692382812, 5.110809326171875, -0.842193603515625, 0.773590087890625, 0.5657234191894531, -6.233978271484375, -0.0575408935546875, 4.573249816894531, 0.9908599853515625, 3.7136077880859375, 4.083221435546875, 7.8900146484375, 5.51373291015625, 3.905731201171875, -0.891815185546875, -8.537017822265625, -0.7106208801269531], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000062.npy"}
{"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 3.5188941955566406, "std": 7.5347208976745605, "min": -16.351806640625, "p10": -4.201014709472656, "median": 2.4015750885009766, "p90": 13.303045654296874, "max": 22.58203125, "pos_frac": 0.6484375, "sample": [-0.5382232666015625, 2.93408203125, 0.3212127685546875, -13.44647216796875, -10.3984375, -0.7328872680664062, 5.023529052734375, 0.026458740234375, 0.7620468139648438, 8.80194091796875, -10.30322265625, -3.4344635009765625, -4.2972412109375, -0.8924179077148438, 8.264114379882812, 1.2156295776367188, 9.587921142578125, 12.15447998046875, 15.883209228515625, 16.06298828125, -3.055633544921875, 0.167327880859375, 16.31353759765625, -2.69671630859375, -1.9808349609375, 12.4464111328125, 9.875213623046875, -4.317352294921875, 2.3288497924804688, 6.7324981689453125, -0.1336345672607422, 4.99981689453125, -3.361236572265625, 1.187713623046875, 2.0590381622314453, 21.30609130859375, -3.288330078125, 6.79400634765625, -8.719314575195312, 3.59808349609375, -3.51904296875, 1.269491195678711, 5.21112060546875, -4.1597747802734375, 2.61041259765625, 9.254806518554688, 15.78948974609375, -15.392578125, -12.93865966796875, 7.52197265625, -0.6360416412353516, 13.24505615234375, 7.7768096923828125, 0.0, -1.3630847930908203, 11.823638916015625, 3.665679931640625, -2.1043243408203125, 3.5802993774414062, 3.9385604858398438, 2.4687767028808594, 1.634002685546875, 7.850860595703125, 22.58203125, -0.787628173828125, -0.7462692260742188, 5.144378662109375, 13.4383544921875, -0.9592132568359375, 8.576080322265625, 21.6356201171875, 2.30096435546875, 8.3515625, 5.53875732421875, 16.8553466796875, -3.089935302734375, 12.614768981933594, 4.607208251953125, 2.6057281494140625, -4.5872344970703125, -16.351806640625, 16.258087158203125, 6.0240631103515625, 1.6915740966796875, 8.551887512207031, 5.026023864746094, 7.990528106689453, 0.5809955596923828, 2.6721038818359375, 1.7518539428710938, 10.17236328125, -0.04229736328125, -10.815185546875, 10.45672607421875, 1.799001693725586, 11.895172119140625, -2.1200790405273438, 15.27789306640625, 10.165153503417969, 12.43145751953125, -1.12188720703125, -0.3532257080078125, -0.703460693359375, -0.83563232421875, 2.3343734741210938, -0.5138473510742188, 3.564422607421875, -0.6159439086914062, 15.168975830078125, 4.2650146484375, 5.63946533203125, -1.7374114990234375, -0.723052978515625, 4.5900421142578125, 1.813852310180664, 20.3658447265625, 1.2884674072265625, 4.033721923828125, -0.9161376953125, 2.7333450317382812, -6.9422149658203125, -0.35955810546875, 5.203765869140625, 10.982421875, -5.356201171875, 1.6166858673095703, 13.203216552734375, 11.5201416015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000063.npy"}
{"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 4.214529037475586, "std": 8.705911636352539, "min": -23.72509765625, "p10": -4.511560058593749, "median": 3.8832855224609375, "p90": 14.085281372070312, "max": 29.482421875, "pos_frac": 0.6953125, "sample": [8.614990234375, 1.13677978515625, -19.5242919921875, 19.383453369140625, -2.707855224609375, 11.992828369140625, 3.5978851318359375, 7.2314453125, 25.98297119140625, 7.74725341796875, 8.5760498046875, 6.416015625, 16.20159912109375, 7.3531494140625, 4.920219421386719, -10.048934936523438, 28.20166015625, 5.856536865234375, 0.753936767578125, -6.701629638671875, 8.567794799804688, 5.459358215332031, -0.68560791015625, -4.035308837890625, 2.6358261108398438, -14.482620239257812, 2.6233253479003906, 3.8627471923828125, 9.377685546875, 14.0986328125, 6.007568359375, -1.621246337890625, 10.12725830078125, 3.473114013671875, 8.69451904296875, 1.32574462890625, -2.34112548828125, 6.9521484375, 4.591743469238281, 0.1537322998046875, 9.911163330078125, -1.3756942749023438, -3.7127304077148438, 7.710578918457031, 7.282745361328125, 24.42767333984375, 17.455810546875, -2.9196395874023438, 5.84405517578125, 5.696380615234375, 5.93890380859375, 0.83935546875, -2.441183090209961, 3.63836669921875, 2.437032699584961, -2.6824569702148438, -8.740005493164062, 5.9244384765625, 11.439682006835938, 5.9359588623046875, 13.136474609375, 9.155776977539062, -7.0880126953125, 4.909027099609375, 0.093505859375, -2.70562744140625, 2.552093505859375, -6.052978515625, -5.389892578125, 9.663665771484375, -3.48358154296875, 6.5855255126953125, 15.1275634765625, 9.670578002929688, -0.29993438720703125, 13.30450439453125, 5.9961090087890625, 0.34539794921875, -1.5297012329101562, -2.0479888916015625, -3.3262710571289062, 13.805328369140625, 0.10312652587890625, -1.4120025634765625, 4.0694580078125, 16.833740234375, 3.3141937255859375, 1.4496536254882812, 3.9038238525390625, -2.4039306640625, 11.318252563476562, 29.482421875, 7.245697021484375, 19.98455810546875, 2.9425430297851562, -5.888824462890625, -18.857818603515625, 5.08953857421875, -3.5044097900390625, -7.6075439453125, 3.1370849609375, 14.079559326171875, 3.8379974365234375, 0.0, 0.321807861328125, 0.6938533782958984, 5.248016357421875, 11.225509643554688, 10.297897338867188, -6.192901611328125, 10.966827392578125, 0.47673797607421875, -0.37587738037109375, 16.287734985351562, 3.9298248291015625, -2.9678955078125, -2.4834671020507812, 22.97412109375, -4.1351318359375, 2.4513397216796875, 0.0, 12.881591796875, 13.95526123046875, 7.9732513427734375, 11.509681701660156, 8.071075439453125, -23.72509765625, -1.84295654296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000064.npy"}
{"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 3.3062353134155273, "std": 7.676756381988525, "min": -16.66522216796875, "p10": -5.1194824218749995, "median": 2.708292007446289, "p90": 14.03816680908203, "max": 21.189849853515625, "pos_frac": 0.625, "sample": [-1.8941650390625, -3.57342529296875, 5.351318359375, 1.2279624938964844, 13.860687255859375, 4.885858535766602, 1.7167892456054688, 11.0068359375, 11.1939697265625, -10.429534912109375, 0.7684326171875, -13.211944580078125, 5.1397705078125, 7.90130615234375, -4.315723419189453, 15.700439453125, 15.374908447265625, -2.2889404296875, 2.4051589965820312, -0.51385498046875, -2.463287353515625, -1.0071754455566406, 6.646820068359375, 14.452285766601562, 3.011425018310547, 2.3120651245117188, 8.48748779296875, 2.34442138671875, -4.8563232421875, -0.37995147705078125, 7.4727935791015625, -1.3938751220703125, -2.4033966064453125, 9.973114013671875, 19.18603515625, -2.162841796875, -0.017465591430664062, 5.684295654296875, -0.9853591918945312, 1.6544303894042969, 3.061065673828125, -4.6352386474609375, -5.7335205078125, 6.85906982421875, 19.7242431640625, 0.9269256591796875, 12.417327880859375, 4.892852783203125, 13.287841796875, 4.608634948730469, 3.3704910278320312, 12.387481689453125, -14.3292236328125, 6.233367919921875, 3.1645355224609375, -6.313232421875, 10.36090087890625, 3.526123046875, 0.9682064056396484, 6.692619323730469, -0.1956787109375, -1.8719482421875, 0.76922607421875, -1.4571952819824219, 5.2985076904296875, -0.77264404296875, 0.27370452880859375, -8.957260131835938, -6.8519287109375, -0.667633056640625, 8.639167785644531, 11.2425537109375, -4.2052001953125, 3.3037872314453125, 9.506591796875, 7.517536163330078, 0.3782501220703125, 0.8929901123046875, -0.1881103515625, -0.0199737548828125, -12.69085693359375, -1.9066162109375, 0.0, -4.236541748046875, 18.60418701171875, -0.19698333740234375, -6.475830078125, 3.8719711303710938, 15.247940063476562, 5.914794921875, 9.107147216796875, 21.189849853515625, 13.773681640625, 9.7840576171875, 20.009658813476562, -4.54962158203125, 4.8080596923828125, 7.418792724609375, -0.9679965972900391, 11.205322265625, 9.0018310546875, 9.748641967773438, -14.73651123046875, 4.19873046875, 3.0608367919921875, 7.8617401123046875, -1.0436553955078125, 18.862060546875, 16.14910888671875, 15.00732421875, 3.13348388671875, -0.49880218505859375, 7.1554412841796875, 2.345611572265625, -8.70709228515625, -3.434375762939453, 8.327560424804688, -16.66522216796875, 0.0687713623046875, 4.744316101074219, -6.08734130859375, -3.8508071899414062, 15.26568603515625, 6.50372314453125, 8.5595703125, -0.7911891937255859, 1.7654342651367188, -2.62432861328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000065.npy"}
{"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 4.20259952545166, "std": 10.360640525817871, "min": -18.72320556640625, "p10": -7.68332061767578, "median": 3.8300018310546875, "p90": 20.023614501953123, "max": 31.2042236328125, "pos_frac": 0.65625, "sample": [6.154319763183594, 1.36029052734375, 23.928192138671875, -6.5675048828125, -2.9122695922851562, 14.58489990234375, -5.1640472412109375, -0.15694046020507812, 7.518074035644531, 1.36212158203125, -1.0497703552246094, -2.6519393920898438, 4.228279113769531, 9.188507080078125, 4.915626525878906, -0.1603546142578125, -1.0914554595947266, 3.8904876708984375, 26.88409423828125, 7.373432159423828, -3.3297119140625, -15.017425537109375, 8.940292358398438, 9.766265869140625, -5.931549072265625, 8.31829833984375, 3.1309051513671875, 3.529022216796875, -2.3727378845214844, 8.948989868164062, 15.3721923828125, -1.5572662353515625, 4.3951263427734375, -14.34576416015625, 5.246253967285156, -0.5085716247558594, 6.8076171875, 27.002349853515625, 5.454742431640625, -2.9821014404296875, -12.789581298828125, 10.765993118286133, 21.631561279296875, 4.6016387939453125, 3.6022911071777344, 6.134101867675781, -14.655029296875, 28.7490234375, 26.8013916015625, 2.16754150390625, -13.32763671875, -0.039764404296875, 3.650146484375, 11.9599609375, 21.88055419921875, 2.203338623046875, 9.074539184570312, 10.332321166992188, -18.72320556640625, 3.130035400390625, 4.4118499755859375, 9.6588134765625, 20.226470947265625, 7.946868896484375, 18.270904541015625, -18.417526245117188, 6.3939208984375, -7.263916015625, 9.17816162109375, -1.7588653564453125, -5.373992919921875, -0.9969482421875, 31.2042236328125, -4.988922119140625, 12.603485107421875, 4.4006805419921875, 9.1219482421875, -2.4744529724121094, 19.936676025390625, -16.829208374023438, 5.748321533203125, 6.597465515136719, -2.7058868408203125, 13.081939697265625, -3.6821136474609375, 9.752769470214844, 4.9561309814453125, 3.5173873901367188, 10.218994140625, 12.37310791015625, 3.7695159912109375, 4.5269775390625, 5.3431396484375, -18.403076171875, 1.4149322509765625, -5.6028594970703125, -7.3597869873046875, -8.780471801757812, 7.2835540771484375, 3.656341552734375, 20.792999267578125, 1.9462432861328125, 0.3314208984375, 0.0, 6.16632080078125, -1.3414306640625, -1.0940093994140625, 20.92169189453125, 6.7606658935546875, -7.282470703125, 8.2513427734375, 1.6407470703125, 1.166748046875, -13.509490966796875, 0.34100341796875, 13.1966552734375, 1.6845550537109375, 23.41650390625, 8.63836669921875, 4.95135498046875, -8.438232421875, -0.19061279296875, 15.11859130859375, -8.842456817626953, 0.19805908203125, -1.2411670684814453, 28.732635498046875, 15.009918212890625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000066.npy"}
{"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 3.655371904373169, "std": 9.1157865524292, "min": -19.557891845703125, "p10": -7.166667175292967, "median": 3.2019786834716797, "p90": 15.0153076171875, "max": 28.21258544921875, "pos_frac": 0.671875, "sample": [2.034332275390625, 17.527862548828125, 5.1212158203125, 28.21258544921875, -0.9698162078857422, -11.20654296875, 2.326488494873047, 11.72613525390625, 9.66943359375, -6.6344451904296875, 3.9399948120117188, 23.60601806640625, 4.748451232910156, 11.612548828125, 17.78350830078125, 11.29913330078125, 5.1400604248046875, 17.9141845703125, 4.503173828125, -6.0478668212890625, 3.7809906005859375, -0.0319976806640625, 14.990798950195312, -1.9592437744140625, 4.460990905761719, 11.424484252929688, -13.0997314453125, 4.40911865234375, 6.881126403808594, -0.0615234375, -13.777862548828125, 1.452627182006836, -3.4425048828125, 6.819221496582031, 14.97119140625, -5.8403778076171875, 0.0, 1.9146957397460938, 10.962921142578125, 3.703397750854492, -3.2479705810546875, 8.966522216796875, 4.614875793457031, 13.67218017578125, 3.2316627502441406, -1.1192913055419922, 8.6295166015625, -6.59326171875, -8.10296630859375, 0.0051708221435546875, -14.6759033203125, 20.75421142578125, 1.8487625122070312, 10.93756103515625, 1.61175537109375, 3.5890884399414062, 20.37518310546875, -13.2135009765625, 0.1400909423828125, 5.6561126708984375, 6.995758056640625, 9.214473724365234, 14.9454345703125, 4.858795166015625, 9.375404357910156, -0.1035614013671875, 3.0479469299316406, -10.605224609375, 0.40106201171875, 2.933307647705078, 9.77313232421875, -5.30596923828125, -3.6585769653320312, 17.198516845703125, -19.557891845703125, 26.86956787109375, 16.39556884765625, 11.3336181640625, 1.9382553100585938, -1.6631698608398438, 11.188873291015625, -5.855731964111328, 13.52630615234375, -3.708343505859375, -10.890045166015625, -0.39461326599121094, 14.1414794921875, -6.6465606689453125, 1.7357406616210938, 15.072494506835938, 13.2135009765625, -15.78582763671875, 8.833251953125, 16.090896606445312, -11.39874267578125, 11.022125244140625, -1.632476806640625, 4.1994476318359375, 7.907135009765625, 10.733673095703125, 14.242706298828125, 0.09166717529296875, 0.3001861572265625, -1.4685897827148438, 2.1338958740234375, -0.9485931396484375, -3.928943634033203, -18.731689453125, -8.713821411132812, 3.165027618408203, 8.060897827148438, 8.5838623046875, 7.953239440917969, -2.670032501220703, -0.09099197387695312, 17.275634765625, 1.41009521484375, 3.1722946166992188, 3.448760986328125, -2.8812713623046875, 4.71014404296875, -0.3276863098144531, 1.32025146484375, 3.511871337890625, 2.5388031005859375, -6.7653961181640625, 7.16607666015625, 2.645580291748047], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000067.npy"}
{"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 4.6750264167785645, "std": 10.961543083190918, "min": -22.817962646484375, "p10": -8.098049926757811, "median": 3.922576904296875, "p90": 16.13986816406249, "max": 38.017333984375, "pos_frac": 0.6796875, "sample": [10.945388793945312, 1.433572769165039, 0.822967529296875, -5.05169677734375, -4.438945770263672, 0.2503814697265625, -0.59112548828125, 14.915054321289062, -4.473411560058594, 4.6174468994140625, 5.499092102050781, -2.2268829345703125, -2.517333984375, 17.882476806640625, 14.914962768554688, -8.584228515625, 0.4114837646484375, 3.8115615844726562, 3.931884765625, 9.434955596923828, 14.804183959960938, 2.9128646850585938, -6.490653991699219, -10.937164306640625, 37.9876708984375, 5.346477508544922, 4.592397689819336, 10.392822265625, -0.021741867065429688, -17.78533935546875, 7.2544708251953125, 20.28814697265625, 37.724639892578125, 14.67425537109375, 0.3245697021484375, 20.006927490234375, -2.122344970703125, -2.75604248046875, -9.421493530273438, -15.02197265625, 4.29803466796875, -1.4168777465820312, 10.068374633789062, 1.44195556640625, 3.91326904296875, -22.817962646484375, -0.12139892578125, 5.129730224609375, 14.86663818359375, 12.7568359375, 4.001953125, -6.380058288574219, -0.33172607421875, 11.098236083984375, 31.45086669921875, -6.51544189453125, 11.744873046875, 19.752471923828125, -1.4701156616210938, 5.870338439941406, 12.709224700927734, 6.28533935546875, 0.9719734191894531, 11.6346435546875, 5.047243118286133, -0.5842666625976562, -18.168548583984375, 18.323699951171875, -2.935302734375, 38.017333984375, 0.39821624755859375, -9.40771484375, 9.630767822265625, 2.3643951416015625, -10.28472900390625, 4.349342346191406, -4.2835693359375, 4.368873596191406, 36.95928955078125, -2.3396949768066406, -6.902740478515625, 7.275260925292969, 1.4147758483886719, 4.3852996826171875, 4.03363037109375, 4.09027099609375, -7.99932861328125, 15.393035888671875, -8.41448974609375, 4.17047119140625, 8.25689697265625, 2.7502059936523438, 5.679931640625, -0.54766845703125, 1.923583984375, 18.87786865234375, -6.2541961669921875, -8.328399658203125, 8.778076171875, 11.333984375, 13.875335693359375, -0.5881061553955078, 2.2848129272460938, 13.80047607421875, 15.312469482421875, 6.53204345703125, 14.23712158203125, 18.031646728515625, 2.765625, 8.930343627929688, 11.184814453125, -9.803489685058594, 12.42913818359375, 1.6301765441894531, -7.048454284667969, 4.32733154296875, 12.140869140625, 2.0219039916992188, 0.888275146484375, 7.782722473144531, 2.8820648193359375, -4.4906158447265625, 2.736236572265625, 36.19354248046875, 8.65618896484375, -9.438079833984375, -2.4229774475097656, 0.49869537353515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000068.npy"}
{"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 5.988138198852539, "std": 10.020265579223633, "min": -23.82940673828125, "p10": -5.112701416015625, "median": 4.36602783203125, "p90": 18.58336639404296, "max": 32.31878662109375, "pos_frac": 0.7421875, "sample": [10.639701843261719, 4.373016357421875, 7.496269226074219, 11.537246704101562, 4.359039306640625, -12.395050048828125, 1.5186080932617188, 7.447834014892578, 6.86297607421875, 2.2696380615234375, 12.101806640625, 3.815654754638672, 27.345550537109375, 1.6675796508789062, 20.298171997070312, -0.09136581420898438, 22.607177734375, 1.817495346069336, 12.1405029296875, -1.859375, -2.942138671875, 0.7007904052734375, -5.562889099121094, -4.901283264160156, 9.26019287109375, 27.878082275390625, 11.225616455078125, 15.9449462890625, -0.8407859802246094, 15.609710693359375, -23.82940673828125, 12.42596435546875, 3.0191574096679688, 1.4653778076171875, 4.6786041259765625, 0.4774932861328125, 1.337686538696289, -4.21807861328125, 21.027099609375, -5.977630615234375, 32.31878662109375, -14.797027587890625, 1.7650470733642578, 4.3221588134765625, 1.0986328125, 6.288015365600586, -5.87164306640625, 11.19354248046875, 7.471702575683594, -3.020965576171875, 17.84844970703125, -2.10919189453125, 11.721923828125, 21.824615478515625, 9.33642578125, -9.553985595703125, -14.604660034179688, 2.3260345458984375, 11.46575927734375, 1.6684036254882812, -4.79400634765625, 11.250579833984375, 3.0429954528808594, 14.735885620117188, -3.1878509521484375, 9.236312866210938, 4.890960693359375, 12.59942626953125, 3.424957275390625, 12.31884765625, 21.26519775390625, 9.709877014160156, 0.745269775390625, -0.63250732421875, -0.5169677734375, -4.33135986328125, 17.71331787109375, 3.483165740966797, 14.356185913085938, 9.485023498535156, -4.561323165893555, -10.043853759765625, 8.32811164855957, 2.082599639892578, -2.9499969482421875, 0.9789886474609375, 10.874755859375, 6.993682861328125, 13.521942138671875, 22.142578125, 15.37615966796875, 29.35809326171875, 7.628288269042969, 11.21722412109375, -4.178070068359375, -8.076416015625, -0.5236434936523438, 2.4142913818359375, 25.751861572265625, 0.9979400634765625, 11.742950439453125, 9.898956298828125, 31.305755615234375, 9.15301513671875, 10.245529174804688, 16.837509155273438, -8.3170166015625, 10.90325927734375, 1.4136962890625, 4.1273651123046875, -5.360931396484375, -5.006317138671875, 3.549732208251953, 28.43157958984375, 17.841094970703125, -4.0391845703125, -4.706611633300781, -7.989776611328125, 14.06130599975586, 7.79931640625, 12.47198486328125, 2.7611083984375, 2.2449188232421875, 0.5283355712890625, 2.9415130615234375, 9.81011962890625, 12.387847900390625, 7.893096923828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000069.npy"}
{"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 5.835497856140137, "std": 11.36013412475586, "min": -30.514892578125, "p10": -5.579417037963865, "median": 4.340179443359375, "p90": 19.69727935791015, "max": 38.42840576171875, "pos_frac": 0.7265625, "sample": [-17.98992919921875, 12.93951416015625, 4.972034454345703, -3.9795913696289062, 1.123565673828125, 0.9814453125, 6.739837646484375, 6.989143371582031, -14.91241455078125, 31.509735107421875, -6.596179962158203, 0.48851776123046875, -3.7962188720703125, 13.08056640625, 24.50732421875, 8.58746337890625, 12.678359985351562, 6.028961181640625, 0.22222900390625, 11.41876220703125, 1.84039306640625, 18.268951416015625, 2.3355712890625, 2.8200225830078125, -13.50811767578125, 15.522079467773438, 2.9906234741210938, 12.824066162109375, 6.2458343505859375, 0.4727935791015625, 23.18499755859375, 9.887809753417969, 3.8459548950195312, -10.432342529296875, -0.14703369140625, 4.290130615234375, 16.918426513671875, -8.798606872558594, -11.583740234375, -2.01416015625, 1.8261299133300781, -0.8823490142822266, 29.16314697265625, 5.324043273925781, 8.119342803955078, 0.177032470703125, 6.676979064941406, 14.16033935546875, -3.9919815063476562, -11.998138427734375, 0.861083984375, -3.554290771484375, -10.584381103515625, -0.32370758056640625, -3.2402000427246094, 22.85986328125, 11.798828125, 0.13653564453125, 10.175750732421875, 0.8871517181396484, 9.332321166992188, 15.1968994140625, -0.8104324340820312, 9.4188232421875, 6.29150390625, 12.21990966796875, 7.188087463378906, 5.4808349609375, 3.1454734802246094, -3.575469970703125, -8.9344482421875, 2.22308349609375, -1.3050460815429688, 8.558837890625, 4.044158935546875, 18.19720458984375, 7.8037109375, 12.263450622558594, 26.13446044921875, 15.967529296875, 28.26751708984375, 10.050262451171875, 4.6448211669921875, 14.1729736328125, 12.5693359375, -1.5430908203125, 0.9480743408203125, 1.881683349609375, -2.7648162841796875, 29.82745361328125, 21.200790405273438, 17.434417724609375, 0.0, 1.4834709167480469, 17.4921875, 19.05291748046875, 18.170883178710938, 14.66802978515625, 8.965179443359375, 33.8486328125, 5.472412109375, 29.85565185546875, 1.9700927734375, 1.4428787231445312, 8.194534301757812, -3.2595367431640625, 23.04052734375, 2.939605712890625, 9.81024169921875, -6.65106201171875, -3.0765228271484375, -3.605998992919922, 2.4156570434570312, 12.057846069335938, -26.379119873046875, -30.514892578125, -5.1436614990234375, 0.1491241455078125, 4.390228271484375, 12.632843017578125, 15.973724365234375, 10.277488708496094, 2.493011474609375, -2.877767562866211, 38.42840576171875, 0.0, -0.190948486328125, 0.339447021484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000070.npy"}
{"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 5.12759256362915, "std": 12.845118522644043, "min": -40.499603271484375, "p10": -9.830062866210938, "median": 5.701457977294922, "p90": 23.021888732910153, "max": 34.301055908203125, "pos_frac": 0.6953125, "sample": [-4.9290618896484375, 30.27874755859375, -0.070098876953125, -13.1083984375, 6.929107666015625, -23.906890869140625, -5.140007019042969, 24.69793701171875, -20.188629150390625, 10.5078125, -0.5158920288085938, 7.214469909667969, -8.532440185546875, 2.19805908203125, 4.991596221923828, 13.683212280273438, 2.5912322998046875, 6.80322265625, 11.311248779296875, 16.93341064453125, 6.340187072753906, -9.739288330078125, 7.017547607421875, 33.47393798828125, 6.5957489013671875, -2.1424942016601562, 5.147605895996094, 28.941558837890625, 18.14417266845703, 14.20721435546875, -11.708206176757812, 3.726531982421875, 6.0010223388671875, 30.38726806640625, -22.167449951171875, 9.37518310546875, 18.3397216796875, 0.4675922393798828, 24.9368896484375, -16.555206298828125, -0.73162841796875, -2.209259033203125, -8.9844970703125, 5.33650016784668, 1.9429855346679688, 0.41436767578125, 5.9130859375, -40.499603271484375, -8.978759765625, -1.851470947265625, 7.09454345703125, 22.3306884765625, -1.223358154296875, 7.635829925537109, 3.9210128784179688, 13.053253173828125, 25.6610107421875, 3.119293212890625, -2.037294387817383, -9.6507568359375, 9.807327270507812, -6.61383056640625, 23.52587890625, 34.301055908203125, 1.246194839477539, 6.13714599609375, 10.401498794555664, -13.563262939453125, 12.56500244140625, 1.8234977722167969, -0.15509414672851562, 26.15777587890625, 11.0096435546875, -10.0418701171875, -5.747444152832031, 16.65582275390625, 25.50885009765625, -7.97808837890625, -10.704833984375, 3.6017627716064453, -11.4542236328125, -5.76776123046875, 0.047637939453125, 8.386001586914062, 11.067535400390625, 22.217529296875, 0.07513427734375, 8.6575927734375, 13.21185302734375, 7.959197998046875, 6.133548736572266, 7.485198974609375, -0.783721923828125, -0.25191307067871094, 2.7287445068359375, 25.2872314453125, -0.22314453125, 18.0546875, 13.666542053222656, 11.37042236328125, 10.8768310546875, -15.331039428710938, 0.6201457977294922, 15.030242919921875, 16.029693603515625, -2.9106369018554688, 4.5887603759765625, 4.7667999267578125, -8.310333251953125, 25.941177368164062, 9.80255126953125, 4.283946990966797, 2.23321533203125, 12.298774719238281, 7.68719482421875, 7.0289306640625, 10.553199768066406, 22.805892944335938, 1.4958343505859375, 9.19488525390625, 5.23028564453125, 7.058891296386719, 8.076171875, 5.489830017089844, -0.6539402008056641, 21.747817993164062, -35.728759765625, 7.8262481689453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000071.npy"}
{"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 5.061861038208008, "std": 11.255148887634277, "min": -35.63427734375, "p10": -6.976168823242188, "median": 4.180385589599609, "p90": 20.834426879882812, "max": 30.66943359375, "pos_frac": 0.6953125, "sample": [13.02264404296875, 6.0868377685546875, -1.9188079833984375, -7.574005126953125, 26.93798828125, -2.526153564453125, 7.2896881103515625, 17.4951171875, 3.3177490234375, 1.42938232421875, 3.608795166015625, -2.12847900390625, 1.74853515625, 5.3325347900390625, 15.400390625, 18.19970703125, -8.609081268310547, 3.54376220703125, -15.513679504394531, 3.524383544921875, 12.122848510742188, 20.98876953125, 3.456207275390625, 20.768280029296875, 8.143936157226562, 3.3135910034179688, -5.6536407470703125, 10.240219116210938, -7.2010345458984375, -1.1116943359375, 7.38702392578125, -30.364990234375, -6.311126708984375, 0.96502685546875, 7.1343231201171875, 5.17041015625, 1.7796478271484375, -3.984468460083008, 15.396255493164062, 5.793670654296875, 4.088623046875, -7.119232177734375, 21.055389404296875, 2.34326171875, 12.62420654296875, 11.767833709716797, 5.31024169921875, -2.4852066040039062, -21.38525390625, -8.86712646484375, -1.9626617431640625, 9.55517578125, -17.0865478515625, 0.0, 22.2177734375, 16.141204833984375, 22.8572998046875, -1.9227104187011719, 26.877471923828125, 16.68731689453125, -2.5251617431640625, -6.211273193359375, 4.364513397216797, 16.79814910888672, 7.081718444824219, 12.711410522460938, 4.64556884765625, 3.5598373413085938, 0.7339954376220703, 3.95233154296875, -6.360565185546875, 5.97468376159668, 6.3359375, 10.657524108886719, 21.085357666015625, 4.272148132324219, 3.1202220916748047, 8.152557373046875, -8.276092529296875, 16.619617462158203, 4.850128173828125, 3.649972915649414, -6.325687408447266, -1.387542724609375, 2.368865966796875, -5.771148681640625, 10.268508911132812, 4.511444091796875, 0.6125087738037109, 13.277252197265625, -13.98297119140625, 6.15716552734375, -0.676910400390625, 7.70294189453125, 2.971771240234375, 14.70821762084961, 18.044204711914062, 1.745849609375, 9.335617065429688, 14.1494140625, 11.395843505859375, 26.449234008789062, 0.0, 19.625320434570312, -1.2047195434570312, -1.1212615966796875, 28.623809814453125, 30.66943359375, 12.351516723632812, 8.165290832519531, 2.0362625122070312, -1.8756866455078125, -3.0305633544921875, -35.63427734375, 25.44091796875, 1.220672607421875, -3.3436012268066406, 0.0, 3.9130401611328125, 14.200607299804688, 0.224273681640625, 9.383544921875, -20.246246337890625, 21.86468505859375, 8.525558471679688, -6.91485595703125, 15.61199951171875, 21.2877197265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000072.npy"}
{"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 4.714159965515137, "std": 12.896790504455566, "min": -54.279205322265625, "p10": -10.774922943115234, "median": 4.820804595947266, "p90": 19.97305450439453, "max": 37.0623779296875, "pos_frac": 0.6875, "sample": [6.539960861206055, 7.247711181640625, 15.353302001953125, 14.20587158203125, -8.810272216796875, 5.7565155029296875, -12.790435791015625, 0.620513916015625, 1.5680694580078125, 15.927017211914062, 6.49822998046875, 11.1119384765625, 2.190460205078125, 24.02825927734375, -0.9141635894775391, -1.2174072265625, -26.85028076171875, -16.070159912109375, -1.6442337036132812, 1.602325439453125, 5.404296875, 1.2204437255859375, 0.14179229736328125, 8.376434326171875, 9.415435791015625, 1.946533203125, 15.85723876953125, 3.1006126403808594, 8.440780639648438, 13.182540893554688, -17.51287841796875, -19.49340057373047, 15.740997314453125, 30.56695556640625, -0.8779296875, 5.59246826171875, -11.038154602050781, 34.57342529296875, 14.545677185058594, 6.99505615234375, 12.978668212890625, 4.6619720458984375, -1.2283477783203125, -0.06734466552734375, 30.69482421875, -0.5741119384765625, -6.7670440673828125, -6.54638671875, 7.828948974609375, 31.662078857421875, -10.662109375, -3.5513916015625, 21.874725341796875, -54.279205322265625, 23.5089111328125, 4.120426177978516, 4.6714019775390625, -17.971206665039062, 20.760543823242188, -11.924896240234375, 0.56536865234375, 7.91119384765625, 3.978973388671875, 5.71575927734375, -8.02197265625, 15.089111328125, 11.613948822021484, 0.5115966796875, 23.56365966796875, 0.3233814239501953, -2.1526412963867188, -14.546142578125, -1.677276611328125, 6.22698974609375, -3.040740966796875, 0.3004608154296875, -15.807907104492188, -15.286712646484375, -3.7246627807617188, 4.970207214355469, 9.2489013671875, 19.63555908203125, 12.342758178710938, 1.5965576171875, -5.550079345703125, 13.838409423828125, -2.4849319458007812, 4.174652099609375, 1.5516891479492188, 12.158905029296875, 5.073514938354492, 9.336494445800781, 21.235015869140625, 13.022064208984375, 13.753387451171875, 16.30322265625, 3.405740737915039, 6.154144287109375, -18.299713134765625, 16.223648071289062, -3.67828369140625, 16.7171630859375, 10.050895690917969, 14.33428955078125, 11.2476806640625, 12.926483154296875, -0.680084228515625, 4.6234130859375, 31.37005615234375, 16.25, 10.79620361328125, -7.14056396484375, 5.710731506347656, -2.063894271850586, 3.370941162109375, -3.1015625, 3.871959686279297, 37.0623779296875, -8.251556396484375, -3.0642547607421875, 26.422149658203125, 4.192802429199219, 10.81939697265625, 6.8248748779296875, 6.713245391845703, 8.015380859375, 11.68902587890625, -10.57098388671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000073.npy"}
{"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 4.489516258239746, "std": 12.214616775512695, "min": -25.994537353515625, "p10": -11.98047637939453, "median": 3.8205718994140625, "p90": 19.9022216796875, "max": 37.8553466796875, "pos_frac": 0.6640625, "sample": [2.51434326171875, 18.1690673828125, -5.084014892578125, 1.0267333984375, 18.7265625, 0.11907958984375, -14.678131103515625, 3.682373046875, 12.120384216308594, 12.6175537109375, -7.84503173828125, 22.076065063476562, 0.0, 10.9173583984375, -2.0827789306640625, 25.56683349609375, 18.6988525390625, 28.4117431640625, -14.1796875, -5.4213714599609375, 2.032135009765625, 9.667678833007812, 1.842041015625, 9.188751220703125, 22.62738037109375, -20.64471435546875, -14.9547119140625, -0.267974853515625, -20.702392578125, -11.512069702148438, -8.18560791015625, -14.8167724609375, -6.72882080078125, -4.11895751953125, 19.143157958984375, 19.084442138671875, -3.4764862060546875, 9.374015808105469, 0.8810081481933594, 13.149459838867188, -0.8493995666503906, -7.06182861328125, -10.4964599609375, -17.691986083984375, -0.1552581787109375, 19.8304443359375, 1.929656982421875, -9.422027587890625, -1.6414794921875, 4.2888031005859375, -13.07342529296875, -20.97540283203125, 8.055572509765625, 10.138214111328125, -1.412017822265625, -3.9973068237304688, 3.2091293334960938, 2.8216400146484375, -23.13079833984375, 20.0697021484375, 5.6099090576171875, 7.259033203125, -2.6865234375, 22.23583984375, 12.79510498046875, 0.720367431640625, 17.09716796875, 5.038848876953125, 4.6629486083984375, 37.8553466796875, 10.993499755859375, 1.5104522705078125, 0.07128143310546875, 1.954864501953125, 7.680023193359375, 3.9940948486328125, 7.560028076171875, 3.2716064453125, -7.4954376220703125, 21.01251220703125, 8.127059936523438, -0.099700927734375, 6.3471832275390625, -25.994537353515625, 15.97216796875, 2.77996826171875, 3.78424072265625, 5.9273681640625, 2.1756210327148438, -3.245504379272461, -0.18241119384765625, -4.6268157958984375, 2.5028076171875, 15.093673706054688, 12.584648132324219, 9.072265625, 11.573455810546875, -14.895751953125, -6.384613037109375, -3.348052978515625, 13.488189697265625, -4.3768463134765625, 7.7882232666015625, 22.129058837890625, 12.84326171875, 12.30877685546875, 14.051651000976562, -3.8453369140625, 11.92852783203125, -13.520599365234375, 2.8926620483398438, 4.0015716552734375, 9.27105712890625, 4.645576477050781, 33.3900146484375, 3.856903076171875, 5.134681701660156, 12.082290649414062, 11.962417602539062, -3.3934326171875, 14.755096435546875, 6.357818603515625, 6.6488494873046875, 22.832229614257812, 33.744293212890625, 16.652374267578125, 0.004241943359375, 31.34722900390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000074.npy"}
{"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 6.687081336975098, "std": 11.60981559753418, "min": -28.534889221191406, "p10": -7.205963325500488, "median": 6.0423431396484375, "p90": 21.2022705078125, "max": 37.31060791015625, "pos_frac": 0.71875, "sample": [-4.2769927978515625, 6.901424407958984, 3.0328750610351562, 3.880859375, -12.403900146484375, 12.123138427734375, 18.299179077148438, -7.6688232421875, -3.74652099609375, 4.768806457519531, 7.438580513000488, -16.540237426757812, 4.715847015380859, 0.988372802734375, 21.30694580078125, -3.3874340057373047, 30.6102294921875, -0.78387451171875, 22.5069580078125, 11.307182312011719, 15.728118896484375, 16.44230079650879, 18.850677490234375, -18.589813232421875, -3.8977890014648438, 17.8582763671875, -7.4468994140625, 15.001190185546875, 3.504669189453125, -0.790802001953125, 9.5899658203125, 24.963455200195312, 13.43804931640625, 6.9147491455078125, -1.2174148559570312, -7.102705001831055, 15.38165283203125, 7.7760009765625, 14.586418151855469, -1.480712890625, 19.707611083984375, 11.720039367675781, 3.6375732421875, -16.2662353515625, 10.214736938476562, 4.1373291015625, -0.6945590972900391, 2.4467926025390625, 10.105377197265625, -0.9375686645507812, 34.05316162109375, 15.450164794921875, 12.180656433105469, 11.15301513671875, 6.954345703125, 14.102195739746094, -13.25286865234375, 21.15740966796875, -4.471710205078125, 11.85870361328125, 16.631591796875, 4.983421325683594, 0.0447998046875, 10.704940795898438, 2.769054412841797, 6.9849395751953125, 21.8426513671875, -2.0145416259765625, 10.691001892089844, 16.857940673828125, 9.009857177734375, 5.23052978515625, -0.2855110168457031, 37.31060791015625, 26.60906982421875, -0.5063247680664062, -17.061981201171875, -28.534889221191406, 3.847991943359375, 10.61214828491211, 27.037689208984375, 19.629051208496094, 12.5721435546875, -13.70703125, -2.4923019409179688, 23.081924438476562, 0.9426002502441406, -10.980377197265625, 4.784492492675781, 8.504348754882812, 6.56646728515625, -3.2138824462890625, 12.05865478515625, 14.1072998046875, 18.38818359375, 22.7174072265625, 0.057403564453125, 15.4547119140625, 1.40313720703125, 11.904533386230469, -14.413742065429688, 3.5175323486328125, 27.750946044921875, 8.887451171875, 18.66253662109375, 23.76239013671875, 0.4691886901855469, 19.00396728515625, 1.0695571899414062, -6.601104736328125, 15.91802978515625, 0.46088218688964844, 5.518218994140625, 9.76639175415039, 4.784482955932617, 19.0535888671875, -2.393798828125, -0.00164794921875, 14.75982666015625, -0.424072265625, 3.9399871826171875, -3.726348876953125, 5.213108062744141, 17.52825927734375, -1.8055267333984375, 2.035858154296875, -16.418502807617188, 3.24700927734375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000075.npy"}
{"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 7.2250285148620605, "std": 13.353407859802246, "min": -29.2080078125, "p10": -5.658474731445312, "median": 5.075248718261719, "p90": 25.711285400390622, "max": 44.1328125, "pos_frac": 0.7109375, "sample": [17.66375732421875, -3.747711181640625, 9.00771713256836, 4.934600830078125, 1.9941787719726562, 11.526611328125, 9.1868896484375, 5.391117095947266, 6.612497329711914, 4.0717010498046875, 38.861083984375, -0.7228145599365234, 9.947311401367188, -1.802520751953125, 14.5469970703125, 15.698272705078125, -1.426605224609375, -10.646408081054688, 6.46722412109375, 2.255584716796875, 38.8902587890625, 1.5622711181640625, 1.114501953125, -1.6291046142578125, 3.847442626953125, 28.9842529296875, -2.9238739013671875, 10.478668212890625, 4.047515869140625, -1.91864013671875, 11.807891845703125, -2.1381149291992188, 8.909255981445312, 19.154571533203125, -27.5301513671875, 38.039794921875, 9.139480590820312, 15.756988525390625, 10.114990234375, 9.81903076171875, 24.384429931640625, 21.867950439453125, 15.946624755859375, 18.126312255859375, -4.55401611328125, 3.3981666564941406, 9.164764404296875, 5.8607025146484375, -5.67840576171875, 18.851165771484375, 0.0, 1.101318359375, 5.623199462890625, 1.1665573120117188, 28.105712890625, -13.57012939453125, 44.1328125, -29.2080078125, -5.649932861328125, -14.783966064453125, 5.2158966064453125, 43.01141357421875, -8.124786376953125, 25.5787353515625, 10.632404327392578, 1.0146026611328125, -3.33349609375, 1.845001220703125, 12.721687316894531, -7.628746032714844, 1.6486358642578125, 22.198394775390625, -4.17340087890625, -0.841888427734375, -12.01812744140625, 0.0, 21.5325927734375, 10.83428955078125, 14.371078491210938, 32.806121826171875, 13.850250244140625, -1.0540809631347656, 3.845245361328125, 4.679573059082031, 6.3971405029296875, 8.009246826171875, 13.300834655761719, 24.3914794921875, 7.0395660400390625, 8.560256958007812, 10.391998291015625, 1.2830963134765625, 2.2275848388671875, 11.234466552734375, -2.033487319946289, 2.3235244750976562, -10.781135559082031, -4.9586181640625, 5.246337890625, -0.36234283447265625, -7.7244873046875, 2.8336334228515625, 23.535003662109375, 31.41693115234375, 22.4844970703125, 26.149139404296875, 38.1077880859375, 3.61224365234375, 1.00531005859375, -11.653854370117188, -0.5404052734375, -4.837383270263672, 31.953155517578125, -4.1712799072265625, -21.84552001953125, 1.7214202880859375, 18.977294921875, 8.6942138671875, 26.02056884765625, 1.834228515625, 2.183643341064453, 9.87396240234375, 5.530181884765625, 9.722694396972656, -4.540184020996094, 2.2209014892578125, -1.38330078125, 3.110137939453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000076.npy"}
{"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 5.858853816986084, "std": 14.533392906188965, "min": -36.17724609375, "p10": -9.900816345214844, "median": 4.68145751953125, "p90": 26.120245361328124, "max": 51.076568603515625, "pos_frac": 0.65625, "sample": [9.455242156982422, 3.3243255615234375, 6.079612731933594, 12.756011962890625, 3.645263671875, -0.5899505615234375, 18.623779296875, 16.462188720703125, -3.5971336364746094, 17.006607055664062, -4.6338653564453125, -0.7343215942382812, 13.712615966796875, 16.885162353515625, 26.33544921875, 9.92759895324707, 2.499664306640625, 14.466102600097656, -32.33831787109375, 15.259017944335938, -3.1199874877929688, -4.91546630859375, -9.172760009765625, -11.888916015625, 32.901336669921875, 20.653579711914062, 15.836685180664062, -0.5900421142578125, 14.8436279296875, 14.854248046875, -5.421142578125, 5.561492919921875, 4.2156829833984375, 20.09124755859375, -2.5496826171875, -1.6109447479248047, 8.181922912597656, -2.08599853515625, 12.37158203125, 0.9673080444335938, 0.079833984375, -3.5028457641601562, 26.976165771484375, 11.45037841796875, 5.721565246582031, 11.293853759765625, 3.328216552734375, 37.34002685546875, -7.520111083984375, 7.2095947265625, 5.10968017578125, -0.54296875, 27.754730224609375, 51.076568603515625, 15.06005859375, 9.995491027832031, -21.21112060546875, 0.717132568359375, 5.9569244384765625, 0.7650165557861328, -6.2689971923828125, -9.837997436523438, 29.3232421875, 44.2100830078125, -7.3223876953125, 15.545074462890625, 6.050846099853516, 5.0465087890625, 24.688690185546875, 7.495250701904297, 13.642501831054688, 8.731529235839844, -36.17724609375, -16.552642822265625, -10.047393798828125, -2.36749267578125, 10.30572509765625, 3.637298583984375, -2.574798583984375, 2.3268890380859375, -7.193450927734375, 18.755706787109375, -10.922409057617188, -15.3994140625, 2.6836013793945312, 13.05938720703125, 31.044281005859375, -17.701629638671875, -13.051910400390625, 15.839851379394531, 10.934028625488281, 3.823822021484375, -5.475688934326172, 6.40423583984375, 14.755859375, -8.0255126953125, -9.054901123046875, -15.333587646484375, 17.411163330078125, 4.4786834716796875, -12.3372802734375, 4.8842315673828125, -0.728851318359375, 10.132568359375, 14.515701293945312, 3.306976318359375, -1.5341796875, 26.36163330078125, 34.1629638671875, -1.6370086669921875, 6.237457275390625, -21.174102783203125, -9.49066162109375, 45.826507568359375, 0.68450927734375, 0.91357421875, -9.799896240234375, 12.157318115234375, -4.714256286621094, 2.72601318359375, 6.65899658203125, 0.1170654296875, 25.03265380859375, 26.02801513671875, 14.355438232421875, 26.65594482421875, -8.182395935058594, 1.1605224609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000077.npy"}
{"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 8.714691162109375, "std": 14.747476577758789, "min": -23.720504760742188, "p10": -8.744131469726561, "median": 7.090230941772461, "p90": 29.69997100830078, "max": 62.1763916015625, "pos_frac": 0.7265625, "sample": [-3.3363494873046875, 11.32708740234375, -14.239501953125, 17.484596252441406, 32.91473388671875, 1.3676528930664062, 10.936988830566406, 22.591796875, -11.42352294921875, 14.762374877929688, -0.22469329833984375, 2.760906219482422, 13.118305206298828, -8.154022216796875, -2.1667709350585938, 11.26544189453125, 5.393959045410156, 15.407913208007812, 29.5364990234375, 0.4390716552734375, 0.0, 25.45623779296875, -17.616973876953125, 10.517189025878906, 5.726631164550781, 5.0405120849609375, 22.2174072265625, 1.6527042388916016, -10.398162841796875, 11.452919006347656, -6.8855133056640625, 20.6785888671875, 24.7462158203125, 0.7288970947265625, 21.363494873046875, -13.366424560546875, 16.4681396484375, 34.7845458984375, -11.514862060546875, -23.720504760742188, 9.344749450683594, 18.0782470703125, 5.395240783691406, 62.1763916015625, 36.048583984375, -6.416259765625, 10.761627197265625, 8.527999877929688, 20.164886474609375, -0.4034576416015625, 17.55853271484375, -14.16461181640625, 4.822227478027344, 7.198810577392578, 17.929779052734375, -2.017824172973633, 4.744659423828125, 10.387359619140625, 11.23382568359375, -3.0896987915039062, 37.3759765625, -2.3873291015625, 8.83135986328125, 18.033218383789062, 7.720741271972656, -14.579429626464844, 14.443679809570312, 12.51202392578125, 11.8953857421875, -4.232696533203125, 38.593994140625, 14.51513671875, 5.574676513671875, 6.981651306152344, 37.932586669921875, 16.9320068359375, 5.398689270019531, 8.344932556152344, -1.183145523071289, 1.1682052612304688, -17.233352661132812, -6.112852096557617, 30.081405639648438, 31.110748291015625, -4.8291473388671875, 6.6072235107421875, 0.5740013122558594, 2.0218505859375, -9.990707397460938, 24.79974365234375, 3.1347179412841797, -8.209884643554688, 22.14453125, -5.9496002197265625, -15.62481689453125, 0.385284423828125, 4.756622314453125, -17.459869384765625, 18.835693359375, 13.651458740234375, -1.8777961730957031, 2.2876739501953125, -8.012741088867188, 38.639251708984375, 9.469512939453125, 10.699325561523438, 13.735198974609375, -5.29949951171875, 5.736427307128906, 1.102874755859375, 33.747589111328125, 9.30078125, -0.32275390625, 25.928848266601562, 19.37310791015625, 4.470806121826172, 25.2100830078125, 32.8253173828125, 23.770233154296875, -7.4210052490234375, 1.639495849609375, 1.1307449340820312, 39.88525390625, 2.193927764892578, 15.012588500976562, 8.757736206054688, 6.239349365234375, 25.31884765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000078.npy"}
{"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 8.329183578491211, "std": 15.607481956481934, "min": -48.3323974609375, "p10": -7.70934944152832, "median": 6.073974609375, "p90": 30.722682189941402, "max": 58.853302001953125, "pos_frac": 0.7109375, "sample": [-3.721038818359375, 6.455543518066406, 58.853302001953125, -10.836090087890625, 9.51641845703125, -3.551483154296875, 1.989898681640625, -0.19384002685546875, 2.5618896484375, 2.6023101806640625, 37.43841552734375, -1.9243316650390625, 0.0, 4.19671630859375, -8.089221954345703, -6.130578994750977, -3.1180419921875, -6.215507507324219, -17.734756469726562, 0.76629638671875, 13.963577270507812, -21.297576904296875, -1.4562225341796875, 0.7979011535644531, -9.74444580078125, 6.502899169921875, 8.70806884765625, 0.0, 25.463729858398438, 7.569601058959961, -22.5137939453125, 13.283294677734375, 18.022262573242188, 10.910430908203125, 12.826080322265625, 1.8265609741210938, 2.03497314453125, 10.763702392578125, 7.676918029785156, -0.39813232421875, 19.167434692382812, -19.70738983154297, 17.214019775390625, -1.9014854431152344, 1.10540771484375, 21.092330932617188, 6.34820556640625, 35.407073974609375, 0.32756805419921875, 4.9410858154296875, 13.407745361328125, -8.234375, -48.3323974609375, -15.060089111328125, 12.383712768554688, 23.652374267578125, 5.028900146484375, 24.19970703125, 20.135833740234375, 47.872222900390625, -3.5697059631347656, 4.045989990234375, 15.589920043945312, 27.188888549804688, 29.08990478515625, -6.9600830078125, 20.500534057617188, 5.972625732421875, 7.4546356201171875, 17.017181396484375, -10.47509765625, 13.347785949707031, 5.888034820556641, 4.5460205078125, -0.023681640625, -7.326904296875, 25.1463623046875, 6.079345703125, 28.9105224609375, 4.343711853027344, -4.165809631347656, 37.62469482421875, 31.5328369140625, 18.871124267578125, 7.1883544921875, 15.483154296875, 8.170753479003906, 1.8028488159179688, -0.7605133056640625, 3.194772720336914, 1.4555644989013672, 10.852577209472656, 6.068603515625, 7.646270751953125, 5.236078262329102, 37.370361328125, 2.392099380493164, 35.78071594238281, -3.38409423828125, 33.33856201171875, 27.67706298828125, 30.375473022460938, 9.6087646484375, 9.617034912109375, 0.70440673828125, 4.540489196777344, 15.635498046875, 7.667060852050781, 31.666656494140625, 0.7361068725585938, -3.82989501953125, 36.178314208984375, -2.888824462890625, 0.1383819580078125, 11.235250473022461, 16.589569091796875, 17.056747436523438, 26.90350341796875, -7.546546936035156, 13.974105834960938, -1.0967178344726562, 18.196929931640625, -14.0379638671875, 34.516845703125, 8.172164916992188, -9.597412109375, -7.237060546875, 41.860870361328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000079.npy"}
{"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 6.570462226867676, "std": 14.086621284484863, "min": -28.1173095703125, "p10": -8.21035385131836, "median": 5.399452209472656, "p90": 25.55225830078125, "max": 58.1041259765625, "pos_frac": 0.65625, "sample": [4.637432098388672, 13.917228698730469, -3.0066680908203125, 13.605907440185547, 6.5814208984375, -4.9608917236328125, -3.55120849609375, -26.57342529296875, 23.3050537109375, -3.6500701904296875, 1.6956405639648438, 0.01446533203125, 1.989105224609375, 9.35296630859375, 4.2278594970703125, -6.638954162597656, 0.9014739990234375, 23.333831787109375, 12.772979736328125, 8.43341064453125, 36.033111572265625, -4.436859130859375, 6.6156768798828125, 0.4984130859375, 8.43792724609375, 19.254638671875, 14.1015625, -7.438018798828125, 10.146604537963867, 26.59234619140625, 5.4327392578125, 2.353912353515625, 1.84088134765625, -3.1100711822509766, -2.053955078125, 3.1931304931640625, -28.1173095703125, 25.549072265625, 0.0, 28.75054931640625, -4.15692138671875, 20.570068359375, -1.2047882080078125, 3.55413818359375, 8.58294677734375, -18.1177978515625, -3.385040283203125, 12.142608642578125, 2.5903472900390625, -6.213085174560547, -3.9325294494628906, -2.58026123046875, 9.462608337402344, 20.33203125, 13.87042236328125, -1.1868877410888672, -2.8394412994384766, -0.064453125, -1.686737060546875, 30.463912963867188, 5.6783294677734375, 7.39764404296875, 6.052202224731445, -7.159088134765625, -4.35003662109375, 5.548675537109375, 18.096939086914062, 10.930274963378906, 3.7067337036132812, 8.518226623535156, -8.221275329589844, 7.573211669921875, 9.867076873779297, -10.104217529296875, 58.1041259765625, 2.60986328125, -2.86822509765625, 2.12109375, 14.40234375, -17.844696044921875, 16.578125, 7.783241271972656, -10.675323486328125, 31.45318603515625, 17.62530517578125, 23.101173400878906, -13.890350341796875, 15.5238037109375, -1.8813247680664062, 26.667083740234375, 24.560195922851562, 9.22216796875, 27.470626831054688, 6.79132080078125, 6.004343032836914, 25.5596923828125, -1.9289398193359375, 5.3661651611328125, -19.617691040039062, -1.2728424072265625, -16.414306640625, 35.778411865234375, -15.749931335449219, 24.5625, 23.57354736328125, 20.7967529296875, 2.6983489990234375, -6.9268798828125, -3.743804931640625, 27.889923095703125, -9.20611572265625, 10.007125854492188, 0.35311126708984375, -7.482940673828125, 5.573883056640625, -8.3695068359375, 9.6455078125, -1.3825607299804688, 6.109577178955078, -8.205673217773438, 23.851593017578125, 38.016326904296875, 14.7059326171875, 12.548614501953125, 43.85394287109375, 1.2209186553955078, 2.9475631713867188, 13.6351318359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000080.npy"}
{"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 7.149898052215576, "std": 17.293418884277344, "min": -42.98345947265625, "p10": -12.377877807617187, "median": 5.895341873168945, "p90": 30.929101562499998, "max": 55.37542724609375, "pos_frac": 0.6796875, "sample": [0.6001777648925781, -1.3280353546142578, 20.3724365234375, -20.61505126953125, 3.2977142333984375, 0.5372695922851562, -16.41290283203125, 7.527046203613281, 45.19500732421875, 31.0234375, 19.296066284179688, 21.2054443359375, 14.103179931640625, -6.367156982421875, 17.98297119140625, 0.5520973205566406, 12.905410766601562, 28.038986206054688, 10.6448974609375, 11.486602783203125, 8.644271850585938, 39.54302978515625, 8.600021362304688, -6.027000427246094, 6.699272155761719, 1.578765869140625, -2.1967010498046875, 17.848220825195312, 24.0313720703125, 32.88140869140625, 29.54693603515625, -6.991119384765625, 20.03215789794922, 3.98907470703125, -12.534912109375, 5.8723297119140625, -16.424530029296875, 42.745147705078125, 29.3095703125, 9.881668090820312, -11.454925537109375, -11.973785400390625, 23.51593017578125, -2.5051193237304688, -27.274032592773438, -0.4978485107421875, -42.98345947265625, -11.0653076171875, 18.498992919921875, 55.37542724609375, 6.655487060546875, 8.442916870117188, -9.134414672851562, 23.626190185546875, -15.628536224365234, -4.7762603759765625, 8.5020751953125, 27.178192138671875, 4.6533355712890625, -36.36759948730469, -11.156982421875, 0.46270751953125, 2.936431884765625, 5.637115478515625, -3.1327133178710938, 10.877941131591797, -3.869873046875, -0.5460433959960938, 7.482818603515625, 5.8182373046875, -24.217483520507812, 9.412395477294922, -20.286468505859375, 30.888671875, 30.37225341796875, 23.80810546875, -2.27215576171875, 14.7489013671875, 4.116003036499023, 14.70098876953125, -23.619049072265625, 9.96380615234375, 35.1307373046875, -9.814346313476562, 9.3699951171875, 9.8807373046875, 4.790092468261719, -12.310577392578125, -0.04901123046875, 5.918354034423828, -18.1793212890625, -8.821182250976562, -24.0323486328125, -6.6307220458984375, 2.9804840087890625, 2.0327796936035156, 6.6365966796875, 1.5869903564453125, 34.395263671875, 36.54499816894531, 35.27825927734375, 20.21282958984375, 12.408111572265625, -4.791748046875, 25.776885986328125, 7.120676040649414, -1.6956787109375, 0.003692626953125, 7.310009002685547, 16.15020751953125, -11.57693862915039, 15.75836181640625, 2.306669235229492, -2.0750274658203125, 3.4902191162109375, 23.78668212890625, 27.07611083984375, -2.7454376220703125, 33.769287109375, 9.36505126953125, 7.47479248046875, 36.53326416015625, 38.888916015625, 20.319091796875, 5.01544189453125, 1.8189468383789062, 2.4156532287597656, -5.6243743896484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000081.npy"}
{"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 7.897035121917725, "std": 16.952478408813477, "min": -35.174163818359375, "p10": -11.544372558593748, "median": 7.133628845214844, "p90": 25.114340209960936, "max": 59.609710693359375, "pos_frac": 0.7109375, "sample": [5.96075439453125, 25.307647705078125, -0.20855712890625, -12.5274658203125, 4.7996978759765625, 9.97906494140625, 11.93536376953125, 58.410064697265625, 15.9737548828125, -6.211647033691406, -4.552764892578125, 8.3416748046875, -7.766319274902344, -21.888885498046875, -1.222412109375, 21.430374145507812, 26.538055419921875, 11.957839965820312, 17.095703125, 17.4859619140625, 3.6103286743164062, -14.4007568359375, 15.439643859863281, -4.558570861816406, 2.2975120544433594, 26.65472412109375, -28.289306640625, 10.977294921875, -3.238555908203125, 4.0285491943359375, 55.8388671875, -4.195465087890625, 21.302261352539062, 13.492385864257812, 28.459716796875, 15.42340087890625, 0.6930770874023438, 8.4158935546875, 14.271636962890625, -35.174163818359375, -15.962677001953125, -1.9218292236328125, 6.065155029296875, 10.710891723632812, -14.106986999511719, 51.574554443359375, -16.731475830078125, 46.45880126953125, 45.5850830078125, -8.510498046875, -3.81658935546875, 59.609710693359375, 15.391403198242188, 10.221565246582031, 14.72698974609375, 0.16397857666015625, 6.086669921875, -4.376708984375, 4.00006103515625, 2.41424560546875, 21.53826904296875, -18.672210693359375, -29.92938232421875, 4.126363754272461, -13.60162353515625, 4.587554931640625, -7.1392669677734375, -5.8209991455078125, 28.0401611328125, 13.277664184570312, -1.3745136260986328, 21.9744873046875, 7.527618408203125, 8.420875549316406, -9.599151611328125, 8.956924438476562, 5.4092559814453125, 4.901277542114258, 8.947746276855469, 15.237335205078125, 6.188236236572266, 53.67852783203125, -8.727706909179688, 22.318115234375, 7.5606231689453125, 12.630599975585938, 19.85174560546875, 14.196449279785156, 7.091285705566406, 9.74200439453125, 7.175971984863281, 2.44781494140625, 9.647876739501953, 23.77288818359375, 4.256259918212891, 7.96380615234375, -11.123046875, 19.82086181640625, 19.83709716796875, 22.62103271484375, 12.016571044921875, 1.2564315795898438, 5.8076019287109375, 12.491989135742188, -3.422088623046875, 6.948799133300781, 2.8731460571289062, 0.35400390625, -5.98724365234375, 5.744049072265625, -5.372947692871094, 14.993072509765625, 17.990894317626953, 18.159950256347656, -16.12841796875, 10.2730712890625, 42.236419677734375, -1.142303466796875, 18.347503662109375, 15.181900024414062, -8.58673095703125, 24.56524658203125, 25.031494140625, 7.24053955078125, 2.304197311401367, 3.0048274993896484, -9.46820068359375, -23.120803833007812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000082.npy"}
{"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 8.615978240966797, "std": 19.18703842163086, "min": -57.962432861328125, "p10": -10.779671478271483, "median": 7.119285583496094, "p90": 34.47160034179687, "max": 53.212890625, "pos_frac": 0.7265625, "sample": [11.62750244140625, 0.0, 33.299530029296875, -0.977447509765625, 14.685836791992188, 1.2048110961914062, -10.126091003417969, 28.739288330078125, 10.837554931640625, -6.003662109375, -8.2384033203125, -57.962432861328125, 5.987480163574219, -41.4906005859375, 0.56378173828125, 2.38812255859375, 18.495132446289062, 49.3717041015625, 10.156600952148438, 11.259626388549805, 16.870513916015625, 19.849830627441406, 2.1306915283203125, 19.947662353515625, 26.9239501953125, -12.41546630859375, 6.5429229736328125, -0.30210304260253906, -8.8758544921875, 41.563873291015625, 38.19439697265625, -14.318145751953125, 22.0947265625, 4.1723785400390625, -11.405410766601562, 31.91107177734375, 1.4220123291015625, 5.492645263671875, 43.01234436035156, 7.595489501953125, 9.975349426269531, -2.5255813598632812, 5.959529876708984, 2.1367950439453125, -4.4644622802734375, 34.16455078125, 4.554290771484375, 2.8082427978515625, 11.697509765625, -4.340118408203125, 13.364151000976562, 21.22113037109375, 46.45623779296875, 17.814697265625, -50.04119873046875, 0.55975341796875, 18.46258544921875, 9.527587890625, 35.18804931640625, -5.945098876953125, -3.5015716552734375, 20.774490356445312, 13.53369140625, 28.747543334960938, 4.255126953125, 4.5733642578125, -5.04425048828125, 5.162353515625, -23.306350708007812, 13.600372314453125, -20.697235107421875, 13.569183349609375, 9.235595703125, -0.587646484375, -13.8729248046875, 8.720176696777344, 53.212890625, -20.390167236328125, 12.89288330078125, 26.898696899414062, -10.511497497558594, 31.68731689453125, 14.705978393554688, 1.797027587890625, -45.5408935546875, -17.037139892578125, 16.757431030273438, 43.014556884765625, -27.867477416992188, 26.3712158203125, 0.04605674743652344, -6.509284973144531, 3.5283203125, 1.1962890625, 9.80706787109375, 44.1290283203125, 37.2548828125, 9.806640625, 12.187789916992188, 6.245212554931641, 30.72259521484375, -0.0098876953125, 17.73907470703125, 27.429656982421875, 2.968658447265625, -3.9041061401367188, 10.161285400390625, 16.713905334472656, 30.972503662109375, -6.08966064453125, 16.95965576171875, -5.120658874511719, 21.80523681640625, 37.40582275390625, 2.26611328125, 5.549163818359375, 10.217086791992188, 36.58160400390625, 43.149993896484375, -2.57196044921875, 2.629894256591797, 0.5270309448242188, 13.07330322265625, -1.3636627197265625, 6.6430816650390625, 13.169754028320312, 2.155914306640625, 13.419273376464844], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000083.npy"}
{"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 7.316062927246094, "std": 17.621902465820312, "min": -50.453155517578125, "p10": -14.06184387207031, "median": 7.760883331298828, "p90": 29.230175781249997, "max": 58.6611328125, "pos_frac": 0.6875, "sample": [1.4488258361816406, 15.16015625, 45.825225830078125, -3.5989303588867188, -4.1083984375, -19.55126953125, 12.50335693359375, 15.351303100585938, 11.016693115234375, -9.54962158203125, 17.546539306640625, 17.143798828125, 5.171913146972656, -0.6026153564453125, 8.57861328125, 10.999229431152344, 0.7476730346679688, -4.652099609375, 4.957004547119141, 23.814315795898438, -5.0329742431640625, 20.0760498046875, 9.56597900390625, 13.61181640625, 21.25714111328125, 12.24359130859375, -1.2193374633789062, 15.6199951171875, 18.52313232421875, 12.731201171875, -3.932952880859375, -23.74493408203125, -50.453155517578125, 31.00439453125, 11.510543823242188, 3.7657928466796875, 15.1058349609375, -1.9975662231445312, 29.52984619140625, 2.8854446411132812, -1.246551513671875, -0.46234130859375, 15.074951171875, -6.9931488037109375, 21.95062255859375, -24.71038818359375, 9.225425720214844, -0.6167373657226562, 19.004608154296875, 13.849029541015625, 58.6611328125, -27.10125732421875, 33.2861328125, 14.383148193359375, 11.325958251953125, -17.825668334960938, 14.97869873046875, -9.72613525390625, 13.620254516601562, 6.309333801269531, 11.201904296875, -8.429794311523438, 4.79595947265625, 8.810447692871094, 30.92095947265625, -1.802978515625, 8.403076171875, -2.9058990478515625, 2.461151123046875, 6.69287109375, -13.537063598632812, 1.35455322265625, 3.771453857421875, 1.973114013671875, 0.27719879150390625, 28.523468017578125, 13.871204376220703, 37.881500244140625, -30.195404052734375, 12.182281494140625, -0.617828369140625, 3.33935546875, -7.293968200683594, 4.358100891113281, -34.93475341796875, -1.1076431274414062, -7.790924072265625, 15.94989013671875, 45.79559326171875, 4.733245849609375, 27.84735107421875, 5.9295654296875, -20.015655517578125, 12.539337158203125, -7.486778259277344, 7.429054260253906, 16.377685546875, 3.7012500762939453, 14.745254516601562, 10.8623046875, 8.09271240234375, -3.1998672485351562, 5.784698486328125, -6.017219543457031, 17.699867248535156, 15.546546936035156, -11.50640869140625, 35.135162353515625, -30.54852294921875, 26.451602935791016, 33.29937744140625, 5.00897216796875, 3.9738731384277344, 41.67559814453125, 44.3306884765625, 29.10174560546875, -16.236328125, -15.286331176757812, 26.89251708984375, 13.63116455078125, 3.111156463623047, 16.487380981445312, 52.091400146484375, -13.045745849609375, 13.479293823242188, 15.106658935546875, 13.37158203125, -20.824554443359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000084.npy"}
{"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 5.754773139953613, "std": 16.470088958740234, "min": -35.02294921875, "p10": -15.050311279296874, "median": 5.782918930053711, "p90": 27.879425048828125, "max": 54.95799255371094, "pos_frac": 0.6484375, "sample": [5.07757568359375, -8.9215087890625, -0.30507659912109375, 34.39739990234375, 6.571445465087891, 21.99919891357422, -5.4130706787109375, -10.048370361328125, 34.313873291015625, 19.58709716796875, 4.229766845703125, -1.9323577880859375, 6.684539794921875, 21.377700805664062, 20.3739013671875, -4.528076171875, 25.052642822265625, 22.17275619506836, 27.743789672851562, 9.929481506347656, 5.581695556640625, 0.564666748046875, 38.56434631347656, 32.609283447265625, -2.3111190795898438, -12.33251953125, 15.145767211914062, -6.865043640136719, -11.82379150390625, 35.0853271484375, 13.4796142578125, 1.61993408203125, -8.463165283203125, 12.6724853515625, 20.074432373046875, 30.736068725585938, -12.187408447265625, 24.581390380859375, 0.29753875732421875, -3.752899169921875, 30.72686767578125, 17.813934326171875, -14.615325927734375, 17.70563507080078, 23.7718505859375, -35.02294921875, 6.419868469238281, 7.760490417480469, 4.324453353881836, 32.52935791015625, 1.629913330078125, -18.68096923828125, 7.744657516479492, 12.44952392578125, 5.761867523193359, -4.0161590576171875, 16.36737060546875, 4.017368316650391, 10.309158325195312, 11.028602600097656, -17.84649658203125, 11.921073913574219, 9.747848510742188, -4.1992950439453125, 24.49597930908203, 18.990386962890625, 36.374237060546875, 18.77593994140625, -9.607437133789062, -22.677474975585938, -7.5697021484375, 17.75259017944336, -27.456207275390625, -8.112701416015625, -11.399826049804688, -6.66265869140625, 24.10430908203125, 12.185272216796875, 2.1221351623535156, 15.620040893554688, 5.92523193359375, 10.3944091796875, 4.6529083251953125, 16.86260986328125, 35.9586181640625, -29.859130859375, -16.065277099609375, 21.7265625, 11.634952545166016, -3.7457733154296875, -18.124771118164062, 9.640071868896484, 2.208221435546875, 3.1938838958740234, 28.195907592773438, -2.6999855041503906, -11.527885437011719, 6.9686126708984375, 9.470901489257812, 5.8039703369140625, -9.303764343261719, 14.826751708984375, 1.8998966217041016, -3.31866455078125, -14.2974853515625, -19.45880126953125, -27.039901733398438, 3.180145263671875, 31.674407958984375, 4.688011169433594, -16.09075927734375, -27.3846435546875, -1.975341796875, 18.836746215820312, -18.185806274414062, 3.03741455078125, 17.737594604492188, 8.217079162597656, 54.95799255371094, -11.109649658203125, -4.844696044921875, 10.70587158203125, -7.65362548828125, 6.22027587890625, 5.6681060791015625, -10.920696258544922, 10.17584228515625, 9.53184700012207], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000085.npy"}
{"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 8.895225524902344, "std": 17.8807430267334, "min": -46.706573486328125, "p10": -9.748001098632812, "median": 9.983550071716309, "p90": 29.17225646972656, "max": 60.416839599609375, "pos_frac": 0.6640625, "sample": [16.924224853515625, -46.706573486328125, 18.410400390625, -18.14947509765625, 20.563316345214844, -4.4567718505859375, 22.21063232421875, 7.8670806884765625, 21.964645385742188, 9.3577880859375, 60.416839599609375, -0.792572021484375, 6.601320266723633, 11.500885009765625, 14.5474853515625, -5.940692901611328, -12.56439208984375, -5.2205352783203125, -1.64569091796875, -3.1273460388183594, 5.737052917480469, 10.609312057495117, -1.2755126953125, -1.6775360107421875, 38.99591064453125, -2.656646728515625, -29.650146484375, 8.9044189453125, 36.266845703125, 18.637969970703125, -9.806427001953125, 22.815185546875, -5.2143096923828125, 28.122314453125, 50.16607666015625, 24.50494384765625, 21.04833984375, -40.866851806640625, 19.7657470703125, 34.7545166015625, -0.7256698608398438, -13.08404541015625, 17.7044677734375, -7.669563293457031, 28.465896606445312, 0.14333343505859375, 24.248329162597656, 26.68951416015625, 28.84088134765625, 4.9805145263671875, -24.7581787109375, 8.91302490234375, 17.310302734375, 11.271041870117188, 19.356231689453125, -4.588470458984375, 31.110809326171875, -25.146728515625, 12.534351348876953, 16.032791137695312, 31.5087890625, 30.2266845703125, 5.1873779296875, 23.291641235351562, 19.35919189453125, 6.62548828125, 0.0, -2.417694091796875, -0.27857208251953125, 16.2933349609375, 32.88484191894531, 23.81488037109375, -27.436614990234375, 24.58502960205078, 2.301534652709961, -2.1183624267578125, -4.975090026855469, -9.72296142578125, 40.59393310546875, 28.19525146484375, 1.3351058959960938, 13.679733276367188, 21.31280517578125, -1.3478546142578125, 6.469963073730469, -18.043914794921875, -38.45513916015625, 13.378982543945312, -4.344696044921875, 17.152862548828125, 1.9011306762695312, 11.769378662109375, 4.0277557373046875, 1.9240798950195312, 14.68658447265625, -22.66748046875, 16.98699951171875, 17.03302001953125, 15.0557861328125, 11.95892333984375, -5.957305908203125, -7.71710205078125, 1.721832275390625, 28.294525146484375, 38.9212646484375, 15.97418212890625, -3.0869522094726562, 49.739013671875, -0.7253799438476562, 15.76995849609375, -0.7116241455078125, 12.019790649414062, 29.945465087890625, 2.574981689453125, 27.696319580078125, -6.97308349609375, -2.6142730712890625, 22.675338745117188, -0.4050788879394531, 8.05126953125, 2.535430908203125, 15.90811538696289, 12.21978759765625, 14.415313720703125, -0.554443359375, 7.34515380859375, 16.300933837890625, 10.9219970703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000086.npy"}
{"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 10.25738525390625, "std": 19.6875, "min": -50.831085205078125, "p10": -8.96411552429199, "median": 5.220245361328125, "p90": 36.45030517578124, "max": 68.51904296875, "pos_frac": 0.6875, "sample": [11.762924194335938, -1.1958999633789062, -5.336883544921875, 24.78564453125, 26.187881469726562, 0.0, 9.621990203857422, -7.752784729003906, 17.872451782226562, 11.2755126953125, 0.336334228515625, 23.95672607421875, -19.087005615234375, 19.276065826416016, 54.511749267578125, -2.356548309326172, 0.194976806640625, 8.539306640625, 42.632110595703125, 8.323959350585938, 1.7001953125, -19.894973754882812, -0.44293212890625, 22.971664428710938, 0.75616455078125, -9.954460144042969, -7.30609130859375, 15.96588134765625, 46.51861572265625, 6.896602630615234, 28.612701416015625, -25.9423828125, 18.33648681640625, 34.6490478515625, 23.49871826171875, 13.677627563476562, 19.49164581298828, 0.610595703125, 24.857666015625, 11.5025634765625, -2.3601608276367188, 3.230010986328125, 25.6495361328125, -2.228271484375, -0.7108230590820312, -3.019359588623047, 52.04933166503906, -16.3017578125, 49.231597900390625, -0.759979248046875, 12.665618896484375, 0.19977569580078125, 34.3857421875, 28.93450927734375, 59.791351318359375, 57.59564208984375, 16.218276977539062, 21.99005126953125, 68.51904296875, 38.7764892578125, -16.055908203125, -1.4673919677734375, 3.1025218963623047, 20.420974731445312, 15.689712524414062, 48.27459716796875, 13.40252685546875, 0.0, 2.971221923828125, 16.7088623046875, 3.6445693969726562, -0.931488037109375, 41.10797119140625, 24.171798706054688, 25.1951904296875, 4.86572265625, 9.4822998046875, 0.8136444091796875, 26.271209716796875, -1.5216903686523438, -3.872528076171875, -6.812591552734375, -8.175773620605469, -12.859405517578125, -8.756065368652344, 24.845703125, 2.080322265625, 14.831521987915039, 10.629058837890625, 26.71221923828125, 10.28070068359375, 3.1672439575195312, 9.710895538330078, -9.945037841796875, 4.49700927734375, 5.57476806640625, 19.374618530273438, 7.410554885864258, -20.4525146484375, 25.352630615234375, 2.432720184326172, 26.402145385742188, -9.449565887451172, -7.4703369140625, 25.860305786132812, -50.831085205078125, 0.23699188232421875, -45.4451904296875, 2.714139938354492, 1.2929534912109375, 33.164337158203125, 41.701568603515625, 1.9292449951171875, -6.7120513916015625, -7.015655517578125, -0.2489471435546875, 41.862152099609375, 4.6627655029296875, -1.414154052734375, 0.2300567626953125, 20.119842529296875, -1.943572998046875, -12.557037353515625, 2.4181785583496094, 12.159187316894531, 0.240386962890625, 35.453369140625, -0.49755859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000087.npy"}
{"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 8.898430824279785, "std": 19.64641761779785, "min": -67.06271362304688, "p10": -11.689435577392578, "median": 4.477697372436523, "p90": 38.146017456054686, "max": 64.60870361328125, "pos_frac": 0.6796875, "sample": [3.0273818969726562, 30.84710693359375, 33.99542236328125, -3.3064651489257812, 43.39422607421875, 3.0879592895507812, -25.192047119140625, -0.2362060546875, 27.782485961914062, 11.539199829101562, 9.548149108886719, -11.547332763671875, 3.582183837890625, 64.60870361328125, -9.56887435913086, 8.622711181640625, -1.2323780059814453, 8.888641357421875, 4.419822692871094, -14.60491943359375, 52.66163635253906, -11.979156494140625, 38.06646728515625, -8.401382446289062, -3.76898193359375, 18.220260620117188, -11.800514221191406, -18.712745666503906, 51.68585205078125, -19.744979858398438, -19.79248046875, 39.53736877441406, -6.7233734130859375, 21.65472412109375, -3.855438232421875, 4.479564666748047, 38.331634521484375, 0.0, 1.1128807067871094, 9.22320556640625, 11.72955322265625, 4.14544677734375, 43.69496154785156, -6.8193359375, 55.257354736328125, 41.026519775390625, 11.76129150390625, -14.475387573242188, 4.415973663330078, 6.701263427734375, 0.0, 13.341949462890625, 10.587005615234375, 22.1214599609375, 26.214569091796875, -10.825286865234375, 1.3553123474121094, -0.8666839599609375, 11.78692626953125, 19.547088623046875, 29.178298950195312, 1.7084426879882812, -3.4568405151367188, -23.5423583984375, 25.029415130615234, 0.35418701171875, 1.1691513061523438, 23.516571044921875, 2.9807701110839844, 25.96917724609375, 12.970348358154297, -12.105682373046875, 13.544677734375, 1.1854629516601562, 4.475830078125, -32.70831298828125, 17.325714111328125, 3.9440364837646484, -2.504058837890625, -11.641830444335938, 18.602737426757812, 27.832565307617188, 15.38763427734375, -2.022706985473633, 6.392364501953125, 1.0172119140625, -2.9956512451171875, 3.0280609130859375, 35.946624755859375, 20.868026733398438, -4.451671600341797, 10.425750732421875, 1.6176300048828125, 15.645477294921875, 20.6651611328125, 46.61552429199219, -0.5135498046875, -8.664764404296875, 5.807586669921875, 41.7137451171875, 5.1051025390625, 8.49774169921875, -1.2881336212158203, 22.857086181640625, -8.470443725585938, 41.287139892578125, 29.14300537109375, 2.5150146484375, 17.41705322265625, -13.378662109375, 27.667282104492188, 1.1823501586914062, 6.6790771484375, 6.30120849609375, 52.61712646484375, 8.291143417358398, -8.719594955444336, 4.385524749755859, 1.9831695556640625, -7.3642578125, -67.06271362304688, 1.30584716796875, -3.42462158203125, 20.1514892578125, 10.828048706054688, 13.16851806640625, 34.64317321777344, -6.179588317871094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000088.npy"}
{"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 9.274109840393066, "std": 19.400217056274414, "min": -49.5089111328125, "p10": -12.950053405761718, "median": 6.817835807800293, "p90": 35.72154235839842, "max": 61.18280029296875, "pos_frac": 0.75, "sample": [1.0766143798828125, -28.754104614257812, 27.537506103515625, 10.235092163085938, 6.23114013671875, 41.13397216796875, 0.6331939697265625, 14.818473815917969, 6.02081298828125, 20.546737670898438, 2.0159912109375, 9.989654541015625, 5.75555419921875, 25.84954833984375, 6.2794647216796875, 22.044479370117188, 4.485317230224609, -22.651992797851562, 12.867362976074219, 7.570648193359375, 15.564422607421875, 23.5977783203125, 16.7005615234375, 2.3614768981933594, 22.68683624267578, -1.5998077392578125, 13.92694091796875, 1.23126220703125, 4.8509368896484375, 7.7703857421875, 0.695037841796875, 3.6354904174804688, 1.516143798828125, 34.38482666015625, 34.17071533203125, 61.18280029296875, 52.8741455078125, -32.883331298828125, -2.4555816650390625, 11.334381103515625, -49.5089111328125, -14.918212890625, 1.7733154296875, -37.5433349609375, 6.446746826171875, 24.25970458984375, 41.35845947265625, -25.2860107421875, -1.3823089599609375, -4.4496307373046875, 0.2860260009765625, 18.89031982421875, 7.4384918212890625, -4.368316650390625, 21.568161010742188, 24.605926513671875, 31.240509033203125, -3.6969051361083984, 19.91814422607422, 30.53155517578125, 7.640632629394531, 25.172454833984375, 17.28240966796875, 15.082305908203125, 12.842041015625, 12.607040405273438, -12.697830200195312, -9.025390625, 0.27835845947265625, -4.6439208984375, 4.382316589355469, 4.989715576171875, 1.9742889404296875, 24.991226196289062, -33.90223693847656, -1.173828125, 10.016197204589844, 28.713104248046875, 46.837158203125, -22.28619384765625, 6.55340576171875, 14.74151611328125, 17.031005859375, 38.840545654296875, 12.334259033203125, 51.780853271484375, -1.1627388000488281, 44.592437744140625, -5.6518402099609375, 41.43389892578125, 18.36248016357422, 5.12451171875, 45.63063049316406, 1.5423660278320312, 3.9445343017578125, 18.47386932373047, 39.8223876953125, 3.0828628540039062, 6.964424133300781, 6.657684326171875, 25.924346923828125, 3.351877212524414, -2.8900299072265625, -8.192901611328125, 4.2927093505859375, 8.799934387207031, -16.226408004760742, -6.2890777587890625, 12.8834228515625, 9.668136596679688, 15.116287231445312, 6.671247482299805, 25.761215209960938, 41.322784423828125, -13.85247802734375, 13.942092895507812, 9.00341796875, -13.53857421875, 40.78486633300781, -43.892425537109375, -2.4705848693847656, 34.300750732421875, 6.2616119384765625, -0.22650146484375, 2.301380157470703, 12.089447021484375, -4.627960205078125, -4.756072998046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000089.npy"}
{"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 8.21851921081543, "std": 18.486713409423828, "min": -46.8265380859375, "p10": -9.82995262145996, "median": 7.291355133056641, "p90": 29.320762634277344, "max": 55.50390625, "pos_frac": 0.640625, "sample": [14.581893920898438, 2.501953125, -4.1887359619140625, 52.97906494140625, -9.624443054199219, -2.4022216796875, -8.646240234375, 6.536582946777344, -6.629547119140625, 23.86395263671875, 14.377265930175781, 14.145378112792969, -32.95646667480469, -3.7525177001953125, -8.640411376953125, 15.1729736328125, 36.149749755859375, 38.873138427734375, 17.62530517578125, 10.975700378417969, 17.481460571289062, -4.9561767578125, 9.183700561523438, -7.507774353027344, -38.26741027832031, -9.57452392578125, 16.83049774169922, 22.27593994140625, 17.9534912109375, 0.97979736328125, -10.888595581054688, 11.4544677734375, 40.78550720214844, 10.873779296875, 21.11041259765625, 26.924942016601562, -8.860443115234375, -1.07586669921875, -5.83209228515625, 12.224761962890625, -10.30947494506836, 6.474456787109375, -0.28887176513671875, -41.30206298828125, 18.7508544921875, 55.50390625, 24.798431396484375, 11.1240234375, 26.320648193359375, -1.07940673828125, 25.745651245117188, -7.1995849609375, -2.5407867431640625, -0.5673904418945312, 3.283782958984375, 14.085433959960938, -18.53021240234375, -3.9374237060546875, 6.443359375, 13.988578796386719, 0.752410888671875, 0.0, 25.202972412109375, 32.80134582519531, 23.637916564941406, -8.467849731445312, 16.913299560546875, 8.362205505371094, -3.94854736328125, 27.69622802734375, -2.8202056884765625, 14.88336181640625, 6.2637939453125, 42.9666748046875, 19.334197998046875, 12.481109619140625, -0.8000316619873047, 4.0866851806640625, 7.382911682128906, 26.28143310546875, 18.9779052734375, -46.8265380859375, 23.3551025390625, 0.34771728515625, -1.5141830444335938, 5.00042724609375, 27.032272338867188, 3.0912246704101562, 10.233552932739258, -6.01171875, 22.780807495117188, -14.080810546875, 24.464935302734375, 36.620361328125, -7.8232574462890625, 12.614349365234375, -9.416488647460938, 7.199798583984375, 15.853973388671875, 26.235977172851562, -4.1072998046875, 25.55959701538086, 21.93743896484375, 10.313705444335938, -14.219375610351562, 54.863189697265625, 4.7581024169921875, 22.573516845703125, -16.785247802734375, -35.632080078125, 13.636810302734375, -4.9806671142578125, -9.007476806640625, 31.48675537109375, -1.743499755859375, 2.2901268005371094, 29.477783203125, 4.42840576171875, 29.38629150390625, 1.08612060546875, -12.66339111328125, -9.07823371887207, 42.92913818359375, 22.659759521484375, 2.1877689361572266, 17.3931884765625, -20.040069580078125, 29.292678833007812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000090.npy"}
{"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 9.532127380371094, "std": 20.62017250061035, "min": -34.51472473144531, "p10": -13.396549987792966, "median": 6.9461669921875, "p90": 39.14651184082031, "max": 59.93092346191406, "pos_frac": 0.6171875, "sample": [6.114013671875, -7.50042724609375, -4.373485565185547, 50.815765380859375, 43.876739501953125, 53.795989990234375, 2.6222305297851562, 18.093597412109375, 14.708847045898438, 11.197357177734375, 15.279342651367188, 18.01678466796875, 28.511886596679688, 18.025978088378906, 14.033946990966797, 24.726913452148438, 15.89093017578125, -26.759307861328125, 6.238311767578125, -1.3838539123535156, 16.89691162109375, 22.214279174804688, 7.283882141113281, 12.509490966796875, -1.2927837371826172, 10.856201171875, 9.022733688354492, -4.7504730224609375, -1.814321517944336, -23.8680419921875, 24.4102783203125, -3.4591751098632812, -5.1721649169921875, -30.93450927734375, -34.51472473144531, 26.04913330078125, -9.442497253417969, 17.250030517578125, 27.919349670410156, 21.094085693359375, -20.2615966796875, 1.3047370910644531, -8.92852783203125, 7.1214599609375, 17.361419677734375, 55.392608642578125, -12.857254028320312, -14.6549072265625, 0.0, -27.27618408203125, -1.9993896484375, -10.700531005859375, -25.37481689453125, 6.246429443359375, 4.6907196044921875, 8.51556396484375, -5.1105194091796875, -6.3037567138671875, 16.832275390625, 34.21905517578125, 22.255035400390625, -30.807785034179688, 14.453811645507812, 6.574638366699219, -1.9095935821533203, -11.1041259765625, 59.93092346191406, -3.2366943359375, -1.7413177490234375, 31.672943115234375, 7.2406005859375, -10.5206298828125, 55.969879150390625, 18.20177459716797, 45.0963134765625, 6.7646942138671875, 30.75366973876953, -0.3879852294921875, 17.02655029296875, 47.341339111328125, -29.3306884765625, -6.731109619140625, 45.7572021484375, 30.789344787597656, 22.426544189453125, 0.0, 22.280532836914062, -3.97265625, 19.63031005859375, 15.99224853515625, 2.45703125, 34.719512939453125, -0.76788330078125, -22.2144775390625, -6.497222900390625, 23.08447265625, -24.04388427734375, -1.8501472473144531, 4.66632080078125, 37.79827880859375, 33.1666259765625, 8.192703247070312, 38.97601318359375, 3.67889404296875, 12.54901123046875, -8.444061279296875, -10.3250732421875, 23.620010375976562, -2.0830078125, 49.0367431640625, -5.245147705078125, 56.18023681640625, 40.7330322265625, -1.9937934875488281, 5.879119873046875, 14.835708618164062, 5.36749267578125, -15.03582763671875, 10.631591796875, -3.8685302734375, 34.86228942871094, -4.68255615234375, 6.7708740234375, 5.31562614440918, 9.07568359375, 7.68292236328125, 39.544342041015625, -2.4822998046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000091.npy"}
{"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 10.072809219360352, "std": 21.87788200378418, "min": -53.46868896484375, "p10": -17.577096557617185, "median": 8.026497840881348, "p90": 39.55935363769531, "max": 64.59494018554688, "pos_frac": 0.71875, "sample": [-22.386383056640625, 4.330146789550781, -28.31243896484375, 14.049148559570312, 15.615478515625, 17.776870727539062, 22.362075805664062, 9.90985107421875, 33.6268310546875, 3.004669189453125, 0.6661148071289062, 7.6467132568359375, 51.23870849609375, 0.213348388671875, 47.589569091796875, 32.5234375, 42.43377685546875, -9.12969970703125, 18.16864013671875, 0.9878807067871094, -23.11481475830078, 24.61260986328125, 37.04942321777344, 31.267059326171875, 4.75750732421875, 7.28363037109375, 13.944992065429688, 25.1007080078125, 11.779876708984375, -6.57489013671875, 2.44317626953125, 25.443984985351562, 31.781524658203125, 9.207000732421875, 3.320953369140625, -15.816238403320312, 3.82110595703125, 12.575241088867188, 28.152740478515625, 12.80303955078125, -2.0099334716796875, -9.898193359375, 0.37309837341308594, -53.46868896484375, 18.721267700195312, 7.9959716796875, 13.478179931640625, 1.234893798828125, 9.743865966796875, -7.034355163574219, -21.07147216796875, -17.0279541015625, -38.429412841796875, 6.065704345703125, 42.97686767578125, -9.667526245117188, -25.139297485351562, -9.48648452758789, -24.907440185546875, 18.3048095703125, 7.5415802001953125, 24.67577362060547, 9.773345947265625, 16.639602661132812, 33.12579345703125, 11.601295471191406, 28.678165435791016, 0.0, -18.858428955078125, 46.3228759765625, -21.92315673828125, 1.787078857421875, -0.711883544921875, -5.984102249145508, 9.944961547851562, 20.609725952148438, -2.8156471252441406, 20.005157470703125, 34.23492431640625, -1.8106250762939453, -16.420013427734375, 48.60137939453125, 29.091064453125, -52.31005859375, -9.39654541015625, 3.149139404296875, 31.6195068359375, 39.243133544921875, 48.030548095703125, 21.050827026367188, 16.186370849609375, 20.562362670898438, -7.166393280029297, -8.52862548828125, 2.4705810546875, 56.2960205078125, -7.546783447265625, 1.878631591796875, 40.2017822265625, 7.347747802734375, 11.44677734375, 31.909759521484375, -33.654052734375, 25.629684448242188, 22.009307861328125, -7.285369873046875, 64.59494018554688, 17.43890380859375, 11.415267944335938, 1.552978515625, 49.1883544921875, 6.540473937988281, 21.391036987304688, -24.532135009765625, 27.97576904296875, 30.4090576171875, 3.4309425354003906, 5.9312744140625, -4.334564208984375, -0.610992431640625, -5.4664306640625, 8.057024002075195, 6.343076705932617, 7.039196014404297, 50.77435302734375, 41.323577880859375, 39.284027099609375, 1.4129409790039062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000092.npy"}
{"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 9.660725593566895, "std": 20.289119720458984, "min": -40.81475830078125, "p10": -14.377615356445311, "median": 8.270828247070312, "p90": 34.87097473144531, "max": 74.43988037109375, "pos_frac": 0.6796875, "sample": [-1.2688255310058594, 26.6881103515625, 2.828338623046875, 59.52423095703125, -15.359298706054688, 20.469833374023438, -1.4523086547851562, 12.569534301757812, 10.04473876953125, 11.759862899780273, 10.7008056640625, -19.548355102539062, 16.115753173828125, 1.61676025390625, 29.92230224609375, 33.529541015625, 4.9854736328125, 35.28875732421875, -8.418861389160156, 12.227422714233398, 16.6832275390625, 22.006118774414062, 8.124748229980469, 14.527122497558594, 37.190704345703125, 13.579421997070312, -3.5741424560546875, 12.646774291992188, 25.6259765625, 19.896148681640625, 19.194046020507812, -29.422348022460938, 7.793132781982422, -6.998443603515625, 8.70806884765625, 26.582672119140625, 13.134857177734375, 45.78216552734375, 4.7627410888671875, 37.209228515625, -11.850250244140625, 26.624183654785156, 8.920158386230469, -5.8744964599609375, -8.049942016601562, 6.073516845703125, 3.572662353515625, 63.5574951171875, 3.9371337890625, 8.226150512695312, 15.437736511230469, 20.74652099609375, 28.810150146484375, -8.73663330078125, 19.3671875, 16.248779296875, -2.3222503662109375, -10.407119750976562, 40.82916259765625, 3.155569076538086, 20.5543212890625, -25.79791259765625, 35.7037353515625, -2.8214263916015625, 6.4050750732421875, -27.53533935546875, 34.691925048828125, 27.59283447265625, -20.312049865722656, 8.315505981445312, 13.42432975769043, 4.446413040161133, 1.20367431640625, 7.1134796142578125, 18.389862060546875, -20.78607177734375, 34.35150146484375, -7.875823974609375, 32.58538818359375, -40.81475830078125, 49.95751953125, 18.497314453125, 25.351654052734375, -22.4056396484375, 11.155609130859375, 6.826576232910156, 0.0, -10.266845703125, -0.4146385192871094, 20.387351989746094, 40.776397705078125, 31.55517578125, -12.903533935546875, -13.956893920898438, 22.952362060546875, -32.97625732421875, 12.261703491210938, -1.5834732055664062, -7.2421875, -6.276277542114258, 18.242935180664062, 19.5294189453125, -9.938720703125, 0.019378662109375, 18.417404174804688, 0.0, 3.2086257934570312, -6.631446838378906, -21.827499389648438, -8.811725616455078, -2.7517852783203125, 1.4329643249511719, 21.465103149414062, 5.914955139160156, -29.08929443359375, 74.43988037109375, 49.23541259765625, 4.761383056640625, -15.986541748046875, -9.497146606445312, 11.906036376953125, 20.783966064453125, 5.71484375, 55.9505615234375, -0.5281028747558594, 20.559066772460938, 5.243690490722656, 20.337188720703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000093.npy"}
{"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 9.39173698425293, "std": 22.732980728149414, "min": -73.84686279296875, "p10": -22.262741088867188, "median": 8.282943725585938, "p90": 39.493090820312496, "max": 61.081298828125, "pos_frac": 0.6953125, "sample": [-6.595306396484375, 33.391883850097656, -22.048095703125, 36.636474609375, 35.310272216796875, 45.15435791015625, 51.89231872558594, 4.829418182373047, 8.287445068359375, 20.759841918945312, -0.7578125, -17.856353759765625, -40.40132141113281, -2.7608184814453125, 34.10899353027344, -9.55712890625, 38.995391845703125, 7.49212646484375, 8.473068237304688, 16.489181518554688, 31.56817626953125, 16.309951782226562, 8.508758544921875, -6.2501983642578125, 53.78010559082031, 23.98223876953125, 2.5514678955078125, -0.968475341796875, -11.64874267578125, 3.7928466796875, 45.108551025390625, 40.87921142578125, -34.468505859375, 49.967529296875, 19.439865112304688, 12.643707275390625, 26.91302490234375, 7.55390739440918, -23.477195739746094, 57.987548828125, 13.029403686523438, 46.06072998046875, 18.365966796875, -10.49005126953125, 43.17901611328125, 14.518829345703125, 20.728538513183594, -30.550445556640625, 15.399368286132812, -6.027078628540039, -3.586334228515625, 10.452327728271484, 5.333835601806641, 19.26104736328125, 12.179878234863281, -6.2657470703125, 32.92799377441406, -25.615753173828125, 12.76812744140625, -9.61114501953125, -6.4295654296875, 3.602752685546875, 1.225555419921875, -8.534660339355469, 4.37054443359375, 14.717193603515625, 19.766510009765625, 22.241287231445312, -0.6899795532226562, 27.406448364257812, 2.43426513671875, 3.037677764892578, 3.3991241455078125, 20.663986206054688, 37.43536376953125, -30.37518310546875, 8.2784423828125, -4.337699890136719, 14.308074951171875, 0.940643310546875, 7.5718231201171875, 27.035614013671875, -11.031112670898438, -4.389495849609375, -27.42352294921875, 40.654388427734375, 41.02314758300781, -24.037109375, 25.837799072265625, -28.522216796875, -73.84686279296875, 32.61607360839844, 4.9714813232421875, 22.5495662689209, 15.719589233398438, 20.397415161132812, 27.00244140625, 9.8121337890625, 52.812835693359375, -39.181365966796875, -1.1013717651367188, -18.167877197265625, 19.931304931640625, 6.0049591064453125, 12.046432495117188, 23.296768188476562, 5.149833679199219, 15.529510498046875, 31.262664794921875, 7.440940856933594, 2.2959136962890625, 4.70416259765625, 11.280197143554688, -27.63397216796875, -1.355712890625, 2.62799072265625, 20.472900390625, 4.4373931884765625, -7.8573455810546875, -17.63140869140625, 34.71368408203125, 8.66851806640625, 61.081298828125, 31.373069763183594, -22.763580322265625, 3.98651123046875, -1.0367774963378906, 2.276641845703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000094.npy"}
{"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 10.054659843444824, "std": 24.095857620239258, "min": -71.3333740234375, "p10": -13.87144317626953, "median": 8.508098602294922, "p90": 41.58292541503906, "max": 75.1402587890625, "pos_frac": 0.6953125, "sample": [-7.231937408447266, -3.274261474609375, -2.0270767211914062, 68.92098999023438, 13.349029541015625, -1.462310791015625, 14.303924560546875, 8.5721435546875, 42.82025146484375, 18.283660888671875, 3.6131591796875, 1.857025146484375, 3.6597900390625, 6.532958984375, -31.043731689453125, 16.37286376953125, 9.476303100585938, 10.106941223144531, 19.53955078125, 8.336181640625, 27.02667236328125, -0.4227142333984375, -34.467041015625, 15.33447265625, 4.951995849609375, 9.074050903320312, 68.85467529296875, -15.578521728515625, 48.81732177734375, -57.57373046875, 52.10284423828125, -13.184112548828125, 21.174057006835938, -35.31434631347656, 21.70343017578125, 29.70574951171875, -2.888275146484375, 2.175048828125, 32.751922607421875, 14.895767211914062, 20.42127227783203, 19.0263671875, -2.108184814453125, 9.389413833618164, 3.39715576171875, 16.521392822265625, 13.654674530029297, 28.159332275390625, 43.9403076171875, 8.1953125, 8.718719482421875, 31.31938934326172, 23.252578735351562, 0.8950653076171875, 18.63189697265625, -11.046661376953125, 12.88623046875, 23.09424591064453, 30.9808349609375, 19.729095458984375, 23.02764892578125, -0.333587646484375, -2.37969970703125, -4.433502197265625, 17.288848876953125, 71.2308349609375, -7.9020538330078125, 24.56573486328125, 8.444053649902344, 7.439617156982422, -11.910621643066406, 27.40106201171875, 18.7158203125, 43.475250244140625, -10.58306884765625, -4.651519775390625, 2.6947193145751953, 41.052642822265625, 1.1248931884765625, 14.035736083984375, -5.43463134765625, 23.681880950927734, 15.671783447265625, 5.686672210693359, 22.169021606445312, 47.581298828125, 37.04137420654297, 7.6125335693359375, -7.6472625732421875, -28.101654052734375, 16.384994506835938, 22.510513305664062, -6.9638214111328125, 19.64942169189453, -39.5625, 5.81524658203125, 26.0753173828125, 18.31414794921875, -29.4437255859375, 16.91131591796875, -7.093780517578125, -3.042804718017578, -7.5795135498046875, 3.585247039794922, 3.9806976318359375, -24.918685913085938, 1.1630439758300781, -8.392631530761719, 71.024169921875, -25.39080810546875, -15.375167846679688, 2.389190673828125, -4.410614013671875, 24.18646240234375, 5.8909912109375, 30.546417236328125, -3.2827911376953125, 19.77092742919922, 13.101776123046875, -33.9439697265625, -13.22698974609375, 0.65753173828125, 6.8682861328125, 75.1402587890625, 5.84063720703125, 56.70928955078125, -71.3333740234375, 48.97479248046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000095.npy"}
{"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 10.272958755493164, "std": 24.28507423400879, "min": -71.7269287109375, "p10": -21.59243927001953, "median": 7.880657196044922, "p90": 39.93215026855468, "max": 81.6788330078125, "pos_frac": 0.671875, "sample": [-8.001220703125, -40.119873046875, -3.548736572265625, 19.747875213623047, 41.38604736328125, 26.84056854248047, -24.223527908325195, -3.63543701171875, 25.36590576171875, 6.070037841796875, 10.478567123413086, 0.18355560302734375, -13.31793212890625, 4.069732666015625, 18.542152404785156, 7.304264068603516, 23.552734375, -22.172607421875, 18.504318237304688, 8.61578369140625, 1.145172119140625, 6.3970947265625, 42.67242431640625, 28.588760375976562, -8.195709228515625, 13.268768310546875, 65.4061279296875, 24.03125, 3.0796585083007812, 16.3245849609375, 81.6788330078125, 20.802398681640625, 39.309051513671875, 29.679073333740234, 26.131591796875, 7.82879638671875, -1.3202171325683594, 1.7103500366210938, -13.65716552734375, -0.9938507080078125, 24.748550415039062, 22.014114379882812, -27.526123046875, 47.8670654296875, -6.388153076171875, 7.932518005371094, -9.298530578613281, 4.800136566162109, 22.60297393798828, -4.612213134765625, 2.998443603515625, -2.371978759765625, 26.17559814453125, -29.43310546875, 6.3671875, -36.91816711425781, 14.695938110351562, 21.40692138671875, 7.715484619140625, 28.22003173828125, -10.959190368652344, -13.49200439453125, 23.034515380859375, 20.1097412109375, 5.90692138671875, 54.77178955078125, 56.23968505859375, 15.3096923828125, 2.7164154052734375, 45.2030029296875, 35.36083984375, 17.25189208984375, 20.68891143798828, -25.81524658203125, 75.15963745117188, 9.16510009765625, 19.757568359375, -2.85015869140625, 65.549560546875, -9.768966674804688, 7.2785797119140625, -2.91082763671875, 3.6447525024414062, -12.687255859375, -16.073776245117188, -22.575103759765625, -2.2092742919921875, 27.173095703125, 36.10736083984375, -11.033248901367188, -1.9317970275878906, -3.10479736328125, 6.027904510498047, 31.501220703125, 50.83453369140625, -24.120506286621094, 15.6527099609375, 24.31468963623047, 10.498611450195312, 30.078338623046875, -43.398895263671875, 31.891754150390625, 10.403076171875, 1.5197162628173828, -6.6338043212890625, 17.27069091796875, -10.617340087890625, 58.3115234375, 10.860214233398438, 19.4814453125, 49.03717041015625, 0.0, -21.343795776367188, 6.259632110595703, 4.526945114135742, 23.086566925048828, -29.518035888671875, 19.322265625, 21.230133056640625, 33.8277587890625, -25.151161193847656, 28.6341552734375, -4.4369354248046875, 20.16278076171875, -1.47039794921875, 17.275421142578125, 5.80401611328125, -71.7269287109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000096.npy"}
{"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 11.97391128540039, "std": 23.063297271728516, "min": -51.013885498046875, "p10": -17.772270202636715, "median": 11.12359619140625, "p90": 43.11524810791015, "max": 74.03652954101562, "pos_frac": 0.7265625, "sample": [12.739408493041992, 23.39654541015625, 17.075164794921875, -13.3165283203125, 8.849834442138672, 9.127384185791016, -10.904273986816406, 9.563289642333984, 20.29876708984375, 34.50048828125, -0.8348121643066406, 46.4542236328125, 54.63722229003906, 49.3070068359375, 1.049112319946289, 3.5496826171875, 39.15972900390625, 3.2358856201171875, -3.01910400390625, 22.69354248046875, 12.821136474609375, 0.7688446044921875, -15.150238037109375, 2.5546875, 32.396453857421875, -30.29656982421875, -1.929351806640625, -10.102584838867188, 5.786823272705078, 22.842681884765625, -28.918067932128906, 19.740196228027344, -10.344070434570312, -34.35662841796875, 23.657730102539062, 58.83036804199219, 40.810577392578125, -3.35400390625, -24.007644653320312, 9.429779052734375, 1.700836181640625, 9.24725341796875, 9.08905029296875, 30.65057373046875, -26.037689208984375, 26.27239990234375, 15.917083740234375, 32.59754180908203, 12.60858154296875, 6.8006591796875, 26.795440673828125, 5.22393798828125, 18.17669677734375, 5.80572509765625, 22.4564208984375, 45.25663757324219, -12.292160034179688, 22.440109252929688, 10.734954833984375, -0.96807861328125, 29.084625244140625, 17.022918701171875, -16.335845947265625, 17.5869083404541, 24.2845458984375, 0.0, 51.658447265625, 25.39080810546875, 12.450347900390625, 40.75189208984375, -15.331375122070312, 54.101654052734375, 29.034652709960938, 7.4886474609375, 0.4098777770996094, 52.72314453125, 17.7108154296875, -42.467987060546875, 23.22396469116211, 34.6038818359375, 12.33941650390625, 6.3494873046875, 9.888275146484375, 11.512237548828125, 47.387298583984375, 11.724594116210938, 9.800802230834961, 74.03652954101562, 25.79651641845703, -9.289657592773438, 17.919513702392578, -30.575439453125, 18.4921875, 5.0037384033203125, 19.579742431640625, -20.606124877929688, 0.0, 11.88665771484375, 18.84552001953125, -1.0846099853515625, 8.849136352539062, 65.389892578125, 9.998703002929688, 4.788543701171875, 1.634002685546875, 51.359619140625, -20.217880249023438, -7.878631591796875, 63.2281494140625, 18.60382080078125, -6.00372314453125, 27.090728759765625, -16.724151611328125, -51.013885498046875, 14.650665283203125, 23.84033203125, 16.018600463867188, 42.197509765625, -6.9803314208984375, -24.909454345703125, 8.074516296386719, -23.668685913085938, 2.874725341796875, 25.32415771484375, 32.99493408203125, -22.58770751953125, -1.789581298828125, 33.899444580078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000097.npy"}
{"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 12.209274291992188, "std": 25.612689971923828, "min": -67.13580322265625, "p10": -13.787100219726561, "median": 9.163986206054688, "p90": 48.34057922363281, "max": 94.65606689453125, "pos_frac": 0.6640625, "sample": [31.082107543945312, 59.49481201171875, 49.162353515625, 13.094566345214844, 74.72921752929688, 33.399810791015625, 37.029388427734375, -35.47406005859375, 13.18017578125, -5.416938781738281, 58.53936767578125, -33.5740966796875, -7.491485595703125, -0.40533447265625, 28.436798095703125, 25.2730712890625, 9.919692993164062, -12.622314453125, -17.019012451171875, -3.6966018676757812, 4.800018310546875, -5.4527587890625, -11.17938232421875, 15.33941650390625, -1.0183181762695312, -7.932435989379883, 9.597137451171875, -9.24871826171875, -12.6939697265625, -42.139556884765625, 3.686220169067383, -67.13580322265625, -20.95294189453125, -8.25140380859375, 7.2662200927734375, -10.785964965820312, -11.634084701538086, 56.8001708984375, 31.684234619140625, 11.242195129394531, 6.4985198974609375, 10.096527099609375, -18.931716918945312, 16.391769409179688, -3.3584136962890625, 47.9957275390625, 10.1806640625, 28.197479248046875, 1.93157958984375, 7.780979156494141, 49.5767822265625, 94.65606689453125, -7.666023254394531, -0.82525634765625, 1.6449089050292969, -5.6898345947265625, 65.1483154296875, 23.209625244140625, -15.5699462890625, 49.145233154296875, -42.90245056152344, -5.9839019775390625, -13.53497314453125, -5.905670166015625, 58.48358154296875, 0.7047119140625, 37.77789306640625, 3.71112060546875, 21.853130340576172, -5.4884490966796875, 32.88165283203125, 34.741180419921875, -17.468292236328125, -5.3899688720703125, 45.73736572265625, 10.758102416992188, 22.153228759765625, 15.9012451171875, 11.032196044921875, 33.166473388671875, 8.574440002441406, 31.853271484375, -14.375396728515625, 7.678798675537109, 35.403656005859375, 13.7803955078125, -0.117767333984375, 3.2913436889648438, 29.668212890625, 21.1097412109375, 51.114227294921875, 34.21466064453125, 25.710037231445312, -7.205810546875, 27.738380432128906, 3.12078857421875, 6.59942626953125, 24.958541870117188, 1.3744888305664062, 38.237823486328125, 8.7308349609375, 30.449493408203125, 67.517822265625, 18.766815185546875, 9.627410888671875, -12.950267791748047, -17.253952026367188, 21.038131713867188, 10.073768615722656, 23.399505615234375, 3.40472412109375, -4.36767578125, 7.977294921875, 7.272941589355469, -0.34124755859375, 22.751380920410156, 2.0367889404296875, -4.6842041015625, 19.021617889404297, 14.47503662109375, 28.007102966308594, -3.37249755859375, 76.4224853515625, 33.61036682128906, 27.12164306640625, 16.559173583984375, -33.022247314453125, 5.5126953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000098.npy"}
{"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 12.604412078857422, "std": 28.356422424316406, "min": -56.12347412109375, "p10": -22.35593109130859, "median": 11.145051956176758, "p90": 50.09259948730469, "max": 84.01937866210938, "pos_frac": 0.6875, "sample": [-50.437713623046875, -2.9261703491210938, -39.5726318359375, 55.89935302734375, -2.3935470581054688, 9.243408203125, 17.55597686767578, 65.490234375, -0.426483154296875, -4.5531005859375, 34.81304931640625, -16.417999267578125, -17.046356201171875, 16.237350463867188, -7.9201812744140625, 20.255279541015625, 55.823150634765625, 39.64472961425781, 44.7117919921875, 10.456710815429688, 23.17047882080078, 14.097965240478516, 22.80572509765625, -56.12347412109375, 67.8907470703125, 37.0670166015625, 7.962594985961914, -11.965057373046875, 13.652328491210938, 30.39471435546875, 44.757049560546875, -4.644195556640625, 60.59088134765625, -0.685699462890625, -9.989898681640625, 7.51708984375, -3.269378662109375, -5.832550048828125, 32.30206298828125, 4.754627227783203, 41.54115295410156, 43.375640869140625, 26.058441162109375, 2.153564453125, 56.75749206542969, -21.35736083984375, 21.203330993652344, -0.08374595642089844, 10.192855834960938, 17.635040283203125, -31.994705200195312, -52.9898681640625, -50.601806640625, 28.077041625976562, 13.930633544921875, -6.71771240234375, 28.282135009765625, -17.46435546875, 39.621612548828125, 21.65545654296875, 15.67413330078125, -12.867118835449219, 50.176361083984375, 1.2774505615234375, -15.300399780273438, 22.79302978515625, -44.325958251953125, 15.932830810546875, 1.3732147216796875, 49.121429443359375, 3.17230224609375, 30.13983154296875, 14.668441772460938, -3.7731170654296875, 6.64752197265625, 37.03070068359375, 26.06709098815918, 66.85421752929688, 18.440628051757812, -8.85647201538086, 23.066253662109375, 15.894508361816406, -9.96002197265625, 4.2321014404296875, 32.42481994628906, 7.416814804077148, 50.05670166015625, -20.851318359375, 28.627166748046875, 10.31884765625, -39.245574951171875, 7.177452087402344, -24.685928344726562, 18.168914794921875, 35.67620849609375, 11.833393096923828, 20.2855224609375, -26.823974609375, 4.4918670654296875, -25.92425537109375, 44.034576416015625, 3.0400009155273438, 0.0, 5.106563568115234, -6.959564208984375, 46.26310729980469, 0.6547317504882812, -28.156944274902344, 58.21453857421875, 43.326446533203125, 65.09124755859375, 20.695449829101562, -42.913604736328125, 43.1309814453125, 7.672780990600586, 12.829261779785156, 14.54095458984375, 84.01937866210938, -9.90020751953125, 1.354461669921875, 9.206260681152344, -20.17626953125, 53.938140869140625, 9.774513244628906, 39.675537109375, 21.32452392578125, 64.13070678710938, 4.85687255859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000099.npy"}
{"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 12.455517768859863, "std": 25.55400276184082, "min": -45.5797119140625, "p10": -18.961975097656246, "median": 10.594268798828125, "p90": 47.940814208984364, "max": 77.96780395507812, "pos_frac": 0.6875, "sample": [14.72332763671875, 18.38555908203125, 30.85174560546875, -0.5307769775390625, 35.419921875, 3.76446533203125, -36.074493408203125, 45.55926513671875, -4.78814697265625, 14.538742065429688, -6.19097900390625, 11.176605224609375, 1.4072113037109375, 31.3966064453125, 24.483062744140625, 68.77853393554688, -5.253263473510742, -8.554962158203125, -26.5689697265625, 57.587127685546875, 10.560546875, -2.22296142578125, 12.2708740234375, 3.1027584075927734, -9.427841186523438, 6.57275390625, 21.15106201171875, 11.546958923339844, -13.148078918457031, 58.06341552734375, -26.89300537109375, -12.2213134765625, -2.244476318359375, 17.857505798339844, -8.58343505859375, 12.395904541015625, 11.58343505859375, 46.70703125, 3.2946243286132812, 38.11338806152344, 1.107421875, 15.246170043945312, 7.2916259765625, 24.23785400390625, -2.1824874877929688, 20.7559814453125, 29.296154022216797, 54.96295166015625, 50.81964111328125, 31.942413330078125, -20.48419952392578, -12.08258056640625, -17.918373107910156, -4.638710021972656, 10.521743774414062, 37.50334167480469, -29.1744384765625, -2.0101318359375, -45.5797119140625, -1.2466201782226562, 3.7026138305664062, 56.738800048828125, -14.192489624023438, 26.997848510742188, 38.788177490234375, 4.7594451904296875, -15.200607299804688, -26.071090698242188, 9.880126953125, 16.25946044921875, 25.79803466796875, 15.00006103515625, 10.62799072265625, 5.854103088378906, 4.162376403808594, -16.805511474609375, -24.528030395507812, -2.034271240234375, 28.919448852539062, 16.040401458740234, -35.245147705078125, 23.177963256835938, 51.279998779296875, -32.13946533203125, 17.291168212890625, 8.567831039428711, 6.394172668457031, -3.9546737670898438, 39.8138427734375, 77.96780395507812, 26.67547607421875, 3.34234619140625, -12.328109741210938, 7.9698486328125, 69.26614379882812, 16.82733917236328, -3.801177978515625, 13.767494201660156, 2.266988754272461, 23.856338500976562, 72.88943481445312, 2.414459228515625, 63.41790771484375, 38.663330078125, 31.7315673828125, -22.276107788085938, 38.26448059082031, 66.67550659179688, 8.645488739013672, 21.981353759765625, 62.552734375, 6.630767822265625, 17.34416961669922, -18.309593200683594, 14.662384033203125, 14.571334838867188, 0.09090614318847656, 15.687408447265625, 38.595001220703125, -26.64544677734375, 2.3188400268554688, 38.249755859375, 35.195037841796875, -15.22359848022461, 29.308128356933594, -4.3355560302734375, 31.997589111328125, -33.44187927246094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000100.npy"}
{"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 14.953869819641113, "std": 27.556543350219727, "min": -54.1815185546875, "p10": -22.014662170410155, "median": 11.27446460723877, "p90": 49.946582031249996, "max": 97.2178955078125, "pos_frac": 0.734375, "sample": [40.62774658203125, -54.1815185546875, 7.994110107421875, 24.49493408203125, 28.275497436523438, -26.039703369140625, 31.855323791503906, 1.6849365234375, -11.9908447265625, 30.133209228515625, 43.7098388671875, 42.323638916015625, 13.81243896484375, 57.581878662109375, 97.2178955078125, 62.03350830078125, 6.802120208740234, 8.0465087890625, -2.3781890869140625, 37.70819091796875, 54.990753173828125, 29.317764282226562, 58.168212890625, 31.05401611328125, 4.7418365478515625, 40.214080810546875, 16.355804443359375, 16.5474853515625, -0.3600311279296875, 64.85223388671875, -2.5835113525390625, 16.287933349609375, 7.4031219482421875, 2.3168182373046875, 36.64739990234375, 13.129119873046875, 62.79402160644531, 27.060699462890625, 34.46775817871094, 0.50677490234375, -17.28424072265625, -2.499176025390625, 30.440673828125, 48.4730224609375, 42.31581115722656, 34.34089660644531, 7.472808837890625, 35.2227783203125, 0.8473072052001953, -21.92877197265625, 0.3340301513671875, 38.500274658203125, -35.140594482421875, 23.8585205078125, 56.89794921875, 44.66912841796875, 64.85220336914062, -39.62480163574219, -4.467374801635742, 9.308975219726562, -23.069869995117188, 4.559051513671875, 13.484588623046875, 32.613746643066406, -9.56951904296875, -9.15266227722168, 13.990394592285156, 40.53514099121094, 6.1219329833984375, 56.15266418457031, 10.28375244140625, 12.265176773071289, 6.353515625, 6.577117919921875, 8.924272537231445, 3.7315673828125, 51.83013916015625, -1.6801204681396484, 3.3759231567382812, -45.795379638671875, -45.294677734375, 31.607818603515625, -18.06854820251465, 32.491790771484375, -2.279693603515625, 0.5212631225585938, 26.19830322265625, 30.97796630859375, 6.3644561767578125, 5.8559417724609375, 37.483253479003906, -2.2587833404541016, 8.460914611816406, 33.46771240234375, 14.958648681640625, 16.228668212890625, 4.699951171875, 59.45654296875, 31.945999145507812, 22.3525390625, -32.33807373046875, -41.590545654296875, 25.303041458129883, 7.838043212890625, 29.229446411132812, -11.0279541015625, 8.019287109375, 19.365097045898438, 95.9000244140625, 2.499359130859375, 22.395294189453125, -2.3562393188476562, 49.13934326171875, -1.8599853515625, 3.5761489868164062, 46.277870178222656, 9.411651611328125, -14.867996215820312, -25.1778564453125, 18.28234100341797, -28.671630859375, -1.622406005859375, -22.215072631835938, -0.5082435607910156, -27.010478973388672, -1.6040210723876953, 23.7203369140625, 17.0738525390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000101.npy"}
{"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 14.892481803894043, "std": 26.153095245361328, "min": -77.32077026367188, "p10": -16.725236511230467, "median": 14.34583854675293, "p90": 44.850294494628905, "max": 65.42979431152344, "pos_frac": 0.7265625, "sample": [0.7400054931640625, -15.243072509765625, 43.16063690185547, 5.991020202636719, 17.931015014648438, 54.90631103515625, 42.673370361328125, -12.855997085571289, 13.121185302734375, 48.884521484375, 17.67431640625, 14.32955551147461, 44.522796630859375, 9.17877197265625, 1.9777679443359375, 35.15480041503906, 36.475738525390625, -14.849609375, 24.695907592773438, 14.157371520996094, -19.132415771484375, 65.42979431152344, 27.452972412109375, 54.964141845703125, 36.142303466796875, -7.568695068359375, 6.837932586669922, -24.509307861328125, 7.5980224609375, -3.8207244873046875, -20.133575439453125, 45.61445617675781, 10.521186828613281, 42.048919677734375, -13.98382568359375, 42.56591796875, 20.726638793945312, -3.8288040161132812, 8.504638671875, -2.94818115234375, -25.5374755859375, -3.6600418090820312, 2.8433074951171875, 12.643943786621094, 10.400970458984375, 12.302719116210938, 23.025917053222656, -21.914337158203125, -39.091705322265625, -64.20213317871094, 34.76812744140625, -30.817916870117188, 19.609451293945312, 8.52203369140625, 26.174774169921875, 25.584442138671875, 27.883163452148438, 62.32879638671875, 20.612579345703125, 32.65521240234375, 14.273860931396484, -12.47515869140625, 35.689056396484375, 39.81982421875, 14.177093505859375, -15.693588256835938, -26.784408569335938, 6.71307373046875, -13.907123565673828, 0.4347801208496094, 37.844268798828125, 31.05913543701172, 32.453636169433594, 42.56549072265625, 16.453176498413086, 44.25323486328125, 19.99756622314453, 32.02531433105469, 59.258056640625, -35.369171142578125, -5.318258285522461, 38.85565185546875, 0.0, 5.1363983154296875, -2.268951416015625, 53.08038330078125, 0.0, 25.472190856933594, 16.97454833984375, 1.3363513946533203, 21.248245239257812, 25.69451904296875, -2.7570037841796875, 0.9427490234375, 62.89892578125, -1.981964111328125, 15.702239990234375, 4.591468811035156, 14.311004638671875, 51.5018310546875, 38.887603759765625, 14.36212158203125, 23.48029327392578, 55.77813720703125, 21.013275146484375, 27.91845703125, -0.1295452117919922, 33.03255081176758, -4.627483367919922, 28.773880004882812, -77.32077026367188, 40.561065673828125, 12.591293334960938, 5.97358512878418, 29.7130126953125, -27.496978759765625, 2.406280517578125, 2.2415313720703125, 65.1722412109375, -2.1745758056640625, 53.8419189453125, 43.1029052734375, 25.980987548828125, 40.0643310546875, 15.016098022460938, 42.69061279296875, -6.751312255859375, -35.311920166015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000102.npy"}
{"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 13.574668884277344, "std": 26.902713775634766, "min": -53.61334228515625, "p10": -12.537142181396485, "median": 7.600368499755859, "p90": 52.55492553710937, "max": 90.04513549804688, "pos_frac": 0.7265625, "sample": [-33.319427490234375, -3.200775146484375, 74.2603759765625, 31.24945068359375, 54.6212158203125, 11.087530136108398, -12.452796936035156, 1.6462631225585938, 87.34231567382812, 7.271881103515625, -1.309295654296875, 41.16644287109375, 5.66851806640625, 7.231719970703125, 0.0, 12.657196044921875, -19.639892578125, -15.280548095703125, -12.73394775390625, 15.637969970703125, 7.301677703857422, -0.4031524658203125, 9.914016723632812, -41.8038330078125, 5.067962646484375, 13.17724609375, 13.197425842285156, 12.400115966796875, 19.840377807617188, 16.01537322998047, 6.689117431640625, 8.02630615234375, 3.14849853515625, -10.35443115234375, 8.44476318359375, 0.5705947875976562, -1.0568695068359375, 55.538848876953125, 43.73719787597656, 24.68022918701172, 0.6821041107177734, 63.50590515136719, -17.8648681640625, 83.20260620117188, 0.7733306884765625, 64.01853942871094, 19.809341430664062, 21.598358154296875, 34.81951904296875, 73.845947265625, 54.819610595703125, 16.804534912109375, 31.8505859375, -29.38153076171875, 9.072845458984375, 4.291694641113281, -0.00275421142578125, 25.163238525390625, 31.66943359375, 27.49725341796875, 90.04513549804688, 9.563232421875, 9.724075317382812, 56.97651672363281, 4.6742095947265625, 51.66937255859375, 12.879013061523438, 51.18492126464844, 22.543075561523438, 23.062332153320312, 6.628028869628906, 8.686767578125, 1.4827728271484375, -11.56866455078125, -9.015594482421875, -1.105947494506836, -17.13137435913086, -53.61334228515625, 4.326385498046875, -9.2349853515625, 36.700225830078125, 41.04606628417969, 2.335062026977539, 39.10191345214844, -1.8302955627441406, 7.313751220703125, 14.10284423828125, -7.525177001953125, 51.569000244140625, -10.4683837890625, 7.108917236328125, 9.123291015625, 25.87939453125, 3.39459228515625, 1.5170135498046875, 2.5167388916015625, 89.04501342773438, 12.504226684570312, 0.30023193359375, -33.526824951171875, 0.0, 7.462608337402344, 5.8894195556640625, 38.514862060546875, 3.071624755859375, 59.04315185546875, -19.42413330078125, 7.738128662109375, -11.973838806152344, 33.15796661376953, 34.664886474609375, 13.53668212890625, 15.204360961914062, 40.216583251953125, -7.2140655517578125, 1.3060359954833984, 27.63580322265625, 18.02197265625, 5.174468994140625, 37.759521484375, -0.74676513671875, 12.058235168457031, -3.15802001953125, -11.095367431640625, -9.56964111328125, -49.26893615722656, -20.219650268554688, 5.57684326171875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000103.npy"}
{"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 9.962066650390625, "std": 26.963048934936523, "min": -57.31549072265625, "p10": -22.34587249755859, "median": 8.499284744262695, "p90": 45.51801681518555, "max": 93.42617797851562, "pos_frac": 0.6484375, "sample": [6.564178466796875, -4.901031494140625, 11.454353332519531, 51.6761474609375, -56.0843505859375, 13.69500732421875, 12.916778564453125, 9.40235710144043, 47.45330810546875, -1.50848388671875, 54.763427734375, -14.911334991455078, 10.967376708984375, 1.505340576171875, 32.745147705078125, 20.487632751464844, 73.52975463867188, 53.136322021484375, 9.10577392578125, -10.526618957519531, -1.497772216796875, 0.714447021484375, 12.80242919921875, -49.399993896484375, 18.581787109375, -21.918106079101562, 1.8930511474609375, -1.4474220275878906, 10.940214157104492, 46.091888427734375, 15.156509399414062, 27.705352783203125, -24.263580322265625, 2.69805908203125, 16.304351806640625, -40.343994140625, 5.922952651977539, -2.165740966796875, 53.936370849609375, 4.66217041015625, 32.803253173828125, 13.537353515625, -10.90130615234375, -11.254058837890625, -18.406097412109375, 22.46484375, -12.4190673828125, 33.901336669921875, -0.107696533203125, -10.61041259765625, -34.544586181640625, 5.354705810546875, 27.89447021484375, -13.58935546875, 56.74639892578125, 12.395164489746094, 38.037811279296875, -12.220169067382812, -13.02813720703125, -0.50469970703125, -18.345369338989258, 31.385398864746094, 4.176055908203125, 8.72930908203125, 4.034423828125, 21.301651000976562, -3.8260421752929688, 25.500030517578125, 14.837776184082031, -37.722198486328125, 34.19866943359375, 30.924896240234375, 17.177871704101562, -57.31549072265625, 76.0489501953125, -3.5600433349609375, 39.6529541015625, 42.63898468017578, 40.166481018066406, 49.40045166015625, 10.48602294921875, -44.8377685546875, -13.411834716796875, 30.989547729492188, 32.922943115234375, -11.743896484375, 3.944591522216797, 32.87518310546875, -5.11859130859375, 34.75331115722656, 93.42617797851562, -3.4228057861328125, 21.539825439453125, 20.2249755859375, -3.5403976440429688, 6.298557281494141, -44.585479736328125, 2.834714889526367, 41.725921630859375, 24.452239990234375, 45.272071838378906, 46.994140625, -10.812957763671875, 36.59942626953125, -22.643951416015625, -9.01947021484375, -27.664215087890625, 3.688201904296875, -6.7044830322265625, -27.13812255859375, 0.0, 25.775115966796875, 8.26926040649414, 18.576095581054688, 5.1054840087890625, 16.799774169921875, 29.6492919921875, -29.084518432617188, 0.8414382934570312, 7.0316009521484375, -22.218124389648438, 32.1776123046875, 1.1084556579589844, -1.941314697265625, 57.57720947265625, 9.4063720703125, 21.373123168945312, 13.515235900878906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000104.npy"}
{"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 20.56756591796875, "std": 28.785106658935547, "min": -75.958984375, "p10": -12.750831604003906, "median": 18.25808620452881, "p90": 53.62482604980468, "max": 104.78424072265625, "pos_frac": 0.796875, "sample": [11.41796875, 17.8541259765625, 50.03050231933594, 8.70748519897461, -6.20733642578125, 23.118011474609375, 45.96429443359375, 41.07111358642578, 27.001174926757812, -6.510246276855469, 32.24163818359375, 10.08544921875, 49.8946533203125, 46.168548583984375, 0.143585205078125, 63.6968994140625, 19.63311004638672, -12.17071533203125, 46.65667724609375, 53.22857666015625, 1.9466743469238281, 7.63134765625, 14.649322509765625, 13.333564758300781, 56.20195007324219, 37.84454345703125, -50.768890380859375, 4.930816650390625, 2.5358123779296875, 63.065673828125, 37.93070983886719, 12.437408447265625, 2.0626449584960938, 56.46772766113281, 7.8094329833984375, 38.229736328125, 47.75724792480469, -7.168243408203125, 46.54876708984375, -28.21759033203125, 9.67608642578125, 58.423828125, 40.757354736328125, 50.6749267578125, 3.20245361328125, 30.568084716796875, 37.0733642578125, -17.827362060546875, 43.395904541015625, 73.9498291015625, -1.67022705078125, 41.378753662109375, 20.576019287109375, 0.49212646484375, -5.808319091796875, 22.028106689453125, 16.429229736328125, 43.399169921875, 14.526908874511719, -26.5030517578125, -3.814208984375, -14.5146484375, -4.7906494140625, 33.87742614746094, 46.17413330078125, 15.826416015625, 48.079681396484375, 3.897216796875, 15.764328002929688, -8.210906982421875, 24.79400634765625, -62.524383544921875, 23.100093841552734, 47.760528564453125, 18.73809814453125, -12.562713623046875, 45.79931640625, 46.054168701171875, 104.78424072265625, -2.8587570190429688, 12.99267578125, 17.735092163085938, 36.22724914550781, 9.843034744262695, 15.659393310546875, 31.7510986328125, -37.04705810546875, 55.232635498046875, 25.509674072265625, -42.050262451171875, 14.374166488647461, 51.97123718261719, 58.26055908203125, -24.066986083984375, 7.711723327636719, 0.00970458984375, 18.662046432495117, -13.189773559570312, 10.5047607421875, 0.0, 7.322772979736328, 17.7723388671875, -5.373149871826172, 16.264541625976562, 39.946266174316406, 6.2133636474609375, 69.11880493164062, 54.549407958984375, 32.37187194824219, 16.88542938232422, 30.325851440429688, 79.44282531738281, 0.1824188232421875, 33.302330017089844, 47.745758056640625, 25.3375244140625, -13.623794555664062, 40.90570068359375, 12.459205627441406, -75.958984375, 35.15925598144531, 65.08865356445312, 20.903907775878906, 8.438880920410156, -23.252288818359375, 49.79278564453125, 43.284034729003906, 40.58087158203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000105.npy"}
{"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 12.505602836608887, "std": 30.364370346069336, "min": -70.94778442382812, "p10": -24.49443664550781, "median": 12.008636474609375, "p90": 49.72729187011718, "max": 82.38995361328125, "pos_frac": 0.65625, "sample": [3.0671234130859375, -3.22271728515625, 21.690826416015625, 16.56756591796875, 11.399871826171875, 48.5062255859375, -11.111129760742188, 18.091522216796875, -9.56561279296875, 31.186355590820312, -36.72430419921875, -4.979255676269531, 0.9450225830078125, 53.829864501953125, 59.31201171875, 57.657806396484375, 33.831298828125, -16.84033203125, -18.379913330078125, -9.045562744140625, 22.459144592285156, -6.0312347412109375, 31.76751708984375, 42.05812072753906, 23.712249755859375, 12.373992919921875, 27.290098190307617, -14.16680908203125, 29.271652221679688, 52.576446533203125, 3.874237060546875, 21.99371337890625, 12.1845703125, 10.677078247070312, 2.361011505126953, -10.519485473632812, 45.636474609375, -37.92622375488281, 17.313507080078125, 38.72132110595703, 44.111053466796875, -23.03179931640625, 29.224441528320312, -32.85736083984375, -13.6495361328125, 27.04046630859375, -24.60186767578125, 0.039764404296875, 38.728759765625, 30.60913848876953, 2.8892822265625, 3.27423095703125, -15.294986724853516, 62.309722900390625, -68.58714294433594, -58.94110107421875, 39.07537841796875, 73.76934814453125, 74.68421936035156, 40.380645751953125, -1.26470947265625, -4.581268310546875, -45.612030029296875, -70.94778442382812, 12.97430419921875, -11.978263854980469, -3.421905517578125, -2.49853515625, 53.98944091796875, -25.897308349609375, 42.655548095703125, 31.082427978515625, -14.963417053222656, 20.736099243164062, -31.9952392578125, 2.4314441680908203, -16.310028076171875, -17.773239135742188, 25.6080322265625, 13.117446899414062, 9.165786743164062, 82.38995361328125, 19.40631103515625, 10.627754211425781, -7.93182373046875, 18.63909912109375, 67.50970458984375, 3.887969970703125, 41.85064697265625, -36.00732421875, 33.725608825683594, -8.990028381347656, 30.131591796875, 3.335540771484375, -0.9036712646484375, 22.5113525390625, 73.41935729980469, 29.816680908203125, 6.6166839599609375, 46.204071044921875, 72.6905517578125, -32.6734619140625, -2.701955795288086, -13.021255493164062, 21.280975341796875, 11.83270263671875, 44.656166076660156, -24.448394775390625, 9.6280517578125, 10.90838623046875, -33.0770263671875, -13.94281005859375, 71.02020263671875, -0.9024143218994141, -9.613006591796875, 26.5875244140625, 12.224517822265625, -17.605148315429688, 46.9610595703125, 14.4083251953125, 43.128662109375, 27.38720703125, 2.0665740966796875, 30.381866455078125, 41.29713439941406, 18.424888610839844, 41.36537170410156, 0.679534912109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000106.npy"}
{"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 19.525043487548828, "std": 28.765731811523438, "min": -104.5789794921875, "p10": -11.01279830932617, "median": 15.93707275390625, "p90": 57.84528503417969, "max": 91.84930419921875, "pos_frac": 0.7421875, "sample": [15.976806640625, -7.6590576171875, 9.635345458984375, 15.512260437011719, 1.852203369140625, -8.7799072265625, 30.547393798828125, 58.6412353515625, 14.813934326171875, 24.175125122070312, -28.130081176757812, 8.251876831054688, 25.839820861816406, 50.271697998046875, 3.2022247314453125, -5.96356201171875, 29.369842529296875, -18.880615234375, 31.163864135742188, 68.84576416015625, -3.6252288818359375, 50.06486511230469, 35.95482635498047, 11.99072265625, -13.22589111328125, 44.987640380859375, 23.124298095703125, -24.6563720703125, 49.42915344238281, -15.57440185546875, 66.8372802734375, 4.186912536621094, 10.493621826171875, 17.936691284179688, 52.01617431640625, 48.049530029296875, 28.324859619140625, 16.818695068359375, 91.3514404296875, 4.909759521484375, 44.41339111328125, -2.68414306640625, 5.199920654296875, 40.323394775390625, 20.52349853515625, 8.096433639526367, -13.504547119140625, 17.302734375, 30.5477294921875, 11.256851196289062, 11.133586883544922, -2.3515167236328125, 19.697174072265625, 34.42182922363281, 36.295196533203125, -12.272026062011719, 1.70123291015625, 28.071327209472656, 39.931732177734375, 36.24592590332031, 3.61297607421875, 8.82843017578125, -1.87933349609375, 35.0877685546875, 0.3467559814453125, -18.07428741455078, 11.793960571289062, 57.94677734375, 46.402923583984375, 64.810302734375, -18.703369140625, 59.93450927734375, 1.99603271484375, 53.30810546875, 76.68017578125, -3.2444992065429688, 0.928314208984375, 54.37757873535156, 39.07005310058594, 32.9288215637207, 43.860687255859375, 25.942520141601562, 5.718292236328125, -10.473129272460938, 15.8973388671875, 17.873260498046875, 32.608238220214844, -15.757720947265625, 91.84930419921875, 0.16650390625, 57.801788330078125, 53.834503173828125, -1.5421142578125, 44.578125, -5.4575653076171875, 2.2290897369384766, 28.90277099609375, 2.11297607421875, 9.9849853515625, 7.425601959228516, -104.5789794921875, -21.793182373046875, 19.71307373046875, 57.64263916015625, 23.703773498535156, -2.6390304565429688, 90.18777465820312, 6.496826171875, 37.95533752441406, -0.1873016357421875, 62.02598571777344, -17.733428955078125, 16.970245361328125, 5.1887054443359375, -4.59075927734375, 36.4822998046875, 27.659637451171875, -1.9302978515625, -7.84234619140625, -0.12738037109375, -9.72821044921875, 26.711349487304688, -9.70904541015625, 39.29975891113281, -2.2282257080078125, 76.83860778808594, 11.304065704345703, 57.97564697265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000107.npy"}
{"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 18.106571197509766, "std": 32.253639221191406, "min": -71.15103149414062, "p10": -22.375543212890623, "median": 15.016395568847656, "p90": 60.47447204589843, "max": 114.71331787109375, "pos_frac": 0.7578125, "sample": [51.8690185546875, -20.070465087890625, -6.6255035400390625, 23.917678833007812, 48.27013397216797, 11.664459228515625, 13.795482635498047, 16.127681732177734, -0.08244705200195312, 58.4400634765625, 0.005279541015625, -12.19140625, -26.410629272460938, 36.604766845703125, 1.03289794921875, -35.165985107421875, 67.08392333984375, -5.7491607666015625, -3.1375465393066406, 28.852813720703125, 5.30389404296875, 21.611541748046875, 27.038787841796875, -6.022705078125, 38.04718017578125, 16.899246215820312, 70.9521484375, -21.360626220703125, 0.0, -41.516143798828125, 41.22076416015625, 0.78607177734375, 35.32464599609375, 16.542083740234375, 88.84835815429688, 0.915374755859375, 36.9068603515625, 22.413787841796875, 2.4793701171875, 0.3082160949707031, 1.9679374694824219, -28.8485107421875, 4.92706298828125, -24.9754638671875, 18.716705322265625, 59.680206298828125, 35.7572021484375, -41.73382568359375, -34.438812255859375, 84.74563598632812, 36.780029296875, 16.106422424316406, 31.334579467773438, 108.83184814453125, 9.911453247070312, 13.1451416015625, 0.0, 40.548797607421875, 2.4106311798095703, 1.5863380432128906, 0.8185997009277344, 67.693115234375, 33.20307922363281, 11.341278076171875, -67.96458435058594, -37.191741943359375, 42.59600830078125, 18.493345260620117, 14.913253784179688, 25.183197021484375, 42.095611572265625, 24.850723266601562, -2.729766845703125, -3.2001953125, -11.467811584472656, 42.115966796875, 56.99896240234375, -5.939945220947266, 114.71331787109375, 53.74961853027344, 20.041717529296875, -26.813037872314453, 77.56005859375, 35.868896484375, 20.27716827392578, 1.0515899658203125, 39.827972412109375, 4.0111846923828125, 5.097137451171875, 42.6932373046875, 3.53399658203125, 41.73736572265625, -3.96978759765625, 18.317184448242188, 11.970190048217773, 11.516357421875, 74.68804931640625, 50.6171875, 81.87646484375, 62.3277587890625, 10.705528259277344, 6.17620849609375, 9.476715087890625, 0.0, -2.2014312744140625, 12.900802612304688, 2.3042984008789062, 66.0250244140625, 15.119537353515625, 75.67434692382812, 36.158477783203125, -71.15103149414062, 33.604896545410156, 11.595125198364258, 17.658340454101562, 50.95294189453125, 55.77972412109375, 15.194259643554688, 7.209129333496094, 17.06572723388672, -27.312728881835938, -13.833976745605469, -24.743682861328125, 6.4658966064453125, 20.119110107421875, 6.853515625, 25.00421142578125, 20.954315185546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000108.npy"}
{"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 20.06651496887207, "std": 32.09856033325195, "min": -47.60711669921875, "p10": -17.710577392578124, "median": 15.83013916015625, "p90": 64.64503784179686, "max": 126.7510986328125, "pos_frac": 0.765625, "sample": [-4.070182800292969, -9.171600341796875, -27.27472686767578, 44.630157470703125, 59.957672119140625, 17.052688598632812, 27.93804931640625, 5.113922119140625, -43.379638671875, 15.827239990234375, 24.43487548828125, 116.26123046875, 1.6210517883300781, -15.716552734375, -10.931365966796875, 5.19140625, -12.842056274414062, 8.074951171875, 68.26651000976562, 0.4795188903808594, 10.876562118530273, 13.016006469726562, -21.413848876953125, -0.974822998046875, -15.43478775024414, 15.13043212890625, 2.53179931640625, 11.664398193359375, -0.029033660888671875, 36.53326416015625, 17.141448974609375, 2.9088134765625, 47.956817626953125, 71.95046997070312, 32.950714111328125, 54.68805694580078, 38.10319519042969, 67.56634521484375, 0.5775299072265625, 28.298065185546875, 41.8504638671875, 39.394378662109375, 7.6353302001953125, -1.9552898406982422, 21.93780517578125, 74.74221801757812, 1.7541656494140625, 66.16143798828125, 25.80316925048828, 10.536003112792969, 66.10379028320312, 2.16180419921875, 29.29296875, 24.862579345703125, -45.3719482421875, 13.945098876953125, 30.908203125, 3.0085792541503906, 34.61949157714844, 45.440673828125, 64.04782104492188, 14.121231079101562, -2.32354736328125, 33.9674072265625, 17.3809814453125, 9.941932678222656, -36.43597412109375, 21.10076904296875, 20.074966430664062, 23.934249877929688, 23.623931884765625, -22.58782958984375, -30.816314697265625, 21.430206298828125, 32.897979736328125, 42.6890869140625, 56.688140869140625, -29.57763671875, 0.7426719665527344, 66.85064697265625, 7.82684326171875, 35.70103454589844, 55.39910888671875, 21.60871124267578, 5.657623291015625, 8.063652038574219, 35.37431335449219, 1.0495109558105469, -20.386993408203125, 87.1922607421875, 46.46075439453125, 43.926544189453125, 9.09830093383789, 100.97296142578125, 5.543357849121094, 15.833038330078125, 1.6515464782714844, 126.7510986328125, -13.875717163085938, 30.991676330566406, -17.50372314453125, 74.76751708984375, 0.0, 17.38188934326172, 66.03854370117188, 12.929061889648438, 63.00726318359375, 29.66876220703125, -0.2251739501953125, 44.024810791015625, 6.83917236328125, 8.778717041015625, -18.1932373046875, -8.177509307861328, 44.066162109375, -2.3676300048828125, 37.69621276855469, 52.248046875, -44.794921875, -47.60711669921875, -32.537994384765625, 35.16632080078125, 22.752853393554688, 62.19844055175781, 13.268798828125, -13.98492431640625, 8.859588623046875, 43.28814697265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000109.npy"}
{"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 18.84237289428711, "std": 34.00063705444336, "min": -74.25079345703125, "p10": -20.292192077636717, "median": 15.659122467041016, "p90": 64.41622314453123, "max": 109.76513671875, "pos_frac": 0.7265625, "sample": [13.473068237304688, -74.25079345703125, -30.811840057373047, 61.031768798828125, 5.7274322509765625, 11.231220245361328, -18.184967041015625, 11.55755615234375, -14.80255126953125, 99.59469604492188, 26.6456298828125, 3.2183380126953125, -1.3882827758789062, 44.15496826171875, -2.2678909301757812, -7.551948547363281, 20.259490966796875, 18.86730194091797, 48.154205322265625, 2.952880859375, -20.108917236328125, 25.127685546875, -1.9639625549316406, 45.122100830078125, 21.783050537109375, 9.063983917236328, -17.433856964111328, 40.327972412109375, 17.5609130859375, -5.8286590576171875, 87.18344116210938, 61.291290283203125, 81.511474609375, 3.531229019165039, 69.8074951171875, 16.609161376953125, 58.69966125488281, 52.511810302734375, 62.47589111328125, -2.21026611328125, 14.794898986816406, 1.76123046875, -20.719833374023438, 34.0386962890625, -17.3326416015625, 10.369365692138672, 23.245498657226562, -28.370452880859375, 12.020004272460938, 4.584754943847656, 48.91706848144531, -49.126617431640625, 19.1748046875, 13.7403564453125, 44.482696533203125, 3.81903076171875, 6.5919342041015625, -55.170806884765625, 34.92341613769531, 109.76513671875, 0.6701278686523438, 16.523345947265625, 33.58525085449219, -1.51727294921875, -2.282989501953125, 41.45574951171875, -33.74676513671875, 54.730499267578125, 53.27679443359375, -23.4703369140625, 0.027660369873046875, 26.64599609375, 17.014976501464844, 31.965797424316406, 18.87653350830078, 18.704940795898438, -3.2033538818359375, -25.898300170898438, -46.64411926269531, 42.0445556640625, 0.0, -13.8350830078125, 42.2579345703125, 14.134757995605469, -33.607051849365234, 40.803497314453125, -0.34353065490722656, -45.106903076171875, 62.325775146484375, 10.651588439941406, 32.213348388671875, 77.23638916015625, 19.81060791015625, 22.065155029296875, 23.033729553222656, 31.5126953125, 43.13433837890625, 2.0863494873046875, 4.348846435546875, 50.5118408203125, 73.59344482421875, 5.1149749755859375, 10.104045867919922, 72.10211181640625, -62.100189208984375, 34.937255859375, 10.519546508789062, -10.628616333007812, 68.94366455078125, 56.098480224609375, 1.9451446533203125, 79.16986083984375, -0.479248046875, 7.1270294189453125, 27.027374267578125, 75.66082763671875, 81.77764892578125, 4.170082092285156, 43.39466857910156, 35.10797119140625, -8.692001342773438, 77.34762573242188, 14.356964111328125, 17.37856674194336, -0.2354736328125, 26.57623291015625, -18.060165405273438, 53.3922119140625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000110.npy"}
{"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 19.914920806884766, "std": 35.21064376831055, "min": -79.02984619140625, "p10": -22.25198516845703, "median": 17.928607940673828, "p90": 60.62474670410154, "max": 114.25042724609375, "pos_frac": 0.7578125, "sample": [1.7993392944335938, -19.464996337890625, 47.0224609375, -48.523406982421875, 110.7049560546875, -2.1206283569335938, 50.748382568359375, 81.80303955078125, -2.092601776123047, 5.0863189697265625, 72.34658813476562, 5.0208740234375, 18.38787841796875, 18.65692138671875, -8.85748291015625, 54.525325775146484, 19.037132263183594, 69.59951782226562, -35.019927978515625, -17.688613891601562, 19.981719970703125, 29.394941329956055, 26.81591796875, 27.1141357421875, 36.93988037109375, 4.892547607421875, 10.090629577636719, 53.307342529296875, -7.9764404296875, 37.52465057373047, 30.824989318847656, 42.5821533203125, -9.63360595703125, 96.84893798828125, -7.68670654296875, 3.925933837890625, -11.734344482421875, 42.69952392578125, 0.050994873046875, 68.72915649414062, 18.23017120361328, 27.699111938476562, -3.6266555786132812, 50.528778076171875, 8.13134765625, 0.95172119140625, 108.83090209960938, -9.688201904296875, 66.158203125, 47.85828399658203, 3.836212158203125, 114.25042724609375, -53.07080078125, 2.6468505859375, 5.45513916015625, 18.2203369140625, 9.925048828125, 55.21696472167969, -8.658073425292969, 94.5340576171875, 21.25738525390625, 10.312362670898438, 11.045700073242188, 21.624847412109375, 6.5691070556640625, -39.44432067871094, 3.866485595703125, 48.089599609375, 40.30615234375, 8.1446533203125, 24.4019775390625, 3.8978424072265625, 13.777999877929688, 13.831192016601562, 52.477783203125, 5.044685363769531, -2.530517578125, 44.039093017578125, 3.4144287109375, -27.546630859375, 7.265693664550781, 42.58978271484375, -27.40350341796875, 3.7283935546875, 19.6541748046875, -23.587387084960938, -13.806533813476562, 2.67437744140625, 57.24922180175781, 47.61309814453125, 56.19535827636719, 36.695556640625, 49.68608093261719, -1.20880126953125, 46.71086120605469, 30.646133422851562, 11.466085433959961, 1.1758384704589844, 87.67596435546875, 23.559234619140625, 31.016700744628906, 7.094383239746094, -53.212554931640625, -79.02984619140625, -40.78717041015625, 54.078338623046875, -20.466094970703125, 36.149139404296875, 11.165847778320312, 31.072219848632812, 81.29719543457031, -2.94647216796875, 57.63677978515625, 58.253265380859375, 66.67376708984375, 35.78070068359375, 53.7034912109375, -45.1065673828125, 42.564605712890625, 17.636878967285156, 25.94397735595703, -34.97442626953125, 25.88299560546875, 12.302703857421875, 23.705810546875, 4.37310791015625, -21.916183471679688, -23.0355224609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000111.npy"}
{"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 12.755572319030762, "std": 37.60680389404297, "min": -79.4412841796875, "p10": -33.74142761230468, "median": 9.063597679138184, "p90": 60.81135864257812, "max": 109.94760131835938, "pos_frac": 0.6015625, "sample": [48.149993896484375, 8.642959594726562, -9.464584350585938, 16.032470703125, -5.2595062255859375, 24.57318115234375, -3.582855224609375, -75.61688232421875, -16.53022003173828, 3.6718292236328125, 9.484235763549805, -2.1878623962402344, -51.18670654296875, 67.48727416992188, -6.8919677734375, 12.414878845214844, 51.6517333984375, 52.877960205078125, -0.41793060302734375, 45.830810546875, -33.50382995605469, 33.285430908203125, 21.174224853515625, 49.99102783203125, -79.4412841796875, -6.18011474609375, -12.444602966308594, 24.202957153320312, 60.59156799316406, 33.521751403808594, -34.29582214355469, 43.347862243652344, -46.398223876953125, 17.828079223632812, 17.02997589111328, -46.5313720703125, -4.603790283203125, 23.695594787597656, 7.3642425537109375, -8.127189636230469, 40.642974853515625, 6.300048828125, 52.9560546875, 66.93185424804688, 26.669952392578125, -14.949882507324219, 50.546844482421875, 7.9474945068359375, 78.05191040039062, -27.996429443359375, 17.621231079101562, -1.808441162109375, -3.9737014770507812, 3.1923828125, -10.40570068359375, -0.4199352264404297, -43.6668701171875, -8.551422119140625, -63.154876708984375, -6.2432403564453125, -6.8057708740234375, -12.073036193847656, 33.71870422363281, 104.47308349609375, 13.25286865234375, -28.700302124023438, 56.356048583984375, 61.32420349121094, 6.2835693359375, 37.029029846191406, 23.039310455322266, -1.8734359741210938, 40.929931640625, 34.1768798828125, 12.350555419921875, 109.5550537109375, -3.7684059143066406, 4.8132476806640625, 73.29608154296875, 16.342281341552734, -11.1512451171875, 44.296234130859375, 13.700828552246094, 0.12013626098632812, 22.81195068359375, 41.42059326171875, 23.98095703125, 5.063568115234375, 10.247634887695312, -18.179489135742188, 109.94760131835938, -19.136886596679688, 84.3548583984375, 1.0025825500488281, -56.886810302734375, 21.939468383789062, -29.611595153808594, -1.6428604125976562, -58.61590576171875, -10.062141418457031, -41.199371337890625, 42.19781494140625, -14.034713745117188, -26.1875, 33.18785095214844, 65.44235229492188, -14.658843994140625, 70.19964599609375, 23.357284545898438, 53.243896484375, 65.7969970703125, -2.87066650390625, 2.4398040771484375, 43.847900390625, 28.712600708007812, 2.6652984619140625, -40.20538330078125, 11.51348876953125, 18.954483032226562, -21.627777099609375, 57.3114013671875, -24.826873779296875, 57.43792724609375, 55.32676696777344, 70.65797424316406, -70.69485473632812, -0.17671966552734375, 39.70747375488281], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000112.npy"}
{"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 23.56146240234375, "std": 39.415470123291016, "min": -89.40182495117188, "p10": -18.417981529235835, "median": 23.562911987304688, "p90": 69.07917175292968, "max": 161.590087890625, "pos_frac": 0.7421875, "sample": [-5.0084381103515625, 38.960052490234375, 51.22509765625, -2.5410194396972656, 31.883880615234375, -3.3424148559570312, -11.046844482421875, 67.5322265625, 35.92523193359375, 20.596710205078125, 27.602859497070312, -12.763107299804688, 68.14935302734375, -11.14992904663086, -17.4820556640625, 2.2914371490478516, 1.6439971923828125, 0.15579986572265625, -4.951934814453125, 3.912599563598633, 116.7325439453125, 29.9307861328125, -1.4022674560546875, 82.67242431640625, 26.972412109375, -34.081207275390625, 32.92118835449219, 36.89471435546875, 14.673004150390625, -25.931869506835938, 17.264312744140625, -14.6214599609375, 161.590087890625, 43.73016357421875, 48.1466064453125, 14.076690673828125, 36.35536193847656, -34.01792907714844, 54.996337890625, -66.064453125, 14.05224609375, 35.69630432128906, 4.4925384521484375, 58.33244323730469, 88.54257202148438, 23.47144317626953, -15.910736083984375, 1.1281967163085938, -0.614288330078125, 31.351593017578125, 20.141006469726562, 34.70668411254883, -48.257293701171875, 49.47157287597656, 39.52276611328125, 39.537261962890625, 32.51308059692383, 28.311370849609375, 31.042755126953125, -13.789901733398438, 4.84637451171875, 27.148818969726562, -20.601808547973633, 88.27810668945312, -5.775810241699219, 36.09991455078125, 4.602012634277344, 19.9666748046875, -29.388336181640625, 25.8531494140625, 10.44976806640625, 157.88330078125, -2.929697036743164, 47.446075439453125, -34.08915710449219, 1.0297317504882812, 42.889617919921875, 60.155975341796875, 67.73654174804688, 4.052825927734375, 45.84959411621094, -25.5577392578125, 14.386260986328125, 58.17822265625, 85.03506469726562, 39.5650634765625, 13.843605041503906, -6.736846923828125, 0.7637176513671875, 6.126495361328125, 71.24874877929688, 73.93537139892578, 30.526214599609375, 23.373931884765625, -34.34074401855469, 58.7640380859375, -0.008209228515625, 48.4185791015625, 97.12091064453125, 44.682159423828125, 23.654380798339844, -13.62933349609375, 33.755767822265625, 38.04539489746094, 92.2564697265625, 54.3857421875, 91.04949951171875, 97.04562377929688, -15.5283203125, 8.888885498046875, 37.48042297363281, 48.47947692871094, 2.599761962890625, 63.5908203125, -11.487205505371094, -89.40182495117188, 1.353790283203125, 2.9787445068359375, 41.28318786621094, 34.45069885253906, 0.209136962890625, 49.139678955078125, -36.291046142578125, 39.56280517578125, 18.94634246826172, 1.5884246826171875, 28.720703125, -54.261993408203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000113.npy"}
{"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 13.136994361877441, "std": 34.79951095581055, "min": -77.81463623046875, "p10": -30.231922912597653, "median": 12.839027404785156, "p90": 56.92694091796875, "max": 94.23333740234375, "pos_frac": 0.6875, "sample": [11.85086441040039, 28.17425537109375, 27.32940673828125, 45.72111129760742, -9.109397888183594, 6.94732666015625, 2.4823951721191406, 18.365406036376953, 18.461074829101562, -12.322021484375, -12.534439086914062, 55.29490661621094, 18.316192626953125, 58.2186279296875, 4.393524169921875, 48.88836669921875, -24.424640655517578, 7.025970458984375, -52.349273681640625, 8.115982055664062, 2.794097900390625, -72.31208801269531, 14.346733093261719, 56.2015380859375, 29.941322326660156, 29.15802001953125, 20.406600952148438, 39.13597106933594, 49.7176513671875, 32.3624267578125, 20.861984252929688, 48.78660583496094, 11.053848266601562, 18.137130737304688, 26.810546875, -77.81463623046875, 61.74787902832031, -53.947296142578125, 51.00360107421875, -3.239654541015625, 28.788986206054688, -8.191070556640625, 5.396352767944336, -17.30413818359375, 12.25372314453125, 7.64739990234375, 23.780197143554688, 30.237548828125, 25.899322509765625, -12.619384765625, -7.187744140625, -8.175704956054688, 34.733856201171875, -38.7796630859375, -41.27073669433594, 2.6039581298828125, -5.174835205078125, 57.82135009765625, 57.25250244140625, -41.8433837890625, 61.25862121582031, -74.38565063476562, 63.029815673828125, -42.190460205078125, -11.27587890625, 93.75253295898438, 40.612030029296875, -27.874664306640625, 18.66809844970703, 84.4945068359375, -77.402587890625, 39.594207763671875, -19.802337646484375, 48.86262512207031, -5.96563720703125, 48.425537109375, -1.0403366088867188, -10.095817565917969, 17.490676879882812, -54.530242919921875, 12.805099487304688, 13.721755981445312, 15.44940185546875, -10.3126220703125, -1.0166988372802734, -7.224651336669922, -33.379791259765625, 31.557464599609375, 0.0, 71.18353271484375, 12.542808532714844, -18.57558822631836, 49.14970397949219, 3.7597503662109375, 4.447113037109375, 56.78741455078125, 2.322021484375, 33.2650146484375, 1.163970947265625, 12.139968872070312, 46.25334930419922, 47.594696044921875, 18.65771484375, 0.9340667724609375, 94.23333740234375, 55.12593078613281, 17.4930419921875, 5.068428039550781, -29.614120483398438, 0.0, 86.96234130859375, -24.170013427734375, -28.88330078125, 38.411376953125, -24.78961181640625, 58.26695251464844, 12.872955322265625, -31.6734619140625, 3.7299156188964844, 14.154983520507812, 24.385101318359375, 58.157470703125, 23.836217880249023, 6.79034423828125, 17.9215087890625, 22.395477294921875, 10.488767623901367, 55.654632568359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000114.npy"}
{"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 13.019362449645996, "std": 37.95793533325195, "min": -83.4703369140625, "p10": -28.73639678955078, "median": 9.290069580078125, "p90": 63.6924331665039, "max": 141.4671630859375, "pos_frac": 0.609375, "sample": [109.68405151367188, -29.365570068359375, 42.918731689453125, -11.905597686767578, -3.56158447265625, -30.082321166992188, -17.78045654296875, 26.174530029296875, -17.53329086303711, 18.923526763916016, -11.79437255859375, 25.827117919921875, -1.3995361328125, 1.7237930297851562, 30.792884826660156, -20.139427185058594, -1.0107498168945312, -28.561050415039062, 49.3824462890625, 22.665451049804688, -32.83470153808594, 17.458099365234375, 6.985954284667969, 10.166015625, -5.363744735717773, -1.000030517578125, 26.6348876953125, 21.56940460205078, 22.529541015625, 9.10089111328125, -5.8976593017578125, 24.82257080078125, -3.38470458984375, -0.7014694213867188, 3.0370426177978516, 9.479248046875, -15.4105224609375, -41.16249084472656, -17.56824493408203, 115.52880859375, 11.181770324707031, 51.38543701171875, 56.8818359375, 25.03583526611328, -39.72344970703125, -27.231765747070312, 46.60282897949219, -9.496047973632812, 46.514373779296875, 81.98931884765625, 17.486724853515625, 129.8616943359375, -31.75958251953125, 11.1610107421875, -17.39935302734375, -3.0001220703125, 66.03854370117188, 37.4490966796875, 69.54621887207031, -69.91732788085938, -57.933563232421875, -7.942840576171875, 52.5252685546875, 26.76446533203125, -29.145538330078125, 25.18768310546875, 5.5019683837890625, 16.88226318359375, 40.8009033203125, 81.1181640625, 0.29836463928222656, 11.36810302734375, 79.23628997802734, 23.184814453125, 25.419008255004883, 17.447053909301758, 90.5831298828125, 14.375457763671875, -23.988800048828125, 41.532806396484375, -1.58453369140625, -15.793609619140625, -11.95068359375, 81.7650146484375, 35.37287902832031, -0.52728271484375, -83.4703369140625, 16.045913696289062, 10.327983856201172, 64.4354248046875, -29.873294830322266, 17.607803344726562, -15.9844970703125, 84.07736206054688, 22.194976806640625, -8.441848754882812, -4.091712951660156, -49.068603515625, -12.617584228515625, 8.24462890625, 39.99951171875, -6.361171722412109, 11.008094787597656, 7.54167366027832, 25.138458251953125, 63.37400817871094, 141.4671630859375, 3.2210693359375, -26.1046142578125, -25.640380859375, 6.0939788818359375, 3.3261642456054688, 23.765426635742188, -14.610641479492188, 8.134319305419922, -9.742523193359375, -15.232879638671875, 28.488723754882812, 8.886436462402344, 53.317108154296875, -68.925537109375, 3.7968826293945312, 38.6685791015625, 21.24053955078125, 12.257415771484375, 13.393218994140625, -13.223968505859375, 41.76385498046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000115.npy"}
{"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 17.517780303955078, "std": 33.55915832519531, "min": -108.77633666992188, "p10": -16.184280395507812, "median": 11.294370651245117, "p90": 62.33706512451172, "max": 105.85031127929688, "pos_frac": 0.6875, "sample": [-2.142547607421875, -18.68682861328125, 30.259567260742188, -2.690582275390625, 10.36199951171875, 28.285980224609375, 47.21086120605469, 44.00779724121094, 84.471923828125, 28.728759765625, 30.767852783203125, 8.494873046875, 47.4130859375, 22.543899536132812, -5.4462890625, 6.410572052001953, 38.504371643066406, 46.06951904296875, -5.386325836181641, 17.20050811767578, -64.10104370117188, -5.184640884399414, 61.740623474121094, 38.99462890625, 3.368936538696289, -29.979583740234375, 43.80943298339844, -4.039581298828125, 31.815467834472656, -31.0308837890625, 1.5574989318847656, 38.62408447265625, -11.162521362304688, 70.4403076171875, 4.448600769042969, 17.308074951171875, 26.07404327392578, 19.2935791015625, -15.461090087890625, -7.0338897705078125, -5.957305908203125, 26.373565673828125, 26.735931396484375, -19.805065155029297, 7.967529296875, 11.69833755493164, -10.6361083984375, 41.367218017578125, -22.339935302734375, 43.85235595703125, -108.77633666992188, 20.908157348632812, 76.38294982910156, -15.87725830078125, -9.410507202148438, 52.8377685546875, 29.01812744140625, 69.43606567382812, 3.232330322265625, 7.7570343017578125, 14.139190673828125, 50.48725128173828, 8.6402587890625, -12.328582763671875, 4.555736541748047, 0.0, 3.6131858825683594, 81.938232421875, -5.43060302734375, 8.851226806640625, 105.85031127929688, 62.177337646484375, 8.320068359375, -16.900665283203125, -0.2265777587890625, 20.66223907470703, -8.873565673828125, 10.679100036621094, 10.890403747558594, 44.431915283203125, 12.732063293457031, 57.5648193359375, 32.6412353515625, 0.0, -2.7444915771484375, 18.191558837890625, -2.7256622314453125, 16.028671264648438, 81.82699584960938, 5.431732177734375, 5.843414306640625, 92.1558837890625, 55.83154296875, -46.2803955078125, 9.3509521484375, 63.66790771484375, -2.9759521484375, 18.925464630126953, 10.704652786254883, 3.423828125, -1.8083229064941406, 3.91546630859375, -15.087905883789062, 62.70976257324219, 41.33612060546875, -38.67210388183594, 16.406982421875, 60.635833740234375, -5.51025390625, -9.276100158691406, 19.973312377929688, 25.583816528320312, 80.3829345703125, -3.2040863037109375, 44.08091735839844, 51.680328369140625, 21.667922973632812, 33.896728515625, 84.57366943359375, 37.35532760620117, -66.73924255371094, 31.307769775390625, 5.6170196533203125, 78.36016845703125, -27.121856689453125, 7.176910400390625, -17.75494384765625, 29.09295654296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000116.npy"}
{"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 19.361888885498047, "std": 33.944480895996094, "min": -58.629547119140625, "p10": -22.28042297363281, "median": 17.46963882446289, "p90": 66.59281311035156, "max": 110.96322631835938, "pos_frac": 0.703125, "sample": [5.540218353271484, 22.629356384277344, 15.828895568847656, 42.27006530761719, 17.088424682617188, 24.674835205078125, 4.603668212890625, -1.5191783905029297, 17.390838623046875, -24.170623779296875, 33.74787902832031, -25.405075073242188, 11.82806396484375, 39.35491943359375, 26.202312469482422, 35.746978759765625, 19.110275268554688, 21.175521850585938, 78.43319702148438, -1.613525390625, 56.142059326171875, -24.312774658203125, -0.0858001708984375, 0.6332035064697266, 16.31598663330078, 38.539794921875, -2.499847412109375, -49.690582275390625, 37.67840576171875, 76.17599487304688, 39.802757263183594, 71.646240234375, 77.71954345703125, 22.65708351135254, 17.548439025878906, -5.09588623046875, 50.44929504394531, -2.5716552734375, 1.4024810791015625, -19.078857421875, 85.44680786132812, -3.5858688354492188, 43.75604248046875, 46.107666015625, 0.0, 45.503326416015625, -12.658050537109375, -4.675445556640625, 73.706787109375, 110.96322631835938, 2.0630035400390625, 89.32791137695312, -47.63988494873047, 93.73373413085938, 42.99462890625, 29.536575317382812, -12.8304443359375, 42.16993713378906, 14.5316162109375, 24.019012451171875, 32.897308349609375, 42.790771484375, 20.550506591796875, 52.55052947998047, 10.885162353515625, 33.6256103515625, 45.666961669921875, 2.5950469970703125, -9.100814819335938, -13.38580322265625, 30.256027221679688, -50.30633544921875, -6.458597183227539, 8.923446655273438, 100.00979614257812, -58.629547119140625, 68.75128173828125, 42.514923095703125, 58.45550537109375, 30.919937133789062, 5.645961761474609, 18.54119873046875, 1.323028564453125, 78.96295166015625, 7.5181732177734375, -41.3497314453125, -39.968902587890625, -16.761199951171875, 24.658111572265625, -2.5292415618896484, 36.041107177734375, 20.0823974609375, 58.967437744140625, 74.48712921142578, 5.711635589599609, 23.596832275390625, -47.959014892578125, 30.540069580078125, -21.717208862304688, 30.20758056640625, -35.73381042480469, -5.9302978515625, 1.1532669067382812, -23.594589233398438, -40.321746826171875, -1.721282958984375, 55.10296630859375, 32.15489959716797, 48.481201171875, 0.3570899963378906, 5.638923645019531, 35.107208251953125, 55.071044921875, 10.066253662109375, 13.147651672363281, -13.200408935546875, 12.004779815673828, 6.722709655761719, 13.364913940429688, -19.546844482421875, -4.2071990966796875, 57.726318359375, -7.15625, 28.939254760742188, -1.7313079833984375, 43.210205078125, 65.66775512695312, 29.275634765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000117.npy"}
{"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 22.025394439697266, "std": 39.685726165771484, "min": -78.64389038085938, "p10": -25.420905303955077, "median": 16.40521240234375, "p90": 73.69514770507813, "max": 132.33889770507812, "pos_frac": 0.6875, "sample": [56.849456787109375, -10.157058715820312, 42.64305114746094, -16.3179931640625, 60.06263732910156, 50.421478271484375, -78.64389038085938, -3.3378067016601562, 56.225799560546875, 50.07769775390625, 19.768035888671875, 10.741729736328125, 37.689430236816406, 7.1684417724609375, -36.121337890625, 94.18222045898438, 52.235626220703125, 80.35750579833984, 75.31197357177734, -14.168182373046875, 26.47015380859375, 59.8519287109375, 89.7038345336914, 2.128662109375, -37.7930908203125, -26.380836486816406, 70.23141479492188, -9.911163330078125, -63.88526916503906, 73.356201171875, 34.544525146484375, 65.90199279785156, 86.54052734375, 36.04852294921875, 30.267215728759766, 48.60368347167969, 1.26129150390625, 11.087043762207031, 43.669334411621094, 34.22930145263672, 0.011383056640625, 49.781951904296875, 64.48890686035156, -1.7652244567871094, 6.6887359619140625, -18.173629760742188, 74.48602294921875, 49.8028564453125, 26.31622314453125, 2.1738510131835938, 29.433486938476562, -3.584716796875, 60.95068359375, 57.5013427734375, 52.572296142578125, 99.32623291015625, -25.009506225585938, 4.856353759765625, 10.522989273071289, -3.3768310546875, -2.1492233276367188, 0.0, -17.67877197265625, 16.00098419189453, -12.77020263671875, -16.68279266357422, -50.456329345703125, 16.80944061279297, 39.42320251464844, -30.673812866210938, -4.594966888427734, 61.5784912109375, -8.093093872070312, 43.436248779296875, -2.649444580078125, 66.4261474609375, 65.10781860351562, 12.86260986328125, -35.932769775390625, 108.48910522460938, -6.9830474853515625, 30.400665283203125, -23.566375732421875, 6.6952056884765625, 84.91409301757812, 34.525634765625, -64.11407470703125, -24.190139770507812, -13.35992431640625, -33.62445068359375, 108.41241455078125, -2.666412353515625, 64.26692199707031, 21.3311767578125, 6.12762451171875, 19.695213317871094, 17.918651580810547, -13.398719787597656, 43.9000244140625, 4.59539794921875, 20.613269805908203, 75.83993530273438, 4.2939453125, 44.520660400390625, 13.83905029296875, 20.81158447265625, 11.727027893066406, -24.97705078125, 44.31519317626953, 69.82991027832031, -13.369583129882812, -28.875, 52.65168762207031, 53.65757751464844, 9.228759765625, -27.239166259765625, -43.715911865234375, 13.411376953125, 40.378265380859375, 87.41122436523438, 10.680488586425781, 60.926239013671875, 1.3193912506103516, 4.2906036376953125, 0.0, 9.395355224609375, 18.696868896484375, 132.33889770507812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000118.npy"}
{"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 16.379058837890625, "std": 40.44919204711914, "min": -84.70684814453125, "p10": -33.9830997467041, "median": 12.52337646484375, "p90": 67.30905151367185, "max": 125.7664794921875, "pos_frac": 0.65625, "sample": [-62.26312255859375, -39.07173156738281, 40.75543212890625, 38.94354248046875, 3.7934799194335938, 21.130813598632812, -50.83540344238281, -28.479537963867188, -3.9384765625, -8.5994873046875, 46.44903564453125, -21.424972534179688, 32.356201171875, 59.3251953125, -22.1226806640625, 18.530059814453125, -4.4741668701171875, 24.947509765625, 26.157501220703125, 4.293609619140625, -12.136688232421875, 50.91984558105469, 16.21599006652832, -28.30724334716797, 12.747802734375, 15.879501342773438, 27.148147583007812, 11.40582275390625, 75.04939270019531, 9.693695068359375, 19.799474716186523, -41.16307067871094, -15.257339477539062, 43.951141357421875, 49.63214111328125, 1.8499603271484375, 39.10401916503906, 16.076339721679688, 10.932281494140625, 2.2640037536621094, 44.885955810546875, -46.098907470703125, -1.78656005859375, -5.042755126953125, 82.28707885742188, 2.481168746948242, 10.924728393554688, -52.14898681640625, -6.806884765625, -29.4041748046875, 106.78097534179688, 33.823028564453125, -31.030197143554688, 58.900421142578125, -18.393661499023438, -10.19677734375, 40.940216064453125, 6.977710723876953, 13.683395385742188, 56.7806396484375, 11.2950439453125, 6.0198974609375, 74.11380004882812, 95.4866943359375, 91.20339965820312, 26.55078125, 113.92642211914062, 20.574893951416016, 61.974456787109375, 8.693511962890625, 7.603786468505859, 11.393844604492188, 24.008987426757812, 0.0, 64.39273071289062, -20.6434326171875, 6.94256591796875, -77.15304565429688, 31.5882568359375, -43.852783203125, 41.37461471557617, 59.9334716796875, 10.5028076171875, -38.033203125, 37.220428466796875, -5.877227783203125, 60.5257568359375, -16.27939224243164, 19.775619506835938, 1.4297332763671875, 38.977874755859375, -48.8841552734375, -11.227081298828125, 125.7664794921875, -4.2301483154296875, 99.47982788085938, 39.6502685546875, -7.1544189453125, 29.009490966796875, -73.27792358398438, 96.16351318359375, 53.95098876953125, 34.8880615234375, 27.188064575195312, -8.368438720703125, 51.54473876953125, -32.24734115600586, 12.2989501953125, 3.2268218994140625, -43.66229248046875, -1.7664031982421875, -84.70684814453125, 89.55349731445312, 49.45465087890625, 83.22183227539062, -10.725982666015625, -0.4455718994140625, 39.699737548828125, 29.814300537109375, 30.350921630859375, 34.15203857421875, 12.775894165039062, 44.869873046875, -6.183677673339844, 82.79948425292969, 27.596176147460938, -7.5709228515625, -22.9896240234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000119.npy"}
{"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 17.26837158203125, "std": 39.49259567260742, "min": -82.89697265625, "p10": -26.49259033203125, "median": 17.65471649169922, "p90": 61.27396850585937, "max": 126.87725830078125, "pos_frac": 0.640625, "sample": [-23.027069091796875, 56.206756591796875, -8.905723571777344, 33.9241943359375, -10.380683898925781, 54.857635498046875, 62.0028076171875, -71.77900695800781, 17.741302490234375, 19.6025390625, -7.042049407958984, -26.060745239257812, 28.204666137695312, -7.578948974609375, -61.191253662109375, 4.60968017578125, 17.58062744140625, -7.708507537841797, -1.166259765625, 7.321952819824219, -21.693511962890625, 95.34326171875, 120.14459228515625, 37.555877685546875, 47.35272216796875, 40.53545379638672, 4.383331298828125, -7.28302001953125, -16.539989471435547, 47.739501953125, 103.418701171875, 32.8516845703125, 103.45745849609375, -15.140716552734375, -67.92520141601562, 16.56396484375, -0.3240509033203125, 34.709197998046875, 44.19415283203125, 22.025070190429688, 25.23614501953125, 7.088932037353516, 17.1240234375, -19.61273193359375, 0.0, -10.59075927734375, -36.806060791015625, -17.416091918945312, 2.4846420288085938, 30.7669677734375, -6.2487945556640625, -5.477657318115234, 101.87115478515625, 39.97552490234375, 36.11517333984375, 24.830535888671875, -8.312355041503906, 36.6260986328125, 27.301101684570312, 0.9252471923828125, -11.894981384277344, 0.0, 13.193084716796875, 46.96015930175781, 67.49102783203125, 35.144866943359375, 43.071746826171875, 90.2738037109375, -11.05865478515625, 54.95942687988281, 59.2003173828125, 17.728805541992188, -18.917274475097656, 34.43670654296875, 33.608497619628906, -23.134231567382812, 4.1033935546875, 42.49908447265625, 23.6202392578125, 26.40069580078125, 45.569122314453125, -45.323577880859375, -43.35235595703125, -32.22929382324219, -24.480941772460938, 75.322265625, -4.4026947021484375, 49.8890380859375, 83.499755859375, 55.70109558105469, 6.771522521972656, 46.508941650390625, -39.25440979003906, 41.408416748046875, 42.88084411621094, 6.9530181884765625, -14.977401733398438, 26.191696166992188, 29.905189514160156, 39.199745178222656, -45.8234748840332, 2.69219970703125, -7.454032897949219, 31.964996337890625, 41.682281494140625, 0.30246925354003906, 37.9432373046875, -15.938407897949219, 89.89892578125, -82.89697265625, 60.96160888671875, 38.107940673828125, 60.82574462890625, -16.60821533203125, -50.65289306640625, 16.764556884765625, 32.43407440185547, 68.45822143554688, 19.46246337890625, -38.3040771484375, -21.358654022216797, 55.649391174316406, -27.500228881835938, 126.87725830078125, 2.658275604248047, 12.543212890625, -23.15545654296875, 22.888885498046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000120.npy"}
{"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 13.762717247009277, "std": 37.270023345947266, "min": -75.72332763671875, "p10": -26.50553665161133, "median": 8.860939025878906, "p90": 61.217360687255855, "max": 145.93814086914062, "pos_frac": 0.59375, "sample": [-10.030418395996094, -1.092620849609375, -2.1929759979248047, -11.978530883789062, 19.754196166992188, -3.5619964599609375, -17.338897705078125, 36.75055694580078, 12.6240234375, 111.42160034179688, 18.279399871826172, -12.013351440429688, 58.00233459472656, 15.528127670288086, 42.291202545166016, 10.060134887695312, -6.68975830078125, -3.2786102294921875, -6.737972259521484, 84.95011901855469, 34.05076599121094, -23.304458618164062, 62.894866943359375, 16.38507080078125, -75.72332763671875, 0.0, 23.884963989257812, -19.849746704101562, -9.259071350097656, 35.42364501953125, 25.605945587158203, 34.89413833618164, -57.33746337890625, 52.622802734375, 22.327556610107422, -51.128662109375, -27.85814666748047, 42.927207946777344, 50.14361572265625, 25.036270141601562, -18.982666015625, -3.762786865234375, 48.811058044433594, -0.6479911804199219, -4.816261291503906, 6.64398193359375, 43.092742919921875, -15.126190185546875, 1.84295654296875, 0.0, 18.946887969970703, 5.57075309753418, 13.689079284667969, 43.2440185546875, 25.933486938476562, -17.44933319091797, -2.57720947265625, 37.147613525390625, 62.139984130859375, -9.63873291015625, -26.45294189453125, -34.91752624511719, -9.66668701171875, 113.93197631835938, 25.634857177734375, 48.509010314941406, 68.33267211914062, 20.7913818359375, 18.911956787109375, 4.7997589111328125, -36.094017028808594, 10.358154296875, 0.948486328125, 33.36477279663086, -0.8892326354980469, 6.84417724609375, 0.04405784606933594, 25.575912475585938, -3.606048583984375, 44.41112518310547, 25.914154052734375, -10.168132781982422, -4.421783447265625, 60.948463439941406, 89.90765380859375, 15.099678039550781, -65.60986328125, -5.35137939453125, -71.5672607421875, 77.53543090820312, 37.104339599609375, 76.90489196777344, 17.50165557861328, -0.0012359619140625, 89.04621887207031, -11.40570068359375, -28.5244140625, 30.025253295898438, 26.873573303222656, -45.752593994140625, 7.293813705444336, -6.766632080078125, 21.41747283935547, 12.58673095703125, 4.117431640625, 27.72149658203125, 1.65966796875, 44.74615478515625, -30.812057495117188, -26.628257751464844, 20.386444091796875, 145.93814086914062, 34.39894104003906, -11.20599365234375, 47.19970703125, -8.379806518554688, 61.84478759765625, -25.47785186767578, 41.984222412109375, 18.397003173828125, 11.719841003417969, -19.500228881835938, 109.02951049804688, -2.8731307983398438, 0.058864593505859375, -23.992141723632812, -46.332672119140625, 7.6617431640625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000121.npy"}
{"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 17.230777740478516, "std": 41.12491989135742, "min": -95.78143310546875, "p10": -29.090628051757808, "median": 14.818069458007812, "p90": 70.69165649414062, "max": 130.79776000976562, "pos_frac": 0.6640625, "sample": [14.270263671875, -26.922119140625, 40.931182861328125, -24.61749267578125, -13.35986328125, 51.86729431152344, 20.37109375, 9.685867309570312, -40.623382568359375, 32.730255126953125, 22.843185424804688, 0.0, 14.90802001953125, -61.64959716796875, 46.34760284423828, -95.78143310546875, 19.748306274414062, 64.41836547851562, -49.30867004394531, 10.976913452148438, -20.008636474609375, 57.379852294921875, -27.2879638671875, -61.207244873046875, 14.665283203125, 114.8201904296875, -26.969207763671875, 45.0543327331543, 121.75625610351562, -21.388633728027344, 13.832427978515625, -17.928115844726562, 11.864715576171875, -1.2061767578125, 48.650634765625, -17.35523223876953, 29.324859619140625, -56.0848388671875, 18.708709716796875, 31.693496704101562, -53.580963134765625, 37.70697021484375, 21.91522216796875, 68.1309814453125, -10.17296028137207, -74.26930236816406, 19.345008850097656, 15.856185913085938, 75.77157592773438, -21.346221923828125, -7.39697265625, 13.858100891113281, 28.748931884765625, -48.786834716796875, 17.38250732421875, 0.704254150390625, -0.5889606475830078, 8.4937744140625, -61.45550537109375, 24.2054443359375, 20.478302001953125, 58.72125244140625, 38.21199035644531, -7.080810546875, -5.488525390625, 50.530914306640625, 44.84410858154297, 58.46685791015625, 13.21380615234375, 70.0440673828125, 89.8128662109375, 41.37012481689453, 1.7089691162109375, 82.18719482421875, 100.31610107421875, 37.63646697998047, 43.68479919433594, 0.519927978515625, 74.99679565429688, 21.90704345703125, 26.48493194580078, -13.010208129882812, 41.06427001953125, -33.296844482421875, -7.54925537109375, -14.5753173828125, 17.86731719970703, 36.0257568359375, 41.50498962402344, 11.19439697265625, 31.428924560546875, 14.034515380859375, -1.7668323516845703, -52.60106658935547, 5.8531951904296875, -48.912200927734375, 18.784942626953125, 78.23758697509766, 28.60125732421875, -0.186248779296875, 14.473533630371094, -22.79046630859375, 50.141143798828125, 40.157257080078125, 2.978008270263672, 55.76530456542969, 43.091217041015625, 74.10516357421875, 2.64947509765625, -7.266796112060547, -8.441139221191406, -8.578109741210938, -0.76580810546875, 14.728118896484375, 130.79776000976562, -1.7872314453125, 55.78009033203125, -24.305007934570312, 55.415771484375, 102.2877197265625, 16.444915771484375, 63.5379638671875, 0.3803672790527344, 72.20269775390625, -0.06473922729492188, 25.829010009765625, 89.39134216308594, 4.443817138671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000122.npy"}
{"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 22.67681121826172, "std": 38.38883972167969, "min": -129.26318359375, "p10": -17.498777770996092, "median": 18.430274963378906, "p90": 71.76849822998047, "max": 129.03598022460938, "pos_frac": 0.71875, "sample": [49.72770690917969, -9.98907470703125, 80.65351867675781, -21.7132568359375, 26.469573974609375, 5.003448486328125, 2.7847747802734375, -3.4800567626953125, 29.82135009765625, 22.687103271484375, -11.132915496826172, 4.44781494140625, 7.6737518310546875, 9.322845458984375, 9.076202392578125, -13.5286865234375, -3.981109619140625, 16.945220947265625, 103.60906982421875, 71.97593688964844, -37.5054931640625, -2.848785400390625, 45.550933837890625, 15.899618148803711, 83.28237915039062, -13.665817260742188, 54.36260986328125, 52.505706787109375, 67.53363037109375, -0.868072509765625, 44.9456787109375, 32.66316223144531, 21.16741943359375, 14.715606689453125, 36.84269714355469, 56.29718017578125, 32.518280029296875, -129.26318359375, -4.094207763671875, 65.184326171875, 33.92364501953125, -3.47998046875, 3.527923583984375, 41.81976318359375, 44.096832275390625, 83.83868408203125, -6.267162322998047, 7.6056060791015625, 33.31219482421875, 92.82952880859375, -9.155277252197266, 32.320281982421875, 52.387786865234375, 92.53903198242188, 83.95159912109375, 20.253448486328125, -4.3985443115234375, 8.39654541015625, -7.423583984375, 70.14297485351562, 38.791259765625, 10.943681716918945, -100.00323486328125, 49.517547607421875, 27.0477294921875, -4.23028564453125, 61.748016357421875, 4.853546142578125, 71.67959594726562, 33.124610900878906, 47.33306884765625, -19.778564453125, 9.509696960449219, -29.658584594726562, 25.515640258789062, 56.728118896484375, 29.64996337890625, -27.03131103515625, -30.190185546875, 47.875030517578125, -6.8101348876953125, 28.884323120117188, 20.23849105834961, 6.3472900390625, 25.325851440429688, 43.82916259765625, -31.662994384765625, 19.915328979492188, 48.83477783203125, 69.29843139648438, 0.0, -15.717201232910156, 129.03598022460938, 28.849349975585938, -21.67584228515625, 0.0, 23.686798095703125, 20.90106201171875, 3.661958694458008, 59.06306457519531, -9.082542419433594, -17.084991455078125, 16.146621704101562, 112.88421630859375, 16.87554931640625, 12.373344421386719, 29.327484130859375, 55.25848388671875, 67.32994079589844, -18.464279174804688, 11.621124267578125, -14.773406982421875, 93.87054443359375, 12.969161987304688, -30.338134765625, 82.85101318359375, 32.8734130859375, 12.110931396484375, 43.947174072265625, 5.5355682373046875, -8.29425048828125, 0.622467041015625, 15.135055541992188, 9.230072021484375, 91.171875, -27.65191650390625, 15.162277221679688, 53.775970458984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000123.npy"}
{"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 18.97907257080078, "std": 36.4538688659668, "min": -93.2735595703125, "p10": -22.223840713500977, "median": 15.398307800292969, "p90": 66.45431671142578, "max": 108.48092651367188, "pos_frac": 0.734375, "sample": [-2.22784423828125, 40.180877685546875, -93.2735595703125, 17.36444091796875, 10.867460250854492, 3.1915283203125, 50.0390625, 52.34745788574219, 48.2757568359375, 87.60723876953125, 36.981170654296875, -26.26702880859375, 20.839019775390625, -14.323516845703125, 14.99017333984375, 27.691482543945312, 0.80096435546875, 49.77569580078125, -21.709423065185547, 34.6829833984375, 26.44415283203125, -23.014205932617188, 15.207962036132812, 51.87767028808594, 24.00799560546875, -26.9632568359375, -2.78594970703125, -15.736103057861328, 15.7716064453125, 16.748443603515625, 60.63261413574219, -50.556304931640625, -13.918128967285156, 43.881195068359375, 71.8692626953125, 78.66690063476562, 53.45292663574219, 74.08270263671875, 32.67364501953125, 62.4144287109375, 47.267333984375, 11.907203674316406, 92.08255004882812, 4.317230224609375, 21.94169807434082, -2.643512725830078, -6.7995452880859375, 7.38232421875, 39.925079345703125, -21.885112762451172, 3.286529541015625, -47.242713928222656, 28.816673278808594, 0.0, -5.133142471313477, 5.1243133544921875, 26.222198486328125, 2.16265869140625, 11.71148681640625, -26.845260620117188, 49.57929992675781, -48.4058837890625, -62.65887451171875, 56.43310546875, 10.168914794921875, 46.360198974609375, 13.2288818359375, 94.38015747070312, 36.4322509765625, 19.870986938476562, 21.539306640625, 28.417205810546875, -92.3350830078125, -52.42138671875, 38.413841247558594, 13.774436950683594, 7.6024322509765625, 30.35454559326172, 40.52130126953125, 36.77781677246094, 16.289794921875, 75.00729370117188, -3.5948028564453125, -21.216156005859375, -12.381378173828125, 10.476150512695312, -4.039306640625, 64.51126098632812, 108.48092651367188, 51.15053939819336, 14.786041259765625, 36.72981262207031, -11.578842163085938, 39.67597961425781, 3.7214202880859375, 11.563182830810547, 17.59918212890625, 85.02593994140625, 18.607202529907227, 66.71394348144531, 10.455757141113281, -5.368621826171875, 67.078369140625, 0.7506103515625, 10.234527587890625, 89.06304931640625, -20.641326904296875, -30.868011474609375, 8.485298156738281, 12.809814453125, 12.667648315429688, -13.868850708007812, 3.419830322265625, 10.777008056640625, 13.299468994140625, 52.098594665527344, 15.588653564453125, 88.06556701660156, -7.251615524291992, -26.99394989013672, 46.10137939453125, 51.35821533203125, 59.087188720703125, 66.34304809570312, 24.8475341796875, 23.627471923828125, 9.44140625, -21.03777313232422], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000124.npy"}
{"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 18.25252914428711, "std": 44.16659927368164, "min": -81.90152740478516, "p10": -30.488742065429683, "median": 15.165847778320312, "p90": 75.06896362304687, "max": 142.8408203125, "pos_frac": 0.6484375, "sample": [19.933731079101562, -25.52187728881836, -81.90152740478516, 52.62017822265625, 19.740510940551758, 35.75263977050781, 28.357391357421875, -14.475372314453125, 1.7122611999511719, -63.884002685546875, -21.73492431640625, 26.83110809326172, 57.94482421875, 29.26226806640625, -1.460906982421875, 72.64193725585938, -67.78369140625, 28.043441772460938, 62.548004150390625, 11.202644348144531, -13.699249267578125, 8.998619079589844, 31.1961669921875, -11.536262512207031, 5.302154541015625, 29.459007263183594, 23.263687133789062, -2.34710693359375, -15.814453125, 142.8408203125, 12.890838623046875, -37.722320556640625, 35.012298583984375, -11.49949836730957, -7.1751556396484375, 73.77674102783203, 10.73919677734375, 68.2933349609375, 25.558731079101562, -16.068336486816406, -38.66191482543945, -38.35418701171875, -20.194671630859375, -19.988441467285156, -25.288040161132812, 56.12371826171875, -16.3397216796875, 0.433990478515625, -54.65869140625, 24.384490966796875, 43.7039794921875, -61.33845520019531, 63.019866943359375, 142.43460083007812, 26.025985717773438, 20.168529510498047, 5.6083984375, 42.57171630859375, 53.736175537109375, 53.743621826171875, -20.89190673828125, 73.4117431640625, 4.8597259521484375, -2.5524120330810547, -43.332763671875, 23.4744873046875, -8.483810424804688, 10.17718505859375, 5.957061767578125, 19.718154907226562, 108.27850341796875, 32.936561584472656, 100.37342834472656, 95.56475830078125, -5.027902603149414, -71.9571533203125, 31.124404907226562, 37.686676025390625, 72.2568359375, 69.45508575439453, 30.33599853515625, 99.82745361328125, -14.914535522460938, 76.23605346679688, 1.29571533203125, 57.680816650390625, -25.49505615234375, -9.268356323242188, 42.86677551269531, 47.005157470703125, 56.28203582763672, 0.763397216796875, -21.211257934570312, 9.34661865234375, -34.327911376953125, 24.191864013671875, -13.002166748046875, 56.3956298828125, 29.932662963867188, -25.766746520996094, 1.29742431640625, 0.0, 1.2501983642578125, -42.697265625, 83.8875732421875, -28.8433837890625, 17.44085693359375, -3.0569839477539062, 9.937347412109375, 93.89010620117188, -13.277053833007812, -27.426437377929688, 36.992767333984375, 106.9365234375, 6.627307891845703, 54.771636962890625, 84.18975830078125, 23.31622314453125, 23.1903076171875, 28.40911865234375, 74.59567260742188, -76.79629516601562, -11.36163330078125, 99.12860107421875, 33.402099609375, 76.17330932617188, 2.5760269165039062, -17.8614501953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000125.npy"}
{"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 20.019376754760742, "std": 38.32088088989258, "min": -91.2781982421875, "p10": -32.33262176513672, "median": 18.469802856445312, "p90": 67.08594360351562, "max": 120.37945556640625, "pos_frac": 0.7109375, "sample": [44.8995361328125, 66.64187622070312, 24.89550018310547, 30.856796264648438, 41.370361328125, 65.98297119140625, 26.11003875732422, 32.914459228515625, 8.75349235534668, 13.496574401855469, -58.5753173828125, 33.975372314453125, -5.246143341064453, 23.935516357421875, 39.15289306640625, 35.27227783203125, 25.894622802734375, 25.5057373046875, -35.52679443359375, -18.8890380859375, 10.092376708984375, 108.34994506835938, -6.54248046875, -0.3105621337890625, 6.65570068359375, -35.70927429199219, -4.5006103515625, -37.69367980957031, -1.5848388671875, 8.345123291015625, 12.636444091796875, 24.051910400390625, 85.69837951660156, 1.417816162109375, 39.937042236328125, -34.3302001953125, -34.458526611328125, 17.324493408203125, -25.343902587890625, 1.4310760498046875, 51.777549743652344, 33.63453674316406, 43.13189697265625, -7.3013763427734375, 28.745147705078125, 63.69012451171875, 6.994281768798828, 12.248870849609375, -16.751205444335938, -24.163833618164062, 63.600250244140625, -2.5625152587890625, 42.27314758300781, 120.37945556640625, 25.418720245361328, 30.386581420898438, 47.430267333984375, 16.56153106689453, -35.757843017578125, 17.549901962280273, 15.006797790527344, 53.514404296875, 18.941650390625, 35.9296875, -36.844970703125, 53.909423828125, 2.2971878051757812, 17.620132446289062, 40.631072998046875, 94.1383056640625, 57.04949951171875, 15.374420166015625, 21.291473388671875, 50.679931640625, -4.536449432373047, 18.560699462890625, 16.0142822265625, -3.5601806640625, -2.001779556274414, -16.908966064453125, 47.9794921875, 68.12210083007812, -2.505250930786133, 68.69091796875, 86.99264526367188, 31.751922607421875, -47.183380126953125, 16.369842529296875, 42.128173828125, -59.74871826171875, 94.17904663085938, -32.61448669433594, 53.8251953125, 29.845626831054688, -81.53662109375, 0.305419921875, 7.107448577880859, 22.495941162109375, -5.005126953125, 2.12847900390625, 73.41937255859375, 26.178207397460938, 49.67738342285156, 66.07037353515625, -2.90838623046875, 78.42503356933594, -15.779022216796875, 101.7708740234375, 39.976104736328125, 65.32672119140625, -91.2781982421875, 0.26641845703125, 64.86156463623047, 93.12127685546875, 38.34709167480469, 13.113311767578125, -5.978599548339844, 5.0642242431640625, 21.381591796875, -32.211822509765625, 39.642333984375, -29.544708251953125, -1.76800537109375, 18.37890625, 33.47700500488281, 75.8472900390625, 3.495361328125, -30.499298095703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000126.npy"}
{"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 23.129392623901367, "std": 40.802425384521484, "min": -106.77813720703125, "p10": -20.00749053955078, "median": 17.14564609527588, "p90": 80.70072708129882, "max": 149.5782470703125, "pos_frac": 0.71875, "sample": [-9.629423141479492, -9.589950561523438, -13.81494140625, 33.1253662109375, 20.112060546875, 18.92474365234375, 34.357086181640625, 7.941871643066406, -19.6025390625, -106.77813720703125, 32.32098388671875, -16.413162231445312, 59.288543701171875, 46.54387283325195, -1.1943206787109375, -52.845245361328125, 15.6925048828125, 101.21583557128906, -13.730712890625, 12.500732421875, 24.447845458984375, 12.015289306640625, 44.373016357421875, 30.536834716796875, 46.63240051269531, 83.128173828125, 53.3607177734375, 49.56646728515625, -5.253486633300781, 35.776885986328125, 47.874420166015625, 3.097921371459961, 117.33642578125, 7.2138519287109375, 8.8482666015625, 85.8538818359375, -25.286712646484375, -16.756744384765625, -6.547416687011719, 5.974700927734375, 92.2891845703125, 0.30948638916015625, -22.92095947265625, 1.431488037109375, -82.94947814941406, -31.955780029296875, 30.069305419921875, 64.40274047851562, 42.58912658691406, -20.86297607421875, 27.552001953125, 32.54437255859375, 22.98583984375, -19.10149383544922, 37.71685791015625, 55.514801025390625, 65.3095703125, 24.500869750976562, -6.91455078125, 19.448135375976562, 7.57806396484375, 5.167816162109375, 75.24327087402344, 26.22894287109375, 23.310958862304688, 70.59832763671875, -22.49261474609375, 9.037734985351562, 84.6763916015625, 70.58056640625, 26.065231323242188, -1.4073104858398438, -7.147331237792969, -4.874839782714844, 12.463226318359375, -73.24075317382812, 4.99884033203125, -23.117443084716797, 26.724990844726562, -7.422088623046875, 11.854888916015625, 26.142425537109375, 67.69293212890625, 10.222171783447266, 79.36227416992188, 61.850616455078125, 15.332199096679688, 149.5782470703125, 82.70541381835938, 60.435302734375, 90.944091796875, -19.640853881835938, 48.1102294921875, -7.3162841796875, 29.045230865478516, -11.051040649414062, -15.247865676879883, 16.6177978515625, 17.53209114074707, 16.759201049804688, 15.608383178710938, 2.0181427001953125, -11.309982299804688, 16.570236206054688, 79.8415756225586, 5.793479919433594, 124.394287109375, 97.00759887695312, -0.951080322265625, -18.069610595703125, 2.91925048828125, 20.085296630859375, 47.319488525390625, 83.86529541015625, -27.357940673828125, 16.472354888916016, 42.85565185546875, 53.45159912109375, -21.592540740966797, 0.443115234375, 60.758758544921875, 43.3519287109375, 9.024658203125, 50.90519714355469, 38.31060791015625, -27.16455078125, 51.01817321777344, 98.51528930664062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000127.npy"}
{"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 26.177688598632812, "std": 42.068302154541016, "min": -83.134033203125, "p10": -26.255059814453123, "median": 26.390594482421875, "p90": 83.17276763916016, "max": 137.89349365234375, "pos_frac": 0.734375, "sample": [42.863555908203125, 70.16950988769531, 65.02311706542969, 34.1121826171875, -68.10107421875, -53.247467041015625, -25.62200927734375, 36.80616760253906, 75.55020141601562, 38.7437744140625, 51.5216064453125, -5.1949462890625, 79.27694702148438, 13.478561401367188, 125.629150390625, 6.332469940185547, 3.41949462890625, 51.884185791015625, 51.78089904785156, 59.64247131347656, -16.555007934570312, -1.9395065307617188, 82.58259582519531, 101.93954467773438, 13.437713623046875, 30.149139404296875, -17.258148193359375, 26.5338134765625, 46.03131103515625, 33.74627685546875, 24.987808227539062, 14.580987930297852, -34.518096923828125, 87.9931640625, -10.92222785949707, 75.28538513183594, -31.752182006835938, 13.097808837890625, -16.064468383789062, 10.140151977539062, 48.20897674560547, -48.14306640625, -21.771751403808594, 21.073028564453125, 84.8004379272461, -31.6331787109375, 12.150920867919922, 37.789703369140625, 43.96351623535156, 74.87382507324219, 32.36351013183594, 55.467529296875, 8.593399047851562, 119.346435546875, -2.842529296875, 3.6768035888671875, 4.2711181640625, 39.86072540283203, 49.608367919921875, 69.51005554199219, 49.06739807128906, 23.07958984375, -1.146759033203125, 37.990570068359375, 2.983661651611328, 15.340103149414062, 78.11962890625, 52.989044189453125, 26.92333984375, 7.487152099609375, 15.3765869140625, 62.35551452636719, 58.226898193359375, 31.5797119140625, 79.6951904296875, 42.48974609375, 28.694007873535156, -30.297592163085938, 1.842529296875, 12.678749084472656, 84.54983520507812, 39.2291259765625, -20.16290283203125, 44.43501281738281, 55.843505859375, -83.134033203125, 40.172149658203125, -36.71586608886719, 31.874534606933594, 92.017333984375, -27.732177734375, -35.270751953125, 34.70921325683594, 18.409027099609375, -16.14080810546875, -1.6300296783447266, -19.47491455078125, -13.032257080078125, 17.61181640625, 13.593544006347656, 84.600830078125, -24.397750854492188, 10.604248046875, 96.222900390625, 59.673866271972656, 74.04730224609375, -15.49087905883789, 28.863067626953125, 97.38009643554688, 4.423046112060547, 11.0728759765625, 0.0, -61.72407531738281, -5.398834228515625, 111.87799072265625, -21.273651123046875, -43.94279479980469, 17.998779296875, 38.733978271484375, 47.52069091796875, 4.189208984375, 26.24737548828125, 38.60316467285156, 86.1229248046875, 137.89349365234375, 38.95835876464844, 12.188720703125, -5.6100311279296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000128.npy"}
{"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 22.774417877197266, "std": 41.923988342285156, "min": -88.03520202636719, "p10": -24.63309631347656, "median": 16.612762451171875, "p90": 71.62014312744141, "max": 196.56280517578125, "pos_frac": 0.7265625, "sample": [-20.080474853515625, 74.4962158203125, -40.404022216796875, 15.8624267578125, -6.533836364746094, 71.0966796875, 43.93855285644531, 87.1146240234375, 70.33499145507812, 21.417343139648438, 50.0950927734375, 3.979045867919922, 26.5279541015625, 6.732421875, -7.338436126708984, -2.2952117919921875, 52.15858459472656, 75.78453063964844, 9.73370361328125, -26.35913848876953, 37.86841583251953, 58.96727752685547, -9.076129913330078, 5.50048828125, 60.16204833984375, -88.03520202636719, 72.47120666503906, 2.4271507263183594, 196.56280517578125, 86.06689453125, 32.80255126953125, 66.7626953125, 109.76036071777344, 27.1483154296875, 26.707077026367188, -67.6357421875, 66.18923950195312, 30.4742431640625, 127.230224609375, 31.242721557617188, 10.687934875488281, 14.926177978515625, 46.92120361328125, 52.07470703125, -3.72412109375, 78.72177124023438, -23.9132080078125, 8.189228057861328, 5.065040588378906, 25.325218200683594, 57.07749938964844, 34.67869567871094, 26.184005737304688, 67.58514404296875, -10.7694091796875, -1.5331001281738281, 3.6375885009765625, 48.00860595703125, 12.70611572265625, 47.27965545654297, 7.201091766357422, 13.94097900390625, 53.76177978515625, 63.72980499267578, 100.994384765625, -2.45751953125, 15.397705078125, 35.0103759765625, 32.7139892578125, -44.49951171875, 6.260711669921875, 14.776908874511719, 5.8691864013671875, -6.438030242919922, 17.36309814453125, -58.40240478515625, 59.31243133544922, 35.977317810058594, -3.43023681640625, 92.59506225585938, 57.59901428222656, 12.048614501953125, 10.7276611328125, 23.616477966308594, 6.768861770629883, -72.43902587890625, 71.07807922363281, -27.58453369140625, 24.606353759765625, 108.06668090820312, -26.312835693359375, -5.29620361328125, -5.627899169921875, 2.027191162109375, -9.605125427246094, 58.2130126953125, -5.89959716796875, -41.7503662109375, -36.541900634765625, 20.24847412109375, 88.10676574707031, 5.7115936279296875, -0.28106689453125, -10.560478210449219, -70.42073822021484, -38.083526611328125, 12.120254516601562, -7.359283447265625, 21.97614288330078, 31.26300048828125, 29.884963989257812, 19.23367691040039, 27.3094482421875, 15.2169189453125, -1.2790184020996094, 71.25540161132812, 3.7688522338867188, 66.31698608398438, -22.30438232421875, 25.963668823242188, 22.40802001953125, 8.77102279663086, 8.103958129882812, 35.93061828613281, 6.74151611328125, -9.187721252441406, 43.44635009765625, 48.46593475341797], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000129.npy"}
{"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 21.70458984375, "std": 40.099815368652344, "min": -68.08056640625, "p10": -31.860794067382812, "median": 22.4819917678833, "p90": 74.14921875, "max": 146.4219970703125, "pos_frac": 0.6953125, "sample": [39.8197021484375, -2.047027587890625, 26.200592041015625, -25.990753173828125, 24.792327880859375, 37.14369201660156, 29.95733642578125, 60.80126953125, 65.8239974975586, -32.15997314453125, -14.498268127441406, 82.91156005859375, 50.66759490966797, 50.779205322265625, -10.681732177734375, 15.538543701171875, 2.270843505859375, -39.04386901855469, 29.272743225097656, 12.96942138671875, 10.27789306640625, 67.77273559570312, -20.289325714111328, 43.586181640625, -5.139875411987305, 37.92582702636719, -3.32806396484375, 78.4295654296875, 82.68179321289062, 72.80392456054688, -56.9573974609375, 22.602874755859375, 0.0, -16.694976806640625, 44.259735107421875, 24.857452392578125, 51.89532470703125, -53.55767822265625, -13.633617401123047, 55.05072021484375, 62.36209487915039, 10.952438354492188, 73.82684326171875, -10.815948486328125, -56.72735595703125, -68.08056640625, -0.7659912109375, 64.7364501953125, -5.4027099609375, -2.7822494506835938, 52.33610534667969, -38.93916320800781, 13.330802917480469, 58.185829162597656, -46.25201416015625, -19.58831787109375, 40.8803825378418, 11.541961669921875, 74.90142822265625, -34.867431640625, -22.028335571289062, 30.275440216064453, 40.369903564453125, 47.1416015625, 19.65643310546875, 106.20002746582031, 32.6170654296875, -12.58953857421875, -31.732574462890625, 5.6290435791015625, 63.15142822265625, 5.455284118652344, 79.80795288085938, -0.06890869140625, 3.460205078125, 42.802825927734375, -5.454658508300781, 35.110260009765625, -0.7793121337890625, 29.24842071533203, 68.51396179199219, 33.66607666015625, 105.67831420898438, 20.0072021484375, 23.044830322265625, 7.3059539794921875, 23.212158203125, 6.7908935546875, 89.53952026367188, 0.3983573913574219, 13.7763671875, 36.40814208984375, 4.157379150390625, 25.51367950439453, 43.751373291015625, -62.17256164550781, -33.04364013671875, 5.64373779296875, 1.242034912109375, 74.9569091796875, 146.4219970703125, 62.501220703125, 101.89871215820312, 12.709136962890625, -66.93548583984375, 51.71258544921875, 20.91345977783203, 25.88677978515625, 26.830322265625, -6.783538818359375, 61.3248291015625, -9.984376907348633, -21.491302490234375, -0.5096282958984375, 28.818527221679688, 48.58094024658203, 109.93682861328125, 39.0208740234375, 0.4301280975341797, 38.473052978515625, 22.31414031982422, -11.41302490234375, 3.8502960205078125, 28.227073669433594, 39.78271484375, -37.031402587890625, 22.361108779907227, 77.77517700195312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000130.npy"}
{"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 18.630504608154297, "std": 41.1131706237793, "min": -81.6688232421875, "p10": -37.964141845703125, "median": 19.833168029785156, "p90": 71.3203353881836, "max": 116.0615234375, "pos_frac": 0.671875, "sample": [31.505203247070312, 10.480710983276367, 57.93914794921875, -34.26751708984375, 34.15107727050781, 41.97547149658203, 16.42922019958496, 43.129913330078125, -40.102996826171875, 114.56396484375, 35.30474853515625, 49.9046630859375, 74.17291259765625, 51.946044921875, -44.21990966796875, -14.750823974609375, 6.22943115234375, -63.1695556640625, 24.966697692871094, 72.48260498046875, 22.809661865234375, 6.734649658203125, -8.273681640625, -1.944967269897461, 55.34510040283203, -4.567626953125, -5.34130859375, 56.94049072265625, 30.98735809326172, 51.338226318359375, 39.806556701660156, 92.71551513671875, 21.487075805664062, 1.122161865234375, 15.130889892578125, 9.806831359863281, -55.75697326660156, 70.77197265625, -1.48236083984375, 93.43988037109375, 6.0521240234375, 31.604286193847656, 12.291431427001953, -29.522430419921875, 53.0753173828125, 15.249290466308594, 116.0615234375, -37.90362548828125, 32.208343505859375, 22.5257568359375, -4.667442321777344, 1.800537109375, -29.345794677734375, 28.7484130859375, 73.17628479003906, 77.79898071289062, -34.3790283203125, -13.23114013671875, 52.072723388671875, 84.52943420410156, 3.7981109619140625, -23.14923095703125, -18.328125, 40.38899230957031, 27.74249267578125, -11.379276275634766, 21.388961791992188, 50.96697998046875, 102.12728881835938, 24.69854736328125, -81.6688232421875, -39.86161804199219, 54.0389404296875, 1.8495330810546875, -1.5260543823242188, 21.386398315429688, -80.28797912597656, 108.10733032226562, -24.977432250976562, 50.83839416503906, 40.66949462890625, -41.025238037109375, 0.04395294189453125, 38.040435791015625, 43.4801025390625, 51.67193603515625, -38.1053466796875, 41.432525634765625, 83.20779418945312, -27.364105224609375, 12.779953002929688, -11.02508544921875, 30.832000732421875, -12.470264434814453, 37.350677490234375, 15.983802795410156, -0.9465389251708984, 18.029449462890625, 5.572380065917969, 39.650672912597656, 18.279937744140625, -35.159576416015625, 58.6243896484375, 7.746360778808594, 70.82221984863281, 61.00901794433594, -3.72705078125, -42.144287109375, 93.3612060546875, 15.25, 53.87322998046875, 66.72152709960938, 31.080078125, 31.778289794921875, -66.51654052734375, -5.45965576171875, 37.37092590332031, -30.59259033203125, 49.559513092041016, 11.021484375, 40.87908935546875, 29.58165740966797, 38.03778076171875, -13.515777587890625, -16.08526611328125, -47.35380554199219, -38.277587890625, -3.307220458984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000131.npy"}
{"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 25.601276397705078, "std": 44.787757873535156, "min": -78.39326477050781, "p10": -30.30509033203125, "median": 22.351089477539062, "p90": 77.25551681518554, "max": 151.6324462890625, "pos_frac": 0.734375, "sample": [-6.462646484375, 21.5926513671875, 62.46417236328125, 62.20726776123047, 8.981216430664062, -37.042724609375, 56.33130645751953, 33.65106201171875, 2.856201171875, 75.55447387695312, 22.270599365234375, -49.41716003417969, -31.676422119140625, 54.499755859375, 50.09873962402344, -20.95050048828125, 2.20318603515625, 51.98045349121094, 55.832489013671875, 132.85281372070312, 71.44900512695312, 51.17181396484375, 68.87716674804688, 87.65924072265625, 12.577262878417969, 2.4051780700683594, 3.2043418884277344, -59.989410400390625, 36.933807373046875, 63.11820983886719, -1.128814697265625, 127.19802856445312, 30.6234130859375, 12.774993896484375, -36.02777099609375, -25.8839111328125, 27.527095794677734, 31.59454345703125, -15.918228149414062, 4.94403076171875, 128.64056396484375, 58.176483154296875, -35.53517150878906, -63.663330078125, -16.47119140625, -2.4370880126953125, 22.43157958984375, -0.56353759765625, 2.768829345703125, 15.577522277832031, 37.82344055175781, -5.579669952392578, 46.22002410888672, 73.32846069335938, 24.729965209960938, 19.364776611328125, -3.050567626953125, 49.381866455078125, 137.36572265625, 0.80126953125, 55.50274658203125, 37.543426513671875, 37.78570556640625, -36.74810791015625, 46.09455871582031, 12.253570556640625, 3.2438812255859375, 18.3404541015625, -19.5234375, 28.76702880859375, 36.28570556640625, 10.329742431640625, 22.906082153320312, 98.72601318359375, -0.57318115234375, 15.034378051757812, 0.0, 80.7694091796875, 13.405471801757812, -0.9774169921875, 4.3207550048828125, 69.9730224609375, 77.68424224853516, -22.13751220703125, -74.79660034179688, 81.33016967773438, 15.410140991210938, 95.03216552734375, -17.08148193359375, 24.352386474609375, 43.31684875488281, 17.713302612304688, 54.258087158203125, 39.494476318359375, 70.00892639160156, 151.6324462890625, -21.097030639648438, 62.054962158203125, 0.70458984375, 5.55517578125, 46.399658203125, 60.481414794921875, 5.455963134765625, -0.2707080841064453, 43.49322509765625, 98.05111694335938, 14.897003173828125, 22.484283447265625, 0.0, -29.717376708984375, 62.7308349609375, 58.7587890625, -27.878860473632812, 31.21123504638672, -67.94554138183594, 24.9158935546875, 2.72369384765625, 77.07177734375, -51.91242980957031, 37.819091796875, 50.56859588623047, 116.29876708984375, -78.39326477050781, 10.962913513183594, -42.354095458984375, 66.20187377929688, 6.833099365234375, 70.93441772460938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000132.npy"}
{"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 23.17754554748535, "std": 40.635292053222656, "min": -96.64080810546875, "p10": -25.88152313232422, "median": 20.395381927490234, "p90": 75.94194412231445, "max": 117.50006103515625, "pos_frac": 0.6953125, "sample": [22.969970703125, -18.84963607788086, -70.219482421875, 0.0, 16.781753540039062, 54.16679382324219, -0.09301185607910156, 14.196296691894531, -1.4774436950683594, 49.47639465332031, 28.32257080078125, -3.673858642578125, -6.037384033203125, 46.008338928222656, 51.525360107421875, 117.50006103515625, -25.780288696289062, 62.6182861328125, 9.733940124511719, -8.584808349609375, 54.251708984375, 96.944091796875, 107.13821411132812, -7.7193145751953125, 31.5850830078125, 64.63739013671875, 35.013145446777344, 70.18988037109375, 0.0, 10.81695556640625, 74.49298095703125, 30.020538330078125, -13.523506164550781, 6.08013916015625, 67.20022583007812, 94.12054443359375, 18.87530517578125, 33.64117431640625, -58.62213134765625, 37.7086181640625, 51.851318359375, 27.84870147705078, -11.704376220703125, -12.795196533203125, -0.3034515380859375, -7.509490966796875, -77.89503479003906, 27.227859497070312, -1.8491363525390625, 18.3963623046875, 32.6922607421875, 55.0118408203125, 0.12799072265625, 21.003211975097656, 95.37063598632812, 9.953338623046875, -0.65850830078125, 91.56939697265625, -2.7303619384765625, 77.79660034179688, -27.06158447265625, 20.507408142089844, 20.14776611328125, -10.757080078125, 49.63478088378906, 4.638336181640625, 47.73731994628906, 70.8148193359375, 69.3343505859375, 55.691162109375, -26.11773681640625, -6.311492919921875, 44.058349609375, 10.983123779296875, 102.73434448242188, 3.6228809356689453, -10.682846069335938, -36.311798095703125, 9.268455505371094, 106.82080078125, 2.0218429565429688, -35.78001403808594, 31.107131958007812, -57.99920654296875, 55.0789794921875, 4.6383819580078125, 0.8150634765625, 56.493255615234375, 22.316009521484375, 71.12736511230469, -2.735565185546875, 80.69540405273438, 20.96533203125, 28.875091552734375, -2.7701797485351562, 3.41717529296875, 14.355926513671875, 47.63458251953125, 75.44235229492188, 58.32823181152344, 28.489059448242188, 86.89430236816406, 26.914688110351562, 58.79212188720703, -51.727569580078125, 16.80364227294922, 34.29412841796875, 54.57122802734375, -32.896888732910156, 81.77645874023438, 20.283355712890625, 27.363906860351562, 13.75518798828125, -18.3173828125, -37.84822082519531, 68.34016418457031, -96.64080810546875, 18.231159210205078, -7.639854431152344, 33.16424560546875, 0.341949462890625, 22.943206787109375, 59.95756530761719, -29.239913940429688, -11.691009521484375, 3.1667938232421875, 61.91925048828125, 77.10765838623047], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000133.npy"}
{"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 18.206329345703125, "std": 38.248172760009766, "min": -112.4459228515625, "p10": -25.856346130371094, "median": 13.71500015258789, "p90": 67.28502502441405, "max": 122.89578247070312, "pos_frac": 0.71875, "sample": [76.18344116210938, 59.125343322753906, -38.77861022949219, -16.61602783203125, -40.044586181640625, -0.83203125, 26.43548583984375, 6.693572998046875, 4.70855712890625, 35.51007080078125, -112.4459228515625, -15.729995727539062, 2.19647216796875, 19.36553955078125, 68.51248168945312, 1.6967658996582031, 4.947776794433594, 51.90638732910156, 76.65240478515625, 5.048685073852539, 25.33001708984375, 70.6649169921875, 0.0, 64.92669677734375, 60.00537109375, 45.348236083984375, 36.8834228515625, -6.087043762207031, 35.53875732421875, 82.90980529785156, -36.1834831237793, 2.5302581787109375, 37.777191162109375, 9.210369110107422, 30.03601837158203, -12.28839111328125, 47.11859130859375, 45.69615173339844, 61.19854736328125, 22.575439453125, -45.30296325683594, 39.10393524169922, 17.008102416992188, 37.20880126953125, 122.89578247070312, 36.484222412109375, 6.431671142578125, -24.41363525390625, 97.53521728515625, 4.7247314453125, 7.35931396484375, 64.44892120361328, 49.05943298339844, -23.637939453125, 8.899513244628906, -0.31429290771484375, 70.06878662109375, 65.38697814941406, 81.904052734375, 39.42256164550781, -50.13432312011719, -12.500595092773438, 15.42889404296875, -23.7005615234375, -63.64971923828125, 108.01495361328125, -1.525726318359375, 0.50701904296875, 21.329376220703125, 38.30553436279297, 0.10190391540527344, 24.8775634765625, -39.0968017578125, -5.791105270385742, 13.34918212890625, 83.50099182128906, 6.5488128662109375, -6.1761474609375, 6.4275970458984375, -49.8302001953125, 53.408935546875, 39.36387634277344, 39.168128967285156, -1.365966796875, 4.49169921875, 4.2310943603515625, 21.644271850585938, 35.397438049316406, 72.2593994140625, 6.106418609619141, -7.353813171386719, 15.998687744140625, 14.080818176269531, 33.346038818359375, -39.859375, 48.42693328857422, 35.1024169921875, 26.12188720703125, 6.158744812011719, -50.069732666015625, 27.247591018676758, 31.31561279296875, -26.876556396484375, -80.05987548828125, -4.39874267578125, 29.32794189453125, -12.14959716796875, -22.48345947265625, 43.003875732421875, 4.40399169921875, 2.962728500366211, 59.78715515136719, 24.075517654418945, 66.75897216796875, 54.20440673828125, 75.96701049804688, 50.155189514160156, 2.9250640869140625, 7.55755615234375, -9.708633422851562, 10.5960693359375, 9.822235107421875, -25.419113159179688, -8.75970458984375, 51.34674072265625, 49.65827941894531, 4.505836486816406, 0.0], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000134.npy"}
{"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 21.48922348022461, "std": 40.46170425415039, "min": -129.38938903808594, "p10": -19.04376525878906, "median": 17.119932174682617, "p90": 78.59798126220703, "max": 116.1507568359375, "pos_frac": 0.6875, "sample": [-3.459024429321289, -13.273239135742188, 3.1451072692871094, 56.62986755371094, -18.11956787109375, 108.01327514648438, 2.928966522216797, 7.74462890625, 43.87867736816406, 11.252683639526367, 93.07052612304688, 28.88427734375, -22.507675170898438, -21.809326171875, 25.2315673828125, 16.3289794921875, 37.25248718261719, 14.09765625, 28.118560791015625, 8.115707397460938, 27.897689819335938, 17.530452728271484, -5.301506042480469, 6.6181640625, -33.56842041015625, -0.46089744567871094, 45.03741455078125, 80.70591735839844, 50.27107238769531, 88.14556884765625, 43.43165588378906, 6.995452880859375, 58.24560546875, 21.609573364257812, -0.9709930419921875, 70.74078369140625, -1.0290985107421875, 7.49053955078125, 29.290496826171875, -12.602447509765625, 8.550384521484375, 25.857452392578125, -9.966140747070312, 52.97186279296875, 107.71542358398438, 35.33976745605469, 82.19784545898438, 31.226852416992188, 11.441650390625, -21.26171875, 2.975921630859375, 30.348114013671875, 25.358848571777344, -8.2327880859375, -11.469070434570312, -21.200225830078125, 89.22824096679688, 10.5374755859375, 39.669403076171875, 48.97039794921875, 71.4889144897461, 34.78741455078125, 32.218971252441406, -0.45758056640625, -3.5224151611328125, 33.308807373046875, 21.65643310546875, 34.0716552734375, 52.449066162109375, 13.685615539550781, 62.781288146972656, 9.26763916015625, 58.82373046875, 16.70941162109375, 27.1685791015625, 52.119354248046875, 28.554656982421875, -9.986602783203125, -3.2577972412109375, 54.350250244140625, 10.084014892578125, -25.398773193359375, -7.831081390380859, -3.63763427734375, -32.893218994140625, 67.11813354492188, 1.34906005859375, 95.6416015625, 36.811676025390625, 77.11956787109375, 30.480926513671875, 90.79541015625, -9.0867919921875, -119.86251831054688, 116.1507568359375, -13.530899047851562, 34.43060302734375, 45.35181427001953, -13.125762939453125, 55.487823486328125, 39.34407043457031, 15.737548828125, 39.162925720214844, 98.50439453125, 1.5494384765625, -129.38938903808594, -18.116825103759766, -2.3904457092285156, 77.694580078125, 54.114044189453125, 22.89801025390625, -12.822021484375, 8.282623291015625, -5.077178955078125, 0.0, 31.503768920898438, -42.353424072265625, -81.24165344238281, -3.7391128540039062, 51.92015075683594, -9.264862060546875, 89.53924560546875, 14.312774658203125, 80.73309326171875, 36.4549560546875, -35.61347961425781, -32.40753173828125, 3.7520751953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000135.npy"}
{"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 23.35352325439453, "std": 44.56053161621094, "min": -95.14788818359375, "p10": -21.972815704345702, "median": 21.947128295898438, "p90": 73.1973907470703, "max": 175.75128173828125, "pos_frac": 0.734375, "sample": [39.890655517578125, 40.72853088378906, 23.045120239257812, 16.33594512939453, 89.21501922607422, 64.13311767578125, 80.19160461425781, 20.94427490234375, 45.68778991699219, 69.69989013671875, 4.471405029296875, 3.199953079223633, 65.39016723632812, 21.29742431640625, 58.34064865112305, 71.30953979492188, 46.903106689453125, 46.456329345703125, 3.94866943359375, 4.678413391113281, -95.14788818359375, 12.741744995117188, 0.61480712890625, 9.34765625, -25.165985107421875, -1.8900909423828125, -18.713104248046875, -66.9478759765625, 27.080215454101562, 24.028076171875, 8.467266082763672, -5.6963348388671875, -19.7801513671875, 3.64910888671875, 55.015968322753906, -17.9605712890625, -17.68792724609375, 38.60246276855469, 22.596832275390625, -2.877899169921875, 8.732162475585938, -18.14959716796875, 88.974853515625, -64.63763427734375, 30.60546875, -18.667144775390625, 10.84075927734375, 4.142631530761719, 28.34124755859375, 6.1077880859375, 39.4384765625, -77.528564453125, 125.33392333984375, 48.217926025390625, 119.73822021484375, 126.82339477539062, -21.8857421875, 6.220977783203125, 1.9501953125, 53.98640441894531, 109.1123046875, -1.84405517578125, 45.97930908203125, 25.052032470703125, 26.604148864746094, -22.175987243652344, -38.638214111328125, 5.6163330078125, 63.274017333984375, -25.181610107421875, 10.01019287109375, -16.0389404296875, 12.990278244018555, 59.87261962890625, 56.4910888671875, 118.4180908203125, -7.228097915649414, 23.363433837890625, 51.32167434692383, -22.337158203125, 34.99066162109375, -52.374755859375, 38.94171142578125, 103.30801391601562, 60.6539306640625, 36.17918395996094, 15.923164367675781, -0.8217411041259766, 4.323230743408203, 69.92581176757812, 7.366315841674805, 24.2991943359375, -20.113311767578125, 25.49840545654297, -1.171234130859375, 9.632957458496094, 107.41397094726562, 37.71941375732422, 41.015380859375, 26.469146728515625, -1.5603179931640625, 29.324237823486328, 36.95947265625, 6.651126861572266, 110.04290771484375, -86.7109375, -18.867599487304688, 62.05924987792969, 72.61669921875, 64.543701171875, 74.55233764648438, 34.45092010498047, -21.10137939453125, -36.46282958984375, 20.231460571289062, 22.844482421875, 175.75128173828125, 5.210124969482422, -18.91326904296875, 29.6182861328125, 31.0335693359375, 61.86647033691406, 34.917503356933594, -5.0293426513671875, 13.076103210449219, -60.7159423828125, 20.068359375, 44.22357940673828], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000136.npy"}
{"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 20.617076873779297, "std": 41.689395904541016, "min": -69.612548828125, "p10": -29.394261169433587, "median": 13.76611328125, "p90": 78.5652572631836, "max": 126.497802734375, "pos_frac": 0.7109375, "sample": [-69.612548828125, 5.48583984375, 20.186767578125, 4.662261962890625, 78.31001281738281, 37.16339874267578, 122.74850463867188, -15.0023193359375, 7.0814971923828125, 14.215072631835938, 38.36976623535156, 26.359085083007812, 23.69146728515625, 126.497802734375, 3.254180908203125, 4.332672119140625, -2.0792236328125, 80.2806396484375, -10.8082275390625, -34.24127197265625, 54.050743103027344, 18.316146850585938, 41.496337890625, -3.821533203125, -61.5185546875, 11.843772888183594, -2.230621337890625, -27.316970825195312, 71.41165161132812, -8.620529174804688, 106.41888427734375, 15.173507690429688, 21.380905151367188, 103.32394409179688, -7.314453125, -16.7564697265625, 33.27116394042969, 10.360458374023438, 0.612640380859375, -19.215492248535156, 3.603973388671875, 3.8795928955078125, -1.5691108703613281, -12.80718994140625, -46.4764404296875, 23.0501708984375, 19.692337036132812, 20.400619506835938, -58.192962646484375, 7.584209442138672, -9.890411376953125, 67.9049072265625, 9.460067749023438, 15.75189208984375, 17.92755126953125, 38.370513916015625, -11.1690673828125, 75.96380615234375, -21.8671875, 9.303665161132812, 7.821380615234375, -1.6747512817382812, 26.23651123046875, 50.320159912109375, -9.57763671875, 67.03298950195312, -48.10862731933594, 3.39678955078125, 6.84710693359375, -55.261383056640625, 4.93121337890625, 24.88702392578125, 81.7598876953125, -16.093658447265625, 97.04109954833984, 23.05561065673828, -2.4437618255615234, 98.95864868164062, 69.4510498046875, 104.06500244140625, 33.925079345703125, 116.087890625, 13.317153930664062, 91.67694091796875, 27.5284423828125, 39.875946044921875, 61.74664306640625, 66.32318115234375, -6.1506805419921875, 79.16082763671875, 4.3337554931640625, 3.5003662109375, 1.9783248901367188, 10.14697265625, 3.5747737884521484, 112.21417236328125, 63.75067138671875, 9.592334747314453, 58.696815490722656, 34.0498046875, -47.867431640625, -65.90142822265625, 44.86285400390625, -45.54304504394531, -0.807586669921875, 52.561248779296875, -42.86037826538086, 31.5396728515625, 55.74322509765625, 11.388538360595703, 15.790496826171875, 41.00628662109375, -12.47906494140625, 46.2508544921875, 47.93212890625, 60.851722717285156, -41.40803527832031, 4.382026672363281, -52.83099365234375, 12.135795593261719, -5.085960388183594, 30.502330780029297, 36.66461181640625, 32.41194152832031, 55.28477478027344, 32.250518798828125, 46.3577880859375, -14.802955627441406], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000137.npy"}
{"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 23.088045120239258, "std": 46.18060302734375, "min": -79.83840942382812, "p10": -28.672518920898433, "median": 18.003232955932617, "p90": 83.29605331420898, "max": 146.65850830078125, "pos_frac": 0.65625, "sample": [141.95782470703125, 43.474853515625, 69.62567138671875, 2.2043991088867188, -12.426193237304688, 9.38360595703125, 33.0111083984375, 116.96525573730469, -1.718994140625, 115.53298950195312, 20.47174072265625, 6.78460693359375, 71.47494506835938, 56.0523681640625, 11.778932571411133, -19.18177032470703, -10.755615234375, 37.282135009765625, -22.756622314453125, -35.686767578125, 6.147125244140625, 16.654937744140625, -31.888687133789062, 7.1614990234375, -46.91424560546875, 28.101516723632812, 17.37743377685547, 53.87681579589844, -2.2631072998046875, 128.02218627929688, -31.46514892578125, -3.5548095703125, 17.120208740234375, 27.382465362548828, 4.613683700561523, 13.650619506835938, -13.843719482421875, 51.4798583984375, 44.24017333984375, -22.845703125, 39.57566833496094, -1.7979278564453125, 83.88288116455078, 90.6671142578125, 21.43157958984375, 70.75910949707031, -8.365142822265625, 15.389896392822266, 17.11322021484375, -16.757659912109375, 3.466552734375, 24.873825073242188, 47.9176025390625, -27.322601318359375, 107.2601318359375, 29.16461181640625, 58.7935791015625, 63.25602722167969, 29.037696838378906, -12.30035400390625, -45.97581481933594, 10.553131103515625, 7.788421630859375, 9.137039184570312, 68.37646484375, -70.701171875, 87.6473388671875, 34.72380065917969, 52.257049560546875, -7.8641357421875, -43.175048828125, 28.378662109375, -13.248580932617188, 87.98654174804688, 25.4183349609375, 26.681594848632812, -32.92869567871094, 43.8782958984375, 115.09443664550781, -21.37659454345703, 51.048828125, 42.916748046875, 70.47026062011719, 57.60502624511719, -79.83840942382812, -17.830352783203125, 29.991363525390625, -10.754203796386719, -0.12987899780273438, -27.475677490234375, 60.25975036621094, 75.067138671875, 139.6561279296875, 54.44744873046875, -10.155584335327148, 20.4844970703125, 37.545013427734375, 43.736907958984375, -18.186126708984375, -17.775375366210938, 70.46597290039062, -24.917251586914062, 146.65850830078125, -39.511390686035156, 18.93889617919922, -4.551513671875, -16.232391357421875, -24.835784912109375, -49.63612365722656, 47.95307922363281, -17.477783203125, 46.299346923828125, -15.838333129882812, 105.54751586914062, 16.92449951171875, 9.009685516357422, 83.0445556640625, -65.79014587402344, 70.62286376953125, -62.972381591796875, 18.629032135009766, 54.477874755859375, 61.17985534667969, 35.401123046875, 78.04150390625, 15.353912353515625, 20.106948852539062, -17.930252075195312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000138.npy"}
{"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 15.895705223083496, "std": 41.82365798950195, "min": -79.2398681640625, "p10": -39.615699768066406, "median": 14.350341796875, "p90": 70.57380905151366, "max": 126.311767578125, "pos_frac": 0.6328125, "sample": [-34.80608367919922, 14.854263305664062, -46.79901123046875, 26.995254516601562, -30.19342041015625, 45.74517822265625, 34.50397491455078, 37.133056640625, -1.7718963623046875, -13.829216003417969, 54.539306640625, 13.952560424804688, 49.081787109375, 29.192230224609375, -29.1763916015625, 61.23675537109375, 65.35366821289062, 2.0407791137695312, 2.24700927734375, 2.5040855407714844, 5.74542236328125, 34.068511962890625, 8.66839599609375, 28.95159912109375, -23.197998046875, -9.1551513671875, -1.6137809753417969, 34.74143981933594, -4.118114471435547, -67.88418579101562, 2.339506149291992, -5.3944549560546875, 74.12899780273438, 74.48837280273438, 49.647796630859375, -1.4999504089355469, 15.991828918457031, 11.854293823242188, -1.227813720703125, 75.26689147949219, -23.14288330078125, -26.565261840820312, 2.725494384765625, 22.388946533203125, 23.312442779541016, 18.930877685546875, -74.61343383789062, -44.226646423339844, -6.175239562988281, 65.08657836914062, 7.296234130859375, -6.898223876953125, -63.69688415527344, -46.84165954589844, 71.1752700805664, 59.8341064453125, -26.076934814453125, 14.748123168945312, -42.871826171875, -21.223419189453125, 68.80096435546875, -8.359125137329102, 52.703643798828125, 18.3258113861084, 80.94622802734375, 16.634368896484375, 5.564245223999023, -3.921142578125, 41.5533447265625, 71.97563171386719, 7.77545166015625, 49.983665466308594, 15.52801513671875, 110.29238891601562, 26.17253875732422, -40.9891357421875, 70.3160400390625, 118.78713989257812, 31.43414306640625, 57.968536376953125, 51.17613983154297, 13.944732666015625, 32.175689697265625, 57.6842041015625, 48.496368408203125, -10.555389404296875, 109.18353271484375, -79.2398681640625, 9.906158447265625, -8.116201400756836, -5.862701416015625, 94.78176879882812, 24.78411865234375, 32.927825927734375, 21.949951171875, 16.805068969726562, 9.844802856445312, -39.02708435058594, -45.39115905761719, 2.5803966522216797, 60.699371337890625, 21.452056884765625, -19.582305908203125, 18.575963973999023, -32.55418395996094, -11.97021484375, 43.67749786376953, 119.5146484375, 6.8769989013671875, 69.82614135742188, 126.311767578125, 34.247406005859375, -6.13128662109375, -12.599700927734375, -55.02476501464844, -52.711273193359375, 0.0, 24.396575927734375, -2.50299072265625, 87.37377166748047, -27.101181030273438, -42.57647705078125, 15.680160522460938, -26.257492065429688, 17.74395751953125, -3.0811767578125, 36.25457763671875, 26.800209045410156], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000139.npy"}
{"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 24.28836441040039, "std": 38.55758285522461, "min": -67.52577209472656, "p10": -22.807942962646482, "median": 19.38697052001953, "p90": 78.17009277343749, "max": 128.07012939453125, "pos_frac": 0.7109375, "sample": [14.6356201171875, 66.15728759765625, 97.34333801269531, -36.209571838378906, -14.677886962890625, 58.814178466796875, -14.325271606445312, 32.20195007324219, -2.422607421875, 0.7660598754882812, -13.999946594238281, -20.961883544921875, 38.28045654296875, 29.03540802001953, -7.350379943847656, 44.73755645751953, 0.128326416015625, 51.32427978515625, 59.643768310546875, -1.0765380859375, 22.253379821777344, -32.28712463378906, 54.97602844238281, -44.717010498046875, 25.840805053710938, 41.02777099609375, -57.42938232421875, -1.6936798095703125, 37.72132873535156, 73.6596908569336, 98.5836181640625, 68.60699462890625, 98.06146240234375, 60.49659729003906, -36.1121826171875, 57.87982177734375, 35.38922119140625, -4.13763427734375, 87.63795471191406, 54.325103759765625, -3.590972900390625, -47.56135559082031, 37.20477294921875, 6.958209991455078, 20.65997314453125, 14.189697265625, 87.32345581054688, 22.0904541015625, -29.21875, 128.07012939453125, 36.082000732421875, 61.10126495361328, 82.98832702636719, 8.432373046875, -1.8775405883789062, 18.66546630859375, 18.511749267578125, -1.6529617309570312, -24.419677734375, 20.108474731445312, 45.337066650390625, -2.5931549072265625, -14.922744750976562, 58.368896484375, 0.25994873046875, 39.87469482421875, 16.77060317993164, -12.57708740234375, 0.0, -67.52577209472656, 4.35906982421875, -42.151123046875, 115.16958618164062, 27.598068237304688, 22.161361694335938, 65.67953491210938, 52.17327880859375, 80.9464111328125, 5.037681579589844, 86.17025756835938, 20.362747192382812, 8.292190551757812, -0.8059043884277344, 80.2352294921875, 48.641937255859375, 51.08684158325195, 61.756195068359375, 9.070747375488281, 47.53485107421875, 56.04901123046875, 8.056510925292969, 4.52789306640625, 32.860198974609375, 9.195938110351562, 15.909904479980469, 61.826873779296875, 0.0, -2.162137985229492, 49.9810791015625, -23.215370178222656, -11.267852783203125, 62.21307373046875, -22.633331298828125, 50.10009765625, 11.349334716796875, -11.660873413085938, 41.2498779296875, 12.954132080078125, 9.57391357421875, 6.115104675292969, 75.50787353515625, 0.37982177734375, 77.2850341796875, 9.475799560546875, 58.29466247558594, 8.839584350585938, 11.44647216796875, -6.306793212890625, 11.95196533203125, -46.00944519042969, 27.526824951171875, 92.18667602539062, 20.39410400390625, -6.768135070800781, 28.37078857421875, -27.91064453125, 93.69384765625, 37.0255126953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000140.npy"}
{"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 27.077795028686523, "std": 43.398536682128906, "min": -124.86749267578125, "p10": -24.580355834960937, "median": 24.644519805908203, "p90": 84.7807403564453, "max": 119.3607177734375, "pos_frac": 0.7578125, "sample": [85.81468200683594, 22.53753662109375, -24.020965576171875, -16.5047607421875, 41.9664306640625, 41.58953094482422, 15.66937255859375, 53.17799377441406, -36.78460693359375, 118.96514892578125, 5.663665771484375, 30.6676025390625, 76.80412292480469, 106.24899291992188, 39.780487060546875, 66.80682373046875, 29.28350830078125, 61.238189697265625, 51.87440490722656, 86.61129760742188, 96.72235107421875, 16.1004638671875, 22.050453186035156, 83.25401306152344, -17.21233558654785, -7.392707824707031, 76.41854858398438, 20.132217407226562, 13.568527221679688, 108.35443115234375, -0.0482177734375, -25.07012939453125, -32.545135498046875, 84.94482421875, 27.74786376953125, 69.093994140625, -24.370452880859375, -61.797821044921875, 10.5472412109375, -29.4161376953125, 81.20977783203125, 38.54180908203125, 49.412635803222656, -9.4534912109375, -52.02220916748047, 25.82501220703125, 31.57354736328125, -20.269668579101562, 92.12863159179688, 13.584213256835938, 27.005340576171875, 54.867523193359375, 18.683029174804688, 5.12225341796875, 81.4532470703125, 33.487701416015625, -13.129451751708984, 11.221649169921875, 51.64500427246094, -14.42095947265625, 62.01617431640625, 74.75798034667969, -12.70892333984375, -40.474639892578125, 13.399078369140625, 84.71041870117188, 20.061553955078125, 28.731300354003906, -14.517082214355469, 73.82734680175781, 2.121307373046875, 31.684494018554688, 21.533891677856445, 59.348976135253906, 36.92315673828125, 66.73670959472656, 96.97137451171875, 23.44451904296875, 0.0, -0.009613037109375, 49.230316162109375, -10.8785400390625, 2.665924072265625, 97.910888671875, -12.78948974609375, 94.12910461425781, -124.86749267578125, 20.54241943359375, 3.557098388671875, 28.716812133789062, 15.40673828125, 24.62853240966797, 8.044654846191406, -106.06077575683594, -13.2828369140625, 26.604034423828125, 12.93170166015625, 79.70033264160156, 24.660507202148438, 26.42877197265625, -36.864654541015625, 41.199462890625, 51.2548828125, 63.4580078125, 33.97509765625, 11.563512802124023, 53.6678466796875, 61.8712158203125, 56.901397705078125, -39.29072570800781, 3.551271438598633, 5.4794921875, 105.63409423828125, -39.41143798828125, 15.227386474609375, 13.692581176757812, 42.508148193359375, 17.653762817382812, 70.47393798828125, 14.626411437988281, -26.147232055664062, 50.030517578125, 119.3607177734375, -15.280059814453125, 45.492042541503906, 49.967376708984375, 9.588653564453125, 15.000198364257812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000141.npy"}
{"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 21.715831756591797, "std": 46.48557662963867, "min": -119.66464233398438, "p10": -33.824545288085936, "median": 12.455307006835938, "p90": 88.9261260986328, "max": 131.40689086914062, "pos_frac": 0.703125, "sample": [17.6827392578125, -2.98895263671875, 44.48915100097656, 37.309112548828125, 14.759244918823242, -17.5867919921875, 1.530792236328125, -5.43951416015625, 98.72235107421875, 57.07469177246094, 60.58990478515625, -33.832855224609375, -18.386810302734375, 66.01458740234375, 31.822906494140625, 54.2213134765625, -33.82098388671875, -11.496551513671875, 112.55133056640625, 37.907981872558594, 12.365142822265625, 16.062294006347656, 0.0, 79.25090789794922, 40.802978515625, 36.982269287109375, 22.16961669921875, 0.66314697265625, -4.776153564453125, -1.423095703125, -13.977508544921875, 34.081024169921875, -12.32187271118164, -40.481536865234375, 5.1547393798828125, 6.646465301513672, 23.335479736328125, 40.020301818847656, -19.542892456054688, 37.265625, 12.54547119140625, 90.87289428710938, 110.50543212890625, 37.719879150390625, 28.760009765625, -52.72203063964844, 6.490503311157227, 92.88359069824219, 27.942367553710938, 79.6666259765625, 5.32489013671875, -14.317594528198242, 23.0245361328125, 50.65467834472656, -13.90423583984375, 7.565792083740234, 10.391433715820312, 70.05426025390625, -19.0518798828125, 4.032623291015625, 1.9774169921875, 58.51513671875, -20.9752197265625, -58.716400146484375, 111.7080078125, 57.15697479248047, 48.2890625, -22.742599487304688, 77.00554656982422, 1.7941741943359375, 0.8904266357421875, -19.236099243164062, 6.550689697265625, 8.738195419311523, -38.876251220703125, 39.179107666015625, -60.362579345703125, 53.72418975830078, 45.4952392578125, -37.8353271484375, -51.658782958984375, 74.08270263671875, 109.01556396484375, 19.11700439453125, -7.138587951660156, 25.61316680908203, 7.5461578369140625, 110.31524658203125, 52.80772399902344, 7.8403167724609375, -14.541194915771484, 0.8296966552734375, 42.919647216796875, 79.96395874023438, 97.6856689453125, -3.6429901123046875, 63.8817138671875, 8.115020751953125, 59.27142333984375, 131.40689086914062, -74.21334838867188, 25.961578369140625, -55.045074462890625, 1.425283432006836, 8.292850494384766, -69.3636474609375, 100.66404724121094, 12.150054931640625, 72.72515869140625, -1.667816162109375, 108.24420166015625, -82.36077880859375, 5.243438720703125, 40.28718566894531, 53.82279968261719, 3.88677978515625, 43.02685546875, -1.7383155822753906, 18.870941162109375, -18.3873291015625, -119.66464233398438, 88.091796875, 0.6260395050048828, -17.970138549804688, 9.022705078125, 44.160011291503906, 120.49557495117188, 65.49051666259766], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000142.npy"}
{"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 26.848384857177734, "std": 41.68739318847656, "min": -85.6859130859375, "p10": -12.7923583984375, "median": 19.782543182373047, "p90": 89.7405364990234, "max": 137.7640380859375, "pos_frac": 0.7265625, "sample": [48.671653747558594, 101.2745361328125, 104.22698974609375, 46.475830078125, 96.07431030273438, 24.30584716796875, 47.17840576171875, 13.136329650878906, -5.13897705078125, -3.2816905975341797, 85.98025512695312, -8.438114166259766, 51.13616943359375, 9.457275390625, -4.728401184082031, 3.800567626953125, 87.02606201171875, 31.84600830078125, -10.948020935058594, 106.4022216796875, -29.511215209960938, -4.4638824462890625, 43.242645263671875, 80.20362854003906, 34.01686096191406, 48.47148132324219, 12.36175537109375, 74.47805786132812, 137.7640380859375, -24.572784423828125, -1.615234375, 56.08430862426758, 102.03684997558594, 65.67341613769531, 22.338531494140625, -0.030517578125, -76.25030517578125, -4.898826599121094, 62.24906921386719, -0.57135009765625, 11.490951538085938, -32.53022766113281, 3.9637222290039062, 76.5638427734375, 23.966983795166016, -11.75079345703125, 12.711523056030273, -2.7382354736328125, 11.294204711914062, -85.6859130859375, 42.283599853515625, 45.04931640625, 19.423973083496094, 58.91209411621094, 24.1405029296875, 37.1961669921875, 5.13226318359375, 55.15381622314453, -4.3032073974609375, 8.224723815917969, -6.4193115234375, 5.799100875854492, 136.34786987304688, 43.89007568359375, 8.929630279541016, 1.1383819580078125, 76.050048828125, 50.76194763183594, 44.1163330078125, -7.065399169921875, 51.355743408203125, 74.55409240722656, 3.2083740234375, 4.64653205871582, 5.090045928955078, -4.6116943359375, -44.52996826171875, 15.6829833984375, 20.14111328125, 45.083168029785156, 5.793487548828125, -1.2193603515625, 9.36175537109375, 52.76422119140625, 15.8519287109375, -12.68304443359375, -13.04742431640625, -6.8831024169921875, 6.285514831542969, 22.4517822265625, 42.098079681396484, 0.054351806640625, 21.807464599609375, 51.094512939453125, 53.813720703125, 38.626617431640625, 0.20026397705078125, 96.19973754882812, -16.420612335205078, 16.2298583984375, 102.33480834960938, 51.21832275390625, -26.0863037109375, 27.522018432617188, -18.30859375, 121.49575805664062, 103.78327941894531, -0.605712890625, 2.0540218353271484, 11.856414794921875, -2.569976806640625, 47.036834716796875, 3.8644790649414062, 98.15728759765625, 32.78228759765625, 100.27664184570312, 26.98626708984375, 67.97634887695312, -65.16207885742188, 40.576271057128906, -41.178680419921875, 54.6253662109375, -36.9993896484375, -6.12445068359375, 46.792449951171875, 36.678863525390625, 28.23248291015625, 2.866544723510742], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000143.npy"}
{"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 21.07245635986328, "std": 46.51237869262695, "min": -117.766845703125, "p10": -28.615026855468745, "median": 18.134601593017578, "p90": 81.03943786621093, "max": 216.9637451171875, "pos_frac": 0.6640625, "sample": [-2.705333709716797, 55.58759307861328, 60.9893798828125, 41.836326599121094, 3.5991973876953125, 33.80242919921875, -16.515640258789062, -65.57696533203125, 0.23871803283691406, 134.93023681640625, 4.1602325439453125, 0.3022918701171875, 25.99371337890625, 78.12518310546875, 19.459686279296875, 47.1011962890625, 18.26551055908203, 51.932708740234375, 19.666343688964844, 1.757171630859375, -8.827407836914062, 46.79254150390625, 47.57452392578125, 0.10932159423828125, 8.4459228515625, -5.237640380859375, 17.343055725097656, -9.499164581298828, 31.844390869140625, -23.02948570251465, -58.5941162109375, 33.136505126953125, -7.8509521484375, 132.98944091796875, 111.66976928710938, 68.79331970214844, 60.010467529296875, 129.587890625, -1.607696533203125, 51.22209167480469, -4.906513214111328, -58.10154724121094, 31.676467895507812, 6.345947265625, -4.790275573730469, -42.974403381347656, -17.947357177734375, 41.4306640625, -41.331085205078125, -30.79962158203125, 49.00703430175781, 84.23782348632812, 54.34044647216797, -117.766845703125, -94.71287536621094, -36.84965515136719, 10.3441162109375, -13.308120727539062, -20.330734252929688, 18.003692626953125, 34.4329833984375, 34.19940185546875, -6.012237548828125, -3.743865966796875, 47.44972229003906, 11.14617919921875, 67.3580322265625, 88.35647583007812, 51.13526916503906, 61.78570556640625, 12.487655639648438, 25.391876220703125, 90.30838012695312, 1.226226806640625, 26.7093505859375, 54.216339111328125, -2.904266357421875, 61.70301055908203, 41.704345703125, 216.9637451171875, 33.752227783203125, 30.71078109741211, -37.520172119140625, -27.67877197265625, 55.73994445800781, 57.903778076171875, -1.527099609375, 29.823028564453125, -5.4841156005859375, 85.84402465820312, 80.64419555664062, 47.333984375, 30.116195678710938, 0.41036224365234375, 4.777257919311523, 81.961669921875, 7.033905029296875, -4.0069580078125, -13.955078125, 56.673736572265625, -14.54962158203125, -3.0032501220703125, 0.0, -23.4674072265625, -40.702423095703125, -23.76043701171875, -2.74468994140625, -74.92231750488281, 36.548614501953125, 25.77890396118164, 49.034454345703125, 17.65875244140625, 85.44357299804688, -4.429595947265625, 22.679931640625, 1.781463623046875, 22.78564453125, 24.64567756652832, 30.148345947265625, -22.20989990234375, 20.399459838867188, 30.6956787109375, 4.053428649902344, 89.62371826171875, -30.870574951171875, -9.44183349609375, 92.54791259765625, 17.6937255859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000144.npy"}
{"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 29.119651794433594, "std": 43.96269607543945, "min": -79.00465393066406, "p10": -22.09614791870117, "median": 28.33318328857422, "p90": 84.62403259277343, "max": 144.26153564453125, "pos_frac": 0.734375, "sample": [3.2109375, 57.02684020996094, -63.25274658203125, 29.784820556640625, 3.7544078826904297, -25.70440673828125, 69.20256042480469, 41.641021728515625, 99.77676391601562, -49.24688720703125, 2.04718017578125, 105.84880065917969, 24.426490783691406, 82.64923095703125, 18.70928955078125, 55.60791015625, 20.5738525390625, 41.405364990234375, 5.927192687988281, 7.9047393798828125, -21.619956970214844, -17.46319580078125, 51.2379150390625, 23.152803421020508, -7.112388610839844, 87.47613525390625, -29.17835235595703, 110.13601684570312, -7.021636962890625, 30.26885986328125, 2.915130615234375, 61.93211364746094, 32.9498291015625, 59.39503479003906, 144.26153564453125, -2.73944091796875, 104.01040649414062, -70.0268325805664, 14.319610595703125, 0.4075927734375, 24.11175537109375, 22.91302490234375, -1.1432418823242188, 118.49853515625, 117.54794311523438, 27.17767333984375, 60.71026611328125, 65.587890625, 18.603309631347656, 37.5885009765625, 63.23344421386719, -25.53790283203125, 28.10015869140625, 83.40170288085938, 61.29498291015625, 66.32550048828125, -17.041259765625, -8.16463851928711, -66.14883422851562, 9.242485046386719, 30.679046630859375, 74.8896484375, -33.045501708984375, 22.435043334960938, 100.305908203125, -8.437355041503906, 55.9390869140625, 3.916900634765625, 39.0531005859375, 49.984519958496094, 56.74237060546875, -23.207260131835938, 28.012176513671875, 37.05548095703125, 57.3170166015625, 74.53765869140625, 13.101547241210938, 31.545944213867188, -9.28369140625, 36.1761474609375, -70.04322814941406, 49.43009948730469, 46.3619270324707, 24.57891845703125, 137.67636108398438, 28.566207885742188, 51.031593322753906, 44.1046142578125, -28.323623657226562, -15.693275451660156, 3.7212142944335938, 0.0, 53.250030517578125, 33.483489990234375, 32.359527587890625, 13.694717407226562, 75.70693969726562, 106.62356567382812, 51.62150573730469, 48.531517028808594, -39.63585662841797, -79.00465393066406, 25.923187255859375, 23.88977813720703, 18.401817321777344, -21.41485595703125, 53.78026580810547, 40.83782958984375, -2.2799606323242188, -2.0762767791748047, -16.909698486328125, 89.45257568359375, 58.91722106933594, 44.58856201171875, -5.69622802734375, -6.265869140625, 52.962120056152344, 44.493125915527344, 59.350616455078125, 62.262969970703125, 26.560546875, 45.39714050292969, -9.910018920898438, 2.6964569091796875, 134.02871704101562, -12.58819580078125, 35.577880859375, -5.319305419921875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000145.npy"}
{"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 27.25536346435547, "std": 44.39828109741211, "min": -94.00860595703125, "p10": -25.57359619140625, "median": 19.69812774658203, "p90": 80.63703613281248, "max": 165.69351196289062, "pos_frac": 0.75, "sample": [49.96856689453125, 6.813686370849609, 9.055023193359375, -21.9443359375, 14.11865234375, 60.162933349609375, -9.8399658203125, 21.36248779296875, 68.34577178955078, -63.3692626953125, 131.02926635742188, 78.64407348632812, 30.64984130859375, -24.98626708984375, 30.29718017578125, 17.913429260253906, 109.29788208007812, 22.151702880859375, 18.82684326171875, 3.477874755859375, 62.504150390625, 37.295555114746094, 48.308937072753906, -32.513427734375, -38.78961181640625, -5.9439849853515625, 18.921274185180664, 60.497344970703125, 24.431922912597656, 19.2171630859375, 15.7176513671875, 13.002273559570312, 64.08074951171875, 20.35382080078125, -4.47802734375, 69.81405639648438, 57.0849609375, 28.97418212890625, 70.12127685546875, 15.617912292480469, 19.775177001953125, 79.52130126953125, 1.8537445068359375, 165.69351196289062, 0.0, 79.49371337890625, -4.63232421875, -7.614046096801758, 56.61570739746094, 37.03009033203125, 24.6964111328125, -22.028228759765625, 39.33438491821289, 13.052902221679688, 90.07101440429688, 111.83599853515625, -84.99172973632812, 63.356101989746094, 61.746795654296875, 85.55164337158203, 24.870849609375, 37.355194091796875, 19.621078491210938, 61.66827392578125, 18.938079833984375, -20.6292724609375, -18.59320068359375, 14.085830688476562, 11.826080322265625, -50.94258117675781, 68.96728515625, 3.862060546875, 7.533294677734375, 136.4329376220703, -33.322357177734375, 121.56085205078125, 63.87825012207031, 11.391403198242188, -6.350067138671875, 60.459442138671875, 4.475738525390625, 16.829315185546875, 5.028678894042969, -16.79736328125, 45.8336181640625, 5.450595855712891, -7.480072021484375, 7.049774169921875, 0.0, -9.312477111816406, -27.321701049804688, -4.229583740234375, 16.340591430664062, -44.8251953125, 10.02001953125, 95.61627197265625, -23.670654296875, 42.87214660644531, 14.614341735839844, 40.55950927734375, -27.165771484375, 1.630645751953125, 14.514259338378906, 31.22235107421875, 78.99258422851562, -26.94403076171875, 73.79730224609375, 26.032989501953125, 61.00390625, 97.59687805175781, 90.51437377929688, -94.00860595703125, 37.9940185546875, 109.53394317626953, 28.03021240234375, 72.9329833984375, -35.33360290527344, 83.24041748046875, 31.14190673828125, 53.55523681640625, -5.6341552734375, 15.213935852050781, 44.3404541015625, 73.06732177734375, 30.974502563476562, -48.23910140991211, 14.242233276367188, 46.216796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000146.npy"}
{"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 20.80077362060547, "std": 47.19132995605469, "min": -121.10107421875, "p10": -37.27857360839844, "median": 20.095474243164062, "p90": 77.57448120117188, "max": 155.76824951171875, "pos_frac": 0.703125, "sample": [-22.000579833984375, 0.168212890625, 1.892547607421875, 25.388702392578125, -16.625526428222656, -18.89898681640625, 122.75148010253906, -36.941558837890625, 0.0, 109.98443603515625, -18.251663208007812, 13.844535827636719, 23.250335693359375, 31.418624877929688, 8.655519485473633, 31.88219451904297, -8.323822021484375, 58.189605712890625, 64.554443359375, 7.575523376464844, 0.47821044921875, 28.882568359375, 32.618858337402344, 49.491424560546875, -9.744415283203125, 131.84967041015625, 30.614463806152344, 86.65892028808594, 42.267425537109375, 4.991668701171875, 72.74514770507812, 25.603790283203125, 31.536848068237305, 73.48190307617188, 1.551605224609375, -38.06494140625, -32.062156677246094, 1.8172454833984375, 7.42138671875, 47.67478942871094, 24.508056640625, 62.6268196105957, 13.713037490844727, 51.299476623535156, -19.916622161865234, 49.709228515625, -10.534088134765625, -0.24629974365234375, 11.598093032836914, 41.54779815673828, -25.661178588867188, 110.9259033203125, 12.689453125, 57.493865966796875, 77.51394653320312, 24.05878448486328, 155.76824951171875, 41.15486145019531, 50.82780456542969, 18.473434448242188, 38.44061279296875, 50.714752197265625, 95.13917541503906, 45.397300720214844, 103.77719116210938, -24.206268310546875, 23.0166015625, 106.03359985351562, 68.49607849121094, 6.021156311035156, -77.026611328125, 10.203125, -17.281219482421875, -10.67352294921875, -44.09625244140625, -5.8037109375, 13.310396194458008, 33.00855255126953, 0.04121971130371094, 1.1919708251953125, 54.33879089355469, 6.570835113525391, 2.064666748046875, 44.17671203613281, 49.47309875488281, 51.590789794921875, 2.595733642578125, 50.63006591796875, -42.428985595703125, 90.90775299072266, 46.0076904296875, 49.0316162109375, -49.836578369140625, -86.908203125, -108.1817855834961, 1.3765106201171875, 60.469940185546875, 94.28518676757812, -4.329833984375, -121.10107421875, -13.95037841796875, 0.059597015380859375, 77.71572875976562, 19.47509765625, -8.069183349609375, -42.311309814453125, 32.400054931640625, -12.757095336914062, -53.10015869140625, -20.608078002929688, -66.10803985595703, 27.13238525390625, -60.18293762207031, 57.27214431762695, 50.43577575683594, 13.037811279296875, 28.867446899414062, 68.47796630859375, 37.70751953125, -29.883819580078125, 20.715850830078125, 44.57774353027344, 81.326904296875, 69.05865478515625, 73.01242065429688, -0.355804443359375, -49.66992950439453, -4.093475341796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000147.npy"}
{"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 21.794063568115234, "std": 48.8247184753418, "min": -87.82876586914062, "p10": -35.15861206054687, "median": 13.469598770141602, "p90": 84.1431640625, "max": 172.88299560546875, "pos_frac": 0.6484375, "sample": [48.73541259765625, 66.01470947265625, 39.28791809082031, 1.789276123046875, -22.146818161010742, 64.3804931640625, 5.015340805053711, 68.5213623046875, 78.38348388671875, 26.929580688476562, 9.7525634765625, 41.388275146484375, -46.632354736328125, 109.293701171875, -11.582244873046875, 82.93096923828125, 57.2626953125, 0.240966796875, 84.22421264648438, 9.469619750976562, -83.84927368164062, 104.566650390625, -58.089324951171875, 2.8023223876953125, -20.17852783203125, -38.67120361328125, 31.629928588867188, -12.046241760253906, 25.590087890625, -43.74404525756836, 7.6831207275390625, 24.595718383789062, -34.315338134765625, -87.82876586914062, -28.06549072265625, 27.685394287109375, 138.81549072265625, -2.7664947509765625, -65.19577026367188, 34.67181396484375, 172.88299560546875, 6.855968475341797, -39.537139892578125, -5.8760986328125, 4.579376220703125, 0.0064868927001953125, -34.209754943847656, -20.845794677734375, 39.037841796875, 51.99125671386719, 19.319271087646484, 27.63873291015625, 15.593313217163086, 39.2703857421875, -30.165481567382812, 136.67410278320312, 26.72662353515625, 51.55238342285156, 46.150718688964844, 145.1318359375, 58.00579833984375, 57.796897888183594, 11.786128997802734, -3.142333984375, 0.0, 46.784996032714844, 96.21188354492188, 57.04563903808594, 19.007553100585938, 41.58740234375, 15.045032501220703, -26.223217010498047, -24.1070556640625, -38.55183410644531, 74.5302734375, -1.103271484375, -25.878570556640625, 86.301513671875, -20.027877807617188, 79.86640930175781, 1.2219619750976562, -13.13275146484375, 11.8941650390625, 9.3291015625, 113.24235534667969, 37.432952880859375, 8.921207427978516, 25.70416259765625, -16.424942016601562, -37.126251220703125, 43.32099914550781, 10.498283386230469, 4.229270935058594, 8.564132690429688, 78.75935363769531, 37.55873107910156, -48.754302978515625, -1.9924392700195312, -21.332015991210938, 81.86463928222656, -13.374420166015625, 39.69352722167969, 112.96270751953125, 44.177001953125, -2.460723876953125, 5.517910003662109, 15.129478454589844, -23.05487060546875, 54.10552978515625, 44.27033996582031, 81.51268005371094, 22.18834686279297, 17.02911376953125, 84.10842895507812, -47.2117919921875, -10.971733093261719, -80.0826416015625, -2.5364990234375, 97.84463500976562, 0.0, -0.73651123046875, 36.15045166015625, 59.48817443847656, -29.006372451782227, -4.72821044921875, 74.61819458007812, 95.34310913085938, -8.37594223022461], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000148.npy"}
{"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 29.804601669311523, "std": 39.858184814453125, "min": -64.28866577148438, "p10": -20.462626266479493, "median": 26.88184928894043, "p90": 89.30650024414062, "max": 130.50323486328125, "pos_frac": 0.765625, "sample": [54.924713134765625, 6.724273681640625, 37.24493408203125, 40.025779724121094, 95.38385009765625, 69.23687744140625, 14.39599609375, 5.088066101074219, -3.66082763671875, 4.0027618408203125, 50.45635986328125, -8.7890625, 75.57052612304688, 64.70648193359375, 32.8447265625, -0.3734588623046875, 49.8731689453125, 14.963233947753906, -32.3846435546875, -24.538330078125, 51.541168212890625, 26.631912231445312, 69.231201171875, 48.351112365722656, 90.5509262084961, -9.061004638671875, 0.3894634246826172, 14.180206298828125, 77.34490966796875, 1.7288284301757812, 35.37181091308594, 36.097900390625, 29.245513916015625, 13.300399780273438, 0.4478168487548828, -5.549835205078125, 74.709716796875, -3.608591079711914, 7.754730224609375, -4.453666687011719, 70.54595947265625, 130.50323486328125, 5.647346496582031, -21.03028106689453, 47.749053955078125, 67.06427001953125, 29.03955078125, 10.641815185546875, 55.584228515625, 43.394012451171875, 28.778457641601562, -2.83642578125, 15.088409423828125, 36.003753662109375, -30.001007080078125, 97.5888671875, -37.15673828125, 31.13348388671875, 66.24604797363281, -1.4351806640625, 14.16080093383789, 50.943878173828125, -51.813690185546875, 104.7408447265625, -4.122894287109375, 24.73260498046875, 16.874847412109375, -12.72808837890625, 17.576904296875, 0.5242805480957031, 25.034835815429688, 42.40167236328125, 27.131786346435547, 25.55133056640625, 61.34246826171875, 32.146209716796875, 68.79638671875, 19.728042602539062, 121.04568481445312, -1.1001434326171875, 54.275360107421875, 35.06459045410156, 50.09477233886719, 81.88555908203125, 52.87451171875, 30.334228515625, 92.4017333984375, 6.320404052734375, -58.223968505859375, 10.583816528320312, 25.74871826171875, 93.89407348632812, 3.886199951171875, -17.82904052734375, 35.33050537109375, -20.250579833984375, 92.4957275390625, 31.42218017578125, 90.05766296386719, 81.98387145996094, -5.9697418212890625, -23.472991943359375, 17.604339599609375, -14.795318603515625, 112.99896240234375, 32.83544921875, -46.112701416015625, 12.307693481445312, 15.639190673828125, 110.314208984375, 25.63287353515625, -27.50177001953125, 43.7227783203125, 43.183380126953125, 20.29522705078125, 15.761810302734375, 73.4735107421875, 88.98457336425781, -20.957401275634766, -64.28866577148438, 21.534866333007812, 52.931854248046875, -0.949859619140625, 122.19049072265625, 61.00028991699219, 34.40007019042969, -23.672210693359375, 40.131103515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000149.npy"}
{"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 22.989498138427734, "std": 43.04914093017578, "min": -89.52935791015625, "p10": -33.55457763671875, "median": 20.494384765625, "p90": 75.9550308227539, "max": 137.093994140625, "pos_frac": 0.703125, "sample": [19.1092529296875, 32.84712219238281, -11.695072174072266, 60.1466064453125, 35.303802490234375, -43.900604248046875, 10.064079284667969, -13.17999267578125, -39.23919677734375, 85.64599609375, 6.612953186035156, 52.539764404296875, 56.4168701171875, 5.62396240234375, -26.287017822265625, 19.83526611328125, 20.344039916992188, 61.80396270751953, 0.430419921875, 83.9864501953125, 39.920806884765625, -8.102279663085938, 6.055267333984375, 16.012847900390625, 35.41034698486328, 9.090057373046875, -42.49481201171875, -32.7142333984375, 84.99630737304688, 17.69964599609375, 71.63995361328125, -15.463333129882812, 29.95562744140625, 42.57079315185547, 41.41302490234375, 68.69656372070312, 59.747589111328125, -36.68292236328125, 67.71178436279297, 40.81452941894531, 31.99444580078125, 48.74598693847656, -16.961669921875, 76.61558532714844, -18.374725341796875, -40.911415100097656, -15.690872192382812, 18.609130859375, 56.648101806640625, 83.90509033203125, 67.71500396728516, 128.85910034179688, 51.816436767578125, 50.956878662109375, 88.93768310546875, 63.090057373046875, 53.096160888671875, 51.920204162597656, -11.713272094726562, -24.312007904052734, -22.52947998046875, 8.683708190917969, -86.5966796875, 99.49207305908203, 88.10305786132812, 39.81098937988281, -9.36676025390625, 6.3867950439453125, 65.96066284179688, 39.257110595703125, 49.394622802734375, 71.790283203125, 20.57293701171875, 20.41583251953125, 21.575748443603516, -25.55804443359375, -35.515380859375, 63.506622314453125, 15.437870025634766, -6.7754058837890625, 66.20357513427734, -59.53369140625, -14.389678955078125, -23.954879760742188, 0.789031982421875, 79.55010986328125, -89.52935791015625, 137.093994140625, 35.3017578125, 95.5499267578125, -0.01617431640625, -25.37188720703125, 11.713958740234375, 32.262725830078125, -60.928741455078125, -18.423065185546875, 32.16729736328125, -3.1942138671875, -69.14700317382812, -10.972213745117188, 9.81744384765625, 75.15396118164062, 4.19049072265625, 39.90571594238281, 15.700244903564453, 9.786865234375, 19.61865234375, 47.97602844238281, 51.291229248046875, 67.65667724609375, 2.858642578125, 50.583343505859375, 29.737628936767578, 59.44580078125, 20.765151977539062, 9.451576232910156, 39.381507873535156, -40.849365234375, 38.49462890625, 42.68310546875, 0.0, 35.630645751953125, 75.67193603515625, -10.473602294921875, -0.23483657836914062, 19.091796875, 88.3848648071289, -55.914459228515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000150.npy"}
{"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 32.897438049316406, "std": 46.15473556518555, "min": -97.32376098632812, "p10": -23.55294570922851, "median": 34.682865142822266, "p90": 87.44215240478515, "max": 166.01873779296875, "pos_frac": 0.7265625, "sample": [51.13800048828125, -12.770538330078125, 5.465137481689453, -62.008453369140625, -19.930397033691406, 106.41326904296875, 87.04777526855469, -13.253036499023438, 39.459716796875, 40.795745849609375, -7.60614013671875, 67.40621185302734, 26.61358642578125, 42.18243408203125, -27.551345825195312, -30.037246704101562, 14.82940673828125, 64.72407531738281, 67.11627197265625, 10.825035095214844, -58.26048278808594, 12.103759765625, 22.8565673828125, 57.10692596435547, -1.47747802734375, 70.99725341796875, 48.402435302734375, 22.16156768798828, 74.00733947753906, 103.15969848632812, 57.90533447265625, 27.560287475585938, 7.03363037109375, 27.138458251953125, 62.98638153076172, -0.0946807861328125, 43.994415283203125, 67.11422729492188, -64.19281768798828, 153.22607421875, 33.882720947265625, 64.23684692382812, 30.15319061279297, -12.728271484375, 63.03955078125, -0.4793243408203125, 46.56573486328125, -29.797042846679688, 66.6715087890625, 86.8042984008789, 33.170433044433594, -32.52008056640625, 14.123119354248047, -22.161277770996094, 42.079925537109375, 64.55833435058594, 59.7176513671875, 25.536956787109375, 7.6909027099609375, 54.61787414550781, -15.194976806640625, 129.075439453125, 40.33587646484375, -54.30803680419922, 66.77536010742188, 88.36236572265625, 6.02252197265625, -4.8599853515625, 83.0714111328125, 52.95184326171875, -97.32376098632812, -26.8001708984375, 39.79229736328125, 34.633399963378906, -6.619598388671875, 34.732330322265625, 86.41702270507812, 34.88475036621094, 120.4637451171875, -29.17205810546875, -39.6678466796875, -6.74517822265625, 73.82476806640625, 34.476715087890625, 116.49053955078125, 28.01728057861328, 41.32878112792969, -5.914915084838867, 0.0, 7.0317230224609375, 64.7449951171875, 45.27667236328125, 74.37163543701172, -2.23394775390625, 46.367919921875, 20.485580444335938, 43.692138671875, 98.71676635742188, -15.025344848632812, 100.360595703125, -43.06219482421875, 56.31678771972656, 26.833587646484375, 47.36309814453125, 96.68777465820312, 59.2686767578125, 25.723953247070312, 10.118072509765625, 67.33383178710938, 32.086944580078125, 41.116363525390625, 60.2593994140625, -8.97442626953125, -19.18658447265625, 138.099609375, -5.34112548828125, 105.33535766601562, 42.46710205078125, 166.01873779296875, 50.517303466796875, 65.50901794433594, -10.4254150390625, 0.82135009765625, 47.90704345703125, 55.72650146484375, 3.20465087890625, 28.0616455078125, -17.505218505859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000151.npy"}
{"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 27.712186813354492, "std": 46.55085754394531, "min": -76.49430847167969, "p10": -31.06383590698242, "median": 22.198654174804688, "p90": 85.17359619140625, "max": 158.58294677734375, "pos_frac": 0.6953125, "sample": [95.03985595703125, 107.0457763671875, 60.878387451171875, 29.63720703125, 33.789398193359375, 22.2923583984375, 21.672119140625, 39.22149658203125, 74.5479736328125, 52.44480895996094, 39.19499206542969, 104.08526611328125, 105.47059631347656, 11.7320556640625, 55.49072265625, -8.272537231445312, 26.91632080078125, 55.5401611328125, -1.185791015625, -18.181121826171875, -53.518310546875, 48.4873046875, -54.678436279296875, -16.79144287109375, 21.570114135742188, 61.492889404296875, 11.544357299804688, 10.158424377441406, 158.58294677734375, -7.829345703125, -14.343589782714844, 50.18019104003906, 75.61444091796875, 49.6839599609375, 41.849609375, 23.923919677734375, 68.80258178710938, 10.98828125, 81.4886474609375, -2.9357681274414062, -28.44776153564453, 144.02450561523438, -55.483184814453125, 1.97088623046875, 56.81163787841797, 1.4504814147949219, -49.6317138671875, 59.93890380859375, -17.723722457885742, 1.9212646484375, 6.880523681640625, -10.018600463867188, 31.979537963867188, -9.9608154296875, -1.6501483917236328, 22.104949951171875, -0.25518798828125, -46.347320556640625, -32.86925506591797, -1.8660430908203125, 157.51788330078125, -33.35292053222656, -36.76519775390625, 41.54150390625, 115.7623291015625, -12.20233154296875, 44.762054443359375, -10.504051208496094, 41.899818420410156, 16.518798828125, 15.109424591064453, -30.290084838867188, -25.28073501586914, 24.316925048828125, 1.7536163330078125, 65.58355712890625, 46.199256896972656, 63.574951171875, 119.189697265625, 56.191802978515625, 75.55532836914062, -37.505943298339844, -13.727035522460938, -14.108489990234375, 11.5328369140625, -5.2578582763671875, 58.8662109375, 65.06390380859375, 96.34152221679688, 10.82431411743164, 20.371047973632812, -45.14329528808594, -16.443145751953125, 80.69955444335938, 66.76629638671875, 58.24799346923828, 8.431396484375, 21.86767578125, 21.919921875, -26.860595703125, 64.57040405273438, 86.6199951171875, 96.46640014648438, 45.28662109375, -37.6749267578125, 84.5537109375, 21.31725311279297, 8.632001876831055, -3.4151611328125, -8.871788024902344, 35.635650634765625, 41.361419677734375, -58.64251708984375, 37.5616455078125, -20.4505615234375, -76.49430847167969, 70.89237213134766, 29.107757568359375, 71.22608947753906, 96.58528137207031, 69.15151977539062, 36.30010986328125, 60.73577880859375, 50.21935272216797, 5.184417724609375, 18.344459533691406, 3.478271484375, 76.05084228515625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000152.npy"}
{"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 38.87435531616211, "std": 44.27922821044922, "min": -75.31245422363281, "p10": -9.370599365234375, "median": 36.711151123046875, "p90": 96.45783081054687, "max": 158.34234619140625, "pos_frac": 0.78125, "sample": [46.42333984375, 2.6757888793945312, 108.58163452148438, -27.905670166015625, 59.810333251953125, 80.25912475585938, 90.60250854492188, 7.75128173828125, 91.85360717773438, 19.10125732421875, 65.00201416015625, -1.4053802490234375, 6.837646484375, -3.4597854614257812, 66.53453826904297, 32.08660888671875, -0.22137451171875, -13.97930908203125, 31.429534912109375, 99.3929443359375, 107.97726440429688, 74.0467529296875, 69.95014190673828, 1.5067520141601562, 42.562408447265625, -8.222946166992188, 74.828125, -8.416284561157227, 19.205503463745117, 41.895355224609375, -36.967315673828125, 45.12238311767578, 84.08831787109375, 4.673187255859375, 29.406402587890625, 13.145263671875, 28.917938232421875, 82.58135986328125, -14.673540115356445, 0.0, 95.9884033203125, 56.28511047363281, 126.958984375, 47.57344055175781, 77.48687744140625, 12.75201416015625, 53.303253173828125, 16.100372314453125, 8.65460205078125, 11.648681640625, 56.875579833984375, 49.1090087890625, -73.65411376953125, -3.9716567993164062, -22.942047119140625, 131.8751220703125, 62.975128173828125, 68.44058227539062, -13.005615234375, 76.27566528320312, 31.896148681640625, 0.4971160888671875, -11.557418823242188, 109.384765625, 30.086227416992188, 35.708251953125, -14.3997802734375, -4.120941162109375, 66.78094482421875, 72.88003540039062, -8.11944580078125, 47.4039306640625, -4.90802001953125, 49.575347900390625, -2.8964691162109375, 55.55059814453125, -1.98822021484375, 12.897186279296875, 64.12887573242188, 91.260986328125, 39.86457824707031, 6.3898773193359375, 100.27621459960938, 91.23031616210938, 45.15284729003906, 36.864715576171875, 5.375312805175781, -9.71978759765625, 72.74445343017578, 7.01702880859375, 30.9505615234375, 60.93310546875, 88.30252075195312, 101.8458251953125, 123.89752197265625, 26.66675567626953, 5.225671768188477, -2.1990966796875, 23.58428955078125, 0.123931884765625, 45.04571533203125, 59.163909912109375, 110.41276550292969, 6.566902160644531, -5.96044921875, -9.220947265625, 87.44534301757812, 76.95343017578125, 19.285133361816406, 158.34234619140625, 97.55316162109375, 28.012741088867188, 17.931793212890625, 68.04672241210938, 17.745508193969727, 79.90496826171875, 37.077919006347656, 155.5472412109375, 85.79617309570312, 4.670379638671875, -36.3875732421875, 62.858062744140625, 66.20639038085938, 43.659271240234375, -27.149810791015625, -75.31245422363281, 38.85150909423828, 36.557586669921875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000153.npy"}
{"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 30.598081588745117, "std": 54.58898162841797, "min": -141.44195556640625, "p10": -37.790513610839845, "median": 31.624847412109375, "p90": 94.788232421875, "max": 211.01556396484375, "pos_frac": 0.7421875, "sample": [105.73713684082031, 82.97395324707031, 6.12200927734375, 180.541015625, 49.5467529296875, 14.5799560546875, 9.457717895507812, 12.564453125, 19.54852294921875, 49.1151123046875, -31.66314697265625, 63.66690444946289, 33.26483917236328, 44.1031494140625, 77.54263305664062, -76.0059814453125, 54.594390869140625, -65.65896606445312, 211.01556396484375, -18.435028076171875, 51.44493103027344, 21.83282470703125, 0.71075439453125, -3.24859619140625, 6.34149169921875, 18.184524536132812, 22.93701171875, -6.197540283203125, 38.89508056640625, 135.89645385742188, 35.052001953125, -10.73764419555664, 75.74432373046875, 125.0462646484375, 38.250732421875, 21.064687728881836, 38.89300537109375, 95.66827392578125, 61.267425537109375, 75.32525634765625, 67.85995483398438, 73.3924560546875, 93.16012573242188, 67.44418334960938, -37.77606201171875, 134.8043670654297, 105.71041870117188, 34.883262634277344, 15.6209716796875, 11.19287109375, -5.382606506347656, 1.5373611450195312, 26.8662109375, -34.327545166015625, 32.14910888671875, 34.792083740234375, 94.41107177734375, -37.82423400878906, -13.926727294921875, -62.829833984375, -8.592079162597656, 11.4908447265625, 61.233367919921875, 53.22984313964844, 44.64234924316406, -141.44195556640625, -1.0629158020019531, -35.69096374511719, -27.970565795898438, 30.475494384765625, 10.296613693237305, -55.243255615234375, 87.75448608398438, 88.19577026367188, 48.58607482910156, -75.28006744384766, 6.513607025146484, -4.410400390625, 30.289581298828125, 20.022232055664062, -44.1474609375, -5.6920928955078125, -21.95305633544922, 3.0369720458984375, 35.65087890625, -51.978363037109375, 45.554046630859375, 16.32373046875, 18.554962158203125, 1.5188140869140625, 53.921295166015625, -40.19868469238281, 97.9766845703125, 72.8834228515625, 31.1005859375, 24.154266357421875, 40.30718994140625, -31.23443603515625, 97.7962646484375, -0.2319183349609375, 88.61810302734375, 128.64654541015625, -16.86181640625, -109.398681640625, 76.66920471191406, 68.05897521972656, 42.70323181152344, 44.824310302734375, 54.41949462890625, 58.372711181640625, 122.58389282226562, 53.28270721435547, 53.61714172363281, 65.04447937011719, 18.700927734375, 36.14436340332031, 125.31448364257812, 57.9901123046875, 13.510765075683594, 38.940673828125, -61.818206787109375, -48.57769775390625, 75.70083618164062, -5.8220367431640625, 27.107891082763672, 14.56854248046875, 63.71502685546875, 77.37751007080078], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000154.npy"}
{"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 27.467987060546875, "std": 50.45574188232422, "min": -100.194580078125, "p10": -27.571904754638666, "median": 22.937868118286133, "p90": 90.87331237792968, "max": 214.01824951171875, "pos_frac": 0.703125, "sample": [95.029541015625, 0.9569091796875, 17.67242431640625, 1.1246185302734375, 69.81673431396484, 0.2770729064941406, 35.97369384765625, 22.9410400390625, 23.083343505859375, 77.08332824707031, 100.59967041015625, 3.0882091522216797, 42.0694580078125, -7.248443603515625, 25.86090087890625, 93.56430053710938, -5.4276123046875, 101.6185302734375, 86.3875732421875, -7.04669189453125, 12.02130126953125, 121.29209899902344, 68.65458679199219, 133.5584716796875, -22.778244018554688, 85.78363037109375, -38.72979736328125, 0.0, -3.35382080078125, -15.009124755859375, -2.8539276123046875, 170.90887451171875, 78.15383911132812, 73.99861145019531, 9.841766357421875, -100.194580078125, -70.97489929199219, 11.633918762207031, -30.616455078125, -2.1518707275390625, -11.30908203125, 35.051788330078125, 28.1343994140625, 79.15293884277344, 100.00994873046875, 68.3258056640625, 46.76324462890625, 50.928619384765625, 11.131004333496094, 22.934696197509766, -9.747421264648438, -16.5115966796875, 5.138511657714844, -12.4857177734375, 9.541023254394531, -56.2109375, 18.033782958984375, 37.69488525390625, 111.82681274414062, 125.54551696777344, 0.7297210693359375, -24.676895141601562, 56.594207763671875, 7.3468780517578125, -12.596237182617188, -13.601200103759766, 65.69503784179688, 109.34896850585938, 28.770843505859375, 26.14899444580078, 6.5027008056640625, -33.88340377807617, 63.06329345703125, -8.674072265625, -35.20829772949219, 51.161102294921875, 22.280181884765625, 42.25914001464844, 188.92034912109375, -51.197021484375, 43.732269287109375, 54.05226135253906, -42.40118408203125, -6.788055419921875, 7.529571533203125, 65.58552551269531, 29.8758544921875, -20.14227294921875, -3.982696533203125, 21.189468383789062, 28.41009521484375, 29.650283813476562, 27.39666748046875, 20.53173828125, -10.844024658203125, 28.357879638671875, 31.833267211914062, 89.72003173828125, 11.446319580078125, 35.653778076171875, -22.201904296875, 15.838569641113281, -43.54681396484375, 65.45968627929688, 33.47296142578125, 60.7452392578125, 11.999473571777344, 4.330711364746094, -64.771484375, 3.4792251586914062, -57.12322998046875, 57.50959777832031, 26.766948699951172, -11.713348388671875, 59.768890380859375, -2.55645751953125, 69.47392272949219, 43.84151840209961, 24.448333740234375, 214.01824951171875, -37.099395751953125, 20.159774780273438, -26.26709747314453, 43.61761474609375, 32.548095703125, 44.4249267578125, 39.43840026855469, 79.49160766601562], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000155.npy"}
{"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 39.545082092285156, "std": 58.209068298339844, "min": -119.00877380371094, "p10": -30.552821350097652, "median": 31.77655029296875, "p90": 111.78767395019531, "max": 194.2027587890625, "pos_frac": 0.796875, "sample": [86.34705352783203, 96.138427734375, 26.346153259277344, 28.062469482421875, 6.8487548828125, 5.6746978759765625, 69.87098693847656, 61.2618408203125, 78.39146423339844, 28.62982177734375, 75.36290740966797, 41.43196105957031, 58.52679443359375, 47.28570556640625, 26.323081970214844, -119.00877380371094, -59.994537353515625, 40.26593017578125, 16.703636169433594, 12.346672058105469, -8.056121826171875, 60.86962890625, 22.098068237304688, 9.768516540527344, 28.6319580078125, 9.70465087890625, 2.6821823120117188, 24.674766540527344, -33.17909240722656, 75.02178955078125, 111.83969116210938, 0.2762908935546875, 85.06573486328125, 2.321563720703125, 2.024232864379883, -11.373275756835938, -43.452239990234375, -29.427276611328125, 68.84295654296875, -0.3768768310546875, 27.76117706298828, 66.08245849609375, 56.005645751953125, 9.10870361328125, -42.998748779296875, 49.57341003417969, 88.0406723022461, -25.789703369140625, 78.17832946777344, 52.251190185546875, 7.096958160400391, 44.66505432128906, 34.921142578125, 128.8720703125, 157.06625366210938, -7.27581787109375, -43.529571533203125, 166.31634521484375, -35.810333251953125, 37.184730529785156, 3.9810791015625, 161.37054443359375, 19.553115844726562, 81.07000732421875, 6.395744323730469, 99.66598510742188, 99.55538940429688, 100.14218139648438, 93.17861938476562, 1.34869384765625, 17.52374267578125, 49.42008972167969, -8.992164611816406, 58.698272705078125, 87.87168884277344, 51.872100830078125, 165.89898681640625, 19.1259765625, -1.0380516052246094, -9.114561080932617, 126.29873657226562, 126.97039794921875, 101.67355346679688, 0.176177978515625, 1.4550018310546875, 81.46456146240234, 7.320808410644531, 86.39208984375, 44.43896484375, 37.85075378417969, 50.52972412109375, 109.52178955078125, 105.6429443359375, 111.765380859375, 117.36981201171875, 26.81610107421875, -114.36788940429688, 24.50310516357422, 44.7750244140625, -0.2089996337890625, 54.379638671875, 25.194915771484375, -1.968658447265625, -36.926513671875, -117.645263671875, 129.5875244140625, 15.649763107299805, 88.66879272460938, 6.4633026123046875, -59.4215087890625, 16.93719482421875, 55.530242919921875, 112.029052734375, 67.1917724609375, -78.94873046875, -14.758552551269531, 19.08990478515625, 94.38618469238281, 81.3575439453125, 129.41958618164062, 15.564140319824219, 66.66537475585938, 104.29293823242188, 110.42837524414062, -17.71014404296875, 194.2027587890625, -48.92841339111328, 10.632080078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000156.npy"}
{"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 31.05329132080078, "std": 51.432960510253906, "min": -91.473388671875, "p10": -34.59219055175781, "median": 25.17890167236328, "p90": 101.13022460937499, "max": 171.9324951171875, "pos_frac": 0.7109375, "sample": [100.75262451171875, 7.5227508544921875, 99.695068359375, 104.3961181640625, 28.918731689453125, -15.603904724121094, 18.053619384765625, 96.50344848632812, 33.8343505859375, 47.027435302734375, -11.5379638671875, -12.805770874023438, -60.359283447265625, 21.401824951171875, 10.005943298339844, 32.188232421875, -9.684898376464844, -12.781524658203125, 10.831293106079102, -66.65277099609375, -19.546844482421875, 109.84130859375, 25.129104614257812, -25.592193603515625, 112.08438873291016, 6.119697570800781, -66.33981323242188, 15.23974609375, 25.741653442382812, 90.43026733398438, -9.321380615234375, -57.31939697265625, 7.911651611328125, -1.47576904296875, 39.83979797363281, 143.66140747070312, 65.96356201171875, 24.398391723632812, -5.3318328857421875, -65.06735229492188, 171.9324951171875, 63.77046203613281, 48.31719970703125, 4.64703369140625, 124.76910400390625, 25.22869873046875, 24.0255126953125, 19.977741241455078, -3.614898681640625, 0.0, 41.28944396972656, 37.692596435546875, 24.813522338867188, -42.9219970703125, 128.05059814453125, 15.78826904296875, 76.34556579589844, -0.3106689453125, 14.5455322265625, -9.423667907714844, 65.28485107421875, 18.57879638671875, 106.63970947265625, -0.10333251953125, 31.19849395751953, -11.462137222290039, 78.98162841796875, -6.84210205078125, -78.92218017578125, 50.2127685546875, 21.061279296875, 26.18634033203125, -34.5291748046875, -42.593048095703125, -8.776763916015625, 11.7745361328125, -22.56768798828125, 31.10284423828125, 68.4290771484375, 75.29779052734375, 66.94622802734375, 48.92987060546875, -6.7242431640625, -1.72662353515625, 32.2777099609375, 106.81068420410156, 6.87493896484375, 113.08187866210938, 82.57279968261719, 0.0, -36.76971435546875, 77.38510131835938, 86.38191223144531, 69.400390625, 27.42303466796875, 7.0089874267578125, 66.03778076171875, 61.871307373046875, 138.8546142578125, 76.62470245361328, 46.7646484375, -55.38270568847656, 47.8751220703125, 43.75738525390625, 19.787017822265625, 14.853179931640625, 92.3612289428711, 50.65909194946289, 43.32550048828125, -5.104888916015625, -34.739227294921875, 67.38299560546875, 7.41778564453125, 67.96186828613281, 87.59928894042969, 35.135284423828125, 6.572296142578125, 10.054985046386719, -91.473388671875, 102.01129150390625, -57.64935302734375, 40.87329864501953, 7.640296936035156, 74.72097778320312, 26.898849487304688, 94.57460021972656, 145.72381591796875, 80.01266479492188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000157.npy"}
{"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 33.56697082519531, "std": 53.50319290161133, "min": -165.627197265625, "p10": -27.33086547851562, "median": 33.19335746765137, "p90": 95.60675048828125, "max": 190.21917724609375, "pos_frac": 0.7578125, "sample": [25.896509170532227, 43.702423095703125, 32.60882568359375, 40.59149169921875, -26.126983642578125, 30.813461303710938, 1.1159820556640625, -13.201446533203125, -77.46115112304688, 74.921142578125, 23.28021240234375, 17.8443603515625, -24.4334716796875, -67.94927978515625, 118.77078247070312, 29.855926513671875, 62.070213317871094, 52.045188903808594, 64.09002685546875, 69.18045043945312, 15.446624755859375, 69.10293579101562, 93.16604614257812, -0.3106536865234375, 45.56669616699219, 105.72763061523438, 95.96563720703125, -165.627197265625, 37.9293212890625, 35.57582092285156, -44.432098388671875, 87.64064025878906, -4.325347900390625, 50.20387268066406, 26.1981201171875, 86.23519134521484, -2.961639404296875, -88.35279083251953, 50.852378845214844, -34.391510009765625, 68.20849609375, 32.2547607421875, 44.50726318359375, -33.0822868347168, -12.70062255859375, -6.662483215332031, -30.139923095703125, 4.19146728515625, 34.026458740234375, 33.24825668334961, 64.70428466796875, -8.696046829223633, -9.794631958007812, 17.088172912597656, 50.23102569580078, 125.85018920898438, -50.615516662597656, -14.54693603515625, 95.45294189453125, 38.070770263671875, 151.65652465820312, 23.250686645507812, -11.18255615234375, 41.95550537109375, 90.7076416015625, 9.380859375, -13.04498291015625, 0.0, 28.138084411621094, 35.733001708984375, 61.76568603515625, 3.7456207275390625, 6.027435302734375, 56.70703125, 95.16693115234375, 54.041473388671875, -25.119461059570312, 22.82685089111328, -76.46052551269531, -22.730810165405273, 134.15414428710938, 2.732818603515625, -73.42222595214844, 8.422348022460938, 19.875167846679688, 190.21917724609375, 73.53680419921875, 4.160228729248047, 76.64254760742188, 134.38717651367188, 50.621734619140625, 86.49356079101562, 46.906951904296875, 19.524993896484375, 111.11795043945312, 59.4024658203125, 100.27032470703125, 30.83669662475586, 10.15203857421875, 7.701446533203125, 95.98554992675781, 85.67916870117188, 67.673583984375, 91.31283569335938, 116.03456115722656, 55.4617919921875, 41.08624267578125, 94.99334716796875, 43.451629638671875, -24.909610748291016, 28.66058349609375, 36.5086669921875, 80.70527648925781, 33.138458251953125, 55.44915771484375, 26.162796020507812, 75.58468627929688, 0.9102897644042969, -12.610000610351562, -34.18011474609375, 74.56646728515625, 81.74911499023438, -70.8594970703125, 28.373321533203125, 122.029052734375, 56.239593505859375, 32.1354866027832, 10.51828384399414], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000158.npy"}
{"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 22.907712936401367, "std": 58.39735412597656, "min": -133.72003173828125, "p10": -43.85323028564453, "median": 20.014591217041016, "p90": 92.18367843627928, "max": 297.40435791015625, "pos_frac": 0.6875, "sample": [-6.543914794921875, 34.4189453125, 47.13092041015625, 19.4466552734375, 17.06092071533203, -41.4390869140625, -52.623565673828125, 106.915771484375, 22.989532470703125, 86.58102416992188, -59.81336975097656, 18.78412628173828, -133.72003173828125, -40.311134338378906, 12.146369934082031, 297.40435791015625, 20.58252716064453, -3.0655670166015625, -46.71868896484375, 89.99958038330078, 23.411026000976562, 5.9637603759765625, -105.9183349609375, -36.58165740966797, 39.89569091796875, 89.87396240234375, 4.5470733642578125, 75.85685729980469, 31.1981201171875, -13.4324951171875, -0.4932575225830078, 34.317779541015625, 31.601959228515625, -1.5845718383789062, 29.583038330078125, -38.30937194824219, 57.65704345703125, -16.9273681640625, -5.155780792236328, -41.707183837890625, 36.146217346191406, 9.9306640625, 26.650543212890625, 70.08990478515625, 9.680999755859375, 48.33001708984375, 1.9151458740234375, -4.877021789550781, -76.33612060546875, -19.736045837402344, 6.407371520996094, -33.97093200683594, 75.797607421875, 15.818389892578125, 65.18878173828125, 73.73564147949219, 5.366203308105469, 43.85896301269531, 1.984039306640625, 4.808380126953125, 106.66792297363281, 49.10009765625, -23.768234252929688, 15.4674072265625, 76.75091552734375, 55.236358642578125, 25.689361572265625, 22.93145751953125, 23.28662109375, 13.21954345703125, 2.25860595703125, -4.797607421875, -34.22780227661133, 102.30570983886719, 32.610595703125, -60.08453369140625, -44.76252746582031, 35.32012939453125, -47.384796142578125, 21.097015380859375, 56.587310791015625, -51.21697998046875, 62.223663330078125, 119.34637451171875, 126.82511901855469, 9.078140258789062, -15.103019714355469, -26.897171020507812, 5.9674224853515625, 99.12786865234375, 57.97709655761719, 35.66084289550781, -41.380584716796875, 27.695648193359375, 39.18841552734375, 35.57073974609375, 39.688751220703125, 70.2430419921875, -10.117687225341797, 6.8478851318359375, 32.389892578125, -71.37794494628906, -115.636474609375, 28.0238037109375, 129.88287353515625, 174.24200439453125, 65.66119384765625, -41.02485656738281, 121.0392837524414, 12.808174133300781, -43.463531494140625, -31.9970703125, 85.81341552734375, -47.655853271484375, 126.095458984375, 1.462921142578125, 68.57464599609375, 67.37579345703125, 97.2799072265625, 103.77886962890625, 6.508056640625, 59.25767135620117, -23.517105102539062, 83.60040283203125, 57.06572723388672, 7.097442626953125, -1.985565185546875, 52.84861755371094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000159.npy"}
{"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 39.104652404785156, "std": 47.29187774658203, "min": -85.22567749023438, "p10": -17.1716796875, "median": 40.29277038574219, "p90": 94.90125350952148, "max": 170.45989990234375, "pos_frac": 0.75, "sample": [7.042266845703125, 57.044525146484375, 0.0, 106.860107421875, 10.76849365234375, 52.015838623046875, 62.513580322265625, 55.15008544921875, 100.98187255859375, 55.7586669921875, 70.598876953125, 94.08380126953125, 25.983596801757812, 80.773193359375, -13.549896240234375, 38.926300048828125, 136.08499145507812, 34.307830810546875, 25.127731323242188, 20.019927978515625, 56.973236083984375, 45.033111572265625, 73.195556640625, 71.99986267089844, 47.19563293457031, 3.0128326416015625, -1.6688995361328125, 82.21238708496094, -12.727289199829102, -26.03228759765625, 80.76974487304688, 148.29669189453125, -85.22567749023438, 156.03823852539062, 45.30558776855469, -0.6006870269775391, 42.454383850097656, -11.92056655883789, 81.42987060546875, -5.793230056762695, 82.35155487060547, 73.20584106445312, 32.4344482421875, 0.0, 48.74151611328125, -10.36651611328125, 59.534156799316406, -19.689926147460938, -38.446990966796875, 37.43351745605469, 57.30328369140625, 51.559356689453125, 170.45989990234375, 79.10076904296875, -12.991668701171875, -16.9654541015625, 55.110015869140625, -23.84149169921875, 5.699287414550781, 25.395706176757812, 96.205810546875, 94.77452850341797, 35.7396240234375, -56.8221435546875, -10.380218505859375, 90.53190612792969, 72.4180908203125, 95.19694519042969, 14.450544357299805, -31.77728271484375, 73.080322265625, 9.360939025878906, 7.332027435302734, 50.935089111328125, 38.82243347167969, 74.45338439941406, 60.24847412109375, 61.10193634033203, 114.896728515625, 45.37762451171875, -17.407028198242188, -17.070816040039062, 88.0791015625, -26.393688201904297, -4.873565673828125, 94.45504760742188, 18.007179260253906, 19.06854248046875, 74.15158081054688, 17.6719970703125, 19.03314208984375, 65.59490966796875, 89.58773803710938, -27.28753662109375, 48.03106689453125, -39.66461181640625, 1.8943939208984375, -4.497039794921875, 61.48028564453125, 2.6792449951171875, 25.472885131835938, 21.4136962890625, -18.48416519165039, 30.302703857421875, 34.55181884765625, -1.27093505859375, 15.84750747680664, -1.9616241455078125, 41.65924072265625, 93.6182861328125, 131.99615478515625, -12.585968017578125, -1.99090576171875, 77.71177673339844, 19.489776611328125, 55.20288848876953, 149.20907592773438, 11.64263916015625, 52.55400085449219, 108.02732849121094, 68.62747192382812, 50.286102294921875, -51.080810546875, 97.3416748046875, 59.63935852050781, 68.47048950195312, 8.061294555664062, 6.689321517944336], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000160.npy"}
{"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 34.65388488769531, "std": 57.06039810180664, "min": -95.51483154296875, "p10": -35.87456359863281, "median": 33.82870864868164, "p90": 98.52132110595703, "max": 282.7989501953125, "pos_frac": 0.75, "sample": [-1.023681640625, 39.407073974609375, -4.2498779296875, -73.67509460449219, 13.22773551940918, 41.530487060546875, 15.384933471679688, 0.05694580078125, -22.46990966796875, -3.51959228515625, 7.583353042602539, 97.34109497070312, 77.33413696289062, -56.533172607421875, 51.131561279296875, 2.0568618774414062, 111.65277099609375, 3.3004150390625, 30.63665008544922, -8.403167724609375, 7.580047607421875, -0.456512451171875, 140.92483520507812, 85.4541015625, -36.710777282714844, 73.06817626953125, 32.83629608154297, 75.02813720703125, 78.2987060546875, 42.44597625732422, 26.46429443359375, 64.0649185180664, 28.04974365234375, -60.24476623535156, 24.45363998413086, 85.28359985351562, 42.659423828125, 61.87353515625, 4.105628967285156, 48.70280075073242, -5.38653564453125, -93.60499572753906, -43.8477783203125, -11.069267272949219, 5.8952178955078125, 52.86224365234375, -40.197662353515625, 13.6282958984375, 0.0, 88.47364807128906, 3.214935302734375, 65.40130615234375, 19.46105194091797, 4.417928695678711, 19.25079345703125, 82.24581909179688, 105.34065246582031, 79.94534301757812, 42.29046630859375, 31.873504638671875, 3.39471435546875, 104.31787109375, -27.4212646484375, 21.741180419921875, -29.918792724609375, 63.150421142578125, 41.252777099609375, 86.14297485351562, -95.06802368164062, 36.361572265625, 179.52471923828125, 36.91532516479492, 46.51092529296875, 186.69192504882812, -36.4404296875, 95.81927490234375, 89.96209716796875, -67.24673461914062, 6.31036376953125, 93.23064422607422, -50.597618103027344, 0.8697509765625, 282.7989501953125, 113.52734375, 55.94354248046875, 104.68362426757812, 64.35218811035156, 62.501495361328125, 65.05328369140625, -6.129922866821289, 55.447052001953125, 78.4129638671875, -21.958709716796875, 87.69268798828125, 7.1710205078125, 9.2017822265625, 53.919677734375, 18.691864013671875, 54.608306884765625, -2.4161376953125, 20.637725830078125, 109.67416381835938, 45.90667724609375, -35.632049560546875, 76.57007598876953, 55.83113098144531, -28.577346801757812, 15.059326171875, 46.47020721435547, 69.989990234375, 31.91595458984375, 92.1900634765625, -6.825645446777344, 56.041290283203125, 58.52052307128906, -95.51483154296875, 105.02163696289062, 108.81130981445312, 34.82112121582031, 0.0, 36.790283203125, 97.7750244140625, 100.26268005371094, -78.2358627319336, 22.28082275390625, 18.8616943359375, -3.3154296875, 46.5157470703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000161.npy"}
{"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 31.066944122314453, "std": 52.47015380859375, "min": -105.78121948242188, "p10": -26.778192138671873, "median": 28.157470703125, "p90": 94.07016296386718, "max": 228.36447143554688, "pos_frac": 0.7265625, "sample": [-23.950607299804688, 7.210544586181641, 123.44631958007812, 46.89476776123047, -10.616806030273438, 25.28021240234375, -54.3782958984375, 34.35076904296875, 8.324710845947266, 46.30992126464844, 32.67113494873047, 92.1751708984375, 9.262699127197266, 54.20470428466797, -17.9669189453125, -24.489044189453125, 21.3612060546875, 28.03778076171875, -7.149017333984375, -0.4592704772949219, -53.06964111328125, 72.73963928222656, 10.13861083984375, 67.68058013916016, 41.722320556640625, -20.494361877441406, 92.83514404296875, 13.14288330078125, -27.51019287109375, 18.11273956298828, -10.853157043457031, 59.54443359375, 82.38714599609375, -24.843338012695312, 130.43328857421875, 38.45343017578125, 35.22760009765625, -103.00909423828125, -1.4720458984375, 114.26700592041016, -0.7513656616210938, 102.8634033203125, 0.8560619354248047, 32.5631103515625, 145.10821533203125, -92.05239868164062, 105.60653686523438, 14.999053955078125, 73.96426391601562, 84.60275268554688, 49.37873077392578, 37.951904296875, 47.831024169921875, 15.05084228515625, 92.12034606933594, -0.6707344055175781, 7.831047058105469, 62.772705078125, 74.26177978515625, 35.332183837890625, 228.36447143554688, 127.87966918945312, -10.48748779296875, -14.744583129882812, 88.65289306640625, 117.29832458496094, 88.93788146972656, 77.05818939208984, -28.675048828125, 63.60188293457031, 2.8647079467773438, 25.778907775878906, -105.78121948242188, 31.780731201171875, 101.37408447265625, -50.59054946899414, -35.579833984375, 8.7110595703125, 13.4752197265625, 54.33778381347656, -12.108955383300781, 74.87860107421875, 86.71293640136719, 34.18501281738281, 90.58331298828125, 38.652618408203125, 49.020755767822266, 14.00970458984375, 49.89349365234375, 74.16172790527344, -8.875778198242188, 57.69952392578125, 27.60089111328125, -74.50595092773438, 25.545654296875, 17.15911865234375, 66.06857299804688, 24.17327880859375, 28.27716064453125, 75.61398315429688, 45.999488830566406, -28.9735107421875, 72.52703857421875, 2.7537384033203125, 1.773193359375, 115.30908203125, 2.0203857421875, -36.15226745605469, -9.807716369628906, 76.97119140625, -64.13771057128906, -6.8757476806640625, -20.817062377929688, 42.0909423828125, 46.716522216796875, 99.60649108886719, 36.8851432800293, 6.598175048828125, 40.153839111328125, 25.17724609375, 67.66756439208984, -26.4644775390625, 26.241172790527344, 10.298980712890625, -21.97638702392578, 96.95187377929688, -12.910736083984375, 50.36717224121094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000162.npy"}
{"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 33.129638671875, "std": 49.83983612060547, "min": -98.49734497070312, "p10": -30.006149291992188, "median": 29.666377067565918, "p90": 95.04928741455078, "max": 166.1588134765625, "pos_frac": 0.7578125, "sample": [6.524627685546875, 29.758907318115234, 75.54735565185547, -21.833160400390625, 55.89581298828125, -15.940170288085938, 29.440155029296875, 7.781097412109375, 8.3487548828125, -56.731536865234375, 5.9544677734375, 94.01480102539062, 129.00636291503906, 22.91302490234375, 106.56170654296875, -29.885986328125, 85.51593017578125, 26.6702880859375, -3.289947509765625, 31.521713256835938, -24.031494140625, 74.23936462402344, 57.75201416015625, -12.962730407714844, 10.0230712890625, 45.355865478515625, -6.455724716186523, -11.568130493164062, 87.55816650390625, -3.458099365234375, 23.087051391601562, 115.65536499023438, -35.850128173828125, 58.417755126953125, 54.84405517578125, 15.10992431640625, 31.162445068359375, 47.364501953125, -67.1514892578125, -6.410797119140625, 18.170684814453125, 35.87596130371094, 4.719207763671875, 54.154136657714844, -2.613067626953125, 97.46308898925781, -88.36083984375, 80.49853515625, 72.6747055053711, 45.02886962890625, 76.99337768554688, 4.471923828125, 28.753204345703125, 103.47163391113281, 47.75987243652344, 144.5542755126953, 1.954498291015625, 37.18017578125, 98.97940063476562, 67.85757446289062, 12.612640380859375, 6.56170654296875, 27.2069091796875, 2.3610763549804688, -7.905464172363281, -30.286529541015625, 18.3460750579834, 74.85250091552734, 41.071044921875, 88.20597839355469, -60.8175048828125, 89.92573547363281, 65.22674560546875, 101.22369384765625, -28.610595703125, -43.151763916015625, 71.6998291015625, -25.6273193359375, 93.50555419921875, 85.15509033203125, 5.1948699951171875, 61.797607421875, 22.63543701171875, -10.096263885498047, 48.10899353027344, 40.98817443847656, 64.11981201171875, 85.69813537597656, 166.1588134765625, 67.501708984375, 12.663156509399414, 20.3704833984375, 88.23294067382812, 89.8746337890625, 20.37769317626953, -2.7158203125, -80.54608917236328, 35.675289154052734, 115.83364868164062, 23.996734619140625, 78.2523193359375, 83.380859375, -37.380096435546875, 62.859375, -33.02268981933594, 22.76666259765625, 67.39404296875, 4.9196014404296875, 2.3575286865234375, 2.3511810302734375, 37.15985107421875, 54.434417724609375, 111.66336059570312, 24.85445785522461, 57.85377502441406, -24.638916015625, 40.37567138671875, 105.80169677734375, 56.4136962890625, 69.63442993164062, -4.462486267089844, -54.3123779296875, -98.49734497070312, 60.10993957519531, 104.06033325195312, 21.115829467773438, 29.5738468170166, -31.892982482910156], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000163.npy"}
{"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 32.83057403564453, "std": 55.57064437866211, "min": -118.76934814453125, "p10": -28.340866088867188, "median": 24.42822265625, "p90": 107.60027313232422, "max": 216.09844970703125, "pos_frac": 0.7109375, "sample": [0.900970458984375, 38.48809051513672, 5.366945266723633, 138.85415649414062, 5.994831085205078, -5.8775482177734375, 21.119510650634766, -41.19664001464844, 70.87522888183594, 96.72821044921875, 77.274169921875, -7.74072265625, -10.8946533203125, 51.28253173828125, 52.320587158203125, -118.76934814453125, 10.229904174804688, 133.1800537109375, 2.764627456665039, 98.93096923828125, 14.16302490234375, -50.364044189453125, 99.82716369628906, -49.154815673828125, 25.77508544921875, -9.267936706542969, 49.29371643066406, -13.02825927734375, -35.251220703125, 82.18238830566406, 32.83868408203125, 52.18426513671875, 40.16912841796875, 97.84471130371094, -26.41485595703125, 68.8597412109375, -62.41712951660156, 10.09796142578125, 4.50799560546875, -56.193389892578125, -27.066558837890625, 79.71632385253906, 56.432586669921875, 38.023193359375, 216.09844970703125, 120.72024536132812, 53.822967529296875, -1.392333984375, 7.122283935546875, 12.476593017578125, -15.284027099609375, 0.05495452880859375, 15.25299072265625, 4.176731109619141, -94.6767578125, 137.58157348632812, 10.307205200195312, 101.25763702392578, 55.24028015136719, 41.938751220703125, -39.14283752441406, 63.53265380859375, 102.29592895507812, 28.259979248046875, -23.53265380859375, 78.96540832519531, 4.2974700927734375, -34.327239990234375, 65.99224853515625, 14.969146728515625, 119.6700439453125, 110.4429702758789, 87.76077270507812, -9.165283203125, -68.47819519042969, 18.028076171875, 30.886489868164062, 0.29864501953125, -10.688385009765625, -5.851020812988281, 36.3621826171875, 76.26841735839844, -29.432373046875, 55.138816833496094, 17.564651489257812, 38.55982971191406, 111.32308959960938, 83.66159057617188, 81.84393310546875, 64.79034423828125, 0.9694805145263672, 107.31346130371094, -7.255775451660156, 98.75213623046875, -9.479583740234375, 23.08135986328125, 114.0763168334961, 45.218017578125, -22.896697998046875, -0.661346435546875, -4.76605224609375, 48.686553955078125, 17.31494140625, 18.398956298828125, 3.8485565185546875, -42.895599365234375, 123.09051513671875, -27.873077392578125, 87.76860046386719, 37.773162841796875, -8.6146240234375, 57.092193603515625, -22.541400909423828, 62.133026123046875, -21.0709228515625, -25.300628662109375, 34.849830627441406, -6.761077880859375, 103.36572265625, 11.45391845703125, 48.01149368286133, 52.109100341796875, 108.26950073242188, 88.83599853515625, 35.6630859375, 143.7375030517578, 144.810302734375, 8.224824905395508], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000164.npy"}
{"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 24.680496215820312, "std": 48.20048904418945, "min": -72.17855834960938, "p10": -30.062528228759767, "median": 17.78612518310547, "p90": 89.36728439331054, "max": 158.4704132080078, "pos_frac": 0.703125, "sample": [17.513214111328125, 63.0992431640625, 4.805088043212891, 36.957275390625, -61.684814453125, 3.308868408203125, 15.983078002929688, 39.7808837890625, 122.08770751953125, 70.5390625, 63.4713134765625, 11.736419677734375, -72.17855834960938, 42.942726135253906, 71.43194580078125, -8.13433837890625, -54.42041015625, 33.7705078125, -5.4667510986328125, 152.5206298828125, 25.737869262695312, -62.482666015625, 158.4704132080078, 45.295509338378906, 40.48194885253906, -58.30388259887695, -13.89886474609375, -22.233993530273438, 33.064605712890625, 56.2384033203125, -13.686614990234375, 20.15899658203125, -35.2227783203125, -12.298095703125, 4.92376708984375, 18.059036254882812, 58.893836975097656, -2.9379730224609375, -17.496782302856445, -46.81634521484375, 22.98193359375, 24.102127075195312, 118.90728759765625, 15.364501953125, 15.13336181640625, 99.0367431640625, -0.5030517578125, -10.718452453613281, 2.262035369873047, 23.779037475585938, 116.8426513671875, 20.89476776123047, 18.768783569335938, -12.319305419921875, 34.71539306640625, -3.9116973876953125, -57.501922607421875, 29.053857803344727, -21.329132080078125, 1.399810791015625, -18.33221435546875, 48.41827392578125, 40.075233459472656, 72.05453491210938, 21.930694580078125, -30.195770263671875, 41.32167053222656, 13.400192260742188, -11.801116943359375, -16.798797607421875, 14.577972412109375, -18.382980346679688, 12.085235595703125, 12.719348907470703, 70.77813720703125, 2.574981689453125, 97.16545104980469, -1.26068115234375, 40.191864013671875, 80.78924560546875, -10.064804077148438, 72.21112823486328, 89.81845092773438, 4.1011962890625, 7.666484832763672, 17.295501708984375, 21.55066680908203, 82.35980224609375, 32.703369140625, 136.0523681640625, 14.859161376953125, 1.014404296875, 30.793304443359375, 40.360591888427734, 110.15087890625, 35.067657470703125, 12.89666748046875, 0.19830322265625, 86.09547424316406, 10.867912292480469, 84.33392333984375, -15.18621826171875, 98.7592544555664, 89.1739273071289, 118.6503677368164, -24.7999267578125, 19.18167495727539, 65.90357971191406, 62.561981201171875, -68.79922485351562, 2.911092758178711, -6.512851715087891, 124.56568908691406, 32.465667724609375, 84.9619140625, 85.37985229492188, 35.652374267578125, -30.00542449951172, 4.09344482421875, 53.46240997314453, 6.6504669189453125, 72.04368591308594, -61.15818786621094, -38.834075927734375, -25.29620361328125, -18.054718017578125, -39.515655517578125, 18.236541748046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000165.npy"}
{"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 28.706207275390625, "std": 52.7598876953125, "min": -84.57852172851562, "p10": -30.344671630859374, "median": 16.108901977539062, "p90": 101.16102905273438, "max": 205.08807373046875, "pos_frac": 0.703125, "sample": [20.0045166015625, 9.83514404296875, -26.270492553710938, 41.086090087890625, -2.17706298828125, 25.673782348632812, 162.38699340820312, 83.89273071289062, 4.090240478515625, 15.485401153564453, 39.77845001220703, -15.330230712890625, 126.48489379882812, -67.74544525146484, 3.1314010620117188, -20.100555419921875, 38.55079650878906, 90.94198608398438, 13.020092010498047, -14.698799133300781, -8.389907836914062, -58.411651611328125, 32.00297546386719, 68.66416931152344, -50.771759033203125, 38.88629150390625, 4.0329742431640625, 57.221961975097656, -47.8243408203125, 9.066375732421875, 3.6160888671875, 58.95074462890625, 101.0804443359375, 151.8834228515625, 107.63182067871094, 91.816162109375, 14.562522888183594, 11.483673095703125, -35.58831024169922, -66.7691650390625, 2.5300159454345703, -13.10601806640625, -0.2248382568359375, -9.0172119140625, -27.78656005859375, 108.64788818359375, 67.12237548828125, 92.71446228027344, -30.182769775390625, 47.801361083984375, 33.00562286376953, 119.19107055664062, -0.738494873046875, 14.724853515625, 155.52294921875, -84.57852172851562, -26.1058349609375, 101.34906005859375, 16.440673828125, 79.23379516601562, -23.73125457763672, -15.562774658203125, 22.260986328125, 41.25859069824219, 47.1004638671875, -63.52667236328125, 80.342529296875, 16.4976806640625, 49.073638916015625, -7.078460693359375, 29.215866088867188, 56.6300048828125, 113.48394775390625, -11.435861587524414, 15.163818359375, 50.23283386230469, 59.10478973388672, 1.5166168212890625, 64.89887237548828, -21.247772216796875, -31.373971939086914, 41.753387451171875, 14.579605102539062, 2.241180419921875, 9.632659912109375, 92.53947448730469, 37.75714111328125, 9.839340209960938, -30.722442626953125, -46.26268005371094, -11.400421142578125, 2.0034103393554688, 3.9425086975097656, 91.15054321289062, 101.041259765625, 73.13102722167969, 26.076171875, 47.077354431152344, 60.51219177246094, 35.880767822265625, 49.1072998046875, -4.199857711791992, -7.6544952392578125, 34.282615661621094, -0.527587890625, 53.00274658203125, 58.28663635253906, 99.162841796875, 7.834857940673828, 109.86929321289062, -9.157958984375, -78.3145751953125, 15.777130126953125, 61.1246337890625, 107.28826904296875, -39.73493194580078, -1.5673828125, -8.0447998046875, 50.4774169921875, 3.214141845703125, 119.299072265625, 3.2770843505859375, 205.08807373046875, 7.954904556274414, 8.939712524414062, 60.884185791015625, 59.924163818359375, 17.476165771484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000166.npy"}
{"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 30.1601505279541, "std": 54.689170837402344, "min": -103.73263549804688, "p10": -35.790359497070305, "median": 31.763771057128906, "p90": 93.75550842285155, "max": 195.66555786132812, "pos_frac": 0.7265625, "sample": [78.19258117675781, 5.503456115722656, -4.82012939453125, 48.990692138671875, 38.27314758300781, 65.88192749023438, 75.80023193359375, 194.33291625976562, 76.10148620605469, 117.41603088378906, 31.408706665039062, -63.46173095703125, 52.85797119140625, 3.1056957244873047, 43.78187561035156, 78.66195678710938, 35.046539306640625, 119.70661926269531, 32.11883544921875, 47.30169677734375, 43.716644287109375, 44.048858642578125, -97.09001922607422, -39.17938232421875, 0.8507919311523438, -8.366729736328125, -13.801422119140625, 62.208587646484375, 35.005767822265625, 77.34294128417969, 56.132293701171875, 118.56483459472656, 86.8691635131836, 110.07672119140625, -0.6950225830078125, 10.299270629882812, 81.30233764648438, -31.890167236328125, 64.00314331054688, -9.619827270507812, 76.32839965820312, -3.616973876953125, -34.337921142578125, 37.588653564453125, -7.733917236328125, -103.73263549804688, 0.0, 44.28411865234375, 80.68704223632812, 58.819801330566406, -20.001449584960938, 77.962158203125, -74.88958740234375, 71.8033447265625, 70.74984741210938, 2.1571044921875, 99.78905487060547, -9.343414306640625, 12.239410400390625, -3.1604766845703125, -59.293609619140625, -8.5355224609375, 95.54385375976562, 0.02489471435546875, 73.49267578125, 17.573776245117188, 62.21685791015625, -26.26605224609375, 70.75961303710938, 9.330425262451172, 1.3190765380859375, -1.8422622680664062, 23.501197814941406, -28.899017333984375, 86.25939178466797, 46.772979736328125, 69.51821899414062, 1.991424560546875, 106.72613525390625, 38.350860595703125, 10.061668395996094, 99.2728271484375, 61.70794677734375, -54.40069580078125, 91.33749389648438, -31.730987548828125, 36.28236389160156, 0.7502079010009766, 28.306182861328125, 44.953834533691406, 45.38214111328125, 8.971633911132812, 42.27748489379883, 37.4383544921875, 88.41471862792969, 133.3079833984375, -14.965019226074219, -65.67750549316406, 6.3893890380859375, -23.774917602539062, 8.456146240234375, 10.716861724853516, -66.46487426757812, 125.65625, -32.43727111816406, -54.019256591796875, 137.25421142578125, 20.851776123046875, 92.98907470703125, -50.73963928222656, 14.041748046875, 86.441162109375, -78.86027526855469, 10.314682006835938, 26.795028686523438, 14.770339965820312, 15.236083984375, -9.1337890625, 41.9598388671875, 54.62208557128906, 43.77949523925781, 61.724945068359375, 8.302993774414062, 20.067588806152344, -61.11370849609375, 46.614044189453125, 12.586196899414062, 195.66555786132812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000167.npy"}
{"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 33.709434509277344, "std": 55.01049041748047, "min": -169.374755859375, "p10": -29.3632827758789, "median": 34.69656753540039, "p90": 106.16971282958984, "max": 148.69009399414062, "pos_frac": 0.75, "sample": [90.21347045898438, 18.82183837890625, 117.85757446289062, 97.50689697265625, 115.39129638671875, 33.228240966796875, -169.374755859375, 41.495994567871094, 57.302398681640625, 10.379081726074219, 145.26788330078125, -3.20538330078125, -38.089996337890625, 25.708370208740234, -6.532096862792969, -1.3341217041015625, 16.580230712890625, 18.240692138671875, -27.513702392578125, 69.05685424804688, 50.37635040283203, 53.64080810546875, 36.262481689453125, 32.9195556640625, 106.76498413085938, 1.3430156707763672, 15.927139282226562, 64.7586441040039, -39.937103271484375, 0.823486328125, 0.0, 34.22711944580078, 80.44207763671875, 9.639068603515625, 65.9775390625, 81.77069091796875, 21.702972412109375, -18.884796142578125, 48.897979736328125, 4.4466552734375, 87.462158203125, 41.79158020019531, 0.05950927734375, 58.829345703125, -33.67897033691406, 105.0972900390625, 26.1142578125, 43.650665283203125, 130.37606811523438, -52.37744140625, 139.7314453125, -71.5263671875, 4.067756652832031, 68.55242919921875, -73.40142822265625, 63.68950653076172, 66.78085327148438, 61.36505126953125, 90.7227554321289, 10.446792602539062, -58.882537841796875, -21.24011993408203, 6.7696533203125, 55.51225280761719, 68.087890625, 0.5039920806884766, 126.60498046875, 42.96148681640625, 148.69009399414062, 36.434967041015625, 70.97802734375, -0.7056465148925781, -67.44540405273438, 65.84342956542969, 85.546875, -0.5318355560302734, 44.40074920654297, 82.47732543945312, 12.639198303222656, -11.03326416015625, -5.1280517578125, 45.92585754394531, 99.39852905273438, 40.50653076171875, 54.691558837890625, 119.7220458984375, 68.62228393554688, 51.250274658203125, 112.82949829101562, 105.91459655761719, -86.56007385253906, -0.0428466796875, 51.4964599609375, -11.030609130859375, 14.316482543945312, 78.48922729492188, -12.113388061523438, 39.965484619140625, 13.736083984375, -49.8330078125, 81.3367919921875, 25.8797607421875, 83.71702575683594, 100.6595458984375, 51.375816345214844, 36.35247802734375, -12.623550415039062, 33.347076416015625, 51.45084762573242, 4.644599914550781, -11.265869140625, 4.72149658203125, 21.2689208984375, -59.231781005859375, 113.41156005859375, 26.3575439453125, 35.166015625, 119.04446411132812, 88.5087890625, 63.331695556640625, -23.89898681640625, 24.834564208984375, 13.77020263671875, -25.293212890625, -22.235321044921875, 17.175941467285156, 126.47720336914062, -103.0980224609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000168.npy"}
{"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 46.15138626098633, "std": 51.94218063354492, "min": -53.81207275390625, "p10": -10.151427459716796, "median": 44.057762145996094, "p90": 118.24467468261717, "max": 212.04473876953125, "pos_frac": 0.7734375, "sample": [-7.0463409423828125, 123.6915054321289, 117.05059814453125, 68.9134521484375, -0.97113037109375, 2.717144012451172, -7.848428726196289, 42.90425109863281, -10.009307861328125, 8.339202880859375, 121.03085327148438, 80.51053619384766, 104.7979507446289, 60.39959716796875, -43.130035400390625, -7.0085601806640625, 46.304718017578125, -21.02886962890625, 9.04697036743164, 133.2484130859375, -8.198898315429688, 7.4899139404296875, 47.378517150878906, 71.15938568115234, 22.623519897460938, 62.247589111328125, 36.19342041015625, -8.931732177734375, 26.775909423828125, 64.6630859375, 106.28872680664062, 97.8463134765625, 109.0972900390625, 29.475082397460938, 1.74774169921875, 26.488555908203125, -47.87821960449219, -53.81207275390625, 47.167694091796875, -4.2264404296875, 11.158340454101562, 84.97987365722656, 3.2689437866210938, 14.484748840332031, 0.0, -10.483039855957031, -19.673219680786133, 67.57347106933594, 57.27436828613281, 11.241600036621094, -7.420989990234375, 25.646942138671875, 135.56610107421875, 88.426025390625, 6.29170036315918, -6.237548828125, -29.265777587890625, 95.69061279296875, -2.710721969604492, -2.6044921875, 95.22882080078125, 52.45051574707031, -10.90765380859375, -2.0193023681640625, 40.23553466796875, 112.37203979492188, 92.84918212890625, 160.99285888671875, 16.3011474609375, 68.72314453125, 6.2479095458984375, 86.69601440429688, 45.3740234375, 26.744873046875, 50.306671142578125, -24.161529541015625, 17.659439086914062, 34.917724609375, 56.125701904296875, 39.1121826171875, 85.59173583984375, 11.775047302246094, 74.47555541992188, 55.67840576171875, 7.8676300048828125, 46.78192138671875, 62.36651611328125, -10.510528564453125, 68.26718139648438, 22.247314453125, 59.57572937011719, 109.5513916015625, 84.52034759521484, 121.81118774414062, 58.103729248046875, 13.300590515136719, 23.529449462890625, 29.335052490234375, 107.52911376953125, 56.83961486816406, 1.672811508178711, 94.86909484863281, 212.04473876953125, 94.93130493164062, -45.906219482421875, -30.37438201904297, 71.9912109375, 2.7396392822265625, 171.15057373046875, 80.13742065429688, 96.50155639648438, -1.6089286804199219, 128.1767578125, 63.5941162109375, 90.4542236328125, 0.0, 18.58880615234375, 19.135108947753906, 147.892822265625, 75.22349548339844, 45.211273193359375, 0.3180217742919922, 59.342041015625, 96.52127075195312, 143.79104614257812, 121.35958862304688, 131.40057373046875, -14.381629943847656], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000169.npy"}
{"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 35.86933135986328, "std": 62.66512680053711, "min": -189.1866455078125, "p10": -20.615776824951173, "median": 28.1934814453125, "p90": 115.48975524902343, "max": 300.3941650390625, "pos_frac": 0.6953125, "sample": [34.92524719238281, 22.342864990234375, 151.6915740966797, -11.54693603515625, 61.24382019042969, 21.23583984375, -12.402801513671875, 154.52825927734375, 1.4486465454101562, 32.513916015625, 73.81082153320312, 126.29562377929688, 32.71907043457031, 50.07965087890625, -189.1866455078125, 141.62939453125, -55.206390380859375, 46.431854248046875, 8.707916259765625, 3.575763702392578, -18.02911376953125, 2.94647216796875, 54.88316345214844, 134.7657928466797, 7.3452301025390625, 2.2365779876708984, -24.600662231445312, 3.726339340209961, 101.99673461914062, -6.5355224609375, 112.88644409179688, 12.08823013305664, 66.55081176757812, 16.253475189208984, -0.0343017578125, 44.837432861328125, -4.420440673828125, 2.9221954345703125, 68.85186767578125, 96.2176513671875, 5.5157623291015625, -39.77605438232422, -18.409759521484375, 71.59304809570312, 63.028961181640625, 91.74050903320312, 93.22021484375, 3.05865478515625, -8.552825927734375, 55.50862121582031, 134.2389373779297, 67.73896026611328, 132.7342529296875, -87.2545394897461, 51.7911376953125, 34.464111328125, 164.08944702148438, -6.693267822265625, 113.0799560546875, 43.61930847167969, 2.98297119140625, 64.7354736328125, 85.69731140136719, -20.200790405273438, 89.24187469482422, 52.35658264160156, -5.72418212890625, 36.85350036621094, 118.17056274414062, -1.633819580078125, -19.40192413330078, 166.54171752929688, 37.70305633544922, 34.76995849609375, 77.61581420898438, 18.54010009765625, 0.0, 114.54183959960938, 9.904067993164062, 73.23587036132812, -49.289093017578125, 114.61093139648438, 32.5736083984375, 87.961181640625, 108.76461791992188, 29.9193115234375, -48.4722900390625, 26.18341064453125, 81.4075927734375, -50.685577392578125, 60.173492431640625, 40.656646728515625, -17.31756591796875, 44.40625, -21.064743041992188, -50.17889404296875, 26.4676513671875, 39.65983581542969, 21.66900634765625, 300.3941650390625, -7.7302703857421875, -18.103851318359375, 9.928848266601562, -12.081878662109375, -80.26629638671875, 15.045928955078125, 43.59623718261719, 10.722267150878906, -7.21368408203125, 58.84153747558594, -20.423362731933594, -11.953216552734375, 117.54034423828125, -22.95660400390625, -16.475906372070312, -10.339218139648438, 18.692474365234375, 61.66290283203125, 11.72482681274414, -2.1394290924072266, -66.08370208740234, 31.671234130859375, 155.95913696289062, 98.6668701171875, -1.3160629272460938, 79.61212158203125, -0.427032470703125, 106.88980102539062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000170.npy"}
{"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 43.084861755371094, "std": 49.9197883605957, "min": -79.18450927734375, "p10": -17.037005233764646, "median": 38.05502700805664, "p90": 110.91524658203124, "max": 193.52392578125, "pos_frac": 0.7734375, "sample": [131.6978759765625, 49.03802490234375, 46.43965148925781, 20.11370849609375, 95.12857055664062, 52.79402160644531, 98.30929565429688, -79.18450927734375, -29.36008071899414, 18.816116333007812, -4.569183349609375, 14.533134460449219, 112.40048217773438, -13.51947021484375, 18.435516357421875, 68.50027465820312, -14.522712707519531, 60.78665542602539, 45.041717529296875, 179.69943237304688, 52.64886474609375, -21.23306655883789, 28.35406494140625, 132.3861083984375, 51.81181335449219, 29.529006958007812, -30.353790283203125, 42.075958251953125, -30.337890625, 34.14727783203125, -6.510566711425781, 82.09654235839844, -11.747314453125, 76.42485809326172, 77.20257568359375, 78.65814208984375, 69.3721923828125, 63.117919921875, 17.347858428955078, -23.90789031982422, 21.60650634765625, -5.834320068359375, -16.281147003173828, 12.322189331054688, 23.39154052734375, 69.10394287109375, 78.25836181640625, -11.601104736328125, -31.047882080078125, 40.40968322753906, 133.1295166015625, 67.76348876953125, 31.23052978515625, 110.27871704101562, 58.789581298828125, 8.237380981445312, 62.13037109375, 127.35159301757812, 193.52392578125, -5.01702880859375, 70.25601196289062, 33.68934631347656, -2.2052001953125, 104.08566284179688, 29.375946044921875, 98.2794189453125, 43.833099365234375, 88.20115661621094, 125.9486083984375, 101.02151489257812, 31.236968994140625, 76.34112548828125, 34.823974609375, 23.436080932617188, -64.64771270751953, -10.544082641601562, 63.223358154296875, 10.564682006835938, 21.301788330078125, 80.02069091796875, 1.5988883972167969, 5.604766845703125, -27.79742431640625, 35.26976013183594, 9.534210205078125, -28.55322265625, 124.54119873046875, 52.86285400390625, -7.67828369140625, 114.37298583984375, 63.08805847167969, 132.48138427734375, -40.51841354370117, 64.61802673339844, 90.3594970703125, 75.68191528320312, 71.24226379394531, -6.241861343383789, 115.30877685546875, -12.566070556640625, 46.76898193359375, 27.994171142578125, 0.3792572021484375, 26.8800048828125, 4.4282989501953125, -22.23602294921875, 35.85712432861328, 95.78213500976562, 22.68408203125, -4.783962249755859, 40.2529296875, 41.363800048828125, 86.11164855957031, 52.773590087890625, 23.836395263671875, 68.95254516601562, 73.49090576171875, 20.10558319091797, 103.22166442871094, 32.49964904785156, 96.93400573730469, 91.3579330444336, -5.321172714233398, 25.718116760253906, -18.800674438476562, 132.44610595703125, 58.99342346191406, 16.31298828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000171.npy"}
{"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 22.226041793823242, "std": 53.47224807739258, "min": -116.8258056640625, "p10": -52.615744781494136, "median": 21.330154418945312, "p90": 88.21456680297851, "max": 140.7590789794922, "pos_frac": 0.6875, "sample": [35.157684326171875, -44.71832275390625, -22.284805297851562, 11.722381591796875, 89.64300537109375, 15.1065673828125, -4.081634521484375, -37.193695068359375, 26.462814331054688, 10.94293212890625, -0.05350494384765625, 41.827606201171875, -63.0743408203125, -15.903139114379883, 140.7590789794922, 17.676912307739258, 49.583587646484375, 102.79745483398438, 134.2782745361328, -7.603858947753906, -4.2413330078125, -37.43267822265625, 47.8936767578125, 77.68487548828125, -9.237945556640625, 8.91705322265625, 57.417999267578125, 6.35565185546875, -1.32598876953125, 44.803497314453125, 12.25225830078125, 62.525848388671875, -73.2010498046875, -55.85711669921875, 61.3634033203125, 42.320159912109375, -38.82757568359375, 60.577301025390625, -43.50951385498047, 9.042732238769531, 12.454843521118164, 114.30935668945312, -84.80703735351562, 1.315317153930664, 88.5277328491211, -116.8258056640625, 22.87335205078125, 76.89178466796875, 64.040771484375, 28.999542236328125, -1.198944091796875, -67.5174560546875, 10.54052734375, -16.683502197265625, -31.472991943359375, 15.9969482421875, 54.04290771484375, 29.256973266601562, -91.84210205078125, 14.40899658203125, -13.500030517578125, 34.5723876953125, 65.8983154296875, -8.450347900390625, 139.42343139648438, 73.76617431640625, 0.29669189453125, 42.13018798828125, 79.90310668945312, 22.955162048339844, 79.66021728515625, -47.5814208984375, 14.826370239257812, 54.32257080078125, 113.6602783203125, 131.74835205078125, 49.70646667480469, -15.492973327636719, 51.22528076171875, -71.61398315429688, 33.066986083984375, 75.4687728881836, -11.82534408569336, -88.5963134765625, 5.058082580566406, 20.154205322265625, -18.446762084960938, 83.68426513671875, 17.31475830078125, -62.99530029296875, 80.16976928710938, 9.09311294555664, 63.85687255859375, -60.09747314453125, -79.04090881347656, 83.74305725097656, 22.57269287109375, 18.663299560546875, -51.226585388183594, 33.44769287109375, -10.37841796875, 108.5335693359375, 45.56231689453125, 72.4871826171875, 97.20497131347656, 30.188949584960938, -2.4765625, 78.66502380371094, 91.3395004272461, 5.624420166015625, 37.59576416015625, 48.797584533691406, 22.76727294921875, 28.282135009765625, 88.08035278320312, 93.69746398925781, -12.580108642578125, 32.87591552734375, 36.04474639892578, 59.777557373046875, -103.3603515625, 86.79257202148438, 7.665008544921875, 22.506103515625, 0.0, 60.65362548828125, 8.19110107421875, 6.9970703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000172.npy"}
{"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 31.36641502380371, "std": 58.555992126464844, "min": -90.35383605957031, "p10": -42.73724136352539, "median": 22.35051727294922, "p90": 110.76910400390624, "max": 242.60162353515625, "pos_frac": 0.671875, "sample": [-47.213104248046875, -89.7926025390625, -90.35383605957031, 64.88265991210938, 46.4398193359375, -6.0035552978515625, 55.517425537109375, -74.76910400390625, 128.72552490234375, 242.60162353515625, 20.895736694335938, 39.01812744140625, 9.129566192626953, 62.6993408203125, 29.87030029296875, 138.20584106445312, 23.632400512695312, 23.290756225585938, 109.00057983398438, 137.4854736328125, 76.10910034179688, 8.171295166015625, 8.177688598632812, 39.52496337890625, 3.4899444580078125, 7.17828369140625, -36.059173583984375, 7.45623779296875, 71.18194580078125, 71.73381042480469, -57.6387939453125, 0.950439453125, -4.761192321777344, 12.453842163085938, -4.605499267578125, 47.63459777832031, -49.283050537109375, 37.619964599609375, 18.605867385864258, 70.62164306640625, 47.11799621582031, 78.07806396484375, -7.44134521484375, 71.35643005371094, 118.95622253417969, 50.09844970703125, -16.491500854492188, -14.205154418945312, 19.983688354492188, 21.4102783203125, 30.192352294921875, -17.50170135498047, -71.16464233398438, 123.8101806640625, 31.391143798828125, -44.18243408203125, -42.469932556152344, -43.3609619140625, 77.0960464477539, 118.66473388671875, 88.12198638916016, 127.91177368164062, -2.0784225463867188, 115.74275207519531, 55.90850830078125, 80.35293579101562, -70.10652160644531, 18.826202392578125, -1.956695556640625, 99.6650390625, 14.92669677734375, -5.97509765625, 127.60952758789062, 89.9800033569336, 29.978668212890625, 86.81988525390625, 3.0889205932617188, 212.96163940429688, -1.13446044921875, 29.793060302734375, 69.36053466796875, -21.29150390625, 63.5574951171875, 15.6431884765625, 80.60285949707031, 16.240509033203125, 50.327911376953125, 51.34468078613281, 8.0316162109375, 6.93536376953125, 30.49383544921875, -6.134674072265625, 58.051666259765625, 53.177978515625, 179.14599609375, 87.23855590820312, 113.1510009765625, 39.1639404296875, 80.25796508789062, 48.48466491699219, 71.53590393066406, -46.804168701171875, -3.9444122314453125, -32.460906982421875, 0.11062049865722656, 60.48341369628906, 5.1256103515625, 84.72555541992188, -13.605400085449219, 68.1815185546875, -13.450912475585938, -7.8896484375, -31.585205078125, -24.2996826171875, -9.042118072509766, -13.5438232421875, -4.413299560546875, -60.776214599609375, 109.748291015625, -5.067138671875, 37.38037109375, -6.417236328125, 0.0, 61.80173873901367, 0.07379150390625, -12.588714599609375, -45.251861572265625, 39.49217224121094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000173.npy"}
{"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 38.94883728027344, "std": 54.68655014038086, "min": -128.03350830078125, "p10": -28.24736328125, "median": 39.39609909057617, "p90": 111.99164581298827, "max": 170.15972900390625, "pos_frac": 0.78125, "sample": [129.79527282714844, 0.5715675354003906, -1.7957839965820312, 4.909629821777344, 129.11981201171875, 120.6766357421875, 48.733123779296875, -99.67718505859375, 0.517181396484375, -12.662376403808594, 122.89947509765625, 51.3134765625, -0.932342529296875, 73.36475372314453, 38.9912109375, 65.21939086914062, 104.86636352539062, 104.65216064453125, 42.81396484375, 39.800987243652344, -32.57872009277344, 77.8404541015625, -53.87312316894531, 142.32379150390625, 120.655029296875, -8.29986572265625, 68.48814392089844, 54.985382080078125, 54.061988830566406, 55.359527587890625, 19.967758178710938, 5.32177734375, 131.5325927734375, 10.92596435546875, -72.20436096191406, 18.275192260742188, 77.0806884765625, 83.08529663085938, 114.70065307617188, 50.745452880859375, 72.73833465576172, 64.98596954345703, 17.363784790039062, 31.525474548339844, -34.94105529785156, 170.15972900390625, 46.82171630859375, 98.22314453125, 19.812255859375, 123.24465942382812, -13.590606689453125, -47.8876953125, 75.20303344726562, 73.82110595703125, -37.49657440185547, 52.8382568359375, -2.2226905822753906, 3.7958297729492188, 48.9595947265625, 28.2965087890625, 4.1894989013671875, -28.140594482421875, 123.078125, 160.16079711914062, 18.789825439453125, -17.37969970703125, 29.206527709960938, 65.88748168945312, 16.977210998535156, 59.7811279296875, 13.64310073852539, 97.26768493652344, 96.33280181884766, 87.36126708984375, 110.83064270019531, 59.664947509765625, 60.67570495605469, -17.738525390625, 106.15576171875, -128.03350830078125, -30.279251098632812, 71.46621704101562, 17.970916748046875, 93.10498046875, 58.172576904296875, 4.000335693359375, 52.63307189941406, 58.63763427734375, 26.736083984375, 30.43991470336914, -4.8219146728515625, 25.18206787109375, 50.926727294921875, 38.62395477294922, 62.65869140625, 98.88250732421875, -3.764892578125, 93.3955078125, 46.421783447265625, -10.897758483886719, 67.6968994140625, -28.496490478515625, 37.55133056640625, 35.01177978515625, 13.286468505859375, 2.6427459716796875, -7.3708343505859375, 2.4159088134765625, 15.1767578125, -50.893096923828125, 45.33244323730469, -12.298049926757812, 15.28238296508789, 79.6583023071289, 41.6536865234375, -76.5054931640625, 88.78466796875, -17.70819091796875, 23.8609619140625, -94.94906616210938, 28.241744995117188, 90.0341796875, 132.18167114257812, 47.79864501953125, 75.57644653320312, 25.59698486328125, 26.7042236328125, 7.76324462890625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000174.npy"}
{"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 34.47612762451172, "std": 67.57425689697266, "min": -140.71298217773438, "p10": -34.720960998535155, "median": 19.21044921875, "p90": 118.21409835815429, "max": 342.0482177734375, "pos_frac": 0.703125, "sample": [62.51226806640625, 116.00405883789062, 59.6082763671875, 3.241565704345703, -24.598663330078125, -61.36737060546875, 28.7374267578125, 74.63737487792969, 91.93218994140625, -70.111083984375, -4.017333984375, 15.403190612792969, 25.848892211914062, -3.88470458984375, 71.60708618164062, 342.0482177734375, 5.7782440185546875, 70.92173767089844, 63.97053527832031, 124.64958190917969, -31.205810546875, 57.503021240234375, 130.37445068359375, 12.0518798828125, 24.367462158203125, 10.847320556640625, -15.573112487792969, -48.5858154296875, 1.867767333984375, -75.68959045410156, 26.769996643066406, -19.942230224609375, 21.8662109375, 80.3388671875, -34.04222106933594, 129.69699096679688, 18.98736572265625, 41.000640869140625, 73.7874755859375, 101.86944580078125, 11.883392333984375, 7.8880615234375, 18.811492919921875, 36.424560546875, 28.530303955078125, 127.30178833007812, 69.72000122070312, 125.37899780273438, -2.832977294921875, 117.87741088867188, 13.367919921875, 87.23981475830078, -10.4410400390625, 10.269805908203125, -54.107086181640625, -74.044921875, 28.72528076171875, 8.584854125976562, -21.774887084960938, 71.8050537109375, 15.522432327270508, 102.0509033203125, 48.7691650390625, 22.895917892456055, 67.30459594726562, 3.753814697265625, -4.870735168457031, -1.2156181335449219, -4.72515869140625, -18.747589111328125, 45.63140869140625, -7.398042678833008, 3.7728729248046875, 97.70526123046875, 92.59648132324219, -48.152191162109375, -8.732879638671875, 148.0852508544922, 0.0, -140.71298217773438, 118.99970245361328, 41.94708251953125, 223.0723876953125, 5.759033203125, -24.66546630859375, -12.88916015625, 13.846771240234375, -27.164825439453125, -50.28253173828125, 69.57650756835938, 73.62493896484375, 3.4093551635742188, 134.76187133789062, 91.98040771484375, 76.2423095703125, 195.58578491210938, 7.751312255859375, 7.337669372558594, -49.4010009765625, 8.5941162109375, 94.10734558105469, 4.107292175292969, 26.390548706054688, 19.43353271484375, -36.3046875, -75.76089477539062, 94.81759643554688, 12.532928466796875, 14.496673583984375, 95.85787963867188, 52.717811584472656, 94.91184997558594, 53.9422607421875, -26.947647094726562, 132.8714599609375, -7.76824951171875, 27.64904022216797, 91.6796875, 61.421409606933594, -21.0157470703125, 12.66424560546875, 26.47344970703125, 80.62991333007812, -11.771087646484375, -110.11993408203125, 0.0, 57.05458068847656, 227.43453979492188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000175.npy"}
{"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 28.380977630615234, "std": 65.23622131347656, "min": -165.91317749023438, "p10": -61.84932022094726, "median": 28.732315063476562, "p90": 106.50683288574218, "max": 209.12576293945312, "pos_frac": 0.71875, "sample": [-51.426483154296875, 12.953819274902344, 28.266815185546875, -44.48435974121094, 54.50177001953125, 31.803855895996094, -72.26397705078125, -103.57061767578125, 65.91307830810547, 40.049285888671875, -101.40435791015625, 74.65988159179688, 65.96134948730469, 102.26641845703125, 43.63739013671875, 18.257720947265625, 94.86067199707031, 8.485132217407227, -3.22723388671875, -61.31360626220703, -47.584991455078125, 33.0245361328125, 0.8948535919189453, -47.98944091796875, 56.1676025390625, 148.48928833007812, 63.21607971191406, 100.45309448242188, 38.871978759765625, 5.348388671875, 72.2850341796875, -93.95449829101562, -6.649177551269531, 27.8824462890625, 103.64252471923828, 14.44476318359375, 31.721588134765625, 106.45022583007812, 37.44764709472656, -0.3025970458984375, -68.7724380493164, 4.575828552246094, 44.36639404296875, -19.422210693359375, 16.52715301513672, 150.02267456054688, 23.04241180419922, -66.06922912597656, 28.365753173828125, 128.87142944335938, -41.941864013671875, 90.5102310180664, 61.224273681640625, -36.67536926269531, 14.623779296875, 82.65533447265625, -18.403106689453125, 30.183446884155273, 48.89691162109375, 5.813648223876953, 54.06011962890625, 32.66911315917969, 25.5029296875, -39.29585647583008, 106.78813171386719, 106.638916015625, 131.75604248046875, -13.38714599609375, 55.50634765625, 11.1878662109375, -40.55299377441406, 137.9582061767578, 88.21980285644531, -2.386077880859375, 209.12576293945312, 50.26287841796875, 11.161773681640625, -50.306915283203125, 56.929412841796875, 15.760101318359375, 37.270294189453125, 61.32557678222656, 11.148078918457031, 98.38577270507812, 58.696834564208984, 101.96430206298828, -3.8148345947265625, 28.164718627929688, -41.2972412109375, 141.65676879882812, 29.098876953125, 48.167572021484375, -67.29550170898438, 45.143951416015625, -22.976150512695312, -15.566204071044922, 90.81671142578125, -82.45611572265625, -79.05047607421875, 55.6998291015625, -89.51834869384766, 99.92256927490234, 1.8481674194335938, 7.010986328125, 124.96299743652344, -102.3460693359375, 47.686126708984375, 51.74664306640625, -63.09931945800781, 21.15605926513672, 82.60707092285156, 17.582992553710938, 193.842041015625, 95.34280395507812, 144.78097534179688, 46.95426940917969, 15.16644287109375, 56.79723358154297, -31.010345458984375, -165.91317749023438, 95.46511840820312, 35.676361083984375, 27.131120681762695, 0.47226715087890625, 54.74658203125, 112.61436462402344, 4.902427673339844, -22.625328063964844], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000176.npy"}
{"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 29.668132781982422, "std": 69.62863159179688, "min": -157.3231658935547, "p10": -49.510965728759764, "median": 26.2218017578125, "p90": 115.94010009765624, "max": 187.46017456054688, "pos_frac": 0.7109375, "sample": [81.272705078125, 1.8023185729980469, 1.4526138305664062, 145.82981872558594, -10.268875122070312, 111.65554809570312, 63.50456619262695, -30.568389892578125, 103.45367431640625, 39.26383972167969, 72.89862060546875, -45.895381927490234, -132.18505859375, -65.85171508789062, -3.1444549560546875, -37.4056396484375, 19.614303588867188, 74.75808715820312, 76.29753112792969, 52.04779052734375, 37.803375244140625, 48.39128112792969, 26.484283447265625, 75.89605712890625, -54.4039306640625, 25.959320068359375, -19.968521118164062, 113.16574096679688, 60.34515380859375, 111.217041015625, 16.994247436523438, 23.965721130371094, -132.6800537109375, 187.46017456054688, 20.084083557128906, 107.72331237792969, 127.87832641601562, 50.61602783203125, -13.565404891967773, 34.59759521484375, 98.65492248535156, -18.097061157226562, 4.709207534790039, -18.267074584960938, -49.1202392578125, -50.42266082763672, 8.827322006225586, 42.915653228759766, 40.73677062988281, 66.34538269042969, -16.257537841796875, -3.1218109130859375, 181.64068603515625, 39.118194580078125, 19.756317138671875, 93.67857360839844, 12.24371337890625, -13.7508544921875, 75.13131713867188, -26.980712890625, -47.04473876953125, 181.79248046875, 4.964385986328125, 18.952880859375, 99.18527221679688, 41.000823974609375, -43.45878601074219, 15.330284118652344, 59.17695999145508, 91.47076416015625, -6.016082763671875, 147.16845703125, -2.363861083984375, 7.129636764526367, 0.4162559509277344, 30.138214111328125, 73.533203125, -157.3231658935547, 160.04379272460938, 29.698272705078125, 17.127716064453125, -107.705322265625, 5.414192199707031, 47.184814453125, -85.652587890625, 5.043701171875, 43.51744842529297, 1.16632080078125, 22.727783203125, 34.70576477050781, 84.18942260742188, 119.0548095703125, 20.902145385742188, 77.16023254394531, 7.97344970703125, -17.396728515625, 59.4302978515625, -1.02197265625, 89.4422607421875, 32.10251998901367, -41.20698547363281, 2.0074920654296875, 28.175689697265625, 136.84970092773438, 52.29255676269531, -90.73846435546875, 1.47174072265625, -77.70448303222656, -131.94171142578125, -133.62826538085938, -56.5977783203125, 87.20103454589844, 83.1737289428711, 95.58486938476562, 33.45348358154297, 135.1669921875, 169.143798828125, 51.221336364746094, 157.48699951171875, 16.69903564453125, 114.605224609375, 0.9433212280273438, -45.456939697265625, -37.54462814331055, -28.464797973632812, 57.906280517578125, 59.745330810546875, 144.27908325195312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000177.npy"}
{"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 39.807884216308594, "std": 64.84062194824219, "min": -112.32620239257812, "p10": -35.25032577514648, "median": 25.632675170898438, "p90": 128.63888549804688, "max": 240.79656982421875, "pos_frac": 0.7265625, "sample": [88.53382873535156, 6.387470245361328, -19.479782104492188, 5.84039306640625, 5.8187103271484375, 7.784799575805664, 64.92745971679688, 144.53309631347656, -21.611804962158203, 41.912689208984375, 160.81527709960938, 92.36947631835938, 12.273284912109375, 41.89640808105469, -6.9957275390625, 154.17813110351562, 0.0, 154.2108154296875, 108.74566650390625, 74.83409118652344, 64.62530517578125, 147.35885620117188, 12.716442108154297, 128.50164794921875, 76.32281494140625, -25.777271270751953, 7.510009765625, 138.147705078125, -36.73553466796875, 116.71514892578125, -4.984766006469727, -71.2574234008789, 12.73175048828125, 107.47357177734375, -47.319549560546875, 77.439453125, 94.1297607421875, 160.6094970703125, -42.67254638671875, 113.44214630126953, 82.91635131835938, 11.996932983398438, 59.03955078125, 96.01614379882812, 149.08303833007812, 113.1806640625, -96.33270263671875, -0.3069305419921875, -12.71661376953125, 79.67404174804688, 88.02679443359375, 128.9591064453125, -51.54425048828125, -93.65643310546875, -2.216064453125, -21.219680786132812, 31.96820068359375, 53.339786529541016, 81.30029296875, 1.335693359375, 32.27156066894531, 129.61175537109375, 25.007354736328125, 74.093994140625, 17.1522216796875, 46.41088104248047, 57.4561767578125, 78.68255615234375, 46.542449951171875, 23.363418579101562, 123.80340576171875, 45.243621826171875, -4.4116058349609375, 12.399993896484375, -7.143772125244141, 157.36477661132812, -27.46484375, 17.7564697265625, -112.32620239257812, 15.2000732421875, 37.15386962890625, 20.083389282226562, 19.982818603515625, -49.307861328125, -59.2493896484375, -24.856369018554688, 240.79656982421875, 34.931915283203125, 17.675086975097656, 51.09259033203125, 16.679054260253906, 17.155590057373047, 26.25799560546875, 12.979660034179688, 56.87938690185547, 125.406005859375, -16.40106201171875, -8.85638427734375, 3.5600051879882812, 118.96231079101562, 12.837516784667969, -10.914665222167969, -34.613807678222656, 98.71527099609375, 36.14436340332031, -75.77255249023438, -13.837654113769531, 110.62632751464844, 102.5218505859375, -63.31135559082031, 18.388107299804688, 147.64251708984375, -26.20191192626953, 19.552154541015625, 116.53976440429688, 11.375457763671875, 39.35963439941406, 79.47966766357422, 3.9718856811523438, -2.2801513671875, 23.65789794921875, 30.632965087890625, 31.154815673828125, -30.1671142578125, -65.90834045410156, 81.77047729492188, 94.85159301757812, 122.45967102050781], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000178.npy"}
{"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 36.82459259033203, "std": 64.22853088378906, "min": -116.9365234375, "p10": -41.272664642333986, "median": 33.81036376953125, "p90": 121.47467041015622, "max": 175.63377380371094, "pos_frac": 0.7265625, "sample": [69.32144165039062, 7.322774887084961, 95.3701171875, 4.6553955078125, 38.498931884765625, 152.18310546875, 51.865264892578125, 22.00647735595703, -32.6231689453125, 9.917877197265625, 79.03341674804688, -23.4493408203125, 115.54180908203125, -5.0516357421875, 137.04629516601562, -63.363189697265625, 55.82135009765625, 65.757568359375, 34.2137451171875, -4.367671966552734, 41.60870361328125, 52.677886962890625, 12.477447509765625, 25.476364135742188, 33.406982421875, 4.840305328369141, -40.82293701171875, 1.5987968444824219, 68.30670928955078, -82.2914810180664, 97.70696258544922, 10.117652893066406, 0.6076889038085938, -51.552886962890625, 48.629150390625, 50.338348388671875, 118.17254638671875, 56.890289306640625, 71.0411376953125, 6.209003448486328, -35.615509033203125, -38.6883544921875, 5.0580596923828125, 91.08828735351562, -8.13604736328125, 82.75154113769531, 79.59815216064453, 4.379814147949219, 79.26519775390625, -90.78985595703125, -20.42730712890625, -91.2620849609375, 32.46013259887695, 8.918533325195312, 5.136585235595703, 147.62530517578125, 133.71527099609375, 144.30616760253906, 83.864013671875, -65.4886474609375, -56.0194091796875, -17.949310302734375, 134.96173095703125, 28.630783081054688, -10.620655059814453, 56.783203125, 164.15875244140625, -69.046875, 30.25335693359375, 80.66900634765625, 80.1033935546875, 56.103546142578125, -4.8204345703125, 60.06317138671875, 57.83967590332031, 19.6035213470459, 100.31320190429688, 37.32533264160156, -38.300437927246094, 22.537200927734375, 5.883171081542969, 36.26361083984375, 112.0186767578125, -8.106613159179688, 116.6195068359375, -23.80902099609375, 175.63377380371094, 129.17962646484375, 52.15349197387695, -39.716888427734375, 114.62583923339844, 4.977855682373047, 173.86279296875, 42.862335205078125, 50.9249267578125, -4.35968017578125, 85.16403198242188, 75.4857177734375, 32.60386657714844, 5.11383056640625, 150.19125366210938, 3.1937255859375, 85.95170593261719, 86.92988586425781, -116.9365234375, 166.4483642578125, -90.76336669921875, 14.402359008789062, 39.13121032714844, 110.84181213378906, 21.523757934570312, 64.02374267578125, -88.51513671875, 100.99136352539062, -55.3209228515625, -20.70012664794922, 76.95807647705078, 85.82122802734375, -16.41796875, 21.344757080078125, 152.25421142578125, 77.58512878417969, -28.095016479492188, -42.32202911376953, -5.338348388671875, -14.50372314453125, 105.11946105957031, 76.88092803955078], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000179.npy"}
{"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 38.694881439208984, "std": 63.83656311035156, "min": -95.2107925415039, "p10": -29.817980957031246, "median": 27.789260864257812, "p90": 117.20685729980468, "max": 350.5653381347656, "pos_frac": 0.7421875, "sample": [2.273082733154297, -3.1608238220214844, -63.31825256347656, 8.523445129394531, 17.883804321289062, 0.1537322998046875, 69.89019775390625, 26.04779052734375, -11.958572387695312, 145.85055541992188, -9.8128662109375, 85.4100341796875, 117.07180786132812, -95.2107925415039, 25.263092041015625, 70.57356262207031, -14.431488037109375, 59.8194580078125, 16.878345489501953, -64.3564453125, 77.197998046875, 122.20916748046875, 23.417285919189453, 59.4544677734375, 38.45709228515625, 37.51997375488281, 18.013092041015625, 152.1871337890625, 117.52197265625, -70.83088684082031, 2.1422271728515625, 0.41119384765625, 92.83013916015625, 99.76385498046875, 154.36801147460938, 15.660560607910156, 87.12661743164062, 21.522567749023438, -0.3957252502441406, 72.483154296875, 101.463134765625, -80.21063232421875, 36.959190368652344, 19.086103439331055, 134.87152099609375, 70.9610595703125, -75.17086791992188, 103.50270080566406, -9.42626953125, 51.76594543457031, 23.327362060546875, -26.250106811523438, -20.765869140625, 16.35333251953125, 92.96453857421875, 25.08758544921875, 86.12512969970703, 4.1577301025390625, 78.64656066894531, 65.88499450683594, 148.7216796875, -2.9996185302734375, -28.122039794921875, 114.4295654296875, 125.1793212890625, -48.431884765625, -1.1751556396484375, 23.444549560546875, -72.9608154296875, 70.94924926757812, 76.106201171875, -47.6053466796875, 63.152427673339844, 151.09249877929688, 143.2629852294922, 40.590850830078125, -10.058155059814453, -0.3820629119873047, 98.64505004882812, 30.271194458007812, 153.74560546875, 52.11222839355469, -51.92236328125, 49.2587890625, 102.45263671875, 29.926986694335938, 45.67420959472656, -25.572647094726562, 9.795883178710938, 89.73495483398438, 31.068115234375, 58.9501953125, -41.749732971191406, 29.78912353515625, 26.972808837890625, 65.88285827636719, 48.71099853515625, 44.36859130859375, -2.044374465942383, 18.790802001953125, 1.9047927856445312, 83.02816772460938, 106.955322265625, 35.45416259765625, -0.2405834197998047, -9.174575805664062, 33.936805725097656, -11.006072998046875, 350.5653381347656, 15.889266967773438, -33.775177001953125, 7.366489410400391, 6.9684906005859375, -14.0899658203125, 65.31971740722656, 22.473785400390625, 56.300750732421875, 50.138214111328125, -69.73196411132812, 92.720947265625, 1.7430248260498047, 67.76480102539062, 18.658111572265625, 25.582382202148438, 28.605712890625, 206.1641845703125, 5.192138671875, -5.608642578125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000180.npy"}
{"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 36.333045959472656, "std": 66.67182159423828, "min": -136.74017333984375, "p10": -49.20267944335937, "median": 30.81072998046875, "p90": 116.53583068847657, "max": 229.79833984375, "pos_frac": 0.7109375, "sample": [-0.38669586181640625, -8.270599365234375, -63.43603515625, -3.70257568359375, 0.0, 46.8153076171875, 97.04876708984375, 21.55938720703125, 96.40475463867188, 4.0671234130859375, 56.5770263671875, -21.50127410888672, 110.40013122558594, -2.4454727172851562, 128.8056640625, 21.35418701171875, 0.0, 76.56819152832031, 5.473667144775391, -31.8524169921875, 87.45808410644531, 26.21405029296875, 152.294189453125, 61.122955322265625, -116.21723937988281, 146.9124298095703, -75.35037231445312, 114.18669128417969, 6.336036682128906, 9.187591552734375, 118.0928955078125, -6.4181671142578125, 10.09229850769043, 110.62699890136719, 33.39276123046875, 72.03823852539062, 116.09017944335938, 56.62242126464844, 44.054168701171875, 93.470458984375, 17.229698181152344, 65.81442260742188, -64.8623046875, -6.981378555297852, 54.719085693359375, 110.17939758300781, 31.73462677001953, -52.235321044921875, 185.3511962890625, -32.67890167236328, 42.681243896484375, -16.031570434570312, 19.286008834838867, 89.12469482421875, -55.42926025390625, -100.95928955078125, -13.717086791992188, 16.82110595703125, 8.75274658203125, 48.22601318359375, 30.5435791015625, 27.53295135498047, -36.40093994140625, 100.43196105957031, 79.24746704101562, 73.31048583984375, -51.860107421875, 134.73739624023438, 6.37542724609375, 133.13607788085938, 7.74957275390625, 18.425241470336914, 43.50975036621094, 67.5028076171875, 226.38714599609375, 199.85113525390625, -136.74017333984375, 98.01502990722656, -53.346405029296875, 61.927490234375, -48.06378173828125, 32.75604248046875, -0.12288475036621094, -42.897064208984375, -21.366409301757812, 123.838623046875, 52.673248291015625, 9.61798095703125, 17.49853515625, 26.76556396484375, -14.603912353515625, 49.9110107421875, 110.42031860351562, -65.95785522460938, 6.6811370849609375, 42.57823181152344, 81.65032958984375, 74.1553955078125, 76.83985900878906, 78.9603271484375, 229.79833984375, 15.104782104492188, 29.59808349609375, 80.73184204101562, -40.51409912109375, -43.69518280029297, 0.9571685791015625, 117.57568359375, -77.1077880859375, -92.47259521484375, 101.81887817382812, 36.601043701171875, 63.88341522216797, -16.794677734375, 58.11412048339844, 126.49287414550781, -22.556358337402344, 11.907867431640625, 31.077880859375, 58.98358154296875, 18.9630126953125, 70.8226318359375, 78.6783447265625, -29.676128387451172, 30.401123046875, 77.96722412109375, 82.34265899658203, 89.24436950683594], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000181.npy"}
{"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 46.71723937988281, "std": 63.11285400390625, "min": -111.36260986328125, "p10": -25.26643371582031, "median": 46.58808898925781, "p90": 132.5654052734375, "max": 199.76071166992188, "pos_frac": 0.75, "sample": [46.07257080078125, 102.97569274902344, 41.516632080078125, -0.15777015686035156, 26.909255981445312, 48.1591796875, 145.32412719726562, 114.85147094726562, 84.77883911132812, -33.072357177734375, 88.20010375976562, 80.93988037109375, 40.1015625, 61.9346923828125, 75.52850341796875, 12.72998046875, -10.052955627441406, 90.55166625976562, 52.532257080078125, 116.36651611328125, 2.44464111328125, 18.025360107421875, 62.72698974609375, 181.86216735839844, 47.103607177734375, 119.6484375, 10.87957763671875, 89.46849060058594, 131.68426513671875, 74.36062622070312, 94.98485565185547, 88.4205093383789, 8.424362182617188, 92.6551513671875, 26.33531951904297, 6.253654479980469, -44.12872314453125, -64.34201049804688, 2.3047866821289062, -18.397705078125, -56.481109619140625, 74.15707397460938, -111.36260986328125, 38.85060119628906, 154.675537109375, -41.405372619628906, 113.80435180664062, -64.60910034179688, 91.21627807617188, 154.24658203125, 99.00390625, 169.38865661621094, 110.64739990234375, 199.76071166992188, 105.20651245117188, -19.9273681640625, 18.320363998413086, 32.27564239501953, 36.79083251953125, 63.656646728515625, 38.691375732421875, 99.5238037109375, 60.89892578125, 68.57896423339844, 142.61895751953125, -6.254585266113281, 60.8074951171875, 117.495361328125, 134.62139892578125, -4.0034942626953125, -85.7017822265625, 2.602163314819336, 45.70147705078125, 32.559669494628906, -22.99072265625, -6.13262939453125, 83.32400512695312, 94.5743179321289, 6.5746307373046875, -24.30224609375, 18.63629150390625, 29.983505249023438, 5.994232177734375, -13.005401611328125, -24.593475341796875, -3.163768768310547, 61.32072448730469, 47.37266540527344, 53.808990478515625, -12.498382568359375, 129.0814208984375, 85.33636474609375, 100.88427734375, 79.336669921875, -14.799346923828125, -11.646394729614258, 106.70806884765625, 51.04811096191406, 49.4713134765625, 4.5485076904296875, -22.304656982421875, 2.5257568359375, -12.295036315917969, 98.6787338256836, -18.6221923828125, 94.99758911132812, 42.998565673828125, 54.3902587890625, 143.23809814453125, 49.5115966796875, -35.091400146484375, 70.93415832519531, 164.83230590820312, -56.08058166503906, -88.40543365478516, 28.176055908203125, 8.593170166015625, 126.11607360839844, 22.282012939453125, -8.00311279296875, 17.7655029296875, 135.3099365234375, 150.20697021484375, -26.836669921875, 163.12380981445312, 5.263336181640625, 52.17205810546875, -51.80133056640625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000182.npy"}
{"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 41.63987350463867, "std": 69.26705932617188, "min": -93.87954711914062, "p10": -42.36046905517578, "median": 37.29856872558594, "p90": 124.4889419555664, "max": 273.51153564453125, "pos_frac": 0.703125, "sample": [-39.60638427734375, 100.96194458007812, -14.955917358398438, 61.14414978027344, -19.280868530273438, 160.5972900390625, -74.29019165039062, 85.6103515625, 127.35897827148438, 2.5698471069335938, -81.75973510742188, 35.96111297607422, 164.20553588867188, 12.08282470703125, -19.76458740234375, -93.87954711914062, -31.65041732788086, 178.4766082763672, 32.48382568359375, -43.37965393066406, 11.154525756835938, 72.498779296875, 41.741943359375, 8.632568359375, 10.552215576171875, -92.72314453125, 98.7904281616211, 49.767730712890625, 25.26568603515625, 36.341217041015625, -1.687225341796875, -92.84307861328125, 65.31123352050781, 38.25592041015625, 18.428375244140625, 64.05615234375, 34.436798095703125, 85.96707153320312, 51.42626953125, 33.56199645996094, 273.51153564453125, 56.349082946777344, 93.88379669189453, -64.89581298828125, 75.91580200195312, 76.37674713134766, 154.50006103515625, -9.569076538085938, 136.71807861328125, 3.0506668090820312, -10.32379150390625, 101.63082885742188, 74.78846740722656, -38.33441162109375, 0.0, 19.768781661987305, -15.17193603515625, -25.016067504882812, -72.0046157836914, 88.1175537109375, 114.64484405517578, 163.28932189941406, 2.6699371337890625, 121.63079833984375, 101.26611328125, -13.2239990234375, 89.7216796875, 114.05715942382812, 91.86689758300781, -14.06298828125, -3.9970703125, 139.14505004882812, 152.7166748046875, 62.722259521484375, -41.923675537109375, 24.900493621826172, -78.51078796386719, 8.74951171875, 12.98284912109375, 48.75410461425781, 66.03533935546875, 88.62762451171875, 27.05162811279297, -12.698797225952148, 6.306667327880859, 60.38031005859375, 72.438232421875, 96.95668029785156, 55.17097473144531, 82.03448486328125, 6.308441162109375, -7.015838623046875, -9.555927276611328, 14.897972106933594, -26.67669677734375, 46.05970764160156, 88.34364318847656, 78.835205078125, 105.3272705078125, -6.631843566894531, -73.2532958984375, 6.615325927734375, 13.31329345703125, 98.49308776855469, 54.62932586669922, -33.8626708984375, -0.7029647827148438, 116.73091125488281, -23.865806579589844, 239.65347290039062, 49.49153137207031, 110.27877807617188, 87.61712646484375, -5.6735992431640625, 84.86233520507812, -81.1761474609375, 22.417327880859375, 13.496826171875, -58.65869140625, 166.31939697265625, 54.096343994140625, 98.49714660644531, 99.13397216796875, 56.92083740234375, -47.4342041015625, 124.90007019042969, 82.0411376953125, 124.312744140625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000183.npy"}
{"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 57.60023498535156, "std": 72.75071716308594, "min": -141.101318359375, "p10": -35.78144836425781, "median": 57.74110412597656, "p90": 145.8948211669922, "max": 259.597900390625, "pos_frac": 0.796875, "sample": [59.691192626953125, 212.69744873046875, 192.74468994140625, -1.877532958984375, 23.454727172851562, 35.31300354003906, 105.48333740234375, -7.436000823974609, 90.84201049804688, 146.20404052734375, 12.598587036132812, 27.706512451171875, -20.04095458984375, 68.51367950439453, -21.376358032226562, 9.275848388671875, 176.9307861328125, 148.8085479736328, 19.926246643066406, 103.18319702148438, 51.908477783203125, 222.15777587890625, -100.12890625, 59.218109130859375, -39.04814147949219, 122.879150390625, 87.95248413085938, 7.16461181640625, -35.60136413574219, 112.01211547851562, 52.50688171386719, 114.65728759765625, 28.0601806640625, -54.83507537841797, 126.084228515625, 259.597900390625, -1.11505126953125, 113.84242248535156, 108.83001708984375, 114.54122924804688, 61.838134765625, 141.916015625, -43.25941467285156, 110.59188842773438, 115.88916015625, 160.37234497070312, -68.11187744140625, 50.38079833984375, -39.08326721191406, 146.9951171875, 99.0819320678711, -141.101318359375, 73.2451171875, 11.498626708984375, -87.77468872070312, 32.099273681640625, 116.23208618164062, 9.200393676757812, 44.52183532714844, 173.3707275390625, 22.631683349609375, 5.065273284912109, 130.62921142578125, -77.633544921875, 88.939453125, 125.8150634765625, 114.41156005859375, 110.95367431640625, 64.96429443359375, 92.0411376953125, 24.2362060546875, 66.16020202636719, 7.515083312988281, 28.775184631347656, 141.904052734375, 31.263031005859375, 43.465087890625, 156.42547607421875, 132.26695251464844, 89.28567504882812, 157.25494384765625, 32.914794921875, 26.435028076171875, 87.26826477050781, -9.780654907226562, -84.93601989746094, 25.670196533203125, 56.805206298828125, 20.830963134765625, 60.7430419921875, 18.417266845703125, 136.9437255859375, 73.15890502929688, 16.099853515625, 58.677001953125, -18.550048828125, -18.713401794433594, -11.263580322265625, 3.6129913330078125, 91.43768310546875, 67.15978240966797, 72.37673950195312, 64.31005859375, 140.56393432617188, 11.09616470336914, 7.144695281982422, 88.24578857421875, 42.55291748046875, -39.080726623535156, 145.76229858398438, 124.44890594482422, -0.7578125, 91.45050048828125, 26.010406494140625, 113.46190643310547, -28.597137451171875, 94.25105285644531, 41.13520812988281, -36.20164489746094, 111.97296142578125, 15.358123779296875, -56.262054443359375, 129.56436157226562, 12.48248291015625, 207.82833862304688, 128.23345947265625, -28.563323974609375, 3.5120811462402344], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000184.npy"}
{"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 42.28759765625, "std": 68.9209213256836, "min": -98.99960327148438, "p10": -49.11988525390625, "median": 42.292022705078125, "p90": 130.91505126953123, "max": 248.04193115234375, "pos_frac": 0.6953125, "sample": [54.4176025390625, -84.61495971679688, 100.68658447265625, -58.235198974609375, 4.68145751953125, 40.256011962890625, 33.42632293701172, 182.50772094726562, 165.11181640625, -1.059967041015625, 113.2220458984375, 57.29020690917969, 42.32330322265625, 104.04088592529297, 46.29328918457031, 248.04193115234375, 59.3597412109375, -47.18548583984375, 34.494415283203125, 117.064208984375, 101.52009582519531, 47.16571044921875, 70.2088851928711, 50.20977783203125, 19.773223876953125, 43.4248046875, 56.917938232421875, 21.235149383544922, 55.32274627685547, 119.09255981445312, -79.836669921875, 100.60482788085938, 97.58063507080078, 2.75164794921875, -49.543670654296875, 94.8668212890625, 59.40913391113281, 104.51908874511719, 85.42095947265625, 141.42184448242188, 39.1300048828125, 102.8521728515625, -57.50940704345703, 90.0621337890625, -4.69268798828125, 114.71383666992188, 20.4237060546875, -25.150245666503906, 42.68316650390625, 182.114501953125, -13.992218017578125, -23.602676391601562, -74.60260009765625, -77.6119384765625, -0.4950714111328125, -4.263572692871094, 49.83439636230469, 0.0, 1.8811912536621094, 15.131240844726562, -3.1031646728515625, -53.94536590576172, 15.979751586914062, -14.043205261230469, 49.72099304199219, 125.42585754394531, 24.355987548828125, 49.4207763671875, 139.8350830078125, 8.118995666503906, -45.2913818359375, 110.8436279296875, -4.510101318359375, 98.14483642578125, -17.837291717529297, 7.0255279541015625, -54.530670166015625, 38.93316650390625, 14.212728500366211, 136.85757446289062, 128.36825561523438, 36.06645202636719, 33.933990478515625, 0.0, 51.706085205078125, 79.40782165527344, 140.5842742919922, -32.924835205078125, 124.36358642578125, 30.866943359375, 54.36615753173828, 40.9443359375, -54.384918212890625, -18.983367919921875, 69.5098876953125, 176.93209838867188, -66.42019653320312, 73.38897705078125, 42.7938232421875, 52.07763671875, -39.91062927246094, 42.2607421875, 93.68853759765625, -48.938262939453125, 191.30316162109375, -5.5836944580078125, 29.76628875732422, -12.788597106933594, 103.43087768554688, 216.16796875, 50.55426025390625, 43.38898468017578, 54.32887268066406, -76.33148193359375, 148.07318115234375, -47.824493408203125, 79.934326171875, 143.0267333984375, 89.83357238769531, -98.99960327148438, 105.73444366455078, 24.058883666992188, -1.578765869140625, -41.49163818359375, -13.839691162109375, -28.239486694335938, 32.73760223388672, 60.74603271484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000185.npy"}
{"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 42.25104522705078, "std": 68.60285949707031, "min": -128.36569213867188, "p10": -47.5783950805664, "median": 43.39198303222656, "p90": 118.64713134765624, "max": 240.92071533203125, "pos_frac": 0.7421875, "sample": [61.87469482421875, 49.7132568359375, 137.55471801757812, 79.20089721679688, -9.284149169921875, 116.11239624023438, 95.62451171875, -7.25543212890625, 9.27130126953125, 25.041885375976562, 36.48797607421875, 71.11456298828125, 4.13812255859375, 115.68168640136719, -46.943389892578125, 41.806671142578125, 100.30130004882812, -4.338863372802734, 165.01687622070312, 10.815872192382812, 59.760284423828125, -66.10470581054688, -57.45289611816406, 21.69720458984375, 13.356639862060547, 34.507080078125, 89.2142333984375, 56.994659423828125, 10.59515380859375, -35.499176025390625, 20.82929229736328, 74.3553466796875, 80.20114135742188, 41.36431884765625, -34.278472900390625, -124.09173583984375, 32.3037109375, 94.40872192382812, 15.0675048828125, 51.91619873046875, 25.034446716308594, 51.4271240234375, 79.98845672607422, 133.2819061279297, 102.97724151611328, 95.65380859375, -57.6497802734375, 4.794168472290039, 57.35008239746094, 48.933685302734375, 83.43995666503906, 117.9273681640625, 133.90396118164062, 11.8323974609375, -27.701141357421875, -22.731201171875, 55.50439453125, 111.26289367675781, -9.16680908203125, 0.191192626953125, -0.48714447021484375, -30.01080322265625, 115.57708740234375, -89.86395263671875, 90.66468048095703, 7.88702392578125, 148.67437744140625, 19.804779052734375, 71.66754150390625, 93.8609619140625, 47.349266052246094, 53.666595458984375, 120.14129638671875, 83.13986206054688, 75.03973388671875, 60.189239501953125, 114.69046020507812, -101.537109375, 17.80613136291504, 184.17547607421875, 84.07635498046875, -27.06109619140625, 103.01849365234375, 193.7152099609375, 58.03900146484375, -68.50311279296875, 39.8162841796875, -51.567718505859375, -28.0750732421875, -121.3209228515625, -56.9761962890625, 1.71484375, 36.43595886230469, 79.308837890625, 128.42337036132812, 174.40579223632812, -6.850685119628906, 98.18368530273438, -128.36569213867188, -8.311773300170898, 97.78323364257812, 117.77120971679688, -109.15701293945312, -1.0402679443359375, 114.73687744140625, 21.94976806640625, 44.977294921875, 36.18617248535156, 96.96519470214844, 76.41353607177734, 99.506591796875, -27.287429809570312, 20.21954345703125, 26.258052825927734, 145.71234130859375, 0.0, 240.92071533203125, 129.75711059570312, -49.06007385253906, 118.00677490234375, 4.719272613525391, -17.768051147460938, 29.016632080078125, -9.31890869140625, 56.3875732421875, 4.9378662109375, 83.39779663085938, 46.26763916015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000186.npy"}
{"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 36.600303649902344, "std": 76.30025482177734, "min": -271.744140625, "p10": -54.1535400390625, "median": 42.311981201171875, "p90": 137.18639068603517, "max": 179.62991333007812, "pos_frac": 0.6796875, "sample": [69.5511474609375, 139.4552459716797, 79.1546630859375, 82.95034790039062, -271.744140625, 53.734893798828125, 179.62991333007812, -4.292442321777344, -73.6031494140625, 171.60284423828125, 110.82278442382812, 9.862472534179688, -4.53668212890625, 35.63792419433594, 100.13703918457031, 87.96722412109375, 90.29034423828125, 67.18169403076172, 124.72000122070312, 76.61605834960938, 115.2938232421875, 3.8538360595703125, 61.646240234375, -46.509246826171875, -53.226806640625, -25.467071533203125, 122.26057434082031, -89.30987548828125, -64.52239990234375, -91.004150390625, -9.839080810546875, 142.54940795898438, 43.1248779296875, -58.91064453125, -33.86668395996094, 3.6629867553710938, -17.22332763671875, -1.825653076171875, 80.66140747070312, 109.19476318359375, 72.1136474609375, -34.24609375, 117.696533203125, -20.594131469726562, -18.08258056640625, 89.96058654785156, 130.97195434570312, 11.79888916015625, -68.39080810546875, 46.56756591796875, -4.6727294921875, 53.66395568847656, -37.36174774169922, 86.71376037597656, -12.3787841796875, -60.224273681640625, 66.52456665039062, 63.44322204589844, 75.33999633789062, 78.75289916992188, 137.17169189453125, 13.197250366210938, 101.88204956054688, -95.31678771972656, 106.31581115722656, 137.22068786621094, -41.24723815917969, -191.8026123046875, 53.644683837890625, 77.59368896484375, 95.85565185546875, -2.998912811279297, -13.042840957641602, -20.5821533203125, 40.821685791015625, -76.48274230957031, 155.1707000732422, 71.84681701660156, 45.58929443359375, 41.49908447265625, 29.48492431640625, 8.215713500976562, -51.462303161621094, 8.29718017578125, 4.2979888916015625, -26.191871643066406, 5.778045654296875, 37.92735290527344, -187.72894287109375, 163.01992797851562, 156.26466369628906, 147.01473999023438, 156.35202026367188, 21.114158630371094, 34.24989318847656, 148.73883056640625, -11.32769775390625, 63.02166748046875, 6.2033843994140625, 82.24246978759766, -27.68707275390625, 18.1466064453125, 145.28466796875, 76.3248291015625, -3.3891735076904297, 126.155517578125, 10.3853759765625, 15.915115356445312, 75.5721435546875, 84.43136596679688, 61.9146728515625, 74.70263671875, -0.824951171875, 71.39661407470703, 12.097593307495117, 62.19635009765625, 95.80068969726562, 0.30419921875, 100.04048919677734, 35.48731994628906, -56.31591796875, 91.81388854980469, 66.96915435791016, -21.693214416503906, 0.0, 138.79541015625, -26.13787841796875, 72.02690124511719], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000187.npy"}
{"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 43.004638671875, "std": 80.75718688964844, "min": -154.93923950195312, "p10": -72.41099548339842, "median": 38.24579620361328, "p90": 144.06256408691405, "max": 288.52685546875, "pos_frac": 0.71875, "sample": [15.794166564941406, 64.08688354492188, 60.83306884765625, 0.0, 44.66691589355469, 1.884063720703125, 25.347381591796875, 257.7195739746094, 9.895896911621094, 118.5451889038086, 86.07850646972656, 7.8911895751953125, 130.41732788085938, -29.579139709472656, 11.665191650390625, 12.551788330078125, 101.76434326171875, 30.462249755859375, 2.2944107055664062, 141.89035034179688, 156.49432373046875, 144.90292358398438, -82.02462768554688, 10.322040557861328, -83.27259826660156, -110.414794921875, -4.852508544921875, -1.504302978515625, -62.38267517089844, 122.70711517333984, 118.91403198242188, 7.32843017578125, 100.38484191894531, -30.199356079101562, 45.852386474609375, 52.2879638671875, 97.322509765625, 146.5946044921875, 91.12342834472656, -10.171966552734375, 14.290771484375, 50.82952880859375, 6.118953704833984, 132.2606201171875, 180.18695068359375, -79.51873779296875, 72.544921875, 97.53802490234375, 81.57293701171875, 40.110595703125, 77.03668212890625, 73.67196655273438, -92.79779052734375, -21.70538330078125, 61.2947998046875, 200.0225830078125, 146.641845703125, 60.16326141357422, 120.23287963867188, 288.52685546875, 165.34307861328125, 7.9064178466796875, 44.48834228515625, 117.43687438964844, 13.567649841308594, 2.584228515625, 41.10743713378906, 103.85121154785156, 109.5008773803711, 124.10649108886719, 11.30419921875, 124.21566772460938, 151.908447265625, -7.217876434326172, 18.02227783203125, 10.8980712890625, 134.26727294921875, 144.83432006835938, -8.218429565429688, -104.35389709472656, -48.83038330078125, 76.94366455078125, 53.02417755126953, 124.92376708984375, 17.762557983398438, -3.707681655883789, 43.05000305175781, 107.98067474365234, 79.48971557617188, -6.51947021484375, -154.93923950195312, 144.96871948242188, -125.42582702636719, -44.85699462890625, 110.87158203125, -22.0521240234375, 34.924217224121094, -107.30722045898438, 71.33375549316406, 34.069480895996094, 63.395355224609375, 35.98173522949219, -105.46208190917969, -110.60293579101562, 143.7318115234375, 30.61431884765625, 89.85586547851562, 98.12847900390625, 186.2207794189453, -39.21769714355469, -89.65576171875, 6.179332733154297, 103.40485382080078, 36.38099670410156, 127.21328735351562, -26.978729248046875, 109.97352600097656, -13.177440643310547, -3.855224609375, -69.90725708007812, -26.0303955078125, -78.2530517578125, 0.0, 24.503204345703125, 48.43907165527344, 91.31184387207031, 25.796478271484375, -57.296142578125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000188.npy"}
{"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 51.379756927490234, "std": 66.49716186523438, "min": -116.33949279785156, "p10": -27.4593017578125, "median": 44.02979278564453, "p90": 140.38875732421874, "max": 291.44732666015625, "pos_frac": 0.8203125, "sample": [73.62875366210938, 45.55645751953125, -102.03485107421875, 31.579879760742188, 17.435394287109375, 5.166473388671875, 56.199951171875, 99.86859130859375, 97.83025360107422, 73.46011352539062, 63.102752685546875, 89.21542358398438, 125.7130126953125, 47.023101806640625, 119.67244720458984, 121.4720458984375, 150.0430450439453, -43.25794982910156, 43.94606018066406, 55.545166015625, 133.84310913085938, 154.40447998046875, -1.6276016235351562, 110.4427490234375, 43.67010498046875, -28.913055419921875, 42.55072021484375, 48.298828125, 92.31353759765625, 72.51335906982422, -57.51715087890625, 50.1197509765625, 44.28077697753906, 66.65935516357422, -9.613288879394531, 116.3517074584961, 29.59526824951172, 0.4669952392578125, 19.983139038085938, 84.1746597290039, 3.4181079864501953, 120.04931640625, 110.78388977050781, 54.598175048828125, 203.37750244140625, 88.06907653808594, 24.6744384765625, -33.264892578125, 164.13717651367188, -27.247787475585938, -48.31427001953125, -32.96990966796875, -44.298370361328125, 142.82781982421875, 17.57861328125, 21.65057373046875, 218.9180908203125, 11.579696655273438, 17.653060913085938, 31.607009887695312, 110.7613525390625, 49.60301208496094, 101.24749755859375, -27.952835083007812, 5.48968505859375, 13.768630981445312, 103.76758575439453, 28.152320861816406, 10.226356506347656, 74.99178314208984, 34.006629943847656, 10.318061828613281, -37.54534149169922, 88.96310424804688, 3.4975509643554688, -25.77027130126953, 26.06207275390625, 13.036975860595703, 94.081298828125, 85.362548828125, 20.920425415039062, 44.113525390625, 29.732330322265625, -22.75146484375, 76.95155334472656, 2.743316650390625, 69.67410278320312, -116.33949279785156, -7.163978576660156, 0.5955257415771484, 6.0181884765625, 175.40957641601562, 36.902008056640625, -83.55044555664062, 47.972557067871094, 0.0, 291.44732666015625, 146.08554077148438, 93.5538330078125, 179.5775604248047, 27.92132568359375, 97.26956176757812, 99.3038330078125, 91.06179809570312, 12.374618530273438, 0.0, 50.32501220703125, 172.149169921875, 7.4249267578125, 139.34344482421875, 61.27262878417969, 14.062057495117188, 47.3482666015625, -0.49523162841796875, 20.972015380859375, -47.36053466796875, 38.63677215576172, 88.71307373046875, 96.20648193359375, 62.733802795410156, 88.94024658203125, 185.8270263671875, -6.2451629638671875, 176.05258178710938, 34.205810546875, 8.27490234375, 18.16741943359375, 10.172426223754883], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000189.npy"}
{"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 43.112449645996094, "std": 68.01982116699219, "min": -123.83892822265625, "p10": -48.687164306640625, "median": 34.626190185546875, "p90": 130.97602081298828, "max": 221.68789672851562, "pos_frac": 0.7578125, "sample": [-52.72523498535156, 37.13140869140625, 128.24282836914062, 16.6640625, 13.86279296875, 11.462671279907227, -87.53704833984375, -9.922687530517578, -108.224853515625, 12.606658935546875, 130.67799377441406, -52.27741241455078, -42.952117919921875, 95.42376708984375, 77.53607177734375, 27.42938995361328, 98.59368896484375, -53.605438232421875, 41.86167907714844, -50.6014404296875, 0.1281147003173828, -3.024658203125, -62.95447540283203, 134.26199340820312, 114.87345886230469, -21.69390869140625, 140.09982299804688, -12.703643798828125, 91.61530303955078, 100.81048583984375, 152.64260864257812, -96.32005310058594, -21.81871795654297, 27.052719116210938, 100.46673583984375, -47.86676025390625, 170.58157348632812, 81.20858001708984, 140.0055694580078, 85.77752685546875, 137.51272583007812, -99.29580688476562, -15.097076416015625, 216.32186889648438, 23.58466339111328, 33.2303466796875, 88.31414794921875, 124.5578842163086, 102.3702392578125, 131.67141723632812, 42.581382751464844, -29.83667755126953, 82.27082061767578, -10.507545471191406, -7.5872802734375, 89.25767517089844, 11.667789459228516, 10.99786376953125, 11.096343994140625, -15.97821044921875, -18.6856689453125, 84.28529357910156, 36.02203369140625, 64.32958984375, 74.65483093261719, 52.735565185546875, 111.92698669433594, 2.89288330078125, 74.1239013671875, 99.516357421875, 110.96593475341797, 112.09043884277344, 38.537750244140625, 0.0, 45.858642578125, 183.68043518066406, -12.445602416992188, 100.1832046508789, 169.85064697265625, 19.206785202026367, 13.147979736328125, -10.065658569335938, 51.55003356933594, 85.357177734375, 107.4620361328125, 51.673828125, 7.3155059814453125, 52.1822509765625, 55.652374267578125, 24.819000244140625, 10.413314819335938, 221.68789672851562, 126.52259826660156, 83.87081909179688, 1.7601966857910156, -123.83892822265625, 68.87872314453125, -68.80413818359375, 147.43563842773438, -26.639633178710938, 94.82867431640625, 25.34783935546875, 56.1611328125, 106.41534423828125, 59.0423583984375, 93.84564208984375, 19.19586181640625, 2.25189208984375, 23.268798828125, 49.22869873046875, 5.108966827392578, 169.908203125, 9.14940071105957, 19.105682373046875, 17.59450912475586, 30.20452880859375, 47.35260009765625, 17.288660049438477, -51.899139404296875, 25.2340087890625, -11.328094482421875, 4.780889511108398, 27.850271224975586, 98.57930755615234, 2.343292236328125, 99.2071533203125, -52.623931884765625, 66.91796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000190.npy"}
{"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 43.860774993896484, "std": 71.54621124267578, "min": -126.2576904296875, "p10": -35.51924819946287, "median": 33.392391204833984, "p90": 136.07213745117187, "max": 271.38818359375, "pos_frac": 0.75, "sample": [160.61978149414062, 127.47344970703125, 31.453521728515625, -2.2799072265625, 88.05011749267578, 122.70120239257812, 13.0179443359375, 35.256649017333984, -63.889137268066406, -66.21197509765625, -30.44744873046875, 99.18113708496094, 75.62968444824219, 16.59387969970703, 13.931884765625, -1.757904052734375, 24.923587799072266, 42.539947509765625, 13.171239852905273, 55.10691833496094, 17.11785888671875, 141.86141967773438, 105.28117370605469, 2.4746017456054688, 2.5728187561035156, 26.865310668945312, 51.313140869140625, -95.08489990234375, 138.398681640625, 5.971504211425781, 28.372451782226562, 77.91641998291016, -24.815048217773438, 92.81390380859375, 5.58294677734375, -80.17266845703125, 102.08008575439453, -16.757766723632812, -126.2576904296875, 70.74903869628906, 25.222251892089844, 127.33230590820312, 120.94454956054688, 35.323211669921875, 6.826072692871094, 271.38818359375, 44.94189453125, 4.824310302734375, 36.64204406738281, 38.293243408203125, 54.6494140625, 86.13827514648438, 164.87091064453125, 107.85597229003906, 34.72920227050781, -53.66656494140625, 32.055580139160156, 66.71723937988281, -5.8887939453125, 82.09364318847656, 9.790824890136719, 42.60003662109375, 8.972623825073242, 5.221519470214844, 124.69114685058594, 15.776168823242188, 148.22747802734375, 31.470611572265625, 40.345001220703125, -2.117738723754883, 16.03759765625, -11.57958984375, 89.91494750976562, -125.74374389648438, 17.83066177368164, -56.115936279296875, 39.297760009765625, 9.690887451171875, 168.69558715820312, -7.4785919189453125, 16.591064453125, 110.33517456054688, -0.28166770935058594, 125.11819458007812, -84.6435546875, 136.78790283203125, 131.66363525390625, -94.5361328125, 120.89462280273438, 195.07345581054688, 120.6878662109375, 100.59219360351562, 48.46917724609375, -1.5198287963867188, 11.80572509765625, -11.47918701171875, 67.84405517578125, 63.11541748046875, 86.69482421875, 46.08941650390625, 205.42367553710938, 14.006561279296875, 107.10628509521484, 55.88697052001953, 157.533935546875, 162.67593383789062, 29.456573486328125, 9.055130004882812, -47.35344696044922, 29.04498291015625, -81.263427734375, -115.18804931640625, 93.51649475097656, 75.69850158691406, 88.47515869140625, 149.3995361328125, -9.34747314453125, -6.935768127441406, -27.095870971679688, 131.53802490234375, 48.80162811279297, 21.6331787109375, -24.96344757080078, -17.8143310546875, 0.0, 114.0538330078125, -2.401266098022461, 135.765380859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000191.npy"}
{"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 35.87798309326172, "std": 68.03169250488281, "min": -167.97146606445312, "p10": -32.33514595031738, "median": 30.202133178710938, "p90": 118.91979370117186, "max": 243.490966796875, "pos_frac": 0.703125, "sample": [78.52403259277344, 109.7242431640625, 64.35153198242188, -84.20147705078125, 82.08258056640625, 53.044647216796875, -18.730865478515625, 23.83697509765625, 9.2950439453125, 191.0994873046875, 71.08123779296875, 41.318878173828125, 51.51753234863281, 12.6524658203125, 99.26553344726562, 6.1725921630859375, -15.193496704101562, 76.75360107421875, -167.97146606445312, 152.65574645996094, -70.53656005859375, 210.5784149169922, -40.52302551269531, -106.29175567626953, 139.18505859375, 110.83474731445312, 7.176582336425781, 70.66154479980469, -88.8504638671875, 132.4893798828125, -101.3155517578125, 49.46217346191406, 43.98994445800781, 114.01275634765625, 30.925262451171875, 10.52668571472168, 7.499664306640625, 88.61576843261719, 32.663116455078125, 106.94888305664062, 21.6033935546875, -84.4134521484375, -8.793136596679688, 17.00848388671875, 72.14279174804688, -30.039779663085938, 70.7657470703125, -93.55419921875, 0.0, 0.0, 6.161216735839844, -17.2181396484375, -122.96446990966797, 55.84893798828125, -5.910369873046875, 2.3623123168945312, -0.166412353515625, 59.562408447265625, 26.628700256347656, 17.403961181640625, 101.31273651123047, 8.166677474975586, -7.480987548828125, 105.78994750976562, 122.43081665039062, 57.79779052734375, 152.76284790039062, 117.41506958007812, 5.224514007568359, 63.57891845703125, 0.4984130859375, 10.3612060546875, 39.32720947265625, 55.7789306640625, 29.47900390625, 101.57953643798828, 54.15423583984375, 102.61041259765625, 88.07801818847656, -18.300376892089844, -2.9298171997070312, 37.60968017578125, -7.9635772705078125, -19.477783203125, -74.37202453613281, 58.794708251953125, 78.27188110351562, 28.15155029296875, 57.44721221923828, 59.24876403808594, 0.961395263671875, -5.57672119140625, -91.04808044433594, 243.490966796875, 125.89607238769531, 15.416877746582031, -11.818031311035156, -34.532405853271484, 141.33914184570312, 148.17434692382812, 65.46222686767578, 54.345306396484375, -11.520050048828125, -8.025951385498047, -19.746505737304688, -2.165283203125, 34.94720458984375, 16.1025390625, 113.04037475585938, 104.84870910644531, 37.219757080078125, -31.393463134765625, 113.92789459228516, 55.6439208984375, 20.95703125, -3.344940185546875, 89.4954833984375, -4.124629974365234, -6.50994873046875, 24.65771484375, 13.060314178466797, 159.83038330078125, 46.00080108642578, 16.684906005859375, 33.21636962890625, 124.96139526367188, -11.702705383300781, 57.104522705078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000192.npy"}
{"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 42.44292449951172, "std": 74.18934631347656, "min": -163.71652221679688, "p10": -42.470611572265625, "median": 33.46416664123535, "p90": 147.43504638671874, "max": 277.06085205078125, "pos_frac": 0.7421875, "sample": [41.635738372802734, 22.012252807617188, -4.20684814453125, -14.36895751953125, 7.9651336669921875, 76.10256958007812, 61.11833190917969, 43.9580078125, 81.7889175415039, 48.16937255859375, 139.37466430664062, 9.30421257019043, 33.333831787109375, 14.6380615234375, 31.852981567382812, 112.1925048828125, -23.21795654296875, 26.35419464111328, 28.95879364013672, 5.59874153137207, 112.1851806640625, -44.47845458984375, 96.44683837890625, -2.093292236328125, 40.918235778808594, 37.196868896484375, -68.81976318359375, -40.61663818359375, 123.69073486328125, -11.6578369140625, -52.460487365722656, 3.45611572265625, 24.434326171875, 34.46319580078125, 87.79081726074219, 31.061004638671875, 72.8231201171875, -88.15576171875, -48.67449951171875, 66.13164520263672, -64.454345703125, 9.0570068359375, 66.34117889404297, 61.7078857421875, 17.4010009765625, 40.30517578125, 102.92626953125, 96.56011962890625, 159.2005615234375, 150.43350219726562, 67.68244934082031, -29.66845703125, 112.33311462402344, 8.73099136352539, 80.09147644042969, 15.56768798828125, 85.68904876708984, 68.87310791015625, -0.65667724609375, 57.1669921875, 27.191848754882812, 34.710975646972656, 80.30147552490234, -23.85076904296875, 146.14999389648438, 58.9803466796875, 110.12220001220703, 101.06683349609375, 37.356292724609375, -32.1124267578125, -163.71652221679688, -114.4908447265625, 66.11105346679688, 195.73031616210938, -65.7789077758789, 78.55023193359375, 186.79510498046875, 176.0879669189453, 28.718414306640625, 277.06085205078125, 154.60067749023438, 70.41827392578125, 193.10675048828125, -134.06118774414062, 153.253173828125, 131.96075439453125, 33.372379302978516, 11.735694885253906, 18.033309936523438, 24.015380859375, 19.47039794921875, -0.17821502685546875, 121.355224609375, 150.8320770263672, 71.86736297607422, 8.716156005859375, 89.39038848876953, 9.128055572509766, 5.2699127197265625, 166.81411743164062, 55.17626953125, 84.77938079833984, 4.0964813232421875, -55.38417053222656, -19.277374267578125, -4.162483215332031, 214.80499267578125, -41.610107421875, -18.909713745117188, -11.93646240234375, -27.336158752441406, 25.97381591796875, -64.438720703125, 94.99642944335938, 27.665420532226562, 7.836111068725586, 204.64791870117188, -135.58425903320312, 37.87139892578125, -41.489776611328125, -16.573974609375, 90.38214111328125, -0.065185546875, 70.93904113769531, 81.46109008789062, 33.55595397949219, 24.538330078125, 85.15855407714844], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000193.npy"}
{"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 34.53583526611328, "std": 66.95690155029297, "min": -163.7559356689453, "p10": -34.19434661865233, "median": 25.622175216674805, "p90": 123.49757690429686, "max": 291.2279052734375, "pos_frac": 0.703125, "sample": [95.49407958984375, 13.267318725585938, 58.059814453125, -53.30388641357422, 9.157470703125, 131.1014404296875, 4.984832763671875, -27.205657958984375, 28.9775390625, -13.410770416259766, 85.88074493408203, -40.44377136230469, -12.388265609741211, 44.081329345703125, 41.374114990234375, 20.21824836730957, -29.922744750976562, 33.96728515625, 91.48089599609375, 34.18064880371094, 49.053802490234375, -1.6727561950683594, 34.13014221191406, -76.93698120117188, 22.41595458984375, 30.558746337890625, -42.619384765625, 66.78314208984375, 49.45982360839844, -51.9998779296875, -22.78375244140625, 25.02508544921875, 7.72906494140625, 25.30224609375, 56.890045166015625, 9.48992919921875, 184.01177978515625, 75.69938659667969, 95.2032470703125, 143.33131408691406, -82.89140319824219, 109.96697998046875, -53.46136474609375, 64.5211181640625, -23.943115234375, 122.98248291015625, -10.424362182617188, 142.46524047851562, 165.71131896972656, -30.656463623046875, 16.22552490234375, 58.59423828125, 97.18311309814453, 29.415191650390625, 291.2279052734375, 16.000076293945312, 40.017784118652344, 64.12338256835938, 27.047462463378906, 57.104278564453125, 49.9617919921875, 5.498077392578125, -5.554283142089844, 59.713836669921875, 2.7288818359375, 22.340057373046875, 60.290863037109375, -81.4002685546875, 24.3634033203125, 73.45562744140625, 108.5614013671875, -5.718841552734375, -28.262939453125, 145.14276123046875, -19.077362060546875, 112.01080322265625, -5.244342803955078, -110.32713317871094, 51.72564697265625, 166.4032745361328, -44.62530517578125, 81.8865966796875, 124.699462890625, 104.89553833007812, 10.8197021484375, 21.122222900390625, 18.140426635742188, 26.65105438232422, 162.45721435546875, 86.35726928710938, 31.96405029296875, -101.9937744140625, 42.65679931640625, -6.23388671875, -3.595714569091797, 3.41949462890625, 25.94210433959961, -163.7559356689453, 174.98638916015625, 99.82862091064453, -20.569732666015625, 116.6322021484375, 94.21563720703125, 20.537208557128906, -18.924591064453125, -1.5235023498535156, -20.6314697265625, 70.5364990234375, -16.858367919921875, 6.1588134765625, 4.651885986328125, 134.98434448242188, 137.8351593017578, -9.017936706542969, 105.68513488769531, 90.95953369140625, -24.986854553222656, 11.66934585571289, -70.39505767822266, 42.45622253417969, -22.77825927734375, 2.705108642578125, 80.22833251953125, 12.74371337890625, 6.059291839599609, 52.9517822265625, -31.516021728515625, 48.7142333984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000194.npy"}
{"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 32.36897277832031, "std": 68.15023040771484, "min": -123.48863220214844, "p10": -39.1064712524414, "median": 17.377220153808594, "p90": 129.23167114257808, "max": 299.56378173828125, "pos_frac": 0.640625, "sample": [-15.670379638671875, -3.9016571044921875, -4.480926513671875, 84.71551513671875, 156.09298706054688, 51.1614990234375, -37.477935791015625, -14.905319213867188, 35.8814697265625, 25.644927978515625, 64.56768798828125, 23.21190643310547, 4.531970977783203, -9.861898422241211, -5.065055847167969, 169.34976196289062, 8.169952392578125, -6.84228515625, 55.0740966796875, 56.8258056640625, -30.77330780029297, 26.27838134765625, 43.75665283203125, -7.054962158203125, -26.76116943359375, -14.758773803710938, 11.872886657714844, -2.00555419921875, -34.024139404296875, 9.486038208007812, -51.07415771484375, 118.90603637695312, -7.599029541015625, -11.518936157226562, 6.7269287109375, 5.934120178222656, 165.48269653320312, 86.57293701171875, -32.55255126953125, -123.48863220214844, 2.109649658203125, -1.955108642578125, 1.1580276489257812, 13.561004638671875, 164.76583862304688, 73.99526977539062, 143.78106689453125, -47.209930419921875, -8.833038330078125, 7.713409423828125, -32.6783447265625, 40.213531494140625, 83.91696166992188, 144.0482177734375, -33.428680419921875, -3.6258544921875, 19.607818603515625, -2.3405609130859375, 105.74594116210938, 0.30780792236328125, 91.5277099609375, -50.00030517578125, -15.923675537109375, 96.8609619140625, -35.798988342285156, -104.1258544921875, -106.90069580078125, -6.548248291015625, -47.069252014160156, 54.03851318359375, 105.765869140625, 49.43341064453125, 106.39523315429688, -17.281394958496094, 57.56829833984375, 108.94183349609375, -53.29400634765625, 70.12979125976562, 50.564910888671875, 44.08100891113281, 71.23284912109375, 78.68081665039062, 75.01190185546875, 72.33419799804688, -42.90638732910156, -24.912330627441406, 20.253021240234375, -44.128387451171875, 5.469783782958984, 33.68434143066406, 34.86126708984375, 205.47119140625, 58.72808837890625, 32.25146484375, 84.92096710205078, 53.65594482421875, 150.8775634765625, 10.579681396484375, 71.43678283691406, 157.30624389648438, -10.890876770019531, 79.75404357910156, -19.277503967285156, 42.08003234863281, 10.739013671875, 6.0253448486328125, -43.975067138671875, 145.62860107421875, 102.99386596679688, 122.9962158203125, 299.56378173828125, 34.289398193359375, -67.07818603515625, 6.21893310546875, -25.386070251464844, 35.41804122924805, 94.024169921875, -112.26513671875, 3.7440338134765625, 32.13988494873047, 156.8321533203125, -10.784957885742188, 15.146621704101562, 55.962005615234375, 83.20764923095703, 26.04876708984375, 152.57394409179688, -16.954818725585938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000195.npy"}
{"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 41.502105712890625, "std": 71.02318572998047, "min": -216.52578735351562, "p10": -24.579047012329095, "median": 32.76454162597656, "p90": 137.40387573242188, "max": 245.34378051757812, "pos_frac": 0.71875, "sample": [-8.461910247802734, 3.4869384765625, 75.320068359375, 23.340011596679688, 29.01605224609375, 34.43304443359375, -17.70123291015625, 46.48291015625, 64.5526123046875, 2.1197967529296875, -4.607696533203125, 173.27171325683594, 43.55303955078125, 65.7037353515625, 90.32867431640625, -4.94677734375, 179.66156005859375, 28.33526611328125, 93.74769592285156, 37.650146484375, 20.287277221679688, -12.665901184082031, 103.88555908203125, 126.74446105957031, 7.934173583984375, 33.13737487792969, -12.484733581542969, -21.312040328979492, -38.53602600097656, 58.8243408203125, 147.91619873046875, 71.3131103515625, 76.2734375, 32.9283447265625, 57.3818359375, 47.048248291015625, -11.547870635986328, -69.7158203125, 55.10707092285156, 88.00698852539062, -7.95281982421875, 126.69827270507812, 89.94290161132812, -39.817596435546875, 10.202690124511719, 7.882007598876953, 29.057769775390625, 137.37774658203125, -56.65966796875, 70.78277587890625, -55.77880859375, -2.744739532470703, -11.77398681640625, 10.480125427246094, 14.452644348144531, 159.68914794921875, 44.421600341796875, 3.7642784118652344, 43.904052734375, 39.6839599609375, 49.62507629394531, -1.42449951171875, 29.800403594970703, 44.240509033203125, -17.23431396484375, 85.95260620117188, -15.71501350402832, 94.25759887695312, 5.02880859375, -36.90550231933594, 1.357330322265625, 186.2958984375, -4.050037384033203, 28.35608673095703, -2.814788818359375, 102.11317443847656, 1.58770751953125, -5.586273193359375, 39.44408416748047, 26.24774169921875, 22.451759338378906, -216.52578735351562, 10.506622314453125, 3.3861846923828125, -40.80902099609375, 125.76483154296875, 56.35614013671875, 186.13990783691406, 48.647979736328125, 32.600738525390625, 170.33340454101562, -127.91351318359375, 137.46484375, 183.8363037109375, 37.311279296875, -19.37994384765625, -16.65489959716797, -1.22247314453125, 121.195556640625, 134.59378051757812, 68.27435302734375, 54.781982421875, 15.70465087890625, 198.38497924804688, 213.7755584716797, -1.4042778015136719, -28.11301040649414, 86.74861907958984, 31.584259033203125, 35.268524169921875, 56.395965576171875, 245.34378051757812, 151.8226776123047, 110.56698608398438, -44.31610107421875, -23.064491271972656, -20.593425750732422, 72.10223388671875, 59.05137634277344, 96.36370849609375, 10.226470947265625, 3.5597877502441406, -139.87213134765625, 130.49676513671875, 78.73880767822266, -32.837860107421875, 87.45196533203125, 7.742757797241211], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000196.npy"}
{"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 51.68421936035156, "std": 69.15528869628906, "min": -114.21954345703125, "p10": -38.026202392578114, "median": 48.68804168701172, "p90": 146.75671997070313, "max": 271.25634765625, "pos_frac": 0.7734375, "sample": [-68.69822692871094, 45.818817138671875, -53.576568603515625, 65.87812805175781, 155.87200927734375, -20.224655151367188, 116.46804809570312, 78.490966796875, 6.4510040283203125, 136.6051025390625, 124.01614379882812, 121.88494873046875, -18.371627807617188, 54.912750244140625, 79.07159423828125, -0.0034961700439453125, 156.63442993164062, 147.45278930664062, -14.293350219726562, 131.36065673828125, -3.0164241790771484, 52.96418762207031, 109.11560821533203, 47.595123291015625, 69.96556091308594, -46.520751953125, 172.26715087890625, 160.83554077148438, 111.9149169921875, 43.29429626464844, 128.95364379882812, -52.02061462402344, -35.66156005859375, 106.68502807617188, 59.71142578125, 144.7363739013672, 21.658863067626953, -22.25934600830078, 16.07282257080078, 173.62481689453125, 32.71728515625, 271.25634765625, -24.39190673828125, 128.81719970703125, -89.1427001953125, -53.787445068359375, 95.79061889648438, 6.2026519775390625, 31.997573852539062, 26.258285522460938, 74.5189208984375, 18.60070037841797, 67.4898681640625, -20.968292236328125, 56.42523193359375, 11.205421447753906, 146.45840454101562, 42.647979736328125, 54.5548095703125, 151.861328125, 59.43368148803711, 186.4373779296875, 8.895706176757812, 33.638916015625, -45.478302001953125, -68.34085845947266, 100.92556762695312, 74.74343872070312, 8.08078384399414, 22.10150146484375, 74.57351684570312, 68.3597412109375, -4.548828125, 173.7321319580078, -5.06402587890625, 58.675506591796875, 0.247589111328125, 33.527984619140625, -61.935646057128906, 6.6783447265625, -114.21954345703125, 20.082183837890625, 91.08804321289062, 105.388427734375, 3.24310302734375, 112.60519409179688, 20.47406005859375, 74.14984893798828, 55.69476318359375, 15.822998046875, 99.69503784179688, 0.101165771484375, 69.20947265625, 36.19580078125, 125.45458984375, -1.826324462890625, 4.84228515625, -43.543701171875, 4.259376525878906, -67.40399169921875, 15.22100830078125, 49.78096008300781, -16.203346252441406, 120.42550659179688, 160.134765625, 16.780227661132812, 70.26272583007812, 77.39094543457031, 127.65298461914062, 29.25701904296875, 114.8370361328125, 79.46453857421875, 18.47711181640625, -23.360305786132812, 51.6685791015625, 20.82672119140625, 121.9564208984375, -4.5278778076171875, 148.893798828125, 5.928680419921875, 0.0, 185.67971801757812, 73.69573974609375, 79.0166015625, 123.82565307617188, -52.6673583984375, 107.7445068359375, 43.24285888671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000197.npy"}
{"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 50.18617248535156, "std": 75.2215347290039, "min": -136.802734375, "p10": -28.91624374389648, "median": 34.383644104003906, "p90": 151.97572021484373, "max": 223.42254638671875, "pos_frac": 0.7265625, "sample": [138.99151611328125, 18.45416259765625, 22.02520751953125, -21.798324584960938, 120.2506103515625, 20.43817138671875, 204.8629150390625, 154.46746826171875, 130.57342529296875, 68.31350708007812, -14.696670532226562, -26.94481658935547, 28.17572021484375, 121.58892822265625, 25.36296844482422, -20.610794067382812, 136.43875122070312, 125.77398681640625, 34.6373291015625, 89.90869140625, 151.52886962890625, -136.802734375, -43.04473876953125, 130.44149780273438, 133.37847900390625, 153.10699462890625, -14.124267578125, 8.588470458984375, 32.391571044921875, 20.8641357421875, 102.51907348632812, 28.469924926757812, 151.18182373046875, 135.66702270507812, -11.77154541015625, 18.837909698486328, -34.99468994140625, 143.7398681640625, -93.08885192871094, 200.7371826171875, 4.651641845703125, -5.99798583984375, 218.8863525390625, 156.56979370117188, -26.576374053955078, -19.40682601928711, 166.25439453125, -24.198318481445312, 5.161069869995117, 12.1397705078125, 45.40838623046875, -52.852569580078125, 1.8733596801757812, 35.0029296875, 13.596298217773438, 99.44866943359375, 223.42254638671875, -15.51068115234375, 42.79888916015625, 91.392578125, 84.71009063720703, 79.476806640625, -10.262727737426758, 72.95155334472656, 108.269287109375, 50.211883544921875, 4.525857925415039, 153.01837158203125, 47.32820129394531, 22.365501403808594, 87.42979431152344, -59.965728759765625, -24.06842041015625, 89.9283447265625, -124.63531494140625, -14.126556396484375, 33.97938537597656, 106.71804809570312, -28.16321563720703, 45.40069580078125, 38.911590576171875, 220.48849487304688, 55.68353271484375, 57.727294921875, 67.07184600830078, 16.729450225830078, 95.35224151611328, 190.5262451171875, -42.704193115234375, 59.60308837890625, 0.0, 91.8629150390625, 21.402862548828125, -6.32122802734375, -78.98622131347656, 13.915267944335938, -7.481334686279297, 104.65924835205078, 112.9345703125, 14.623199462890625, 205.5958251953125, 18.699234008789062, 57.12812805175781, 103.85449981689453, 89.30204772949219, 103.50346374511719, 65.87615966796875, 0.24439620971679688, -7.357330322265625, -104.51806640625, 96.78866577148438, -2.46173095703125, -9.65837287902832, 5.0895843505859375, -36.888275146484375, -42.775787353515625, 165.344482421875, 32.351226806640625, 132.36212158203125, 19.991180419921875, 86.34408569335938, -30.673309326171875, 51.269317626953125, -20.105466842651367, 34.12995910644531, 143.17996215820312, 108.55267333984375, 1.6676025390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000198.npy"}
{"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 45.65498352050781, "std": 75.06439208984375, "min": -145.80899047851562, "p10": -50.27880859375, "median": 38.12360191345215, "p90": 151.78990936279294, "max": 235.10641479492188, "pos_frac": 0.734375, "sample": [21.601287841796875, 89.05474853515625, 149.4317626953125, 66.35916137695312, 127.81013488769531, 127.88566589355469, 67.77261352539062, -10.801668167114258, 109.47314453125, 73.76043701171875, 235.10641479492188, 157.29225158691406, 7.734283447265625, 91.60452270507812, 168.1900634765625, -78.33636474609375, 59.36016845703125, 46.86222839355469, 39.2945671081543, -5.027658462524414, 76.17760467529297, -29.841655731201172, 20.482088088989258, 78.91242980957031, -62.451019287109375, 87.34873962402344, -11.454269409179688, 21.338943481445312, 76.27996826171875, -84.83407592773438, 70.1080322265625, 104.82869720458984, 119.50299835205078, 12.93865966796875, 18.568527221679688, 170.0320587158203, -2.958911895751953, -55.73239517211914, 74.6479721069336, 0.509521484375, 24.766464233398438, 22.54254150390625, 27.18316650390625, 7.40533447265625, 4.485443115234375, 141.9251708984375, 15.4046630859375, 28.541362762451172, -10.513336181640625, -18.42572021484375, 61.87492370605469, 190.26864624023438, 2.6409912109375, 63.153594970703125, -54.616180419921875, 64.39454650878906, 26.445159912109375, 30.727027893066406, 130.72476196289062, 19.15500831604004, -145.80899047851562, 88.12371826171875, 102.72540283203125, 127.39207458496094, -26.130935668945312, 13.677467346191406, 34.48845672607422, -52.5782470703125, -15.810455322265625, 59.777740478515625, 197.94894409179688, -25.021804809570312, 89.00981903076172, 203.355712890625, 43.622283935546875, -32.9005126953125, 66.54621887207031, 74.52798461914062, 3.338451385498047, 28.138885498046875, 57.78802490234375, 12.940628051757812, 164.35931396484375, -17.449249267578125, 131.5103759765625, 137.64605712890625, -19.700973510742188, 34.22113037109375, 196.22039794921875, 52.10807800292969, -63.67578125, -49.2933349609375, 58.60191345214844, 162.74615478515625, 25.76422119140625, 36.95263671875, -8.327133178710938, -20.22772216796875, 17.121444702148438, 92.8443603515625, 75.59259033203125, -106.14674377441406, 49.51458740234375, 4.4635009765625, 105.68310546875, -119.35223388671875, -20.191749572753906, -0.41835784912109375, -1.250030517578125, 80.13626098632812, 61.68785095214844, 39.51934814453125, 122.98004150390625, -58.02101135253906, 94.07593536376953, 57.62933349609375, 185.3731231689453, 52.02699279785156, -47.01802062988281, 115.97320556640625, 16.127716064453125, -104.88150024414062, 17.924026489257812, -25.02733612060547, 216.08624267578125, 181.52679443359375, 73.82817077636719, -67.49191284179688], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000199.npy"}
{"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 40.557559967041016, "std": 86.49433898925781, "min": -201.94656372070312, "p10": -62.48418426513671, "median": 29.17577362060547, "p90": 153.10333862304688, "max": 261.386474609375, "pos_frac": 0.671875, "sample": [49.512481689453125, -54.99424743652344, 191.31576538085938, 109.57102966308594, -66.8104248046875, 40.048927307128906, -21.915607452392578, 4.8131103515625, -15.994384765625, 85.1184310913086, -69.416015625, 166.08511352539062, 10.1839599609375, -36.15771484375, -93.41897583007812, 139.31155395507812, 96.52191925048828, -31.622344970703125, 86.17581176757812, 174.86520385742188, -54.02410888671875, -25.311187744140625, 141.88299560546875, 66.38491821289062, 152.240234375, 103.37625122070312, -60.63008117675781, -30.097076416015625, 57.299468994140625, 20.193115234375, -99.91647338867188, -201.94656372070312, 56.34272766113281, 132.480712890625, 31.094818115234375, -19.3343505859375, -31.0013427734375, 158.61767578125, 105.46713256835938, -9.73974609375, 0.0, 79.46307373046875, 133.9244384765625, 19.168075561523438, -49.1973876953125, -84.67864990234375, 108.94775390625, 28.499267578125, 213.786376953125, -26.35289764404297, 3.117828369140625, 64.47222900390625, 66.5968017578125, 59.31880187988281, 16.613937377929688, -19.961715698242188, 147.7822265625, -150.31195068359375, 106.3389892578125, -38.650146484375, 58.058685302734375, 56.44940185546875, -4.291999816894531, 61.460418701171875, 150.94677734375, 94.44261169433594, -53.21649169921875, 168.5364990234375, -113.95791625976562, 49.626708984375, 258.2657470703125, 78.6978759765625, 18.685638427734375, -13.843353271484375, 110.06475830078125, 58.87457275390625, -18.849609375, -23.746856689453125, 6.5701904296875, 140.80975341796875, -108.46417236328125, 195.23748779296875, 261.386474609375, 16.468963623046875, 17.89189910888672, 136.749755859375, 2.2919921875, 155.11724853515625, -119.30715942382812, 138.55047607421875, -40.776153564453125, 174.97772216796875, -18.085525512695312, 131.47964477539062, 50.3206787109375, 96.6351318359375, -53.52129364013672, 161.096435546875, -82.6380615234375, -13.00543212890625, 23.105224609375, 86.40484619140625, 86.93902587890625, 12.43914794921875, -81.246337890625, 70.33837890625, 14.63494873046875, -44.73634338378906, -113.671630859375, 6.8919677734375, 14.567131042480469, 129.81483459472656, 98.16378784179688, 201.36495971679688, 47.6815185546875, 83.93582153320312, -10.199493408203125, 25.725677490234375, 82.30172729492188, 29.152816772460938, 29.19873046875, 7.534900665283203, 133.42041015625, -10.10455322265625, 43.01416015625, 9.03521728515625, 80.99908447265625, 13.22235107421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000200.npy"}
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 47.841041564941406, "std": 80.7685546875, "min": -212.63995361328125, "p10": -47.662821197509764, "median": 43.412445068359375, "p90": 157.4559783935547, "max": 236.57269287109375, "pos_frac": 0.703125, "sample": [-68.46673583984375, -27.602142333984375, 112.817138671875, 113.516845703125, 45.94061279296875, 40.58935546875, -27.9444580078125, 11.048828125, 43.287353515625, 131.06179809570312, 58.28509521484375, 66.70892333984375, 11.324722290039062, 102.68965148925781, -92.66763305664062, -70.6043701171875, 172.67718505859375, 188.13607788085938, 103.47024536132812, 14.354522705078125, 14.214141845703125, 80.914306640625, 33.81635284423828, 0.0, 60.32267761230469, -18.744720458984375, 112.04183959960938, 200.64358520507812, 90.86334991455078, 37.963287353515625, 236.57269287109375, 95.8377685546875, -17.36572265625, -33.62675476074219, 0.0, -125.701904296875, -29.18341827392578, -8.313629150390625, -47.648353576660156, -63.40483093261719, -5.6874542236328125, 65.53477478027344, 188.20181274414062, 54.77339172363281, -20.549407958984375, 100.76853942871094, 87.98690795898438, 112.09297943115234, -2.816558837890625, 14.048440933227539, -10.789505004882812, 85.27372741699219, 160.1556396484375, 177.757080078125, -10.945549011230469, 71.47212982177734, 145.64224243164062, 161.26954650878906, 68.170654296875, 96.345703125, 9.205352783203125, 156.29898071289062, 69.72885131835938, 24.126495361328125, 151.29312133789062, 96.32125091552734, 32.399932861328125, 0.0, 30.868247985839844, -117.4483642578125, 50.3087158203125, -10.45452880859375, 0.88409423828125, 171.23431396484375, 132.85638427734375, 63.09288024902344, 53.51130676269531, 9.480964660644531, -14.987060546875, 51.5205078125, 60.18780517578125, 43.10685729980469, 0.029108047485351562, 3.058563232421875, -40.736480712890625, 148.7285919189453, -47.69657897949219, 164.95455932617188, -63.903900146484375, 201.7794189453125, 234.926513671875, -14.4654541015625, 109.70108032226562, 20.343109130859375, 43.53753662109375, -9.28076171875, 46.136260986328125, 131.20388793945312, 24.03156089782715, 179.48382568359375, 117.20379638671875, 101.92425537109375, -78.91647338867188, 44.8277587890625, 36.938209533691406, -15.444953918457031, -105.05862426757812, 107.9052734375, 131.52749633789062, 43.08526611328125, 87.33905029296875, 24.22418212890625, -49.41632080078125, -14.426544189453125, 87.28178405761719, 148.53045654296875, 17.397613525390625, 42.99798583984375, 46.69486618041992, 121.87652587890625, 33.84943389892578, -119.09066772460938, 74.76336669921875, 108.05099487304688, -26.24573516845703, 131.33486938476562, -44.756874084472656, -212.63995361328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000201.npy"}
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 46.53825378417969, "std": 81.2291259765625, "min": -152.73342895507812, "p10": -50.975651550292966, "median": 45.30491638183594, "p90": 147.52232360839844, "max": 231.97012329101562, "pos_frac": 0.7109375, "sample": [224.48150634765625, -145.09349060058594, -50.84696960449219, 48.20237731933594, 169.73477172851562, 83.08342742919922, -10.546615600585938, 204.87255859375, -74.19979858398438, -18.48870849609375, 29.33648681640625, 118.31069946289062, 66.40999603271484, 132.13079833984375, 119.34651947021484, 25.715106964111328, -76.54672241210938, 145.29550170898438, 59.929779052734375, 127.26100158691406, 60.33866882324219, 129.16064453125, -41.91265869140625, 43.68498229980469, 32.8424072265625, 45.225555419921875, 121.28814697265625, 3.6299400329589844, 22.032852172851562, 40.46142578125, -81.2845458984375, 59.337127685546875, -19.465911865234375, 160.94973754882812, 21.636444091796875, 131.0008544921875, 116.60791015625, 149.55099487304688, 12.634223937988281, -57.03242492675781, 74.59074401855469, 203.51654052734375, 83.44263458251953, 142.4390869140625, 17.14422607421875, 78.42813110351562, 1.982757568359375, 41.883636474609375, 139.53582763671875, -64.53056335449219, -18.381134033203125, 98.83368682861328, 139.2286376953125, -25.052961349487305, -120.72854614257812, 35.78157043457031, 153.85543823242188, -51.275909423828125, 11.78887939453125, 3.0923385620117188, 71.67216491699219, 50.27142333984375, 148.03363037109375, -13.412841796875, -39.77435302734375, -34.980224609375, 123.912841796875, -20.33233642578125, -56.919403076171875, 54.53240966796875, 27.942543029785156, 100.7510986328125, 45.82928466796875, 3.8128662109375, 139.58108520507812, 30.162689208984375, 129.34886169433594, 45.38427734375, 231.97012329101562, -30.97491455078125, 0.0, 147.30319213867188, 111.29158020019531, 69.62054443359375, -18.133087158203125, -36.11626434326172, 9.83761978149414, -11.480361938476562, 50.93598937988281, 56.7056884765625, 54.4117431640625, 208.8232421875, 6.791290283203125, 39.405860900878906, 86.22824096679688, 77.899658203125, 91.68962097167969, -103.79910278320312, -25.6727294921875, -19.58575439453125, -149.58132934570312, 104.239990234375, 61.533233642578125, 196.9766387939453, 65.62982177734375, -9.362945556640625, 87.09019470214844, 76.83233642578125, -0.07485198974609375, -25.148822784423828, 75.17996215820312, 14.420135498046875, -36.08375549316406, 33.943145751953125, 109.9779052734375, -90.54061889648438, 63.639312744140625, 112.4161376953125, -152.73342895507812, 5.26483154296875, -29.699426651000977, 204.3385772705078, 79.2677230834961, 15.525459289550781, 5.424003601074219, 113.02523803710938, -30.071693420410156, 177.85098266601562], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000202.npy"}
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 34.70266342163086, "std": 70.02011108398438, "min": -115.51702880859375, "p10": -66.68101806640625, "median": 35.67683029174805, "p90": 116.40422821044922, "max": 241.31524658203125, "pos_frac": 0.6875, "sample": [20.32856559753418, 91.0370864868164, 99.15557861328125, 84.3353500366211, 73.45346069335938, 100.46946716308594, 100.55350494384766, 75.54971313476562, 77.28067016601562, 62.80046081542969, 73.81951904296875, 31.540679931640625, 3.1137828826904297, 85.323974609375, -90.75691223144531, 113.1461181640625, -68.61904907226562, -80.11506652832031, -27.48389434814453, 60.87744140625, 9.652496337890625, 0.0, -65.85043334960938, 84.03228759765625, -87.63973999023438, 27.551429748535156, 56.293182373046875, 39.20989990234375, 28.8992919921875, -40.209556579589844, 106.84121704101562, 131.925537109375, 71.04373168945312, 53.4503173828125, 27.06158447265625, -18.458965301513672, 35.588287353515625, 241.31524658203125, 116.75494384765625, -6.274749755859375, 73.82884216308594, 147.47674560546875, 97.6343994140625, 82.35560607910156, 5.016429901123047, -33.216644287109375, 35.76537322998047, 123.3966064453125, -0.4956207275390625, 50.21293640136719, 2.3925628662109375, 66.744140625, 80.84524536132812, 27.509170532226562, 187.2487030029297, -45.48095703125, 52.04547119140625, 6.570119857788086, 153.8736114501953, 180.52627563476562, -21.826248168945312, 99.4646224975586, 31.080078125, 32.57524108886719, -52.440528869628906, 79.79268646240234, -72.51290893554688, 109.03228759765625, 40.63239288330078, 14.27362060546875, 21.0682373046875, -53.374298095703125, -98.26345825195312, -32.29278564453125, -35.234832763671875, -4.422792434692383, -101.23373413085938, 22.6456298828125, -76.163818359375, 46.078765869140625, 52.86619567871094, -35.130279541015625, -115.51702880859375, 116.25392150878906, 63.02198791503906, -79.78926086425781, 53.6845703125, 39.7564697265625, 186.84609985351562, 85.04168701171875, 33.503074645996094, 84.70294189453125, -43.694427490234375, 124.43832397460938, 93.96380615234375, 81.56552124023438, 112.51220703125, 6.580528259277344, 133.95501708984375, 91.35787963867188, 82.33572387695312, -48.3681640625, -29.73797607421875, 21.60504150390625, 45.165245056152344, -81.50949096679688, -20.677520751953125, -55.17338562011719, -10.87835693359375, 131.565673828125, 0.0, -4.564788818359375, -34.439178466796875, 15.782974243164062, 86.56644439697266, -2.397369384765625, 34.92608642578125, 60.47100830078125, 13.899269104003906, 37.59844970703125, 101.83761596679688, -98.883056640625, 73.41934204101562, 121.93539428710938, 35.1644287109375, -105.48670196533203, -16.417686462402344, 62.15895080566406], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000203.npy"}
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 51.785491943359375, "std": 74.55298614501953, "min": -99.93107604980469, "p10": -45.21339111328125, "median": 44.7366828918457, "p90": 147.2667510986328, "max": 247.39413452148438, "pos_frac": 0.7578125, "sample": [146.26669311523438, 96.62446594238281, 12.18878173828125, 67.76788330078125, 40.58872985839844, 172.25950622558594, 10.130779266357422, 106.22732543945312, 53.15113067626953, 157.23236083984375, 49.7454833984375, 49.184814453125, 63.9976806640625, -28.45281982421875, -83.37548828125, -38.13983154296875, 20.642959594726562, -14.15679931640625, 43.880409240722656, 28.506011962890625, 101.58706665039062, 19.135330200195312, 28.4158935546875, 118.6253433227539, 39.663330078125, 127.98553466796875, -9.284286499023438, -59.81904602050781, 36.61555480957031, 102.39334106445312, 45.59295654296875, 203.29840087890625, 8.859268188476562, 38.15306091308594, 88.561767578125, -96.39556884765625, 22.003253936767578, 247.39413452148438, 42.19157409667969, 110.90902709960938, 149.6002197265625, -66.34991455078125, 165.63018798828125, -27.45996856689453, 84.11105346679688, 141.6182861328125, -99.93107604980469, 167.71673583984375, 121.97055053710938, 183.64663696289062, 118.78675079345703, 6.21539306640625, -80.48968505859375, 136.42852783203125, 34.2777099609375, -3.9140090942382812, 154.4315185546875, 6.2093963623046875, 31.58484649658203, -64.29412841796875, 140.00180053710938, 89.04302978515625, 86.32638549804688, -33.28863525390625, -20.76385498046875, 168.08102416992188, 138.36593627929688, 115.6929931640625, -78.7945556640625, 56.22528076171875, -17.396228790283203, 99.80148315429688, -2.1557159423828125, -38.83422088623047, -5.42802619934082, 4.250244140625, 68.76307678222656, 143.5572509765625, -3.543212890625, 23.22784423828125, 20.151626586914062, -36.5914306640625, -68.70984649658203, 110.2159423828125, 17.932159423828125, -17.843841552734375, 102.68118286132812, 34.82817077636719, 129.2501220703125, 143.48728942871094, 95.26229858398438, 17.605560302734375, -94.70626831054688, 115.3404541015625, -44.31060791015625, 4.45159912109375, 27.595779418945312, -53.09674072265625, 50.87164306640625, 124.76312255859375, 59.073516845703125, 6.6679229736328125, 121.60781860351562, 41.8494873046875, 106.47933197021484, 93.55853271484375, 124.94265747070312, 37.091758728027344, 156.10037231445312, 55.383506774902344, 11.6627197265625, 54.986968994140625, 26.44122314453125, 139.76239013671875, 54.34417724609375, 64.47111511230469, 55.29754638671875, -13.749351501464844, 10.5032958984375, 48.623687744140625, 83.38910675048828, 1.0416107177734375, 229.8823699951172, 98.06649780273438, -47.31988525390625, 159.98455810546875, -76.5567626953125, -17.298583984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000204.npy"}
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 55.225860595703125, "std": 82.67754364013672, "min": -132.66580200195312, "p10": -38.65321655273437, "median": 45.66012191772461, "p90": 152.0146728515625, "max": 388.220947265625, "pos_frac": 0.7265625, "sample": [169.48187255859375, 0.0, 15.195213317871094, 6.711164474487305, -82.82141876220703, 97.03250122070312, -36.055938720703125, -17.639862060546875, -14.654998779296875, 15.628095626831055, 166.4808349609375, 111.28562927246094, 24.705310821533203, 36.09831237792969, 6.874176025390625, -2.7234268188476562, 144.04266357421875, 109.25271606445312, 49.076087951660156, 90.9188232421875, 15.20404052734375, 78.948974609375, -65.6357421875, 120.53662109375, 56.14569091796875, 87.93487548828125, 37.29261779785156, 62.14539337158203, 126.51300048828125, 67.41683959960938, 95.52130126953125, 66.48983764648438, 14.72869873046875, -8.934234619140625, -41.8546142578125, 18.832294464111328, 43.93022155761719, 80.9420166015625, 28.678630828857422, 48.69319152832031, 39.867279052734375, 79.8994140625, -132.66580200195312, 78.83686828613281, 41.074310302734375, 146.5740966796875, 0.0, 60.13720703125, -44.767723083496094, 15.854095458984375, -1.768829345703125, 37.3071174621582, 80.62820434570312, 138.7230224609375, -12.000885009765625, 1.666717529296875, 180.80638122558594, 124.76484680175781, -37.28118896484375, 78.40042114257812, -124.22391510009766, -69.3604736328125, 20.10693359375, 15.90057373046875, -25.957977294921875, -20.79144287109375, 241.06549072265625, 14.428024291992188, -45.19744873046875, 139.54278564453125, 100.34518432617188, 140.26806640625, -78.14724731445312, 147.03823852539062, -5.6912994384765625, 42.55888366699219, 32.645721435546875, 132.81744384765625, 79.57553100585938, -20.5848388671875, -80.33258056640625, 248.73243713378906, 114.46820068359375, 136.59481811523438, 125.57794189453125, 14.396659851074219, -19.13287353515625, -31.009613037109375, 126.97737121582031, 109.27999877929688, 137.58172607421875, 225.63088989257812, 150.570556640625, -20.99835205078125, 92.54217529296875, 141.47410583496094, 174.92352294921875, 79.33222961425781, 50.793739318847656, -84.33810424804688, 388.220947265625, 155.6318359375, -9.226287841796875, -26.578887939453125, 181.67245483398438, 83.36224365234375, -82.52436828613281, 102.72149658203125, 186.57073974609375, 98.91886901855469, 84.2991943359375, 11.309051513671875, -74.71685791015625, 105.14957427978516, 109.06057739257812, 34.646331787109375, -36.103240966796875, 96.517578125, 4.260047912597656, 29.575454711914062, 46.528076171875, 128.1221923828125, -34.70344543457031, 163.0205078125, 15.123458862304688, 155.38427734375, 0.0, 44.79216766357422], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000205.npy"}
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 44.53862762451172, "std": 85.7778091430664, "min": -193.93115234375, "p10": -77.39850463867188, "median": 34.25469970703125, "p90": 158.5122863769531, "max": 246.5819549560547, "pos_frac": 0.7265625, "sample": [99.2562255859375, 49.95238494873047, -68.8983154296875, 26.93145751953125, 179.54177856445312, -97.5793228149414, -90.89312744140625, 43.89227294921875, -13.01763916015625, 51.01947021484375, 123.50480651855469, 136.60528564453125, 203.15032958984375, -15.748970031738281, -118.27328491210938, 54.84893798828125, 197.2022705078125, 86.33082580566406, 78.33711242675781, 209.71240234375, -8.901290893554688, 164.71975708007812, -24.137592315673828, -77.23834228515625, 75.37335205078125, 88.29959106445312, -12.44598388671875, -1.677490234375, 16.378761291503906, 1.335357666015625, 3.28997802734375, 20.54620361328125, -102.23735046386719, -0.4732666015625, 83.8367919921875, -23.942794799804688, 158.7122802734375, 246.5819549560547, 156.45452880859375, 37.18689727783203, 46.3048095703125, 74.2377700805664, 47.24188232421875, 23.036346435546875, 92.36715698242188, 146.85284423828125, 6.399169921875, 189.2255859375, 201.12319946289062, 34.380950927734375, 106.38864135742188, 99.23048400878906, 76.1867446899414, 108.89532470703125, 15.650957107543945, -77.772216796875, 71.72967529296875, 197.68154907226562, -193.93115234375, -24.59326171875, 111.41632080078125, -37.00129699707031, -104.32518768310547, 104.42710876464844, 148.9422607421875, 42.493133544921875, 20.1756591796875, -21.299072265625, -12.658819198608398, 22.3380126953125, 54.346763610839844, 83.342529296875, 137.7518310546875, -123.1343994140625, 210.3372802734375, 36.12953186035156, 1.3522491455078125, 3.7939682006835938, 102.65457153320312, 14.055450439453125, -63.286712646484375, -125.66326904296875, 181.9613800048828, 102.89337158203125, 111.65142822265625, 29.724273681640625, 38.62578582763672, -93.71294403076172, 25.96613311767578, 2.484344482421875, 100.337646484375, 88.2796859741211, 101.07318115234375, -6.731021881103516, -24.65093994140625, 101.36434936523438, 18.0389404296875, -10.8026123046875, 0.86993408203125, 12.29489517211914, 25.442138671875, 10.3804931640625, 158.42657470703125, 106.93486785888672, 34.128448486328125, 136.41793823242188, -106.96673583984375, 1.171783447265625, -43.09295654296875, -32.57379150390625, 12.925689697265625, 183.77450561523438, 125.08793640136719, 28.43548583984375, -1.539337158203125, 86.73248291015625, 107.3465347290039, -98.47715759277344, 145.84622192382812, 152.8836669921875, 63.50775146484375, 12.52899169921875, -47.28997802734375, 105.68153381347656, 33.07380676269531, 19.27447509765625, 4.19390869140625, -87.33921813964844], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000206.npy"}
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 45.057533264160156, "std": 90.32816314697266, "min": -265.79901123046875, "p10": -55.14114685058593, "median": 31.587875366210938, "p90": 156.05694885253905, "max": 337.676025390625, "pos_frac": 0.7109375, "sample": [142.4180908203125, 96.59149169921875, 29.057464599609375, -36.418296813964844, 10.221755981445312, 86.28456115722656, 69.91317749023438, 166.5289764404297, 100.83421325683594, -13.39426040649414, -54.55010986328125, 94.29681396484375, -99.699951171875, 9.371391296386719, 22.698974609375, 192.6619415283203, -0.22638320922851562, 96.46444702148438, 147.95538330078125, 10.418914794921875, 75.87928009033203, -25.707290649414062, -185.01025390625, -10.497817993164062, 61.252994537353516, 75.01336669921875, 44.203582763671875, -19.162628173828125, -142.54483032226562, 16.152313232421875, 151.23121643066406, 11.58203125, 54.53753662109375, -81.41323852539062, 161.33349609375, 115.12399291992188, 24.8050537109375, -0.7764129638671875, 37.62834930419922, 151.98406982421875, -77.64703369140625, 8.593978881835938, 78.0499267578125, -23.415008544921875, -2.56671142578125, -16.166473388671875, 89.50582885742188, -5.55059814453125, -7.02178955078125, 27.878890991210938, -265.79901123046875, -18.744384765625, -36.135711669921875, 8.288330078125, 149.06097412109375, 0.0, 71.58290100097656, -11.599517822265625, 69.92626953125, 199.9425811767578, 114.80313110351562, -18.975982666015625, 17.96137237548828, 34.18603515625, 31.86602783203125, -66.1094970703125, 179.31478881835938, -147.05921936035156, 28.719131469726562, 0.967376708984375, 22.50872802734375, -43.205047607421875, 71.3607177734375, 262.57818603515625, 227.81610107421875, 218.81356811523438, 26.17852783203125, 52.35557556152344, 42.65196228027344, 144.106201171875, 6.172386169433594, -110.336181640625, 78.25933837890625, 86.10041809082031, 16.332414627075195, 337.676025390625, 31.905349731445312, 107.03738403320312, -56.520233154296875, 17.487022399902344, 7.925537109375, 67.59288024902344, -4.30877685546875, 86.35905456542969, 80.8675537109375, 151.36578369140625, 20.3939208984375, 5.6319122314453125, 18.93531036376953, -108.49822998046875, 114.84405517578125, 202.3237762451172, -28.73175048828125, 60.97344207763672, 8.405176162719727, 107.64163208007812, 34.271820068359375, -41.46502685546875, 99.74360656738281, 132.25009155273438, 31.309722900390625, 162.699951171875, 21.21784210205078, 78.98870849609375, 26.52099609375, -2.6157989501953125, 153.92489624023438, 127.98152160644531, 174.456298828125, -84.45440673828125, 116.53048706054688, -127.32830810546875, 125.58642578125, 161.03173828125, -3.0654296875, 110.83018493652344, 49.32243347167969, 87.72705078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000207.npy"}
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 56.24591064453125, "std": 84.85790252685547, "min": -140.38116455078125, "p10": -15.398383331298826, "median": 43.43522644042969, "p90": 159.7709777832031, "max": 475.119873046875, "pos_frac": 0.7421875, "sample": [19.33770751953125, 11.197967529296875, -69.43251037597656, 51.0404052734375, -42.636474609375, -7.818168640136719, -5.911647796630859, 135.03775024414062, -0.5393524169921875, 177.31982421875, -3.192729949951172, 54.6624755859375, 29.758209228515625, 57.03369140625, 84.69621276855469, 239.4873046875, 92.87161254882812, 23.38172149658203, 0.91455078125, 257.43505859375, -140.38116455078125, 144.72946166992188, -42.84202575683594, 103.778564453125, 46.923828125, 87.43423461914062, 175.88980102539062, 112.11929321289062, 20.112083435058594, -54.22789764404297, 227.7271728515625, 77.14262390136719, 72.03178405761719, 142.89276123046875, -12.594352722167969, 145.43344116210938, -129.20904541015625, 79.12060546875, 92.8468017578125, 29.536956787109375, 4.19325065612793, -120.587646484375, 11.1348876953125, 76.80274963378906, 121.04910278320312, -17.56170654296875, 124.735107421875, 49.3934326171875, 0.0, 166.34375, -109.11318969726562, -0.4134063720703125, 26.214263916015625, -12.65545654296875, 41.471588134765625, 164.18994140625, 26.34552001953125, -8.170822143554688, -1.8980255126953125, 231.79949951171875, -13.199020385742188, 95.69831848144531, -6.157127380371094, 119.259521484375, 43.5294189453125, -11.0718994140625, 16.50543212890625, 22.999771118164062, 76.63316345214844, 110.91273498535156, 475.119873046875, 0.0, 92.78766632080078, 154.0955810546875, 72.72456359863281, 54.107879638671875, -10.614250183105469, 188.43972778320312, 175.39715576171875, 147.36239624023438, 34.63658142089844, 175.47918701171875, 147.59445190429688, 112.51824951171875, 4.2736968994140625, 82.37871551513672, 17.3140869140625, -13.985645294189453, 42.272857666015625, 51.615234375, -26.608428955078125, 94.66201782226562, 43.341033935546875, 69.00277709960938, 120.92417907714844, 67.1815185546875, 166.3099365234375, 25.970794677734375, 12.119476318359375, -8.572479248046875, 15.766738891601562, 12.861295700073242, -7.71929931640625, 29.39917755126953, 42.45481872558594, 66.9194107055664, -11.041336059570312, 129.322509765625, 2.1169815063476562, 11.956130981445312, 54.75752258300781, 64.38427734375, 92.74714660644531, 26.75518798828125, 22.86395263671875, -82.13650512695312, 17.784820556640625, 72.4638900756836, -72.00909423828125, 110.360107421875, 157.87713623046875, 103.7659912109375, -14.471244812011719, 94.16455078125, 139.09298706054688, -79.21746826171875, 5.46551513671875, 113.45098876953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000208.npy"}
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 39.051788330078125, "std": 85.30299377441406, "min": -204.4775390625, "p10": -69.00661544799804, "median": 36.269412994384766, "p90": 146.01010131835938, "max": 240.5489501953125, "pos_frac": 0.671875, "sample": [125.43011474609375, 15.5982666015625, 111.56961822509766, 112.14253234863281, -49.1463623046875, 41.27308654785156, 99.42221069335938, -204.4775390625, -93.81304931640625, 82.172119140625, 141.64093017578125, -12.874015808105469, 32.043548583984375, 28.97779083251953, 136.7286376953125, 145.83779907226562, -155.44638061523438, 10.95561408996582, 1.916290283203125, 115.97537231445312, 120.81500244140625, -20.06427001953125, 49.71128845214844, -70.76258087158203, 59.887939453125, 92.89837646484375, 89.87907409667969, 39.2042236328125, -66.77706909179688, 102.4351806640625, 109.93441772460938, -112.67340087890625, 28.538436889648438, -121.15724182128906, -66.29522705078125, -42.21989440917969, 222.88546752929688, 0.4545440673828125, 0.0, 23.411972045898438, -28.44024658203125, 56.5865478515625, 146.41213989257812, -30.702850341796875, -46.6898193359375, 34.6156005859375, 63.014137268066406, 12.557376861572266, -26.56951904296875, 56.13104248046875, 84.03475952148438, 86.83245849609375, 94.83563232421875, 4.50787353515625, -18.776092529296875, 5.697751998901367, 16.8448486328125, 14.310043334960938, 37.641395568847656, -7.561235427856445, 85.52130126953125, -153.42218017578125, -1.5240478515625, 220.64495849609375, 119.05105590820312, -61.82206726074219, -21.26904296875, 47.5213623046875, 93.93527221679688, 112.94366455078125, 53.48443603515625, 153.10745239257812, 22.085281372070312, 43.43109130859375, -34.66099548339844, 0.0, 30.38739013671875, 162.11474609375, -9.249496459960938, 82.70857238769531, 10.832427978515625, 0.583831787109375, 81.89566040039062, -25.429367065429688, -30.800430297851562, 154.57763671875, 109.5909423828125, -21.189308166503906, -68.25405883789062, -85.481201171875, 53.212005615234375, -11.324127197265625, -10.748748779296875, 115.29002380371094, -11.517675399780273, -18.382568359375, 88.42999267578125, 114.77288818359375, 193.14910888671875, -108.46710205078125, -79.1956787109375, 8.614730834960938, -109.78634643554688, 34.897430419921875, -1.635955810546875, -121.96417236328125, 8.055343627929688, 43.80970764160156, 97.86178588867188, 134.27688598632812, 96.45330810546875, 72.49295043945312, 172.40850830078125, 84.85650634765625, 27.938861846923828, 163.7022705078125, 240.5489501953125, 44.1646728515625, 87.89308166503906, 136.6011962890625, 192.05902099609375, -102.32391357421875, 109.31539154052734, 67.45730590820312, 186.72293090820312, 121.69009399414062, -23.295814514160156, 147.96661376953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000209.npy"}
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 50.124977111816406, "std": 80.78539276123047, "min": -152.236083984375, "p10": -50.97598724365234, "median": 44.69782257080078, "p90": 159.60062866210936, "max": 238.7630615234375, "pos_frac": 0.6953125, "sample": [5.4722747802734375, -13.734786987304688, 51.07244873046875, -11.2872314453125, 207.58273315429688, 36.85029602050781, 143.68731689453125, 143.763671875, 161.4813232421875, 93.44137573242188, 22.451416015625, -19.170337677001953, 2.809234619140625, 73.73497009277344, 143.94354248046875, 3.4868698120117188, 125.42584228515625, 69.67642974853516, 47.35260009765625, -17.783416748046875, 9.349788665771484, 47.316680908203125, -60.624664306640625, -5.9225616455078125, -4.5316314697265625, 39.98613739013672, -10.84808349609375, 227.12420654296875, -152.236083984375, 13.304595947265625, 76.94781494140625, 63.63313293457031, -3.155670166015625, -18.57122802734375, 158.79461669921875, 88.66995239257812, 42.597496032714844, 89.55584716796875, -0.8621330261230469, 80.337890625, 43.76197814941406, 141.2155303955078, 13.135477066040039, -32.60398864746094, 57.38745880126953, 41.70068359375, 84.0087890625, -49.54249572753906, 158.14913940429688, 64.31356811523438, 47.16595458984375, -14.323562622070312, 126.69033813476562, 170.9954833984375, 123.97711944580078, 106.72531127929688, 19.0535888671875, -152.14413452148438, 138.06390380859375, -21.14410400390625, 144.85275268554688, 17.456329345703125, 177.213623046875, 119.16734313964844, -17.082672119140625, 191.8444366455078, -54.32080078125, 6.5900726318359375, 114.6009521484375, 75.14459228515625, -63.850616455078125, -80.96052551269531, 103.06326293945312, 38.28495788574219, 150.96701049804688, 217.89669799804688, -1.8819770812988281, 73.83074951171875, -27.990997314453125, -120.29269409179688, -13.838573455810547, 63.62461853027344, -5.4960174560546875, 188.99560546875, -4.097160339355469, 94.35049438476562, 238.7630615234375, 109.22052001953125, -8.806396484375, -54.61280822753906, -54.74974060058594, 15.512413024902344, 4.4578857421875, -79.32183837890625, 47.62085723876953, 45.6336669921875, -27.773239135742188, 204.39617919921875, 55.518890380859375, 68.79508972167969, 101.15277099609375, 175.60122680664062, 48.43968963623047, 89.00117492675781, 215.11602783203125, 106.88035583496094, 20.948165893554688, 29.568267822265625, 14.193572998046875, -40.46051025390625, 61.78947448730469, 51.08428955078125, -6.4776763916015625, 94.02838134765625, -3.9136505126953125, -69.07522583007812, 8.37158203125, 177.57440185546875, 43.53045654296875, -79.20091247558594, 94.9698486328125, -54.77288818359375, 135.36624145507812, 82.901123046875, -37.938690185546875, 10.701416015625, 35.24449157714844, 140.93896484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000210.npy"}
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 60.318023681640625, "std": 72.8712387084961, "min": -163.8197021484375, "p10": -18.540682983398437, "median": 49.08287048339844, "p90": 153.8717834472656, "max": 242.2049560546875, "pos_frac": 0.828125, "sample": [-13.910560607910156, 40.49678039550781, 197.3345947265625, 127.576904296875, 140.48854064941406, 36.00401306152344, 88.83753204345703, 19.254549026489258, 1.896331787109375, 57.6312255859375, 70.39445495605469, 0.222198486328125, 49.326629638671875, 34.177093505859375, -57.281822204589844, 176.90042114257812, 14.699882507324219, 0.7079582214355469, -163.8197021484375, -58.6373291015625, 115.40280151367188, -12.214374542236328, 33.110450744628906, 23.213653564453125, 76.57697296142578, 82.828857421875, 45.832244873046875, -52.258026123046875, 35.11717224121094, 3.02288818359375, -10.0550537109375, 151.0889434814453, 48.839111328125, 139.38970947265625, 73.177490234375, 152.71572875976562, 92.23602294921875, -37.87202072143555, 5.890144348144531, 140.17251586914062, 42.68743896484375, -62.92974853515625, 22.266387939453125, 35.936492919921875, -21.72998046875, 204.42877197265625, 172.94241333007812, 88.72784423828125, 58.8843994140625, 17.069366455078125, 118.289306640625, 0.096038818359375, 9.62420654296875, 12.815139770507812, 242.2049560546875, 133.88430786132812, 193.01898193359375, -41.39628601074219, 64.76185607910156, 82.69415283203125, -22.75579833984375, -11.1630859375, 19.972122192382812, 212.5670166015625, 7.466461181640625, 15.28399658203125, -19.570281982421875, 127.666259765625, 120.1783447265625, 76.96991729736328, -75.52520751953125, -0.570068359375, -4.926647186279297, 21.589075088500977, 1.2751617431640625, 119.26837921142578, 143.82998657226562, -18.09942626953125, 132.3661346435547, 136.6182861328125, 97.99130249023438, 48.10841369628906, 21.47625732421875, 3.8863525390625, 138.97259521484375, 136.63760375976562, 79.967529296875, 15.141693115234375, 18.13111114501953, 1.5149307250976562, -65.63615417480469, 181.70648193359375, 67.114501953125, 14.887969970703125, 187.05303955078125, 92.11326599121094, 129.4791717529297, 156.56924438476562, 34.660240173339844, 169.07061767578125, 120.3272705078125, 132.00405883789062, 91.79067993164062, -3.361572265625, -116.92228698730469, 24.316986083984375, 66.2576904296875, 48.7308349609375, 171.45974731445312, 113.60397338867188, 83.64886474609375, 14.538616180419922, 106.60359954833984, 70.81158447265625, 79.28730773925781, 49.703460693359375, 95.00576782226562, 75.61553955078125, 157.67880249023438, 44.001708984375, 33.04991149902344, 91.02730560302734, -5.601287841796875, 128.0723876953125, 32.87921905517578, 139.39468383789062, 4.635108947753906, 118.069580078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000211.npy"}
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 44.28451919555664, "std": 78.47290802001953, "min": -217.81918334960938, "p10": -41.48590393066406, "median": 45.39101791381836, "p90": 150.92008056640623, "max": 222.37908935546875, "pos_frac": 0.75, "sample": [81.73245239257812, 53.94512939453125, 22.690216064453125, 9.230537414550781, -26.352767944335938, -87.35044860839844, -37.458526611328125, 101.15097045898438, 43.170005798339844, -47.82586669921875, 126.54334259033203, -8.391143798828125, 162.853515625, 3.4058189392089844, 30.8895263671875, 3.9272613525390625, 22.124116897583008, 124.09503173828125, 94.82868957519531, 100.45266723632812, 33.03919982910156, 2.304351806640625, -110.977783203125, 37.75054931640625, 127.13127136230469, 72.96307373046875, 40.493927001953125, -38.712989807128906, 78.1336669921875, 30.074508666992188, 31.387958526611328, -24.979202270507812, 38.04264831542969, -12.093673706054688, 36.38553237915039, 59.0086669921875, 72.25439453125, 187.6087188720703, 69.65328979492188, 73.168701171875, 136.40811157226562, -185.75473022460938, 20.643646240234375, 53.315101623535156, -65.38897705078125, 125.05792236328125, 30.547332763671875, 85.331787109375, -99.96165466308594, -20.574878692626953, 57.467918395996094, 52.210540771484375, -152.439453125, 156.46282958984375, 10.695907592773438, -15.73358154296875, -53.81029510498047, 194.73590087890625, 101.55123901367188, 44.63243865966797, -23.20307159423828, -44.08573913574219, 138.26547241210938, 11.015167236328125, 22.087871551513672, 121.33749389648438, -23.447338104248047, 21.65704345703125, 53.925079345703125, 158.54696655273438, 109.33984375, 160.98155212402344, -11.448745727539062, 87.16738891601562, 160.523681640625, 132.54270935058594, 23.45014190673828, 139.9627685546875, -24.91974639892578, 167.05734252929688, 49.281402587890625, 167.1822509765625, 16.305923461914062, 46.14959716796875, -138.19081115722656, 78.60299682617188, 73.42770385742188, -65.54177856445312, -10.411285400390625, 1.8632354736328125, 105.37753295898438, 122.5206298828125, 46.954498291015625, -4.3030548095703125, 148.54461669921875, 90.38995361328125, -15.603271484375, 168.05340576171875, 52.126312255859375, 51.856964111328125, 88.37165069580078, 222.37908935546875, 24.377288818359375, 1.3196182250976562, 53.06414794921875, -217.81918334960938, 106.32061767578125, 59.526275634765625, 90.98504638671875, -29.80902099609375, 80.63441467285156, -63.36260986328125, 60.6409912109375, 10.338743209838867, 88.26629638671875, -40.37168884277344, 53.19476318359375, -10.857658386230469, 30.447071075439453, -12.332794189453125, 21.572967529296875, 49.96453857421875, 8.426101684570312, 159.65057373046875, 5.99407958984375, 104.45912170410156, 220.24853515625, 85.75241088867188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000212.npy"}
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 46.75450897216797, "std": 86.87726593017578, "min": -213.74069213867188, "p10": -49.3861701965332, "median": 35.678409576416016, "p90": 159.85609130859373, "max": 307.7698974609375, "pos_frac": 0.7109375, "sample": [137.2677001953125, 159.52935791015625, 16.585723876953125, 83.64089965820312, 12.577789306640625, 106.25917053222656, 120.93328857421875, 106.31591796875, 14.824188232421875, 160.61846923828125, 92.01994323730469, 107.88800048828125, 49.26826477050781, 15.385551452636719, 6.19514274597168, 183.9005126953125, -14.941207885742188, 18.40802001953125, 57.546142578125, 101.302978515625, 82.0516357421875, -5.733062744140625, 182.12246704101562, -32.024658203125, 265.6114501953125, 104.45361328125, 1.58917236328125, 37.639678955078125, -2.949554443359375, 15.566650390625, 57.32615661621094, -213.74069213867188, -58.94560241699219, 7.7579803466796875, 149.4359588623047, 8.883132934570312, 77.75682830810547, 40.54084014892578, 80.77609252929688, -81.5777587890625, 204.2518768310547, 132.72630310058594, -38.82975769042969, 52.88584899902344, -41.852630615234375, 236.56005859375, 36.543548583984375, 18.388580322265625, -39.920013427734375, 51.985931396484375, 9.840347290039062, 153.08636474609375, 150.61013793945312, 90.82378387451172, 96.72872924804688, -45.02644348144531, -5.65411376953125, 51.25733184814453, -49.11076354980469, 168.1865234375, 15.89080810546875, 289.5069580078125, -1.2078857421875, -16.802215576171875, 31.3994140625, 72.912109375, 197.775390625, -14.440643310546875, -99.76014709472656, -13.76873779296875, 122.87019348144531, -7.6407470703125, 75.23068237304688, -35.91954040527344, 2.87847900390625, -10.989875793457031, 92.87492370605469, 143.58456420898438, -92.8777847290039, 83.35830688476562, 34.813270568847656, 14.30313491821289, 3.613161087036133, -16.945831298828125, -59.759239196777344, 95.7774658203125, -121.32925415039062, 68.30706787109375, 58.316253662109375, 131.97207641601562, 20.899505615234375, 213.39573669433594, 86.72319030761719, 39.495513916015625, 53.0712890625, -106.61653900146484, 120.0855712890625, -34.87908935546875, -50.76190185546875, 307.7698974609375, -10.692169189453125, 27.361038208007812, 113.95187377929688, 7.1798858642578125, -105.73895263671875, 171.29595947265625, 96.8380126953125, 20.564056396484375, 69.58734130859375, 0.0, 10.247138977050781, 117.32260131835938, 76.46340942382812, 174.10525512695312, 11.70730209350586, -103.15682983398438, 99.12640380859375, 0.9959392547607422, -61.85889434814453, 78.319580078125, 9.61489486694336, 59.4964599609375, -40.03041076660156, 33.89698791503906, 86.24627685546875, -50.028785705566406, -41.52970886230469, -45.38141632080078], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000213.npy"}
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 53.9517822265625, "std": 76.35438537597656, "min": -176.76971435546875, "p10": -34.03656234741211, "median": 42.486305236816406, "p90": 149.3336944580078, "max": 283.473388671875, "pos_frac": 0.7734375, "sample": [42.69239807128906, 90.88162231445312, 4.143585205078125, 14.23443603515625, 131.1240234375, 180.47952270507812, 56.02655029296875, -13.070114135742188, 52.4534912109375, -35.922576904296875, 83.43414306640625, 136.1166534423828, 53.404205322265625, 89.3489990234375, -23.75634765625, 98.08206176757812, -53.21978759765625, 104.63548278808594, -56.78778076171875, 6.07080078125, 12.283882141113281, -74.57745361328125, 29.052772521972656, 95.47647094726562, 147.99652099609375, 10.885223388671875, 54.07396697998047, -44.68548583984375, 79.767333984375, 86.22748565673828, 3.6029052734375, 21.44989013671875, 9.905166625976562, -34.870765686035156, 90.47856140136719, 85.34478759765625, 278.73797607421875, 191.1919403076172, 143.2346954345703, 17.204483032226562, 79.73637390136719, -90.03952026367188, -58.35847473144531, 64.43411254882812, 4.1402740478515625, 38.96805191040039, -3.86187744140625, 195.22933959960938, 70.03924560546875, -2.2668113708496094, 86.38705444335938, 85.04949951171875, -31.039840698242188, 138.92535400390625, -98.21856689453125, 151.7450714111328, 4.023162841796875, 17.22601318359375, -15.51702880859375, 74.96246337890625, 120.52854919433594, 168.7890625, 88.58049011230469, 21.94085693359375, 92.18864440917969, -6.332275390625, 35.256927490234375, 2.5755844116210938, -33.679046630859375, 0.0, -7.9301605224609375, 139.56826782226562, 21.669281005859375, 18.058929443359375, 134.78652954101562, -30.96369171142578, 43.39990234375, -19.013416290283203, 114.58425903320312, 98.39474487304688, -47.21269226074219, -176.76971435546875, 120.309326171875, 54.991180419921875, 45.799468994140625, -2.798004150390625, 120.69806671142578, 16.565616607666016, 5.44049072265625, 20.218307495117188, -54.816192626953125, 48.893531799316406, 149.15530395507812, 72.85855102539062, 149.74993896484375, 23.407272338867188, 16.24664306640625, 67.06329345703125, -7.809356689453125, -48.65745544433594, 178.0426025390625, 142.37225341796875, 17.73052978515625, 140.2042236328125, 14.018341064453125, 283.473388671875, 180.4071044921875, 183.87014770507812, -33.644012451171875, 110.09637451171875, 200.48870849609375, 3.3247222900390625, 42.28021240234375, 20.45611572265625, 65.64321899414062, 145.96817016601562, 122.26052856445312, 34.500213623046875, -4.171142578125, 23.696502685546875, 39.499732971191406, 24.71794891357422, 82.07406616210938, 103.85588073730469, 32.25154113769531, 189.22955322265625, 75.17253112792969, 37.5560302734375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000214.npy"}
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 42.540748596191406, "std": 88.45248413085938, "min": -135.20404052734375, "p10": -66.9789825439453, "median": 26.206028938293457, "p90": 158.14507751464842, "max": 338.6185302734375, "pos_frac": 0.671875, "sample": [44.88911819458008, -47.44905090332031, 47.56842041015625, 129.84072875976562, 203.747314453125, 173.30548095703125, 131.3319091796875, 111.74956512451172, 110.01211547851562, -2.19134521484375, 134.4073486328125, 80.90042114257812, 254.6444091796875, -77.36679077148438, 15.265167236328125, -16.38775634765625, 100.42216491699219, 200.75323486328125, 338.6185302734375, 115.20978546142578, 118.03903198242188, -44.0401611328125, 0.0, -75.13189697265625, -0.8036308288574219, -34.02070236206055, 43.39097595214844, 115.77053833007812, 218.2789306640625, 4.479698181152344, 156.49298095703125, 11.64276123046875, -8.322265625, -116.312744140625, 28.434722900390625, 56.253875732421875, 129.43197631835938, -52.70948791503906, -12.414886474609375, 166.4561767578125, -7.207611083984375, 21.118438720703125, 5.5838623046875, -66.34886169433594, 4.978004455566406, 74.76837158203125, -3.61822509765625, 6.591796875, 161.99996948242188, 12.979202270507812, 88.5630874633789, 294.52294921875, 152.96286010742188, -2.38055419921875, -119.00921630859375, -9.619121551513672, 54.76408386230469, -36.280792236328125, 0.0, 100.75457763671875, -26.46478271484375, 99.58936309814453, 69.73046875, -54.5384521484375, 73.90884399414062, 94.23857879638672, -66.2674560546875, 0.07077789306640625, 120.67364501953125, 36.274261474609375, 36.95184326171875, 22.20940589904785, -12.755340576171875, 129.75518798828125, -135.20404052734375, 113.8140869140625, 36.419677734375, -125.83441925048828, 19.90432357788086, 67.3656005859375, 81.57466125488281, 27.40930938720703, 50.392120361328125, -37.217559814453125, -92.91339111328125, -0.2718353271484375, 178.32867431640625, 15.738555908203125, 116.58929443359375, -63.00408935546875, -30.576385498046875, 15.938018798828125, -7.0131378173828125, -16.10101318359375, 11.575149536132812, 107.80585479736328, -10.435811996459961, 36.31626892089844, 114.454345703125, 119.31884765625, -61.41094970703125, 94.1358642578125, 10.081451416015625, 2.8682594299316406, -78.65968322753906, 50.977996826171875, 24.22882080078125, 171.7688446044922, -68.44926452636719, 49.80853271484375, 34.912841796875, 1.255990982055664, 25.002748489379883, 27.89312744140625, 152.79896545410156, 109.69439697265625, 1.1414031982421875, 55.43586730957031, -114.41973876953125, 1.78131103515625, 178.85336303710938, -80.28482055664062, 68.51385498046875, 108.16143798828125, 6.353668212890625, -72.57379150390625, -78.51300048828125, 216.79934692382812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000215.npy"}
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 50.732765197753906, "std": 92.61478424072266, "min": -188.47604370117188, "p10": -61.65228271484374, "median": 47.353271484375, "p90": 163.003466796875, "max": 378.2462158203125, "pos_frac": 0.7109375, "sample": [-17.998069763183594, 126.06584167480469, 146.7775115966797, 82.21957397460938, 23.7933349609375, 139.37918090820312, -154.64990234375, 67.20623779296875, 80.28958129882812, -5.969703674316406, 158.56649780273438, 116.10406494140625, 96.65316772460938, 23.093154907226562, 40.033447265625, 90.75991821289062, 378.2462158203125, 2.3933944702148438, 157.70169067382812, 125.70626831054688, 50.6953125, -123.401123046875, 24.12548828125, -33.165191650390625, -76.42534637451172, 34.552001953125, -70.6634521484375, -0.6317291259765625, 127.6640625, 122.68414306640625, 17.458465576171875, -124.51174926757812, -68.0784912109375, 34.61468505859375, 139.54959106445312, 28.32745361328125, 158.11090087890625, -188.47604370117188, 211.39920043945312, 158.24559020996094, -112.84535217285156, 202.5337677001953, 144.759033203125, 160.02085876464844, 180.1737518310547, 124.79840087890625, 52.3592529296875, 139.42550659179688, -0.183502197265625, 21.880096435546875, 63.83056640625, -23.010589599609375, 162.767333984375, 66.24490356445312, 1.9377670288085938, 65.31854248046875, 29.551116943359375, 165.4031982421875, -58.67730712890625, -58.898193359375, 0.0, 71.25521850585938, -46.386993408203125, 114.89279174804688, 12.719940185546875, 85.42138671875, 219.3492431640625, -33.846710205078125, -98.90304565429688, -14.859619140625, 106.01824951171875, 93.57989501953125, -123.81378173828125, 32.02655029296875, 168.34417724609375, 197.19171142578125, -1.1796875, 205.25445556640625, 80.7186279296875, -35.38800048828125, -47.22727966308594, -43.005401611328125, 103.8419189453125, 34.016845703125, 131.1422119140625, 51.14704895019531, -26.438308715820312, -34.63664245605469, 163.554443359375, 172.767333984375, -78.43414306640625, -23.374176025390625, 44.273193359375, 111.12554931640625, -18.992462158203125, 50.433349609375, 19.2403564453125, -0.747100830078125, 78.15613555908203, 20.604095458984375, 227.05618286132812, 50.66843032836914, 0.0, 59.3839111328125, 142.14633178710938, -115.22822570800781, 39.84388732910156, 117.72112274169922, 94.62581634521484, -4.8264923095703125, 3.0996227264404297, 83.85098266601562, 58.200408935546875, 53.171630859375, 103.69891357421875, 117.16267395019531, 3.260284423828125, 22.6046142578125, 13.48553466796875, 205.322265625, 30.8343505859375, -163.65414428710938, 103.61837768554688, 57.02583312988281, 13.7314453125, 6.92218017578125, -20.39410400390625, 24.7861328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000216.npy"}
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 63.56452941894531, "std": 89.3263168334961, "min": -170.45037841796875, "p10": -29.270997619628886, "median": 55.95344543457031, "p90": 175.46547241210936, "max": 351.639892578125, "pos_frac": 0.765625, "sample": [29.409217834472656, 24.72186279296875, -64.49664306640625, 54.66548156738281, 114.4266357421875, -18.475997924804688, 11.956512451171875, 264.7276306152344, -86.66291809082031, 61.6737060546875, 183.8371124267578, 12.44085693359375, 68.29576110839844, 114.84664916992188, 21.68768310546875, 188.0947723388672, 25.479164123535156, 108.81291198730469, 115.6573486328125, -12.40142822265625, 52.19915771484375, 51.64080810546875, 79.5604248046875, 130.48291015625, 351.639892578125, 2.3527984619140625, 81.13414001464844, 109.43284606933594, 95.87905883789062, 168.11526489257812, 42.53179931640625, 164.74371337890625, -14.89398193359375, 266.11248779296875, 228.1025390625, -23.138031005859375, 238.52236938476562, 123.19598388671875, 161.12814331054688, 61.532562255859375, -116.69589233398438, 211.03265380859375, 82.47256469726562, 178.18258666992188, -49.96272277832031, 112.36463928222656, 88.52249908447266, 5.39776611328125, 24.31805419921875, 25.56292724609375, 10.520679473876953, -6.538063049316406, -52.999908447265625, 64.23489379882812, 86.4481201171875, -91.02957153320312, 74.21562194824219, 157.88946533203125, 32.69410705566406, -43.58125305175781, -4.117084503173828, 17.68408203125, 59.276702880859375, 104.36026000976562, 21.518707275390625, -3.6222152709960938, 86.70003509521484, 133.06988525390625, -2.4892826080322266, 33.657073974609375, 148.5624542236328, 161.31661987304688, 134.8861541748047, -9.52505874633789, 173.33245849609375, -56.40129089355469, 18.115936279296875, -21.03215789794922, 71.69692993164062, 207.68280029296875, 31.140457153320312, 194.19903564453125, 32.87133026123047, -3.7421340942382812, -72.36907958984375, 174.30099487304688, 149.90371704101562, 214.07586669921875, 4.5020751953125, 3.9632110595703125, 19.582305908203125, 106.9261474609375, -13.310211181640625, 167.443603515625, 99.65505981445312, 94.28216552734375, 64.76700592041016, -3.0822830200195312, 7.7508087158203125, 60.49566650390625, -13.26239013671875, 0.0, 36.936431884765625, 110.01957702636719, -132.81280517578125, 172.36663818359375, 109.57632446289062, -5.5311279296875, 81.6470947265625, 249.81509399414062, 117.87078857421875, 3.2331695556640625, 7.666259765625, 123.07989501953125, 57.24140930175781, 22.149551391601562, 125.27813720703125, 50.601287841796875, -170.45037841796875, -101.15266418457031, 150.43890380859375, -67.56771850585938, 97.37530517578125, 2.80926513671875, 102.56411743164062, -12.794967651367188, 24.99114990234375, 4.122936248779297], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000217.npy"}
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 55.73572540283203, "std": 85.2222900390625, "min": -174.81072998046875, "p10": -33.561972427368154, "median": 43.239776611328125, "p90": 157.98018951416014, "max": 355.71514892578125, "pos_frac": 0.7265625, "sample": [37.88525390625, 45.89654541015625, 42.705535888671875, 135.11807250976562, -30.519744873046875, 42.20843505859375, 13.650665283203125, 355.71514892578125, -1.1410980224609375, 20.089962005615234, -24.48235321044922, -0.7648162841796875, 158.74729919433594, 200.12857055664062, 101.42422485351562, -104.51583862304688, -48.581329345703125, 134.02603149414062, 21.782669067382812, 125.92237854003906, -8.637638092041016, 119.61839294433594, 130.75424194335938, 118.52809143066406, 38.83935546875, -14.214370727539062, -26.946502685546875, 2.0377197265625, -18.768402099609375, 152.36007690429688, 108.56976318359375, 243.2298583984375, -3.5097579956054688, 99.24395751953125, 36.50775146484375, 127.12701416015625, 123.29743194580078, -61.134307861328125, 173.5517578125, 109.87835693359375, 171.28985595703125, -40.66050338745117, -17.666046142578125, 82.35552978515625, 19.562973022460938, 141.44598388671875, -2.6965560913085938, -132.78729248046875, 37.383331298828125, 213.90692138671875, 145.8189697265625, 13.963340759277344, 58.998321533203125, 18.140724182128906, 146.16372680664062, 2.5219058990478516, 248.5860595703125, 92.23526000976562, 98.36747741699219, -88.94464111328125, 26.73211669921875, 18.480880737304688, 44.433685302734375, -3.5222320556640625, 0.0, 75.42971801757812, 75.66290283203125, 4.0281982421875, -21.405807495117188, 116.74191284179688, 104.44760131835938, 86.918212890625, 30.90289306640625, 43.774017333984375, 140.12176513671875, 66.57785034179688, -49.930397033691406, 28.988998413085938, 35.6971435546875, 27.740005493164062, 182.1298828125, 157.65142822265625, 157.35479736328125, 15.693115234375, 39.11053466796875, 54.43389892578125, -55.76953125, -6.927040100097656, 103.186767578125, 46.1490478515625, -25.293563842773438, 21.714309692382812, -11.6783447265625, -174.81072998046875, 102.5274887084961, 52.81360626220703, 160.7583770751953, 256.32952880859375, -97.96173095703125, 96.6541748046875, 35.855316162109375, 72.27499389648438, -14.856132507324219, 83.1817626953125, 64.23760986328125, 57.068634033203125, -168.81581115722656, 111.51754760742188, -66.83609008789062, 123.5340576171875, 19.247528076171875, 132.791015625, 76.2794189453125, 58.08819580078125, -55.2861328125, 208.71429443359375, 126.09625244140625, 35.4466552734375, -11.777984619140625, -15.920074462890625, -1.61090087890625, 85.23773193359375, 80.81512451171875, -9.631866455078125, 75.92059326171875, 169.11370849609375, 35.879058837890625, 18.109039306640625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000218.npy"}
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 57.083457946777344, "std": 89.32341003417969, "min": -222.2679443359375, "p10": -36.7398193359375, "median": 53.579811096191406, "p90": 183.11747436523433, "max": 297.5125732421875, "pos_frac": 0.765625, "sample": [-149.20819091796875, 39.85472106933594, 41.62091827392578, 80.24760437011719, 236.15451049804688, 82.95234680175781, -36.60699462890625, 107.70953369140625, 137.04229736328125, 53.8388671875, 63.87646484375, -6.6675262451171875, -22.52191162109375, -110.54678344726562, 4.653984069824219, 45.50660705566406, 86.61355590820312, -30.719451904296875, 41.37937927246094, 124.44906616210938, 53.32075500488281, 10.423128128051758, 38.410400390625, 131.55178833007812, 101.86380767822266, 96.66806030273438, 57.82313537597656, 10.829507827758789, 96.9888916015625, 17.188125610351562, -112.50053405761719, 74.79684448242188, -133.53585815429688, 64.21183013916016, 13.753204345703125, 120.11189270019531, -47.27557373046875, 133.6094970703125, 104.353515625, 169.64224243164062, 198.6256103515625, 207.71624755859375, 73.90943908691406, -37.04974365234375, 15.761474609375, -33.09295654296875, -73.20703125, 83.072998046875, 116.80801391601562, 174.4952392578125, 0.0, 37.88359069824219, 20.202301025390625, 5.72784423828125, -19.262413024902344, 63.65411376953125, -15.458740234375, 120.23504638671875, 113.58883666992188, -0.12366485595703125, 148.09674072265625, 224.76904296875, 13.878463745117188, -54.925506591796875, -9.183891296386719, -17.52740478515625, 2.77044677734375, 11.490154266357422, 4.7613372802734375, 159.5804901123047, 11.29522705078125, 60.65728759765625, 20.793426513671875, 198.02239990234375, 44.060546875, -6.140777587890625, -62.62530517578125, -5.6221923828125, 48.6978759765625, 219.632568359375, 28.673667907714844, 175.2164764404297, 170.0855712890625, 83.78179931640625, 24.29991340637207, -66.35989379882812, 179.28436279296875, -0.02922821044921875, 237.0360107421875, 5.46746826171875, 77.40390014648438, 56.423614501953125, 57.990478515625, 36.45306396484375, 55.832176208496094, 173.030029296875, 117.69862365722656, 63.8502197265625, 100.27487182617188, 192.0614013671875, 14.369270324707031, 111.99211120605469, 219.47601318359375, -167.0977325439453, 204.17092895507812, 86.99703979492188, 57.79443359375, 14.629013061523438, 66.11306762695312, 51.84400939941406, -7.709148406982422, 3.6956787109375, 57.54791259765625, -222.2679443359375, 194.56759643554688, 248.88287353515625, 22.161697387695312, 97.73936462402344, 149.45181274414062, 66.70054626464844, 13.851318359375, -11.109710693359375, 78.6053466796875, 58.63999938964844, 297.5125732421875, 28.866317749023438, -6.990272521972656, -50.05706787109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000219.npy"}
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 60.62110900878906, "std": 83.29022216796875, "min": -175.22695922851562, "p10": -49.173211669921876, "median": 58.57145309448242, "p90": 163.6856719970703, "max": 285.69781494140625, "pos_frac": 0.7734375, "sample": [143.169189453125, 285.69781494140625, 65.01312255859375, 95.90658569335938, 4.354799270629883, 22.017578125, -24.50799560546875, 75.79702758789062, 35.52935028076172, 111.54074096679688, 65.78594970703125, 42.819244384765625, 165.95986938476562, 164.47943115234375, 8.994537353515625, 106.48065185546875, 113.02001953125, 106.28054809570312, 110.7738037109375, 39.69013977050781, 26.762779235839844, 58.0614013671875, 11.639942169189453, 9.751632690429688, 71.11466979980469, 154.61325073242188, 21.7490234375, 140.9656219482422, -49.72406005859375, 70.03511047363281, 55.580284118652344, 18.160812377929688, -19.2265625, -43.45954895019531, 64.50509643554688, -3.310791015625, 17.99749755859375, 85.36471557617188, 6.228143692016602, 183.39599609375, 139.42184448242188, 112.4482421875, 158.3878936767578, -27.008514404296875, 189.5323486328125, 137.46795654296875, -85.5185546875, -175.22695922851562, 35.9136962890625, 34.0816650390625, 160.1798095703125, 77.16666412353516, 15.39208984375, 249.3294677734375, 86.82952117919922, 42.31011199951172, -48.9371337890625, 3.1478729248046875, -7.90673828125, 141.30587768554688, -27.594558715820312, 115.94828796386719, 163.34548950195312, 119.78767395019531, 21.95861053466797, -32.4320068359375, -6.333526611328125, 69.89599609375, -83.38438415527344, 176.01065063476562, 92.58493041992188, 124.09695434570312, 77.96646118164062, 157.8607177734375, 104.70114135742188, 68.38877868652344, -82.2818603515625, 44.09100341796875, 28.40749168395996, -9.17388916015625, -61.603271484375, 59.081504821777344, 28.123680114746094, 27.60211181640625, 87.44639587402344, 143.29449462890625, -45.69189453125, 57.265899658203125, 202.58425903320312, 180.22830200195312, -61.294281005859375, 31.98834228515625, -53.561798095703125, -70.30116271972656, 100.72164916992188, 131.58462524414062, 193.3846435546875, 36.853431701660156, 51.689239501953125, -6.06121826171875, 140.08428955078125, 127.18547821044922, -55.13897705078125, -141.073486328125, 123.79916381835938, -10.870498657226562, 202.4639892578125, -43.702850341796875, 172.0164794921875, 183.09954833984375, 101.1661376953125, 53.253997802734375, 91.1055908203125, 19.060287475585938, 150.48095703125, 13.32550048828125, 137.50518798828125, 159.91555786132812, 98.87655639648438, 108.343505859375, -89.42681884765625, 41.162109375, 152.33815002441406, 21.099756240844727, 9.699951171875, -81.97637176513672, 104.808837890625, -43.60417175292969], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000220.npy"}
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 48.812232971191406, "std": 91.48057556152344, "min": -187.27285766601562, "p10": -52.86272850036621, "median": 29.505996704101562, "p90": 174.9791275024414, "max": 308.8925476074219, "pos_frac": 0.6953125, "sample": [4.09375, 75.53070068359375, 17.3612060546875, -20.05267333984375, 10.510726928710938, 75.61575317382812, 13.301811218261719, 131.07980346679688, -61.83973693847656, -22.592222213745117, -19.215248107910156, 181.7233428955078, 14.71624755859375, 65.537841796875, 0.0657958984375, 90.56884765625, 107.35293579101562, -80.46480560302734, 101.56487274169922, -22.482994079589844, 85.77200317382812, -2.345733642578125, 308.8925476074219, -40.086509704589844, 133.12484741210938, 172.9882354736328, 95.64126586914062, 226.81613159179688, 94.77384948730469, 49.546966552734375, 31.5240478515625, 14.281570434570312, 274.34423828125, -168.01809692382812, 3.98876953125, -12.57489013671875, 5.15667724609375, 74.89251708984375, 1.873291015625, 236.65753173828125, -17.5670166015625, 41.823577880859375, -31.367294311523438, 31.2603759765625, -77.15516662597656, 16.054473876953125, 159.42071533203125, -68.461181640625, 180.5257110595703, -3.709686279296875, -1.1833877563476562, 153.09927368164062, 23.84539794921875, 30.168609619140625, -12.40455436706543, 125.4378662109375, 13.78643798828125, -15.849822998046875, 71.24275207519531, 126.3074951171875, -17.630401611328125, 34.136627197265625, -17.9761962890625, 24.91527557373047, 63.68548583984375, 69.6645736694336, 204.01144409179688, 66.88883972167969, -1.7689056396484375, 83.38005828857422, 28.8433837890625, -9.228546142578125, 83.88540649414062, 163.32080078125, 112.114013671875, 40.23944091796875, 197.71084594726562, 51.567352294921875, -84.21803283691406, -187.27285766601562, 245.3370361328125, 129.13632202148438, -55.751251220703125, 27.007980346679688, -59.6380615234375, 227.3011016845703, 8.884063720703125, -6.908294677734375, 78.14041137695312, 76.73304748535156, 158.4830322265625, 123.737060546875, -30.502197265625, 8.956695556640625, -90.39981079101562, 79.62077331542969, -78.79074096679688, 155.30987548828125, -31.11572265625, 117.61849975585938, 17.40521240234375, -151.853759765625, -21.73638153076172, 179.62454223632812, 43.2210693359375, 20.16619873046875, 138.872314453125, 192.7625732421875, 32.210479736328125, 11.647186279296875, 14.521072387695312, 137.15875244140625, -32.27166748046875, -51.62479019165039, 8.651962280273438, 0.9314498901367188, 122.6715087890625, 5.38836669921875, 168.81033325195312, -5.574951171875, 97.55496978759766, 96.7835693359375, -46.3839111328125, 223.913330078125, -13.43096923828125, 32.30879211425781, -119.75009155273438, 161.66085815429688], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000221.npy"}
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 52.52268981933594, "std": 79.91925811767578, "min": -201.67724609375, "p10": -41.622468566894526, "median": 47.634965896606445, "p90": 163.23670043945313, "max": 227.1588134765625, "pos_frac": 0.7109375, "sample": [56.43373107910156, 2.401519775390625, 223.24789428710938, 107.34719848632812, -20.35009765625, -17.24798583984375, 43.445465087890625, 98.72062683105469, 50.29345703125, 91.39280700683594, 21.658050537109375, 227.1588134765625, 146.70416259765625, 12.812606811523438, 57.3924560546875, -46.962013244628906, 101.30338287353516, 178.39785766601562, 69.5623779296875, -11.20941162109375, 42.979957580566406, 8.479972839355469, -22.359039306640625, -14.9976806640625, 5.09393310546875, 204.437255859375, 90.18582153320312, -50.45367431640625, -91.44752502441406, 16.360755920410156, -22.376388549804688, 42.87945556640625, 202.121337890625, -60.64753723144531, 75.85621643066406, -32.87174987792969, -14.37432861328125, 26.10662841796875, 157.6946563720703, 215.6688232421875, -0.1808624267578125, -96.96307373046875, 38.661163330078125, -0.23220062255859375, 147.73971557617188, 19.6838321685791, -35.877288818359375, 6.465156555175781, -201.67724609375, 71.14678955078125, 40.957733154296875, 107.34870147705078, -2.8603248596191406, 187.13478088378906, -31.095630645751953, 118.65411376953125, 46.12715148925781, 162.42083740234375, -3.5162391662597656, 12.80523681640625, 120.12063598632812, 50.852325439453125, 67.21884155273438, 82.99749755859375, 76.72152709960938, 12.4266357421875, 140.01531982421875, 96.86984252929688, 101.2591552734375, 31.272628784179688, 157.7606658935547, 147.0037078857422, -34.47526550292969, 123.87249755859375, -54.408294677734375, 135.3887481689453, 127.53863525390625, -32.98603057861328, 149.2096405029297, -43.3673095703125, -46.126487731933594, 106.19961547851562, -1.753622055053711, 80.18630981445312, -102.57749938964844, 4.76593017578125, 83.27029418945312, -33.138702392578125, -12.900260925292969, 139.18527221679688, 165.140380859375, 167.78564453125, 105.30602264404297, 57.862586975097656, 11.388956069946289, -40.87467956542969, 34.08812713623047, 81.24638366699219, 45.08941650390625, 108.060546875, 175.44720458984375, 74.53715515136719, 58.614990234375, 61.189361572265625, 167.08331298828125, -17.80127716064453, 97.83139038085938, 48.55671691894531, -24.939727783203125, 128.765625, 40.634429931640625, 62.910858154296875, -116.17973327636719, 111.81283569335938, -13.313682556152344, 12.681747436523438, -46.18212890625, 182.98159790039062, -70.27865600585938, 109.1075439453125, 50.2119140625, 5.229736328125, 116.85940551757812, 171.07882690429688, 46.671295166015625, -10.982841491699219, 88.5850830078125, 46.71321487426758], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000222.npy"}
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 53.958213806152344, "std": 86.12184143066406, "min": -154.77772521972656, "p10": -64.77842025756836, "median": 53.40629577636719, "p90": 175.31161499023438, "max": 275.8093566894531, "pos_frac": 0.7890625, "sample": [13.040901184082031, 3.121490478515625, 101.8465576171875, 113.62216186523438, 54.60577392578125, -74.76219177246094, -64.13469696044922, 32.34901428222656, 75.27239227294922, 221.30203247070312, 38.322479248046875, 206.46044921875, 34.60639953613281, -25.368560791015625, 30.65606689453125, 86.69720458984375, 191.12466430664062, -127.51129150390625, 73.63377380371094, -111.78012084960938, 99.96673583984375, 174.6109619140625, 10.012786865234375, 54.61968994140625, 37.90849304199219, -154.77772521972656, -4.0971527099609375, 109.60400390625, 56.883636474609375, -39.388153076171875, 113.65817260742188, 24.514251708984375, 98.2774658203125, 28.422332763671875, 170.50927734375, 144.2615966796875, 176.94647216796875, -139.30276489257812, 120.90123748779297, -50.67283630371094, 33.738548278808594, 43.05816650390625, 38.60333251953125, 38.311546325683594, 187.7640838623047, 41.827301025390625, 148.81105041503906, 21.71954345703125, 193.30844116210938, 141.52993774414062, 21.714324951171875, 32.1766357421875, 131.42391967773438, 228.69833374023438, 58.386619567871094, 100.12518310546875, -5.35888671875, 182.3538360595703, 73.53667449951172, 82.1702880859375, 39.62542724609375, 84.80476379394531, 76.12554931640625, 159.31243896484375, 3.0490341186523438, 78.06581115722656, 7.111602783203125, 111.829345703125, 128.8178253173828, 24.36505126953125, 53.656585693359375, 59.712554931640625, 13.027191162109375, 158.60397338867188, -79.5435562133789, -33.28551483154297, -113.18592071533203, -66.28044128417969, 78.96624755859375, 68.19012451171875, 40.76054382324219, 9.911224365234375, -3.1215133666992188, 57.464111328125, 188.28097534179688, 68.1067886352539, 53.156005859375, -14.230659484863281, 112.28921508789062, -11.450702667236328, -3.026580810546875, 13.911842346191406, 26.643783569335938, 249.34716796875, 211.35986328125, 275.8093566894531, 149.4810791015625, 78.25492858886719, 8.624771118164062, 69.62680053710938, 9.33401107788086, 49.59355163574219, 96.43670654296875, -13.340375900268555, 93.08193969726562, 3.485076904296875, 153.89483642578125, 93.79220581054688, 3.8391342163085938, -33.66841125488281, 74.3155517578125, -107.15967559814453, 2.759521484375, 29.752548217773438, 179.41436767578125, 79.91990661621094, -73.79302978515625, -13.457717895507812, 89.08946228027344, 78.52093505859375, -130.60260009765625, 142.74984741210938, 65.318359375, 15.82232666015625, 55.67742919921875, 51.54711151123047, -120.38629150390625, -103.31472778320312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000223.npy"}
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 61.35163497924805, "std": 83.8979721069336, "min": -116.6692123413086, "p10": -33.08531265258788, "median": 49.38755798339844, "p90": 178.04499206542968, "max": 301.972412109375, "pos_frac": 0.7578125, "sample": [182.8575439453125, 223.5284423828125, 29.282867431640625, 25.462203979492188, 15.007257461547852, -72.37789916992188, -81.46725463867188, 154.26498413085938, 25.893096923828125, 143.4404754638672, 72.40914916992188, 145.20187377929688, 9.649078369140625, 17.26299285888672, 156.37893676757812, 38.795166015625, -53.46270751953125, 154.91033935546875, 19.182464599609375, -71.6357421875, 78.49771118164062, -76.32765197753906, 117.75361633300781, 197.45144653320312, 83.39553833007812, 24.063785552978516, 78.62669372558594, 42.91717529296875, 93.38900756835938, 88.608154296875, 182.1942138671875, 118.66024780273438, -22.752685546875, 101.01248168945312, 118.17578125, -12.718536376953125, 36.177276611328125, 112.81027221679688, 133.78213500976562, 75.76339721679688, 37.93562316894531, 110.20333862304688, 18.16663360595703, -17.878467559814453, 46.8189697265625, 97.85362243652344, 103.2501220703125, -85.8747329711914, 105.49505615234375, -21.338315963745117, 122.06585693359375, 94.14291381835938, 0.0, 150.32669067382812, 77.74522399902344, 54.149261474609375, 26.598220825195312, -55.305206298828125, 4.879364013671875, 42.258026123046875, 87.55464172363281, -7.892669677734375, 82.53703308105469, 205.87437438964844, -6.488433837890625, 175.90652465820312, 6.762176513671875, -18.0458984375, -4.82623291015625, -8.982803344726562, 216.1859130859375, 264.8921813964844, 8.499908447265625, 74.74554443359375, 36.08197021484375, -25.8521728515625, -3.6210479736328125, 210.61380004882812, 43.84869384765625, 156.14556884765625, -12.246810913085938, -108.23960876464844, -45.63441467285156, -7.1745452880859375, 102.88330078125, 207.761962890625, 75.77862548828125, 15.41790771484375, 253.4288787841797, -116.6692123413086, 155.62240600585938, 35.30023193359375, 80.59982299804688, 176.26675415039062, 35.7012939453125, 52.050048828125, 82.56524658203125, 51.956146240234375, 108.55859375, 26.03326416015625, -112.75923156738281, 64.00833892822266, -17.966339111328125, 76.5703125, -57.801490783691406, 16.171157836914062, 218.93246459960938, -30.279457092285156, 301.972412109375, 58.646209716796875, 25.18182373046875, -23.965560913085938, 0.49812889099121094, 38.011383056640625, -12.887237548828125, 81.41627502441406, 128.53665161132812, 46.790626525878906, 80.608642578125, 100.56954956054688, 36.776092529296875, 242.57534790039062, 38.48822021484375, 17.5272216796875, 101.27120971679688, 72.86997985839844, -39.63230895996094, 119.424560546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000224.npy"}
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 52.82275390625, "std": 95.59945678710938, "min": -203.6018829345703, "p10": -44.03375549316406, "median": 38.43998718261719, "p90": 194.4032989501953, "max": 337.09149169921875, "pos_frac": 0.703125, "sample": [337.09149169921875, -35.43927001953125, 84.99676513671875, 162.92706298828125, 160.3857421875, 27.4822998046875, -113.30978393554688, -93.45745849609375, -72.28045654296875, 171.52102661132812, 19.872344970703125, 151.22723388671875, 35.230072021484375, -36.113677978515625, 194.49197387695312, 52.58868408203125, 213.92825317382812, 137.42221069335938, 22.521217346191406, 92.44198608398438, -15.5228271484375, -4.27032470703125, -27.58922576904297, -29.95752716064453, 120.10031127929688, 1.44610595703125, 204.4581298828125, 4.1381072998046875, 74.1441650390625, -36.2650146484375, -37.930389404296875, -102.69195556640625, 153.9093475341797, 53.56793212890625, 11.885147094726562, 69.1250228881836, -19.20526123046875, -81.954833984375, 5.82952880859375, 29.92535400390625, 61.9251708984375, 232.27166748046875, 5.3307647705078125, -88.30392456054688, -178.98458862304688, 67.35140991210938, 139.3712921142578, 89.62261962890625, 48.62462615966797, 63.625701904296875, 6.4752655029296875, 55.31854248046875, 73.050048828125, 95.6612548828125, 39.465362548828125, 100.14910888671875, -86.592041015625, -46.56982421875, 99.78768920898438, 133.9816436767578, 3.755401611328125, 235.1034393310547, 203.0628662109375, -0.17132568359375, 106.27606201171875, -4.7175140380859375, -42.842529296875, 100.5479736328125, 39.763450622558594, 182.91213989257812, 35.777496337890625, 89.42236328125, 89.10595703125, 170.0609588623047, 30.63665771484375, 92.17304992675781, 130.0410614013672, 237.98287963867188, -24.06939697265625, 80.56842803955078, -6.9153594970703125, -111.06195068359375, -27.026412963867188, 18.39938735961914, 60.19520568847656, -42.946868896484375, 140.53570556640625, -0.9058685302734375, 70.80514526367188, -203.6018829345703, 29.36376953125, -29.0721435546875, -96.43670654296875, 55.5654296875, 17.196685791015625, -16.2388916015625, 21.437545776367188, -15.393753051757812, 207.21466064453125, 71.24395751953125, 27.792781829833984, 72.47346496582031, 75.21038818359375, 274.66412353515625, -23.9423828125, 131.5316162109375, 120.04015350341797, 7.2475738525390625, 14.4874267578125, 194.36529541015625, -29.7012939453125, 127.18658447265625, 125.84426879882812, 246.6444091796875, 5.161376953125, 68.52647399902344, 87.70822143554688, -39.404296875, 115.86909484863281, -34.87255859375, 28.412841796875, -21.86260986328125, 32.363677978515625, 37.41461181640625, -73.00164794921875, 34.81059265136719, 211.177490234375, 249.18646240234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000225.npy"}
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 46.73792266845703, "std": 89.05424499511719, "min": -206.04135131835938, "p10": -43.94480056762695, "median": 31.57787322998047, "p90": 157.9598876953125, "max": 325.1612548828125, "pos_frac": 0.7109375, "sample": [31.984619140625, 228.36782836914062, -4.364501953125, 15.072479248046875, 132.09674072265625, 245.07855224609375, 3.2614288330078125, -92.61734008789062, 51.73716735839844, 50.497650146484375, 131.25634765625, 122.38482666015625, 17.23345947265625, 93.46694946289062, 6.21234130859375, 29.486797332763672, 3.832366943359375, 15.886444091796875, 150.45904541015625, 127.84597778320312, 18.111724853515625, 19.99219512939453, 70.51960754394531, 102.57799530029297, 87.94830322265625, 37.183746337890625, -16.62289047241211, -206.04135131835938, 157.52499389648438, 42.623687744140625, -42.99413299560547, 216.07794189453125, 55.9705810546875, 133.70120239257812, -24.752777099609375, 80.1776123046875, 153.05337524414062, -184.9515380859375, 12.388519287109375, 138.18252563476562, -23.214950561523438, 1.5534591674804688, -11.067176818847656, -3.6573009490966797, 50.0819091796875, 151.80624389648438, -11.095748901367188, -84.62914276123047, -92.35784912109375, 148.30706787109375, 236.55580139160156, -61.9835205078125, 173.8125, 44.78676986694336, 24.416545867919922, 15.740936279296875, 13.193450927734375, 41.6531982421875, -22.001447677612305, -4.5800018310546875, -30.025802612304688, 125.28862762451172, 28.468368530273438, -93.2755126953125, 127.41352844238281, -9.56991195678711, 62.960731506347656, 325.1612548828125, 1.657968521118164, 46.381683349609375, -15.816680908203125, 14.717018127441406, 57.56785583496094, 14.85565185546875, 10.377033233642578, -17.429595947265625, 21.199508666992188, 131.32012939453125, 4.542655944824219, 181.88436889648438, -3.0364532470703125, 53.001739501953125, 128.58660888671875, -90.02359771728516, 191.087158203125, 36.756256103515625, 6.631317138671875, 110.2830810546875, 58.667236328125, 40.297027587890625, 100.95346069335938, 123.2509765625, 177.6175537109375, 10.24017333984375, 70.44598388671875, 13.653396606445312, 45.210479736328125, -32.819000244140625, 31.171127319335938, 33.3880615234375, 131.2241668701172, 46.18096923828125, 158.97463989257812, -74.89944458007812, -6.21087646484375, 26.383575439453125, 258.1156921386719, 65.35186767578125, 55.3702392578125, -36.259307861328125, -123.83052062988281, 29.542144775390625, -0.0108184814453125, 87.47233581542969, 174.44418334960938, -22.23296356201172, 56.061309814453125, -7.83057975769043, -46.16302490234375, 270.9747314453125, 41.41064453125, -96.09506225585938, -78.10409545898438, 107.40906524658203, -3.0844879150390625, -14.64111328125, 112.39083862304688, -19.702667236328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000226.npy"}
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 60.899742126464844, "std": 91.58000183105469, "min": -164.62542724609375, "p10": -64.59556274414062, "median": 59.627838134765625, "p90": 175.92718505859375, "max": 265.13519287109375, "pos_frac": 0.7265625, "sample": [-80.89286804199219, 190.9652557373047, 175.51913452148438, 6.5036163330078125, 93.08766174316406, -5.638179779052734, 214.69094848632812, 128.700439453125, -34.515892028808594, -76.959716796875, 172.6868896484375, 93.5562744140625, 31.34954833984375, 1.614654541015625, -23.761337280273438, 125.1728286743164, 0.0, 168.61224365234375, -164.62542724609375, -81.02894592285156, 115.21707153320312, -105.7825927734375, 136.3761749267578, 158.74468994140625, 176.87930297851562, 189.53057861328125, 112.1290283203125, -117.92617797851562, 133.2576141357422, -11.338287353515625, -48.79243469238281, 140.95355224609375, -101.46638488769531, 60.205963134765625, 41.8134765625, -31.11053466796875, -3.8710174560546875, 39.696434020996094, 45.643157958984375, 202.78619384765625, 171.6925048828125, 265.13519287109375, 52.719329833984375, 130.66265869140625, 150.1707763671875, -94.62171173095703, 106.032470703125, 88.74365234375, 2.2370147705078125, 54.81158447265625, 206.5037841796875, 69.848876953125, -12.221435546875, -37.460418701171875, 87.74435424804688, 163.38101196289062, 39.6771240234375, 141.55882263183594, 134.5738525390625, 31.516605377197266, 214.7367401123047, -86.42681121826172, 75.72830200195312, 144.56265258789062, -155.52178955078125, -8.8515625, 19.918498992919922, 137.76092529296875, -20.342498779296875, 134.96719360351562, 73.50477600097656, 47.07795715332031, 113.415283203125, -22.227676391601562, 236.3563232421875, 28.55499267578125, 97.20648193359375, 65.9930419921875, -113.47653198242188, 86.20841979980469, 208.45486450195312, 70.02543640136719, 152.55108642578125, 196.93685913085938, -9.2978515625, 152.46987915039062, 29.026229858398438, 59.049713134765625, 31.987625122070312, 11.04071044921875, -66.40576171875, 112.775146484375, -9.255615234375, -45.298404693603516, 93.40274047851562, 165.67266845703125, 132.763916015625, 109.20062255859375, -10.5863037109375, 141.71243286132812, 31.22857666015625, 178.9290771484375, -20.075822830200195, 73.75480651855469, 50.280731201171875, 143.66696166992188, -9.395975112915039, 6.77813720703125, 31.91539764404297, 73.28643798828125, 93.61553955078125, 4.7716522216796875, 181.771240234375, 98.85358428955078, 39.46160888671875, 127.03153991699219, 10.24853515625, 150.59051513671875, 144.0731201171875, -9.9608154296875, -120.40487670898438, -63.81976318359375, -48.50909423828125, 108.17328643798828, 1.966644287109375, 31.384136199951172, 56.235504150390625, 13.256195068359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000227.npy"}
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 61.75171661376953, "std": 85.54605102539062, "min": -167.00152587890625, "p10": -34.0075942993164, "median": 48.99125671386719, "p90": 186.8907501220703, "max": 286.5362854003906, "pos_frac": 0.7578125, "sample": [19.590103149414062, -9.843658447265625, 185.43280029296875, 59.20147705078125, 26.932144165039062, 5.111989974975586, 190.29263305664062, 38.579559326171875, 111.74945068359375, 150.6298370361328, 42.3448600769043, -13.923377990722656, -71.92807006835938, 136.64825439453125, 27.9149169921875, 37.396148681640625, -11.3524169921875, 196.931884765625, -33.65901184082031, 60.62347412109375, 139.427978515625, 35.98104476928711, -123.27142333984375, -15.955841064453125, 55.049041748046875, 38.09391784667969, 20.774887084960938, 16.78282928466797, -60.5816650390625, 61.01097869873047, 31.54827880859375, 23.073165893554688, 13.50628662109375, 54.294921875, 134.52957153320312, 62.534027099609375, 37.62493896484375, 44.054931640625, 39.4425048828125, -34.820953369140625, 139.4111328125, 19.202117919921875, -24.2178955078125, -20.26763916015625, 149.8133544921875, -22.510589599609375, -105.06512451171875, 59.16802978515625, 94.37139892578125, 105.72578430175781, 60.39495849609375, 202.27777099609375, 111.38307189941406, -48.30873107910156, 202.51123046875, 135.07435607910156, 110.32098388671875, 60.6922607421875, 82.7564697265625, 286.5362854003906, -10.164398193359375, 4.532600402832031, -47.33131408691406, -4.7259674072265625, 49.583831787109375, 3.622955322265625, 0.0, 128.13729858398438, 51.823577880859375, -69.77120971679688, 70.5298080444336, 141.791748046875, 105.59526062011719, 15.6199951171875, -17.927459716796875, 167.75360107421875, -72.7149658203125, 48.398681640625, 24.418609619140625, 138.60247802734375, -38.068206787109375, -36.981163024902344, 41.3685302734375, 68.69671630859375, 201.23568725585938, 225.49037170410156, -3.349517822265625, 52.56494140625, 210.18862915039062, 50.036773681640625, 131.20803833007812, 155.144287109375, 55.01985168457031, 104.50839233398438, 134.30511474609375, 55.1370849609375, 25.58056640625, 20.671722412109375, 16.373504638671875, 276.48602294921875, 29.70233154296875, -17.508453369140625, -8.457046508789062, 56.59259796142578, 128.05276489257812, 255.90792846679688, 115.93655395507812, -167.00152587890625, 7.11651611328125, -28.2113037109375, -1.4399566650390625, 231.16326904296875, 22.941696166992188, 51.593833923339844, 158.5438995361328, 208.430908203125, 211.63812255859375, 9.80474853515625, 6.008781433105469, -60.951873779296875, 47.957061767578125, 138.09979248046875, 159.75537109375, 131.2259979248047, 164.233642578125, 136.70147705078125, -19.173019409179688, 141.12100219726562], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000228.npy"}
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 70.11982727050781, "std": 92.82654571533203, "min": -210.55545043945312, "p10": -20.764836692810054, "median": 65.97071838378906, "p90": 195.83565673828124, "max": 255.42730712890625, "pos_frac": 0.8203125, "sample": [152.24462890625, 6.4613037109375, 155.96359252929688, 141.3525390625, -11.847900390625, -128.52536010742188, 42.02647399902344, 48.93211364746094, 37.5377197265625, -123.41461181640625, 226.6552734375, 142.3541259765625, 74.36424255371094, -50.30008316040039, 230.13778686523438, 197.9007568359375, 95.16229248046875, -13.017486572265625, 73.2406234741211, 36.99908447265625, 152.06277465820312, 48.97595977783203, 228.66659545898438, 76.38467407226562, 26.421539306640625, -13.002685546875, 25.7095947265625, 47.21582794189453, 129.3834686279297, 81.75923919677734, 5.8215179443359375, 204.68115234375, 188.03948974609375, 66.19744873046875, 48.65442657470703, 140.61474609375, 113.00439453125, 31.990203857421875, 165.52407836914062, -149.96682739257812, 22.78460693359375, -105.71337890625, 42.11256408691406, 111.04942321777344, 21.177536010742188, 246.62521362304688, 195.82159423828125, 50.543731689453125, 73.24369812011719, 143.8856658935547, 163.08375549316406, -0.6731452941894531, 32.199668884277344, -10.261474609375, 25.95989227294922, 89.65278625488281, 178.38543701171875, 4.616790771484375, 27.4127197265625, -67.908935546875, 32.816734313964844, -95.04657745361328, 177.29910278320312, 84.7357177734375, 2.1939773559570312, -70.66773223876953, 91.62353515625, 42.62464904785156, 136.00489807128906, 34.47991943359375, 80.65022277832031, -12.562797546386719, -4.613975524902344, -16.241867065429688, 106.60861206054688, 120.0955581665039, 31.942474365234375, 63.370147705078125, 78.32180786132812, 62.45794677734375, 245.8531494140625, 153.95523071289062, 127.22412109375, 217.96759033203125, 16.938739776611328, 8.9041748046875, -210.55545043945312, 113.74922180175781, 203.9091796875, 21.765945434570312, 180.01123046875, 61.63935089111328, 14.432504653930664, 0.9434051513671875, 24.565673828125, 175.4259033203125, 65.74398803710938, 195.86846923828125, 30.363372802734375, 68.43971252441406, 95.28070068359375, 32.31233215332031, 179.25674438476562, 156.16754150390625, -24.12921142578125, 104.3184814453125, -189.31382751464844, 27.46246337890625, 83.5777587890625, 237.8776092529297, 40.79393005371094, -87.71186828613281, 85.9422607421875, 90.18704223632812, 114.4281005859375, 106.19558715820312, 169.1480712890625, -13.046356201171875, 140.54029846191406, -50.3529052734375, 243.87710571289062, 18.22747802734375, 105.07040405273438, 126.64634704589844, 85.8525390625, -19.322961807250977, 255.42730712890625, 1.0247879028320312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000229.npy"}
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 63.0064697265625, "std": 97.54784393310547, "min": -252.92999267578125, "p10": -35.36490325927734, "median": 45.20932388305664, "p90": 182.45272521972655, "max": 396.7528991699219, "pos_frac": 0.734375, "sample": [222.54547119140625, 42.931427001953125, 126.3017578125, 225.24560546875, 193.6806640625, 112.01092529296875, 130.4981689453125, -7.682769775390625, 155.20111083984375, 29.84698486328125, 37.294952392578125, 21.086944580078125, 3.9071502685546875, -128.9630126953125, -252.92999267578125, 189.38931274414062, 197.2890625, -3.4114990234375, 11.5401611328125, 77.13116455078125, 2.123504638671875, 30.12865447998047, 230.4725341796875, -67.81829833984375, 0.0, 180.25894165039062, -15.43609619140625, 35.38213348388672, -21.86993408203125, 25.973487854003906, 135.22348022460938, 168.79249572753906, 12.106643676757812, 85.347900390625, 1.0041351318359375, -33.852630615234375, 145.97576904296875, 181.34417724609375, 143.1873779296875, 110.43898010253906, 6.462007522583008, -32.627197265625, 23.3485107421875, 39.857574462890625, 96.0611572265625, 126.72027587890625, 8.495689392089844, 71.504150390625, 145.25473022460938, -14.343719482421875, -9.25424575805664, 69.28535461425781, -98.19012451171875, -15.440887451171875, 110.74234008789062, -47.08361053466797, 115.28640747070312, -9.38726806640625, 113.90792846679688, 55.88409423828125, -97.95841979980469, 54.72349548339844, 162.70675659179688, 70.92111206054688, 7.4760284423828125, 14.4212646484375, -38.96678161621094, 167.03631591796875, 99.3597412109375, 175.2333221435547, 99.09576416015625, 3.930419921875, -26.041763305664062, -46.58057403564453, 88.6890869140625, -28.484283447265625, -7.79693603515625, 17.412017822265625, 153.12994384765625, 144.98638916015625, 116.34471130371094, 131.17373657226562, 33.61561584472656, 96.95063018798828, 49.701019287109375, -191.9556884765625, 37.902313232421875, -38.89353942871094, 185.03933715820312, 103.62823486328125, 219.991943359375, 146.586181640625, -28.732177734375, 28.665435791015625, 142.9124755859375, 122.77886962890625, 3.998149871826172, 10.40863037109375, 43.886009216308594, 396.7528991699219, 125.46217346191406, -4.254419326782227, 103.62974548339844, 46.53263854980469, 2.0927047729492188, 87.08204650878906, -11.18890380859375, 372.8935546875, 169.40554809570312, -21.156740188598633, 52.489288330078125, 224.1134033203125, 230.9566192626953, -23.002464294433594, 102.07403564453125, 71.41207885742188, 4.266271591186523, 140.6201171875, 248.0040283203125, -3.1600723266601562, -43.92650604248047, 90.64910888671875, 2.9898529052734375, -21.61529541015625, 31.97528076171875, 60.57708740234375, -42.149749755859375, -44.17088317871094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000230.npy"}
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 48.583763122558594, "std": 86.37181091308594, "min": -183.30599975585938, "p10": -68.16970138549804, "median": 46.54583740234375, "p90": 149.43274078369137, "max": 295.7249755859375, "pos_frac": 0.734375, "sample": [51.28257751464844, 56.818115234375, -9.9849853515625, -93.06690216064453, 79.33612060546875, 147.29010009765625, 68.0389404296875, -13.831253051757812, -7.715576171875, -79.30268859863281, 173.9278564453125, -42.65879821777344, 101.71505737304688, 82.092529296875, 23.383438110351562, 80.49638366699219, 11.426071166992188, 23.875213623046875, 16.00794219970703, 36.52056884765625, -124.0970458984375, -125.57431030273438, -17.904037475585938, 49.3280029296875, 32.95558166503906, 51.314735412597656, -2.265289306640625, 72.89578247070312, 16.14910888671875, -70.37542724609375, 97.39349365234375, 130.18569946289062, 35.534942626953125, 2.6856040954589844, 30.26385498046875, 7.711708068847656, 295.7249755859375, 215.81573486328125, -58.66253662109375, 131.4783935546875, 123.50970458984375, 92.3441162109375, 124.30392456054688, 15.006622314453125, 186.46347045898438, 76.48318481445312, -43.00701904296875, 38.34111022949219, -21.92681884765625, 60.547119140625, 104.8602294921875, 135.32159423828125, 57.467132568359375, -67.57662963867188, 53.58441162109375, 26.642120361328125, 146.18670654296875, -16.1988525390625, 32.52135467529297, 219.81735229492188, -115.24055480957031, 121.27937316894531, 2.381195068359375, 21.48847198486328, -14.55313491821289, 41.194000244140625, 110.72946166992188, 21.076202392578125, 62.4552001953125, 120.07568359375, 120.4136962890625, 119.30819702148438, 196.25711059570312, -183.30599975585938, -113.06729125976562, 78.800048828125, 16.150897979736328, -13.81854248046875, 118.19091796875, 55.867218017578125, 3.2333984375, 80.46249389648438, 33.95722961425781, -12.990997314453125, -19.632953643798828, 25.724639892578125, -25.478271484375, 137.03732299804688, 68.3712158203125, 34.517333984375, 71.21502685546875, 88.42178344726562, 43.763671875, 123.55778503417969, -35.423828125, 154.43223571777344, 108.06929016113281, 40.990386962890625, 227.19964599609375, -0.06288719177246094, -147.75250244140625, 53.00469970703125, 41.24552917480469, 188.13131713867188, 11.191925048828125, -22.08868408203125, 85.42689514160156, 189.33834838867188, 37.025543212890625, 116.7857666015625, -47.56034851074219, 69.34735107421875, 124.5440673828125, -99.35366821289062, 94.07699584960938, 191.31326293945312, 78.15963745117188, 108.39947509765625, -87.59130859375, 76.54829406738281, -69.55353546142578, 106.90701293945312, 195.85842895507812, -122.22833251953125, 240.01290893554688, 0.0, 13.479660034179688, 60.1055908203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000231.npy"}
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 54.60589599609375, "std": 95.66480255126953, "min": -211.730712890625, "p10": -47.41189880371093, "median": 45.88746643066406, "p90": 182.4342559814453, "max": 269.2107849121094, "pos_frac": 0.703125, "sample": [-211.730712890625, 133.42977905273438, 10.209396362304688, 92.03927612304688, 44.05799865722656, 113.38330078125, 47.378997802734375, 166.92724609375, -11.68548583984375, 101.00967407226562, 90.64173889160156, 197.92041015625, -2.000213623046875, -33.94349670410156, -16.195220947265625, 119.7811050415039, -4.378883361816406, -45.481040954589844, 11.7601318359375, 55.43512725830078, 61.24560546875, 39.84465789794922, -148.48345947265625, 24.27081298828125, 194.23196411132812, 188.31207275390625, 116.35581970214844, 44.4027099609375, -38.60920333862305, 232.24951171875, 38.89630126953125, 118.44274139404297, -71.82424926757812, 99.92327880859375, 247.918701171875, 149.43841552734375, -50.385894775390625, 39.424224853515625, 95.5052490234375, 236.28598022460938, -19.236587524414062, -29.317787170410156, 1.512887954711914, 55.523773193359375, -3.5187835693359375, 6.683738708496094, 56.210365295410156, 47.401611328125, 133.58680725097656, 2.6635093688964844, -6.2459716796875, -44.803131103515625, 114.33824157714844, 17.47576904296875, 51.708351135253906, 128.0079345703125, 82.5638427734375, 187.12689208984375, 29.580429077148438, 131.40293884277344, 58.488975524902344, 254.68597412109375, 178.82244873046875, -14.55328369140625, 53.41896057128906, 250.41969299316406, -6.8770751953125, 62.305084228515625, -8.52386474609375, 66.09564208984375, 53.26226806640625, 153.62152099609375, -2.84710693359375, -186.52658081054688, 209.83956909179688, 162.20257568359375, 0.3852386474609375, 34.39581298828125, -0.322418212890625, 82.9017333984375, -3.1449832916259766, 76.37002563476562, -50.39813232421875, 17.677352905273438, 79.19927978515625, -10.007942199707031, 11.575231552124023, 158.9160614013672, -19.310302734375, -46.1373291015625, 160.1417694091797, 8.17596435546875, 0.0, 126.23187255859375, 29.643287658691406, -7.691432952880859, 43.15858459472656, 107.64444732666016, 47.372222900390625, 90.69586181640625, 25.19220733642578, 150.9735107421875, 159.2275390625, -89.84298706054688, 210.12664794921875, 180.42312622070312, 42.41259765625, -148.38656616210938, 144.61428833007812, 73.16107177734375, 12.1551513671875, 163.01193237304688, -123.58167266845703, -59.75738525390625, -28.581199645996094, -101.6591796875, 269.2107849121094, 2.8585662841796875, 75.53192138671875, -130.36843872070312, 155.73501586914062, 124.55606079101562, -121.42547607421875, 174.71133422851562, 188.682861328125, 19.896896362304688, 0.384307861328125, -19.68463134765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000232.npy"}
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 60.27885055541992, "std": 86.39936065673828, "min": -157.2371826171875, "p10": -33.91006278991699, "median": 43.68254089355469, "p90": 183.8097137451172, "max": 298.0440673828125, "pos_frac": 0.7421875, "sample": [-115.94745635986328, 160.4229736328125, 37.8018798828125, 3.56048583984375, -11.791397094726562, 143.58242797851562, -65.89254760742188, 56.908660888671875, -5.366342544555664, -34.71498107910156, -14.666595458984375, 122.27200317382812, 180.46429443359375, 19.5546875, -4.708404541015625, 49.710205078125, 74.95755004882812, 238.03125, 122.95625305175781, 162.58084106445312, 210.36398315429688, 163.70492553710938, 64.74383544921875, 144.07574462890625, 15.873130798339844, 98.61518859863281, 145.86233520507812, 120.276123046875, 1.26519775390625, -33.56509780883789, 87.80740356445312, -67.70681762695312, 199.0766143798828, 14.021881103515625, 87.64572143554688, 56.32989501953125, 22.2933349609375, -14.800529479980469, 98.7441177368164, 236.83270263671875, -49.558380126953125, -0.7721652984619141, 12.314342498779297, -0.592529296875, -66.1365966796875, 199.23800659179688, -10.10516357421875, 34.71271896362305, -14.55645751953125, 44.259429931640625, 298.0440673828125, 192.50709533691406, 97.52728271484375, -10.851463317871094, 48.53684997558594, 53.56353759765625, 225.5244140625, 22.0491943359375, 93.7696533203125, 8.34832763671875, 79.74566650390625, 140.0592041015625, -88.52493286132812, 64.87968444824219, 109.3107681274414, -31.245849609375, 3.799346923828125, 40.5821533203125, -75.50607299804688, -97.0355224609375, 75.88070678710938, 193.54241943359375, 103.610107421875, 134.45643615722656, 227.7061767578125, 137.29534912109375, 43.10565185546875, 28.278289794921875, 36.44361877441406, -0.42889404296875, 30.53570556640625, -68.55589294433594, 3.96710205078125, 145.40298461914062, -39.64613342285156, 109.74336242675781, 193.71588134765625, 101.54666137695312, 33.993072509765625, 183.57205200195312, 38.04424285888672, 33.6519775390625, -99.0413818359375, 136.4171142578125, -12.688766479492188, -8.75140380859375, 7.017793655395508, 138.9935760498047, -31.94598388671875, 184.3642578125, -7.612548828125, 6.038536071777344, -32.49884796142578, -157.2371826171875, 210.14889526367188, 94.02410888671875, 41.32148742675781, 163.27964782714844, 56.595489501953125, 4.79425048828125, 8.622947692871094, 118.16970825195312, 132.98770141601562, 30.712265014648438, 95.77713012695312, 72.66824340820312, -3.379016876220703, 171.88720703125, 39.47972869873047, 83.69334411621094, 96.98648071289062, 4.847225189208984, -31.464599609375, 87.47683715820312, 172.53692626953125, 72.76162719726562, 7.32415771484375, 20.462921142578125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000233.npy"}
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 54.26837158203125, "std": 100.21244049072266, "min": -174.6597900390625, "p10": -79.93429260253906, "median": 49.479248046875, "p90": 177.98124389648436, "max": 323.516357421875, "pos_frac": 0.6796875, "sample": [-45.902496337890625, -107.1004638671875, -10.66937255859375, 72.70368957519531, 125.11453247070312, 56.00151062011719, 205.31259155273438, 59.215667724609375, -67.29891967773438, 211.00845336914062, 149.70315551757812, -174.6597900390625, 184.322265625, 153.2366943359375, -167.77801513671875, 73.45339965820312, 109.4912109375, -13.33001708984375, 11.666473388671875, 142.7391357421875, 67.28904724121094, -2.481689453125, 70.2215576171875, 120.68875122070312, 31.696197509765625, -109.09214782714844, 181.086181640625, 95.38855743408203, 29.88182830810547, 13.30926513671875, 14.438156127929688, 53.11639404296875, 99.78997802734375, 179.34796142578125, 173.55355834960938, -72.92875671386719, 45.84210205078125, 56.2138671875, -29.100540161132812, -17.95263671875, 110.57977294921875, 16.967620849609375, -7.86138916015625, -111.76109313964844, 94.91548156738281, -14.237747192382812, 21.007125854492188, 120.78701782226562, -108.6733169555664, 28.590614318847656, 16.115982055664062, 27.20166015625, 171.734130859375, 133.52394104003906, -4.59051513671875, 239.4183349609375, 167.6686553955078, 79.6301040649414, 161.22630310058594, -2.81854248046875, 0.5937862396240234, 36.507720947265625, 284.2428894042969, 59.3138427734375, -2.2361927032470703, 40.551177978515625, -74.58087158203125, 0.0, 98.6258544921875, -32.40480041503906, 30.077919006347656, 98.69078063964844, -26.421844482421875, 24.450302124023438, 172.6307373046875, 219.12136840820312, 228.10467529296875, 127.584228515625, -136.1195526123047, 84.24603271484375, -7.937568664550781, 174.6851043701172, 121.22598266601562, -29.98150634765625, 84.85488891601562, 66.71183776855469, 143.5620574951172, 30.9122314453125, -150.10556030273438, 138.0997314453125, -94.865966796875, 278.2104187011719, 38.66900634765625, 85.19329833984375, 136.5145721435547, -33.75000762939453, 129.8726806640625, 104.58905029296875, -20.6309814453125, -1.057373046875, 142.8946990966797, 0.0, 86.28810119628906, 29.616607666015625, 323.516357421875, 215.98968505859375, -79.63095092773438, 32.35272216796875, 177.3955078125, 88.31097412109375, 134.16873168945312, -37.6593017578125, -122.90147399902344, 35.16717529296875, 67.9508056640625, 120.7610855102539, -43.39637756347656, -20.81365966796875, -109.50714111328125, 31.54412841796875, 86.8748779296875, -47.582611083984375, -116.6068115234375, 100.5401840209961, -80.64208984375, 223.53024291992188, 37.39910888671875, 136.07958984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000234.npy"}
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 56.78924560546875, "std": 92.21825408935547, "min": -158.6358642578125, "p10": -62.72856979370117, "median": 42.840858459472656, "p90": 175.70763244628907, "max": 267.0675048828125, "pos_frac": 0.7109375, "sample": [15.214935302734375, 18.94701385498047, -20.1590576171875, 10.814437866210938, 16.429706573486328, -94.91497802734375, -16.24575424194336, -92.12958526611328, 144.19815063476562, 0.18231201171875, 12.439064025878906, 234.24945068359375, 163.22174072265625, 66.8753662109375, -4.317169189453125, 56.069114685058594, 71.20783996582031, 133.06011962890625, -65.37450408935547, -28.5858154296875, 92.28459167480469, 68.24700927734375, 180.39410400390625, 200.6317138671875, 192.09434509277344, 123.50445556640625, 95.80624389648438, -63.18511962890625, 132.82574462890625, -25.876312255859375, 103.982421875, -4.50146484375, 123.31668090820312, 202.16677856445312, 40.273101806640625, 176.89183044433594, 266.61456298828125, 172.7847900390625, 143.29217529296875, 143.28564453125, -17.504989624023438, 123.71890258789062, -62.53290557861328, -65.83212280273438, 160.13705444335938, 139.1136474609375, -155.40374755859375, -15.349845886230469, -111.03353881835938, 10.134773254394531, 231.32022094726562, 87.34072875976562, 5.295747756958008, -98.99789428710938, 141.66455078125, 56.1986083984375, -23.262252807617188, 130.63230895996094, -75.37469482421875, 106.18695068359375, -6.824871063232422, 63.98345947265625, 144.55047607421875, -10.394424438476562, 227.46002197265625, 28.452526092529297, -6.49444580078125, -140.14939880371094, 13.366607666015625, 82.74211883544922, 33.16712188720703, 95.25469970703125, 1.758453369140625, 79.77809143066406, 8.243610382080078, 12.31414794921875, -4.7230224609375, 267.0675048828125, 37.24896240234375, 106.355712890625, 129.73777770996094, -5.819976806640625, -123.75296020507812, 29.00464630126953, 102.24211120605469, -8.821319580078125, 140.67172241210938, 175.2001190185547, 70.8203125, 4.1658477783203125, 200.9169921875, 45.40861511230469, 106.60494232177734, 198.5037841796875, 152.291259765625, 112.80178833007812, 3.9696807861328125, 15.59869384765625, 142.22265625, 30.180906295776367, -10.992151260375977, 171.57122802734375, -9.672369003295898, -18.05322265625, 119.0177001953125, 91.51060485839844, -7.035057067871094, 47.80938720703125, 37.09625244140625, 3.75872802734375, 37.830841064453125, 31.876541137695312, 4.505836486816406, -118.54339599609375, -34.509857177734375, 106.88078308105469, 149.20106506347656, -158.6358642578125, -10.305061340332031, -30.48723602294922, 127.74018859863281, 56.15000915527344, 38.377349853515625, 199.9986572265625, 174.36279296875, 121.49613952636719, 57.8106689453125, -17.30965232849121], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000235.npy"}
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 58.862945556640625, "std": 92.73278045654297, "min": -185.56488037109375, "p10": -49.44457244873046, "median": 45.01325988769531, "p90": 187.8285186767578, "max": 251.69017028808594, "pos_frac": 0.7421875, "sample": [114.87710571289062, 35.35455322265625, -89.39895629882812, 79.43009185791016, 18.361495971679688, 22.973983764648438, 15.964462280273438, 240.70309448242188, -107.18460083007812, -35.181182861328125, -21.59735107421875, 78.27149200439453, -55.153053283691406, 150.74862670898438, 85.36126708984375, 173.7232666015625, -4.040271759033203, 147.5637664794922, -13.32763671875, 89.76412963867188, 43.074615478515625, 12.357818603515625, 52.25091552734375, 23.527170181274414, -55.09046936035156, 133.39463806152344, 27.078750610351562, 65.39830017089844, 20.430328369140625, 206.71542358398438, 170.83236694335938, 31.561431884765625, -17.927711486816406, 8.798271179199219, 2.555145263671875, 65.94570922851562, -87.877197265625, 199.7274169921875, 251.69017028808594, -8.043670654296875, 157.0309295654297, 77.1478042602539, 156.03045654296875, 38.72332763671875, 155.58489990234375, 9.192405700683594, -107.33468627929688, -25.3138427734375, 13.736846923828125, 0.0, 8.1591796875, 9.380889892578125, 188.980224609375, 165.57501220703125, 60.025054931640625, -40.781097412109375, -185.56488037109375, 189.25128173828125, 24.501708984375, -80.61541748046875, 37.520477294921875, 119.25637817382812, 68.38616943359375, 207.13186645507812, -25.166183471679688, -37.45758056640625, 122.01129150390625, -76.24667358398438, 29.038238525390625, 52.75926971435547, 21.127532958984375, 124.40087890625, 92.25177764892578, 169.258056640625, -33.286598205566406, 13.280891418457031, -89.05300903320312, -9.411590576171875, 187.33493041992188, 2.040956497192383, 78.02978515625, 77.536865234375, -47.02490234375, 134.70103454589844, 142.0009765625, 82.64801025390625, 139.43003845214844, 116.90155029296875, 4.160240173339844, 29.309249877929688, 111.52267456054688, 116.02496337890625, 172.76528930664062, 13.762176513671875, 16.5977783203125, 184.78707885742188, 88.70742797851562, 65.16035461425781, 138.66700744628906, 176.69912719726562, 161.67352294921875, 46.951904296875, -28.072174072265625, 25.873672485351562, 0.0, 57.290191650390625, -37.124786376953125, 92.82330322265625, -146.6846466064453, 216.32835388183594, 19.946502685546875, -7.7490081787109375, -105.5796890258789, 91.94700622558594, 174.78256225585938, 199.75401306152344, 245.8369140625, 177.1129150390625, 4.364002227783203, -100.81755065917969, 241.67758178710938, 56.372962951660156, 194.94793701171875, 11.15130615234375, -34.7305908203125, 204.69387817382812, -0.09223175048828125, 66.8918228149414], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000236.npy"}
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 56.669921875, "std": 91.54711151123047, "min": -233.22616577148438, "p10": -52.45880279541015, "median": 39.83268737792969, "p90": 187.14858093261716, "max": 295.16693115234375, "pos_frac": 0.75, "sample": [60.55181884765625, 34.787445068359375, 95.97891235351562, 216.07644653320312, 22.429931640625, 85.43978881835938, 32.980621337890625, 172.9650115966797, 111.09136962890625, 109.70864868164062, -11.570831298828125, -54.59510803222656, 133.82833862304688, -9.682777404785156, -46.44789123535156, -29.227462768554688, 95.04995727539062, -21.335906982421875, 281.5252380371094, 145.01580810546875, 22.997940063476562, 127.96455383300781, 193.23165893554688, 66.59785461425781, 38.72970199584961, -77.240966796875, 39.018218994140625, 0.0, 117.36518859863281, 97.0726318359375, 55.20001983642578, 80.19731140136719, -89.15451049804688, 20.804885864257812, -11.4478759765625, 40.64715576171875, -114.04544067382812, 17.75128173828125, -142.4906005859375, 17.044570922851562, 19.861373901367188, -5.26715087890625, 88.01505279541016, 102.035888671875, 29.66973114013672, -99.61831665039062, 206.29592895507812, 37.963623046875, 59.0689697265625, -3.3819580078125, 53.53660583496094, 232.2689208984375, 87.54679870605469, 19.6328125, 79.41357421875, 166.55776977539062, 148.81124877929688, 174.90509033203125, 43.83726501464844, 55.13673400878906, 201.53070068359375, 33.870079040527344, 15.776336669921875, 51.109893798828125, -40.95677185058594, 10.113922119140625, 120.49826049804688, 161.9269256591797, -71.79562377929688, -77.760009765625, -5.452503204345703, 227.82492065429688, -7.600860595703125, 84.85662841796875, 194.87957763671875, 54.05009841918945, -13.809555053710938, 13.2882080078125, 295.16693115234375, 59.662109375, 64.17318725585938, 39.0013427734375, 227.28135681152344, -28.04833984375, 153.96913146972656, 22.780975341796875, 0.38861846923828125, -23.76202392578125, 275.8547058105469, 12.324859619140625, 28.808460235595703, 185.06723022460938, 0.3637847900390625, 9.341400146484375, -91.82183837890625, 105.02410888671875, 6.98541259765625, 12.27142333984375, 74.50601196289062, 132.7649383544922, -56.763427734375, 14.133934020996094, 142.93157958984375, 0.4828605651855469, 26.81473159790039, 15.250144958496094, 92.15205383300781, -75.73794555664062, -51.543243408203125, -55.45001220703125, 42.40740966796875, 230.71966552734375, 16.409027099609375, 33.428192138671875, -35.20850372314453, 112.95431518554688, -233.22616577148438, 107.43562316894531, 192.00506591796875, 149.064697265625, 118.77639770507812, -8.025787353515625, -2.6038665771484375, 119.02803039550781, 99.19332122802734, 179.52842712402344, 57.324462890625, 86.71160888671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000237.npy"}
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 57.69230651855469, "std": 96.4224853515625, "min": -150.8944549560547, "p10": -47.56091995239256, "median": 39.91069793701172, "p90": 196.5199005126953, "max": 320.4337158203125, "pos_frac": 0.7265625, "sample": [-32.443084716796875, 39.44615173339844, -20.131439208984375, 30.989604949951172, 7.0764617919921875, 202.1160888671875, 171.8603057861328, 41.5050048828125, 46.553680419921875, -103.8095703125, 4.55389404296875, 135.33627319335938, 5.6607666015625, 48.940521240234375, 4.1945953369140625, -117.00491333007812, 198.35113525390625, 205.37179565429688, -85.26375579833984, 129.85614013671875, 172.9541778564453, 142.3748779296875, 86.5673828125, 88.9195556640625, 215.34027099609375, 129.74298095703125, -60.80145263671875, 58.3187255859375, 24.31475830078125, -9.283920288085938, -140.83726501464844, -129.06936645507812, 320.4337158203125, -91.15814208984375, 7.100311279296875, 88.81884765625, 34.24322509765625, 239.4798583984375, 8.105979919433594, 98.798095703125, 127.9954833984375, 62.73236083984375, 42.759002685546875, 309.62091064453125, 60.529541015625, 170.75411987304688, 89.38363647460938, 139.25289916992188, 81.87261962890625, 70.54586791992188, 258.3455810546875, 249.4627685546875, -5.052001953125, -3.6979827880859375, 56.35894775390625, 23.7930908203125, 30.826400756835938, 8.927999496459961, 61.037261962890625, 106.9783935546875, 0.7572174072265625, 29.4405517578125, 1.7962646484375, -77.598388671875, 196.48294067382812, 59.33338928222656, 31.419570922851562, 154.61326599121094, 123.1534423828125, 60.42639923095703, -84.33201599121094, -92.67303466796875, 32.152801513671875, 28.249847412109375, 39.028438568115234, 109.64639282226562, 172.62081909179688, -69.99258422851562, 139.81764221191406, 116.59197998046875, -21.9405517578125, 224.84423828125, 2.0919532775878906, 71.65746307373047, -15.8046875, 40.375244140625, 135.28475952148438, 25.215251922607422, 38.20003128051758, 53.081573486328125, 118.56593322753906, -18.37322998046875, -79.4400634765625, 98.8341064453125, 253.21029663085938, -15.180770874023438, 48.385955810546875, 32.18269348144531, 41.83611297607422, -41.88640594482422, 24.030990600585938, 196.60614013671875, 166.25906372070312, 12.9365234375, -9.716968536376953, -17.508041381835938, -150.8944549560547, 179.90628051757812, -7.9207763671875, -20.008758544921875, 13.858428955078125, 117.641845703125, -21.770145416259766, 27.651779174804688, -3.924957275390625, -23.614944458007812, 55.74501037597656, 150.71075439453125, -14.756645202636719, 4.070892333984375, 143.82933044433594, 306.5889892578125, 57.5380859375, -35.118621826171875, -31.55849838256836, -3.20684814453125, 167.05963134765625, -1.8427734375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000238.npy"}
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 48.139625549316406, "std": 107.99394226074219, "min": -345.10797119140625, "p10": -61.51195983886718, "median": 35.523372650146484, "p90": 186.52792816162108, "max": 319.6341552734375, "pos_frac": 0.671875, "sample": [-53.26548767089844, -59.39768981933594, 230.05599975585938, -40.975341796875, -25.5291748046875, 1.1644287109375, -225.0333251953125, 64.75032806396484, 161.11122131347656, 10.342720031738281, 171.26681518554688, -345.10797119140625, -46.61956787109375, -10.156005859375, 176.93832397460938, 125.6981201171875, 197.99365234375, 204.727294921875, -44.34159851074219, -27.86016845703125, 22.485504150390625, 300.21514892578125, 0.5252456665039062, 88.236083984375, -32.17195129394531, 133.46295166015625, 32.09489440917969, 67.66554260253906, 86.56646728515625, 27.364913940429688, -30.45062255859375, 182.7496337890625, 53.214332580566406, -14.279182434082031, 299.71783447265625, -55.08404541015625, 62.2039794921875, 184.87506103515625, 319.6341552734375, 211.169921875, 116.5694580078125, -111.18170166015625, 99.75334167480469, -58.64051818847656, 185.646728515625, -47.496734619140625, -25.1795654296875, 92.4840087890625, -56.005279541015625, 1.034149169921875, -101.01925659179688, 144.73004150390625, -70.19868469238281, -105.34921264648438, 57.15423583984375, 22.4378662109375, 57.1361083984375, 83.52511596679688, 99.91495513916016, -128.933837890625, 27.69378662109375, 154.778564453125, 8.50163459777832, 99.05659484863281, 122.73004913330078, 97.05633544921875, 128.24588012695312, -73.04257202148438, -1.0519561767578125, 171.72113037109375, -30.175743103027344, 8.347526550292969, 87.90322875976562, 93.65350341796875, 60.142822265625, 2.71978759765625, 101.71658325195312, -0.0390167236328125, -82.10739135742188, -169.13377380371094, -4.627410888671875, 20.66938018798828, 183.27346801757812, -27.6422119140625, 38.695709228515625, 286.385986328125, -157.35916137695312, 30.730628967285156, 31.654800415039062, -42.8126220703125, 131.00173950195312, -6.246315002441406, 61.33563995361328, 72.09419250488281, 32.351036071777344, 14.235832214355469, -24.248458862304688, 56.279632568359375, 128.6944580078125, 76.2987060546875, -87.03229522705078, 219.82037353515625, 41.72285461425781, -59.579986572265625, 89.81304931640625, 8.566715240478516, -66.0198974609375, 52.441253662109375, 82.77874755859375, 274.3601379394531, 241.43557739257812, 77.40505981445312, 116.42854309082031, 235.02285766601562, 150.08169555664062, 0.0, 40.5745849609375, -2.7954559326171875, 188.5840606689453, 4.7847900390625, -26.517730712890625, 17.65148162841797, 75.49636840820312, 41.818389892578125, 22.5157470703125, -19.866500854492188, 2.790203094482422, 67.7696533203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000239.npy"}
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 63.33018493652344, "std": 96.9720458984375, "min": -193.5372314453125, "p10": -34.46073150634765, "median": 56.74699401855469, "p90": 194.26491241455074, "max": 432.7064514160156, "pos_frac": 0.7265625, "sample": [29.291587829589844, 27.440216064453125, 68.6414794921875, 115.93522644042969, -98.79059600830078, 145.2528533935547, 89.74551391601562, 73.41539001464844, 0.0, 153.73333740234375, 59.9566650390625, -193.5372314453125, -80.4859619140625, -1.2366943359375, -43.10216522216797, 114.4090347290039, 147.13958740234375, 59.79127502441406, 114.42202758789062, -1.774627685546875, -1.763427734375, 15.4169921875, 123.5155029296875, 39.705230712890625, 0.03656005859375, 57.78436279296875, 109.30254364013672, -83.09860229492188, 228.54156494140625, 215.92080688476562, 67.2943115234375, -2.28826904296875, 432.7064514160156, 61.50846862792969, -65.51678466796875, 2.267364501953125, 214.47784423828125, 146.21051025390625, 105.75800323486328, 45.926841735839844, 122.10684204101562, 86.660888671875, 227.50100708007812, -22.011444091796875, 81.12890625, 70.28640747070312, 190.84921264648438, 48.84471130371094, 162.0021514892578, 78.88473510742188, 250.65576171875, 97.74847412109375, 0.0, 37.464622497558594, -178.73147583007812, 38.09710693359375, 5.94317626953125, 34.54908752441406, -18.947418212890625, -15.85009765625, 89.13823699951172, 201.957275390625, -24.6771240234375, -32.69671630859375, 251.4169921875, 190.9681854248047, 50.74652099609375, 136.4400177001953, 205.16165161132812, 28.186660766601562, 233.51190185546875, -21.3009033203125, 0.0, 139.38687133789062, 32.907188415527344, -23.635772705078125, 160.63482666015625, -9.6488037109375, 13.773391723632812, 84.708740234375, -110.1610107421875, 55.709625244140625, 145.55538940429688, 59.57878112792969, 75.37179565429688, 176.28286743164062, -61.472564697265625, 215.69509887695312, -29.605438232421875, 138.75234985351562, 10.505844116210938, 5.25927734375, -16.986968994140625, 110.3511962890625, 170.38723754882812, 84.38232421875, -129.25918579101562, 141.6361846923828, 60.781410217285156, 4.537418365478516, 14.401458740234375, 29.599245071411133, 9.234779357910156, -38.57676696777344, 218.921875, 42.42364501953125, 148.8188018798828, 149.06988525390625, 141.74069213867188, 29.6575927734375, 151.9900665283203, 106.84429931640625, 42.163330078125, -50.702735900878906, 67.41520690917969, -10.860496520996094, -142.424560546875, 103.41459655761719, -26.81060791015625, 221.38177490234375, 0.0, 2.8827056884765625, 66.75791931152344, 21.4986572265625, 19.71875, -2.8465137481689453, -5.1707763671875, 190.30450439453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000240.npy"}
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 51.938262939453125, "std": 102.00372314453125, "min": -204.51171875, "p10": -67.66073837280273, "median": 36.50879669189453, "p90": 190.22962341308593, "max": 299.4791259765625, "pos_frac": 0.734375, "sample": [25.4996337890625, 131.22964477539062, 123.55979919433594, 60.273284912109375, -0.6478195190429688, -96.72935485839844, 30.872406005859375, -3.064168930053711, 105.6737060546875, 230.20291137695312, 267.1563415527344, 0.0, -25.86474609375, 262.95416259765625, 95.97640228271484, 299.4791259765625, 44.85736083984375, 6.3447265625, 21.01885986328125, 272.4330139160156, 32.50194549560547, 132.47901916503906, -12.530426025390625, 34.805908203125, 124.4600830078125, 122.78924560546875, -122.4637451171875, 6.4475250244140625, 126.66140747070312, 78.95413208007812, 119.89629364013672, -74.98529052734375, 222.9149627685547, -38.99346923828125, 55.669464111328125, 1.472686767578125, -25.285736083984375, -42.26361083984375, 15.59260368347168, 151.30526733398438, 53.084327697753906, 59.162994384765625, 122.33663940429688, 218.96466064453125, 31.17474365234375, 154.5212860107422, 98.2677001953125, -152.73452758789062, 62.39385986328125, 39.85089111328125, 156.3043670654297, 9.084733963012695, 5.576850891113281, 87.73858642578125, -155.4984588623047, 7.7098388671875, 36.37388610839844, 18.614425659179688, -41.63739013671875, 49.848419189453125, 0.7025146484375, 110.81832885742188, 125.83843994140625, 57.54254150390625, -8.410087585449219, 188.84506225585938, 7.915191650390625, 57.787628173828125, 120.64531707763672, 145.95294189453125, -111.78263854980469, 5.920867919921875, 227.24124145507812, 67.66534423828125, 10.610595703125, -204.51171875, 219.50067138671875, 136.99957275390625, -34.236968994140625, -4.8914794921875, 23.2357177734375, 35.170047760009766, 211.86703491210938, 84.619873046875, 42.9453125, 26.941757202148438, -39.05018615722656, 40.80621337890625, 36.643707275390625, 137.45672607421875, 20.93475341796875, -99.573974609375, 19.481201171875, -49.38751220703125, 83.63034057617188, 141.01637268066406, -31.6214599609375, -161.2625274658203, 193.46026611328125, 2.895538330078125, 88.95477294921875, -164.0078125, 39.135406494140625, 29.0894775390625, 288.309326171875, 176.6038818359375, 8.642646789550781, 13.352958679199219, 175.2969970703125, 162.93063354492188, -0.07257080078125, -130.88088989257812, -49.59004211425781, -6.931282043457031, 144.6094512939453, -4.486732482910156, 129.68673706054688, 3.3820343017578125, -64.52164459228516, 139.70343017578125, 277.3431701660156, 43.87286376953125, 40.1072998046875, -75.16134643554688, 27.31634521484375, -45.75897216796875, 58.750335693359375, -119.73236083984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000241.npy"}
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 63.03213882446289, "std": 93.92060089111328, "min": -134.53497314453125, "p10": -35.992352294921865, "median": 57.10308837890625, "p90": 179.10179443359374, "max": 358.7056884765625, "pos_frac": 0.7421875, "sample": [115.47018432617188, 16.829200744628906, 80.928466796875, 26.04882049560547, 31.940994262695312, -8.178428649902344, 175.42169189453125, 91.18417358398438, 3.33355712890625, 6.7674713134765625, 275.3622131347656, 100.5264892578125, 93.5263671875, -42.09039306640625, 11.745269775390625, 154.97694396972656, 107.71842956542969, 153.8179931640625, -80.52217102050781, 89.55535125732422, 9.71306037902832, 164.11619567871094, 281.6565246582031, 112.19014739990234, 60.96929931640625, 201.42874145507812, 58.604339599609375, 225.89300537109375, 26.442291259765625, 8.160484313964844, 65.7359619140625, -7.465576171875, -14.2674560546875, 182.12295532226562, 25.601959228515625, 315.9725036621094, 358.7056884765625, -11.808013916015625, -118.93124389648438, 69.75472259521484, 105.06973266601562, -33.37890625, -30.1248779296875, 67.3460693359375, -12.656707763671875, 11.398258209228516, 99.67034912109375, 86.15768432617188, 168.76162719726562, -46.40919494628906, -109.84072875976562, -104.73568725585938, -29.960052490234375, 69.16065979003906, -75.79469299316406, 13.014896392822266, 126.57601928710938, -23.193084716796875, -53.30242919921875, 93.85548400878906, 74.97073364257812, 112.95857238769531, -89.03793334960938, 233.1665496826172, -5.046875, -2.749755859375, 179.339599609375, 319.7160949707031, 25.20166015625, 34.99951171875, 57.3951416015625, 82.133056640625, 95.55381774902344, 42.65980529785156, 114.2442626953125, -14.428733825683594, 151.7540740966797, 39.26714324951172, 29.19805908203125, -25.734081268310547, 105.72604370117188, 122.99456787109375, 248.82373046875, 104.14435577392578, -70.60275268554688, 46.67253112792969, 11.592676162719727, 32.82347106933594, 99.82290649414062, 48.198028564453125, 220.9051513671875, 34.41912841796875, 84.4405517578125, 56.28349304199219, -7.349254608154297, 178.9998779296875, 63.73365783691406, 136.65875244140625, 85.29863739013672, -16.77044677734375, -20.568038940429688, 105.05972290039062, -23.3859806060791, -16.427749633789062, 131.8216094970703, -134.53497314453125, 36.9803466796875, -128.9097900390625, 52.671417236328125, 85.12962341308594, 2.234222412109375, -20.413169860839844, 56.81103515625, 179.958251953125, 61.039215087890625, 122.32669067382812, 169.08551025390625, 113.4799575805664, 16.887680053710938, 168.42041015625, -20.232086181640625, 53.16956329345703, 4.6179351806640625, -77.9554443359375, 120.38004302978516, 148.9547119140625, 19.294219970703125, 79.26997375488281], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000242.npy"}
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 69.48601531982422, "std": 106.2991714477539, "min": -253.410888671875, "p10": -40.39987182617187, "median": 60.33057403564453, "p90": 183.25924377441405, "max": 594.169189453125, "pos_frac": 0.71875, "sample": [-162.7769775390625, -29.855606079101562, 94.78575134277344, -94.47267150878906, 182.57138061523438, 30.8328857421875, 56.0338134765625, 184.8642578125, 17.743743896484375, 118.96995544433594, 46.1883544921875, 203.35150146484375, 61.173179626464844, -63.79808044433594, 62.3795166015625, -10.18035888671875, -39.35191345214844, 58.33302307128906, 356.322998046875, -28.013877868652344, 81.30927276611328, 28.9921875, -4.749267578125, 7.464866638183594, 37.25572204589844, 76.9420166015625, 594.169189453125, 133.37811279296875, -5.2739410400390625, 56.221038818359375, -3.0029468536376953, 305.59698486328125, -19.21019744873047, 82.58247375488281, 186.44158935546875, 43.7425537109375, 147.7178955078125, 23.91571044921875, 112.2384033203125, 285.91357421875, -85.4854736328125, 126.93450927734375, 69.80670166015625, 92.94338989257812, 47.22122573852539, 151.69540405273438, 266.9749755859375, -61.532867431640625, 165.58868408203125, 85.78230285644531, 111.37615966796875, -253.410888671875, 15.643577575683594, 113.68499755859375, -30.45843505859375, -123.9696044921875, -89.74104309082031, -3.883697509765625, -5.75433349609375, -11.3018798828125, 179.96856689453125, 91.37615966796875, 143.14291381835938, 48.772735595703125, -16.52056884765625, 139.08987426757812, 123.9207763671875, 13.9718017578125, 135.96932983398438, -47.966552734375, 41.496551513671875, -79.23931884765625, -25.1318359375, 228.9279327392578, 97.00277709960938, 47.94395446777344, 3.010162353515625, 114.4769287109375, 104.36805725097656, 18.58881378173828, -6.090290069580078, 174.95330810546875, 89.07211303710938, 162.25296020507812, 3.883026123046875, 0.0, 3.7814483642578125, 117.88336181640625, 178.63699340820312, 59.48796844482422, 32.46343994140625, 92.87313842773438, 123.11709594726562, -18.644268035888672, 127.12336730957031, 197.11245727539062, 84.25450134277344, 112.16477966308594, -54.540016174316406, 5.17901611328125, -26.83636474609375, -29.70306396484375, -42.84510803222656, 54.319122314453125, -17.85321044921875, 133.11956787109375, 192.62628173828125, 85.09895324707031, -14.727073669433594, 56.954345703125, 141.08221435546875, 51.54658508300781, 48.72325897216797, -47.8333740234375, -8.384750366210938, 176.2374267578125, 153.40969848632812, 214.83920288085938, 86.50442504882812, 81.01449584960938, 76.98176574707031, 169.44314575195312, 271.02325439453125, 140.52401733398438, 144.58438110351562, 102.1273193359375, 86.656982421875, -31.41680908203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000243.npy"}
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 67.416748046875, "std": 95.37513732910156, "min": -288.3774108886719, "p10": -30.273988342285154, "median": 66.58606719970703, "p90": 188.7806884765625, "max": 375.47003173828125, "pos_frac": 0.75, "sample": [-288.3774108886719, 104.87457275390625, 73.75326538085938, -12.075775146484375, 59.70953369140625, 22.134811401367188, 123.5543212890625, -29.239181518554688, 98.55413818359375, 40.53123474121094, 95.9730224609375, 44.695709228515625, 75.42523193359375, 16.200271606445312, 97.32540893554688, 73.44766235351562, 156.25830078125, 114.03176879882812, 120.85745239257812, 108.7479248046875, 159.89410400390625, 46.228538513183594, -70.76123046875, -12.388595581054688, 114.47663879394531, 2.460693359375, -119.4642333984375, 20.23882293701172, 153.48617553710938, -25.135955810546875, 30.446670532226562, -40.890869140625, -9.047348022460938, -4.964569091796875, -145.88330078125, 196.9789276123047, 274.1015625, 35.1673583984375, 137.87457275390625, 134.6685028076172, 8.800247192382812, 91.89166259765625, 221.79421997070312, -24.568889617919922, 171.4302978515625, 28.561920166015625, 7.86907958984375, 23.450851440429688, -5.996063232421875, -11.7086181640625, 71.74310302734375, 101.84326171875, 0.0, 144.94931030273438, 44.70941162109375, -98.34552001953125, 33.643341064453125, 82.84713745117188, -32.68853759765625, 20.512046813964844, 164.6619873046875, -42.91184997558594, -33.8760986328125, 139.16302490234375, 18.61017608642578, 79.9425048828125, 142.95867919921875, 279.57513427734375, 375.47003173828125, 76.30743408203125, 197.38064575195312, 117.63621520996094, -116.70901489257812, 92.7908935546875, 85.46951293945312, -24.481063842773438, 119.59283447265625, -17.869415283203125, 88.55712890625, 137.9770965576172, -1.960693359375, 122.61380004882812, 105.62216186523438, -114.56207275390625, 49.515625, 166.8115234375, -8.594329833984375, 224.9531707763672, 196.71920776367188, 5.708251953125, 30.680023193359375, -18.031585693359375, 205.95700073242188, 186.7374267578125, 154.08737182617188, 192.2385711669922, 215.0834503173828, 17.275318145751953, 180.25613403320312, 188.087158203125, 27.738128662109375, 213.28402709960938, 44.33158874511719, 127.97687530517578, 3.473297119140625, 114.54627227783203, 68.07698059082031, 80.72952270507812, 144.22354125976562, 114.15418243408203, -58.713226318359375, 51.56268310546875, -21.933246612548828, 90.88916015625, 122.4674072265625, 190.39892578125, -17.863872528076172, -24.43761444091797, 17.763389587402344, -26.983139038085938, 172.31094360351562, 30.064483642578125, 177.73622131347656, -75.70028686523438, 53.21142578125, 46.236419677734375, 64.65328979492188, 65.09515380859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000244.npy"}
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 68.50173950195312, "std": 95.99797821044922, "min": -168.04840087890625, "p10": -53.68803405761717, "median": 72.0665283203125, "p90": 198.76248474121093, "max": 259.91546630859375, "pos_frac": 0.7890625, "sample": [32.889251708984375, -27.72943115234375, 88.72695922851562, 30.439781188964844, 2.990234375, 44.47216796875, 109.9610595703125, -9.0457763671875, -29.952056884765625, 131.71542358398438, 148.4693603515625, 46.914276123046875, -4.848419189453125, 116.51324462890625, 250.83216857910156, 4.18487548828125, 172.74462890625, 235.51547241210938, 17.053863525390625, 219.62396240234375, -165.7996368408203, -79.1375732421875, -19.721237182617188, -40.4014892578125, 103.98500061035156, 79.85610961914062, 0.0, 24.79241943359375, 152.8992919921875, 77.215087890625, 221.02655029296875, 138.2296905517578, 146.68499755859375, 122.160400390625, 46.812042236328125, 172.45498657226562, 50.969818115234375, 133.5570068359375, 130.20120239257812, 153.49679565429688, 233.69192504882812, 49.22499084472656, 112.69412231445312, 166.69570922851562, 37.90631103515625, 91.89227294921875, 45.89801025390625, 91.24063110351562, 17.97032928466797, -82.35308074951172, -48.668701171875, 9.448640823364258, 27.23974609375, 257.2789611816406, 136.185791015625, -27.39073944091797, 74.56724548339844, 186.5299835205078, 259.91546630859375, 228.35862731933594, 197.972412109375, 26.644515991210938, 43.25811767578125, -30.466033935546875, 80.0460205078125, 122.61175537109375, 84.20294189453125, -98.33349609375, 80.02423095703125, -145.7863006591797, 16.336471557617188, 93.850830078125, 76.46166229248047, 112.27374267578125, 14.499008178710938, -68.1824951171875, -65.39981079101562, 10.10696792602539, 254.6170654296875, 166.04397583007812, 43.869903564453125, 132.3974609375, 12.5338134765625, 46.33216094970703, 23.3583984375, -156.1399383544922, 16.788299560546875, 171.16046142578125, 119.90797424316406, 47.70408630371094, 176.01919555664062, 81.51480102539062, -105.94975280761719, 164.17941284179688, 24.962860107421875, 215.52230834960938, 157.58334350585938, 97.53314208984375, 200.60598754882812, 69.56581115722656, 28.4332275390625, 64.27223205566406, 86.67291259765625, -119.16006469726562, -3.659900665283203, 174.9439697265625, 96.05673217773438, 99.7188720703125, 121.39636993408203, 88.19468688964844, -8.27923583984375, -31.257728576660156, 123.78167724609375, 58.908538818359375, 202.57965087890625, 34.25970458984375, 40.23878479003906, -78.10100555419922, 86.44326782226562, -124.125244140625, 229.45562744140625, 50.94844055175781, 134.98263549804688, 148.4383544921875, -168.04840087890625, 2.139101028442383, 0.0, 19.679946899414062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000245.npy"}
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 47.2033805847168, "std": 103.81891632080078, "min": -212.18743896484375, "p10": -66.60409698486328, "median": 39.13969421386719, "p90": 194.63575439453123, "max": 282.6739501953125, "pos_frac": 0.671875, "sample": [40.2796630859375, 120.0277099609375, 76.52702331542969, -37.66265869140625, 124.2364501953125, -62.3082275390625, 255.12728881835938, 18.33416748046875, 50.5383415222168, 262.7495422363281, -43.03999328613281, 145.45199584960938, 50.335845947265625, 41.93806076049805, 7.291618347167969, -61.74461364746094, -15.876808166503906, 149.05728149414062, 230.80364990234375, -137.3787841796875, 67.05123901367188, 253.5072021484375, 43.199462890625, 35.643898010253906, 50.380035400390625, -71.7283935546875, 116.16583251953125, -21.990264892578125, 2.558563232421875, 13.01507568359375, 72.16552734375, -49.90997314453125, 279.4530029296875, 40.71836853027344, -93.89398193359375, 20.042129516601562, -131.1441650390625, 154.43011474609375, 32.66668701171875, 2.0554141998291016, 72.24005126953125, -183.21963500976562, 87.39596557617188, 177.8331298828125, -67.47457885742188, -37.83807373046875, 47.677513122558594, -10.900527954101562, -26.14373779296875, 49.1475830078125, 42.860137939453125, 112.35333251953125, -80.07847595214844, 54.83135986328125, 200.98458862304688, -46.091861724853516, 143.57330322265625, -75.67534637451172, -212.18743896484375, -1.4495296478271484, 260.83538818359375, -62.91259765625, 160.18821716308594, 36.613525390625, -66.23103332519531, 33.06554412841797, 121.33248901367188, 126.25787353515625, -43.64404296875, 205.70794677734375, 282.6739501953125, 4.5093994140625, -40.510162353515625, -3.07086181640625, 7.94451904296875, 36.5869140625, 228.9130096435547, -19.428619384765625, -166.48263549804688, 37.999725341796875, -78.7325439453125, -15.9393310546875, 79.96832275390625, 56.569091796875, 8.213603973388672, 33.042755126953125, -192.25363159179688, -21.00592041015625, 54.78814697265625, 28.92852020263672, 59.604034423828125, 218.37030029296875, 13.567985534667969, -25.607147216796875, 22.28656768798828, 203.4163818359375, -34.498085021972656, 75.26922607421875, 0.0, 71.79525756835938, 228.24102783203125, 176.45001220703125, -62.6256103515625, 45.98985290527344, 191.73927307128906, 108.37644958496094, -24.74462890625, 44.57611083984375, 99.73243713378906, 83.3985595703125, 66.90980529785156, -39.277801513671875, 170.716796875, 175.39385986328125, 13.714935302734375, 173.68487548828125, -32.17035675048828, -142.0935821533203, -25.3009033203125, 18.8802490234375, 178.9185791015625, 70.5400390625, 156.01255798339844, 115.9522705078125, 47.092742919921875, 36.27156066894531, -1.3043365478515625, 191.91482543945312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000246.npy"}
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 60.199615478515625, "std": 97.61360168457031, "min": -132.90081787109375, "p10": -71.36180572509765, "median": 48.777076721191406, "p90": 190.56462707519532, "max": 349.4207763671875, "pos_frac": 0.71875, "sample": [153.41468811035156, -82.98486328125, 208.6980743408203, 127.18919372558594, -23.1322021484375, -18.727828979492188, 157.07687377929688, 184.80331420898438, 184.8813934326172, 7.892360687255859, 0.0, 192.61688232421875, 164.08016967773438, 20.11029052734375, 45.48548889160156, 164.51763916015625, -11.348739624023438, 10.632904052734375, -34.792999267578125, 92.36056518554688, 18.41082763671875, 143.21453857421875, -39.947906494140625, -106.93224334716797, 65.2490234375, 72.48054504394531, 71.48956298828125, -8.42144775390625, 106.67669677734375, 156.523193359375, 86.4945068359375, 180.6260528564453, 5.170745849609375, 268.1079406738281, 141.00885009765625, -132.90081787109375, -0.0224151611328125, 236.17005920410156, -11.5108642578125, 14.847061157226562, -51.101402282714844, -71.32215881347656, 67.1234130859375, -11.396987915039062, 15.52252197265625, 96.4422607421875, 1.7093696594238281, 21.732421875, 12.392677307128906, -86.99920654296875, 22.029773712158203, 133.91685485839844, -18.095542907714844, 111.82794189453125, -79.75741577148438, 189.68508911132812, 205.74172973632812, 73.53681945800781, 139.02627563476562, -11.678520202636719, -121.95152282714844, -32.98567199707031, 64.65819549560547, 282.1993408203125, 196.69102478027344, -71.69207763671875, 115.85304260253906, -35.376617431640625, -56.90727233886719, -104.09445190429688, -2.98614501953125, 349.4207763671875, -127.17874145507812, 10.439929962158203, 199.69247436523438, 15.1397705078125, 68.79974365234375, 7.3137054443359375, 63.79656982421875, 209.07752990722656, 101.61929321289062, 36.995399475097656, 225.2481689453125, 10.5294189453125, -71.45431518554688, 5.2974853515625, 21.564125061035156, 38.27635192871094, 148.51339721679688, 74.50396728515625, 63.731201171875, 175.2277069091797, -4.7659912109375, 52.06866455078125, 30.330047607421875, 127.26698303222656, 185.41598510742188, 138.22067260742188, 212.82952880859375, -35.12786865234375, 31.23729705810547, 34.781410217285156, 38.722686767578125, 99.87939453125, 138.5411834716797, -31.591705322265625, 53.5057373046875, -74.11361694335938, 293.75396728515625, 67.77828979492188, -86.46707153320312, 70.05213165283203, 110.49553680419922, 103.50750732421875, 11.174272537231445, -14.819366455078125, 135.43374633789062, 11.49398422241211, -103.55364990234375, 0.21044921875, 128.26205444335938, 63.144744873046875, 142.47740173339844, -1.5378265380859375, 128.4002227783203, 69.8399658203125, 9.459197998046875, 139.43988037109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000247.npy"}
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 61.86365509033203, "std": 97.26870727539062, "min": -250.881103515625, "p10": -67.80684204101563, "median": 64.56965637207031, "p90": 183.82459564208983, "max": 312.2904052734375, "pos_frac": 0.7578125, "sample": [17.925559997558594, 202.37191772460938, 3.6864700317382812, 123.14291381835938, 205.9530029296875, -15.9364013671875, 97.17431640625, -96.6397705078125, 117.3411865234375, 23.861549377441406, 14.097122192382812, 32.74493408203125, 107.36203002929688, -3.1205062866210938, -31.324432373046875, 46.62591552734375, 235.54226684570312, 18.637298583984375, -4.579437255859375, 45.004676818847656, -43.85302734375, 163.82315063476562, 92.85911560058594, 63.486053466796875, 190.614013671875, 138.23577880859375, 6.883785247802734, 27.97222900390625, -0.946563720703125, 177.25894165039062, 73.35188293457031, -114.93138122558594, 211.33670043945312, -17.782180786132812, 9.994062423706055, 90.91484069824219, 205.2974853515625, -250.881103515625, -16.761028289794922, -30.770647048950195, -26.3990478515625, 30.14728546142578, 89.81744384765625, -63.10430908203125, -66.9541015625, 185.2550048828125, 58.862911224365234, -166.15753173828125, 44.01625061035156, 62.019775390625, 34.97752380371094, 268.10491943359375, 54.3011474609375, 110.89027404785156, 289.27880859375, -22.3253173828125, 98.67010498046875, 3.167865753173828, 76.01486206054688, 97.13909912109375, 92.8004150390625, 169.01719665527344, -33.8878173828125, -82.14433288574219, 122.54312133789062, 84.3828125, 218.7994384765625, 245.94859313964844, 28.75135040283203, 117.90786743164062, 69.99714660644531, -69.79656982421875, 48.604373931884766, 118.94213104248047, 13.845703125, -117.24324035644531, 0.0, 107.05105590820312, 192.134033203125, 122.02352905273438, 132.33889770507812, 115.41952514648438, 124.5804672241211, 139.89266967773438, 44.689842224121094, 183.21156311035156, -57.27294921875, -74.9949951171875, 73.13612365722656, 182.41433715820312, 40.16888427734375, 78.36029052734375, 140.21896362304688, 117.14292907714844, 167.9296875, -71.002197265625, 116.24169921875, 65.65325927734375, 57.17022705078125, 100.10891723632812, 3.4507198333740234, -127.99798583984375, 144.00018310546875, 74.68276977539062, 170.22802734375, 182.29629516601562, 105.37197875976562, -95.30601501464844, 97.07644653320312, -98.7255859375, 24.291061401367188, 20.40338134765625, 98.44239807128906, 24.471115112304688, 312.2904052734375, 143.43331909179688, 70.9888916015625, 83.67156982421875, 62.269866943359375, 40.315460205078125, -10.767402648925781, -20.84038543701172, 31.584579467773438, 87.76821899414062, 118.05982971191406, -162.85797119140625, 97.71627807617188, 43.449981689453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000248.npy"}
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 52.993560791015625, "std": 102.95821380615234, "min": -290.98028564453125, "p10": -59.54825744628906, "median": 46.32504081726074, "p90": 189.005859375, "max": 279.26287841796875, "pos_frac": 0.7109375, "sample": [-85.08354187011719, 36.03001022338867, -11.198516845703125, -110.64837646484375, 102.65809631347656, 3.1828155517578125, -12.707244873046875, 41.95252990722656, -74.53828430175781, 26.815383911132812, -31.62206268310547, 5.747001647949219, 132.47003173828125, 156.5662841796875, -43.58512878417969, 183.74700927734375, 60.50323486328125, 101.31710815429688, 16.721588134765625, 161.95654296875, 52.757080078125, -290.98028564453125, 166.26512145996094, 41.0491943359375, -45.641693115234375, 279.26287841796875, 197.59274291992188, 2.7526702880859375, -52.426239013671875, 27.812286376953125, 117.45843505859375, -64.55368041992188, 87.619140625, -92.87002563476562, 18.731719970703125, 48.41736602783203, 193.84716796875, 42.259727478027344, 193.35903930664062, -133.3980712890625, 67.09117126464844, 175.8934326171875, -19.255142211914062, 175.58493041992188, 211.2020263671875, 93.3603515625, -49.28314208984375, -50.59051513671875, 127.455322265625, 49.238372802734375, -7.394805908203125, 120.81708526611328, 194.39413452148438, 147.29391479492188, 190.0047607421875, 42.052978515625, 38.08236312866211, 75.34256744384766, -46.81201171875, 45.42219543457031, 225.48321533203125, -42.81976318359375, -2.048675537109375, 3.236846923828125, 110.41021728515625, 4.139060974121094, 113.29251098632812, -1.5628204345703125, 39.18809509277344, 71.04573059082031, -47.52870178222656, 92.08870697021484, 171.1942138671875, 119.44810485839844, 135.6599884033203, 105.6204833984375, 259.15875244140625, 150.23231506347656, 172.5546417236328, 71.64037322998047, 196.7691650390625, 109.2135009765625, 42.70281219482422, 17.412567138671875, 9.735931396484375, 80.61502075195312, -28.593524932861328, -57.403076171875, 13.81451416015625, 153.39752197265625, 0.9124603271484375, -74.10772705078125, 188.5777587890625, 48.023529052734375, 78.32891845703125, 235.1361083984375, 123.83544921875, 114.52864074707031, -45.857421875, -178.09368896484375, 274.923583984375, -140.01986694335938, 75.27265167236328, 47.27947998046875, 91.24825286865234, 164.47662353515625, -18.11798095703125, -128.52838134765625, -36.5074462890625, -45.82582473754883, -197.09103393554688, 11.215465545654297, 96.09481811523438, 9.929115295410156, 49.57391357421875, -8.8377685546875, 74.04849243164062, 46.0155029296875, 168.70938110351562, -25.0389404296875, 181.95657348632812, 35.183624267578125, 143.24978637695312, -10.8602294921875, -156.17111206054688, 14.780975341796875, 46.634578704833984, 210.7005157470703], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000249.npy"}
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 65.92703247070312, "std": 102.919921875, "min": -211.68507385253906, "p10": -53.66481323242187, "median": 46.542301177978516, "p90": 204.35772857666015, "max": 283.3329162597656, "pos_frac": 0.734375, "sample": [71.02394104003906, 4.447235107421875, -117.03770446777344, 171.9754638671875, 146.69488525390625, 187.96502685546875, 178.8272705078125, -8.849189758300781, 107.19749450683594, -24.31565284729004, 10.671485900878906, 29.1021671295166, 262.24639892578125, 238.971435546875, 25.8074951171875, 282.66900634765625, 69.12850952148438, 46.876068115234375, 174.81536865234375, 264.7354736328125, 206.56826782226562, 44.705322265625, -9.253860473632812, 177.9442901611328, -64.454833984375, -14.708221435546875, -211.68507385253906, 30.309600830078125, 62.374725341796875, 9.11480712890625, 202.89312744140625, -36.6964111328125, 133.17919921875, 283.3329162597656, -32.53429412841797, 72.240966796875, 64.15814208984375, -5.9170989990234375, 213.2320556640625, -79.22271728515625, -10.072505950927734, -66.54417419433594, 248.691162109375, 94.66714477539062, 93.81967163085938, 37.688446044921875, -150.08090209960938, -21.55096435546875, -0.5866413116455078, 196.1100616455078, 199.59848022460938, 73.97944641113281, 57.267120361328125, 35.170806884765625, 61.977294921875, 51.1453857421875, 46.208534240722656, -54.79791259765625, -6.099639892578125, -55.88337707519531, 181.00819396972656, 66.737548828125, 115.04939270019531, 0.5416526794433594, 45.247276306152344, 157.51971435546875, 211.91119384765625, 41.65431213378906, -65.30908203125, 189.4700164794922, 11.212509155273438, 150.27752685546875, 63.46307373046875, 124.48666381835938, 37.3851318359375, -43.40008544921875, 18.085906982421875, 184.99029541015625, 133.38473510742188, 22.8304443359375, 65.83989715576172, -26.2008056640625, 150.53729248046875, -174.30740356445312, 185.9906005859375, 84.94284057617188, 107.63435363769531, -86.19526672363281, 90.02804565429688, 216.14053344726562, 44.379058837890625, 280.3170166015625, -24.87633514404297, 18.994216918945312, -166.81884765625, 32.49482727050781, 51.723663330078125, -4.7779083251953125, -55.53857421875, 203.4103546142578, 154.07937622070312, 191.28770446777344, 92.056884765625, 132.3608856201172, 41.297908782958984, 227.06983947753906, -53.17919921875, -6.406219482421875, 11.464889526367188, 10.468414306640625, 2.02587890625, 16.643280029296875, 278.1883850097656, 17.9154052734375, 119.70040893554688, 124.3940200805664, 134.91921997070312, 23.779693603515625, 23.004974365234375, 103.31639099121094, 6.9447479248046875, 142.65310668945312, 68.75576782226562, -43.36421203613281, -27.85284423828125, -14.55572509765625, 27.69439697265625, -5.507465362548828], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000250.npy"}
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 60.726905822753906, "std": 106.39469146728516, "min": -136.45620727539062, "p10": -54.34358596801757, "median": 36.480224609375, "p90": 179.9275146484375, "max": 520.3704833984375, "pos_frac": 0.71875, "sample": [19.791366577148438, -8.64752197265625, 53.29351806640625, -52.97376251220703, 63.01983642578125, -25.235443115234375, 17.773468017578125, -72.77345275878906, 71.74935913085938, 84.425537109375, 54.781166076660156, 181.7230224609375, 164.59686279296875, 18.000701904296875, -6.4427490234375, 12.813529968261719, 29.093048095703125, -87.93983459472656, 34.31483459472656, 154.3704833984375, -28.853408813476562, 33.40605163574219, 46.415863037109375, 167.580810546875, 74.35330200195312, 180.7080078125, 23.79644775390625, 130.5352325439453, 179.593017578125, 29.841461181640625, 54.93438720703125, 58.298675537109375, -59.457984924316406, 120.0201416015625, 56.814117431640625, -78.08301544189453, 108.9267578125, 25.81277084350586, -28.149246215820312, -136.45620727539062, 108.33805847167969, 89.451416015625, 130.87478637695312, 253.98138427734375, 111.3907470703125, 16.45361328125, -96.83526611328125, -132.7120361328125, 14.857770919799805, 58.93536376953125, -42.36708068847656, -4.52618408203125, 520.3704833984375, 178.71533203125, 117.72489166259766, 73.23382568359375, 15.5067138671875, 76.89836120605469, 34.572265625, -5.828590393066406, 168.29879760742188, -88.06356811523438, 19.07115936279297, 50.404327392578125, 124.75567626953125, -44.064117431640625, 19.99968719482422, 10.565670013427734, 55.3944091796875, -19.779972076416016, 161.7166748046875, -8.430892944335938, 150.57118225097656, 13.477994918823242, 178.05702209472656, 19.471633911132812, -57.53984069824219, 17.308837890625, 15.668968200683594, 26.82470703125, 40.7232666015625, -28.092483520507812, 1.0846748352050781, 90.98182678222656, 91.80083465576172, 27.01061248779297, 128.5814208984375, -41.837066650390625, -114.98509216308594, 48.802642822265625, 19.606800079345703, 434.5765380859375, 160.58306884765625, 95.87808227539062, 422.3367614746094, -1.9313106536865234, 279.25823974609375, 171.72708129882812, 146.89797973632812, -73.82193756103516, 83.26376342773438, 50.929046630859375, 202.89059448242188, -2.453338623046875, 225.62753295898438, 55.17437744140625, 162.00534057617188, -19.982223510742188, 49.23893737792969, 25.116943359375, -6.025726318359375, 169.5362548828125, -39.843170166015625, 49.03546142578125, -6.518226623535156, -68.2110595703125, -62.3851318359375, -27.070289611816406, -24.10228729248047, 216.97006225585938, 227.23779296875, 59.16331481933594, 268.88720703125, 202.21905517578125, -7.77899169921875, 29.943084716796875, 38.38818359375, 24.09002685546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000251.npy"}
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 59.50675964355469, "std": 100.37870025634766, "min": -244.34799194335938, "p10": -57.784204101562494, "median": 51.50007247924805, "p90": 192.65082397460935, "max": 307.8944091796875, "pos_frac": 0.7265625, "sample": [160.59066772460938, 224.1515350341797, 212.78216552734375, 122.93891906738281, 2.0129241943359375, 51.581565856933594, -10.4185791015625, 42.540374755859375, 10.37037467956543, -42.575347900390625, 76.95125579833984, -187.3051300048828, 76.2529296875, 44.777374267578125, 307.18841552734375, 61.549041748046875, 103.33472442626953, 157.69955444335938, 151.75521850585938, 160.64996337890625, -4.267192840576172, 162.77972412109375, 0.333282470703125, 307.8944091796875, -244.34799194335938, -11.474327087402344, -1.8638916015625, 146.732421875, -19.63702392578125, -27.514678955078125, 54.60688018798828, 196.48458862304688, 119.04791259765625, 114.4488525390625, -56.533958435058594, 0.314605712890625, -177.30865478515625, -93.72203826904297, 70.21456909179688, 208.2718505859375, 122.894287109375, 0.3600921630859375, -20.381980895996094, 41.467376708984375, -12.488632202148438, 31.852676391601562, 16.662750244140625, -7.204132080078125, 73.0191650390625, -34.559226989746094, 197.84735107421875, -9.71697998046875, 210.6768798828125, 165.32618713378906, -0.08941650390625, 32.8690185546875, 78.27157592773438, -69.79278564453125, 102.27586364746094, 55.32881164550781, -38.788330078125, -66.23689270019531, 51.4185791015625, 211.76446533203125, 24.941177368164062, 47.43414306640625, 123.618896484375, 136.1785888671875, 170.15570068359375, 226.0543212890625, -9.875, 180.67959594726562, 143.09368896484375, 6.9781494140625, 35.94768524169922, 110.13577270507812, -63.815673828125, 30.731307983398438, 112.3565673828125, 23.937664031982422, 120.50732421875, 53.27294921875, 201.60302734375, 187.20245361328125, 36.719268798828125, -24.968460083007812, 0.0, 105.93132019042969, -161.66140747070312, 167.41000366210938, 0.4766578674316406, 153.52037048339844, 63.536041259765625, 35.7056884765625, 37.301513671875, 183.83602905273438, -5.9921417236328125, 256.4651184082031, 73.18096923828125, 73.57984924316406, -56.66400146484375, 7.569236755371094, 122.2037353515625, -77.64701843261719, 238.60240173339844, 28.35845947265625, 19.232269287109375, -158.65420532226562, 108.490234375, 36.354400634765625, 10.03948974609375, -123.268798828125, 109.44882202148438, -60.39801025390625, 59.8055419921875, -11.40960693359375, 93.73616027832031, 191.00778198242188, -75.90802001953125, 161.75454711914062, 42.45098876953125, 127.18782043457031, 45.26995849609375, -23.21917724609375, 60.7127685546875, 73.3374252319336, 140.34078979492188, 67.8876953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000252.npy"}
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 54.394264221191406, "std": 105.89563751220703, "min": -220.05853271484375, "p10": -71.66204986572265, "median": 43.82670593261719, "p90": 193.67011108398435, "max": 358.5730895996094, "pos_frac": 0.7109375, "sample": [246.15872192382812, 5.8560791015625, 44.132484436035156, 25.447097778320312, 0.31695556640625, -50.69767761230469, -1.426971435546875, 358.5730895996094, 264.77545166015625, 36.449920654296875, 45.47320556640625, 134.28866577148438, -4.92254638671875, -86.25360107421875, 100.68173217773438, -120.76101684570312, 122.13357543945312, -112.84698486328125, 211.43557739257812, 1.312652587890625, 96.29916381835938, 16.829086303710938, 137.82342529296875, 38.159339904785156, 3.6871337890625, -220.05853271484375, 44.05340576171875, 196.77841186523438, 71.45774841308594, 81.7952651977539, 34.53790283203125, -62.89652633666992, -156.50120544433594, 8.619842529296875, 192.33798217773438, 69.30105590820312, -71.65995788574219, 44.348663330078125, -139.5829620361328, 66.93399047851562, 110.88153076171875, 6.406707763671875, 167.57254028320312, 165.629638671875, 53.5714111328125, -67.61155700683594, 130.9178466796875, 43.600006103515625, 26.373550415039062, 6.2935791015625, 257.9568786621094, 39.518218994140625, 238.02587890625, 125.90040588378906, 97.27484130859375, 26.6220703125, -3.2073898315429688, 77.6430892944336, -54.417144775390625, 64.88006591796875, 165.2242889404297, -26.634262084960938, 55.17362976074219, 171.75579833984375, 74.47964477539062, 102.63273620605469, -65.0491943359375, 9.858184814453125, 266.8206787109375, -147.6722412109375, 24.806655883789062, -33.380706787109375, -80.55070495605469, 130.93527221679688, -55.3243408203125, 66.41187286376953, -107.56649780273438, 16.672216415405273, -13.506744384765625, -10.3262939453125, 13.894535064697266, -0.825531005859375, 129.10662841796875, 106.292236328125, 97.0277099609375, -17.337005615234375, -18.952117919921875, -30.984100341796875, -145.12310791015625, 300.6899108886719, 95.42047119140625, -82.16914367675781, 45.95135498046875, -31.472930908203125, 50.238311767578125, 190.25619506835938, -71.66693115234375, 232.70208740234375, 80.45686340332031, 25.840118408203125, 181.8003387451172, -23.496604919433594, 12.4453125, -22.2752685546875, 110.20477294921875, 162.96177673339844, 57.3865966796875, 274.0273132324219, 77.78900146484375, 12.037891387939453, 55.23297119140625, 143.13418579101562, 133.8189697265625, -129.96694946289062, -12.973674774169922, 29.995941162109375, 42.373291015625, 122.09614562988281, 110.2230224609375, 143.58837890625, 17.388153076171875, 260.7038269042969, 224.22796630859375, 145.1891326904297, -34.78533935546875, 17.04839324951172, -19.699729919433594, 171.66238403320312], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000253.npy"}
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 61.004638671875, "std": 96.02066802978516, "min": -149.07391357421875, "p10": -57.628469085693354, "median": 44.875633239746094, "p90": 184.70811462402344, "max": 293.8836975097656, "pos_frac": 0.703125, "sample": [171.93869018554688, -63.41612243652344, -149.07391357421875, 68.64493560791016, -107.93075561523438, 0.2624073028564453, 224.7723846435547, 58.73194122314453, 7.52161979675293, 10.872764587402344, -55.877010345458984, 4.023712158203125, -10.422943115234375, 29.124420166015625, -26.088279724121094, 161.36898803710938, 64.740966796875, 104.34979248046875, -12.118316650390625, -48.7928466796875, -61.84466552734375, 207.39031982421875, 29.338653564453125, 11.648681640625, 32.77496337890625, 6.9287567138671875, 130.8621826171875, -1.740081787109375, 131.6556396484375, 80.98907470703125, 32.79754638671875, 229.08645629882812, -77.04644775390625, 151.3377685546875, 9.601922988891602, 91.19314575195312, -86.38517761230469, 118.86711883544922, 25.795913696289062, -25.787986755371094, -46.86212158203125, 159.1631317138672, 112.330078125, -79.47105407714844, -40.298553466796875, 42.547523498535156, 145.32711791992188, 102.50381469726562, 258.3878173828125, 173.14639282226562, 26.60533905029297, 2.556070327758789, -3.351428985595703, 183.93655395507812, 107.5125732421875, -0.7154541015625, 138.63638305664062, 140.01962280273438, 171.59661865234375, 77.515625, 71.0112533569336, -59.458412170410156, 166.94244384765625, 1.807647705078125, 86.0340576171875, 105.33482360839844, 73.24508666992188, 90.11343383789062, 197.79449462890625, 141.91998291015625, -92.22711181640625, 23.60479736328125, 106.30047607421875, 45.67588806152344, 153.4453887939453, -22.480987548828125, -71.94671630859375, 44.07537841796875, -69.00221252441406, -31.13678741455078, -29.247802734375, -8.5684814453125, 135.88720703125, 186.5084228515625, 124.34660339355469, 156.66067504882812, 108.535400390625, 7.773101806640625, 0.37921142578125, 264.1708068847656, 12.684009552001953, 175.46517944335938, 190.0992431640625, 162.13156127929688, 229.09974670410156, 75.12030029296875, -1.4700984954833984, 293.8836975097656, -52.032440185546875, -26.902252197265625, -32.51173400878906, 171.84872436523438, 79.521240234375, -2.0303707122802734, 0.0, -49.986358642578125, 195.25347900390625, 117.03401184082031, 46.655914306640625, 123.51741027832031, 28.57599639892578, -38.482261657714844, 140.1490936279297, 257.680908203125, 92.3818359375, 7.019775390625, -130.28982543945312, -10.981239318847656, 168.33868408203125, 27.147750854492188, -67.09912109375, -56.844207763671875, 26.081443786621094, 282.53973388671875, 51.945587158203125, 135.51393127441406, 117.07696533203125, 21.776840209960938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000254.npy"}
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 70.84877014160156, "std": 109.55414581298828, "min": -342.720703125, "p10": -49.16154174804687, "median": 58.80889892578125, "p90": 209.89700317382812, "max": 366.44866943359375, "pos_frac": 0.734375, "sample": [-8.802652359008789, -47.398345947265625, 158.8184814453125, 219.4890594482422, 66.91322326660156, -138.00576782226562, -53.275665283203125, 89.03540802001953, -31.64410400390625, 120.35786437988281, 2.828826904296875, -21.199851989746094, 26.57662582397461, 46.089599609375, 366.44866943359375, 349.074462890625, 14.745002746582031, 79.06048583984375, 267.32818603515625, 14.5687255859375, 246.16696166992188, -15.591140747070312, 58.91552734375, 145.18930053710938, 63.722869873046875, 63.79960632324219, 88.3841323852539, 161.7803955078125, -14.156646728515625, 23.500091552734375, 133.71075439453125, 86.60162353515625, 131.90615844726562, 50.897918701171875, 116.53231048583984, 14.03558349609375, 180.71710205078125, 127.31442260742188, -31.863616943359375, -46.0181884765625, 213.1041259765625, 182.68569946289062, 255.79598999023438, 49.87831115722656, -105.88410949707031, -14.473373413085938, -75.803955078125, 122.26885986328125, 39.552276611328125, 100.91777038574219, 114.95526123046875, 200.245849609375, 208.52252197265625, 79.42868041992188, 170.47573852539062, -54.296142578125, 65.0545654296875, -85.58212280273438, -158.59913635253906, 118.05377197265625, 39.5111083984375, 277.89910888671875, 58.7022705078125, 66.70191955566406, 57.5660400390625, 193.57595825195312, 15.87478256225586, -109.21147918701172, 225.34178161621094, 22.036346435546875, 126.281982421875, -23.230133056640625, 23.70165252685547, -94.49382019042969, -2.713134765625, 71.81402587890625, 9.644004821777344, 144.15330505371094, 104.7001953125, 129.3313446044922, 11.474227905273438, 115.63461303710938, -74.04489135742188, 245.07437133789062, 47.2269287109375, 207.44296264648438, 52.5245361328125, 145.70867919921875, -7.5497589111328125, 28.550064086914062, 164.826171875, 23.0399169921875, 148.33505249023438, 274.20404052734375, 10.3983154296875, 257.238525390625, 178.67974853515625, 17.863235473632812, 74.51589965820312, -19.71490478515625, 3.0294227600097656, -72.96524047851562, -7.977760314941406, -14.472000122070312, 201.75506591796875, 141.75244140625, -37.77616882324219, 196.13772583007812, 46.547515869140625, 123.35577392578125, 5.60357666015625, -19.045852661132812, -55.37284851074219, 192.37042236328125, 216.42947387695312, -1.260101318359375, -40.981689453125, -42.678131103515625, 45.655120849609375, 66.23101806640625, 35.034400939941406, 120.92835998535156, -10.167877197265625, 192.2208251953125, -342.720703125, 202.59759521484375, 18.3824462890625, 134.5616912841797], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000255.npy"}
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 62.19213104248047, "std": 121.38816833496094, "min": -560.409423828125, "p10": -83.26708984374999, "median": 69.02613830566406, "p90": 200.30506744384766, "max": 330.7142639160156, "pos_frac": 0.765625, "sample": [12.94622802734375, 53.84637451171875, 196.00961303710938, -40.16156005859375, 178.32754516601562, 109.05844116210938, 132.14236450195312, 99.8526611328125, 143.86569213867188, -22.538375854492188, 128.30812072753906, -28.039688110351562, 17.8634033203125, -95.87371826171875, 1.4859561920166016, 146.59962463378906, 200.1607208251953, -188.8011474609375, 156.64610290527344, 93.48884582519531, 23.812881469726562, 190.5126953125, -41.630210876464844, 203.37799072265625, 56.130126953125, 244.73788452148438, 233.05918884277344, -560.409423828125, 212.84591674804688, 64.2950439453125, 200.64187622070312, 81.60771179199219, -22.11376953125, 24.73114013671875, 92.51846313476562, -28.13177490234375, -78.7454833984375, -106.96234130859375, 123.4862060546875, 73.34156036376953, 285.04046630859375, 104.92959594726562, 68.39698791503906, 148.86895751953125, 65.8111572265625, 69.80963134765625, 39.6676025390625, -8.610015869140625, -137.9390106201172, 219.24295043945312, -5.116142272949219, 65.22628784179688, 179.52755737304688, 152.58831787109375, -216.5780029296875, 12.610614776611328, -180.87811279296875, -34.92742919921875, 117.46595001220703, 117.35885620117188, 202.77842712402344, -2.4676971435546875, 106.91836547851562, 175.48553466796875, 1.4113883972167969, 47.950347900390625, 72.25297546386719, -94.10028839111328, 89.28073120117188, 59.120521545410156, -49.911834716796875, -159.49642944335938, 198.33953857421875, 88.77655029296875, 184.18711853027344, 330.7142639160156, -29.164306640625, 106.0423583984375, 18.068756103515625, 219.82708740234375, 180.702392578125, 146.7174835205078, 178.8245086669922, 12.535919189453125, 96.092529296875, 118.17138671875, -93.8175048828125, 34.03282165527344, 17.65081787109375, 216.66510009765625, 1.3018035888671875, 69.65528869628906, 73.71194458007812, 289.41644287109375, 20.572906494140625, 195.58645629882812, 13.639617919921875, 47.54364013671875, 48.547935485839844, 59.31414794921875, 58.69512939453125, 14.91773796081543, 216.19903564453125, 110.413330078125, 179.09906005859375, 88.31549072265625, 108.555908203125, 70.01589965820312, -141.6472625732422, -191.03106689453125, 58.245391845703125, 164.4025115966797, 174.32257080078125, 57.552947998046875, 31.305734634399414, -152.0838165283203, 128.69845581054688, -58.11822509765625, -70.72095489501953, -77.17677307128906, 134.3437042236328, 4.576904296875, -57.21134948730469, 109.36494445800781, 44.12933349609375, 62.14130401611328, 97.482421875, 158.13589477539062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000256.npy"}
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 68.49266052246094, "std": 103.33313751220703, "min": -192.82815551757812, "p10": -31.008268737792957, "median": 61.02293395996094, "p90": 211.64789733886718, "max": 354.17559814453125, "pos_frac": 0.7265625, "sample": [102.36973571777344, 176.38644409179688, 224.73281860351562, 31.268600463867188, 34.670433044433594, 271.5052490234375, 43.496826171875, 182.03195190429688, 144.20989990234375, 179.92645263671875, 10.614166259765625, 236.56109619140625, 72.79391479492188, 101.1295166015625, 7.247224807739258, 34.347808837890625, 95.025390625, -97.45977783203125, -18.907028198242188, 82.43399047851562, 211.97967529296875, -4.9500579833984375, 22.93498992919922, 354.17559814453125, -15.056488037109375, -8.84136962890625, 92.6710205078125, 99.071533203125, 18.388031005859375, 7.393278121948242, -103.77193450927734, -10.4923095703125, 86.79702758789062, 58.03460693359375, 120.82699584960938, 159.05075073242188, 222.83038330078125, 78.72508239746094, 183.49920654296875, 154.39833068847656, -9.55975341796875, -11.76361083984375, -192.82815551757812, 176.3304443359375, 247.2703857421875, 25.99993896484375, -1.6391754150390625, 12.11260986328125, 47.09175109863281, -183.68380737304688, 75.15670776367188, 218.3486328125, 71.98882293701172, 40.85716247558594, 119.14291381835938, 72.57443237304688, 190.52200317382812, -54.06901550292969, 29.379638671875, 150.41094970703125, -17.540977478027344, 217.59268188476562, 172.619140625, 35.3204460144043, -104.1748275756836, -37.96376037597656, 64.01126098632812, -17.88246726989746, 13.883445739746094, 107.28697204589844, 65.6573486328125, 16.65289306640625, 81.00595092773438, 195.95059204101562, 4.311954498291016, 125.85031127929688, -173.57632446289062, -8.42120361328125, 142.97119140625, 143.587158203125, -12.266128540039062, 123.66743469238281, 247.97833251953125, -23.29425048828125, 81.56239318847656, 2.06988525390625, -16.12933349609375, 182.377685546875, 165.28106689453125, 9.824310302734375, -6.501708984375, 229.64764404296875, -92.93618774414062, 205.69497680664062, 116.3213119506836, 317.2648010253906, 229.20025634765625, 73.3807373046875, -23.510574340820312, -23.133529663085938, -67.04241943359375, 34.77679443359375, 141.086181640625, 85.93463134765625, 9.485260009765625, 103.38031005859375, -18.92791748046875, -11.102890014648438, -120.66839599609375, 211.50570678710938, 13.19512939453125, 131.44326782226562, -24.936660766601562, 211.09588623046875, 39.443359375, -88.98995971679688, 107.56390380859375, -28.02734375, 10.874702453613281, 29.2943115234375, 100.37638854980469, 50.89110565185547, -21.0814151763916, 166.88279724121094, 101.92305755615234, -56.425018310546875, 4.3869476318359375, 171.390380859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000257.npy"}
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 71.9548110961914, "std": 99.1325912475586, "min": -227.2403564453125, "p10": -47.29626007080078, "median": 66.34634399414062, "p90": 193.98507995605468, "max": 354.722412109375, "pos_frac": 0.7890625, "sample": [205.70068359375, -3.224306106567383, 60.84025573730469, -40.901885986328125, 5.8732757568359375, -67.42691040039062, 105.05401611328125, -25.471099853515625, 4.822511672973633, 354.722412109375, 51.68475341796875, 151.54718017578125, 143.13055419921875, 51.4134521484375, 180.27410888671875, 121.09622955322266, -51.86286926269531, 156.71237182617188, 104.114013671875, 100.42790222167969, 272.6121826171875, 143.60580444335938, 9.823158264160156, 23.6109619140625, 210.24343872070312, 60.084564208984375, -227.2403564453125, 148.76242065429688, 116.83551025390625, -45.339141845703125, 157.429443359375, 138.33956909179688, -28.609100341796875, 150.58392333984375, 219.62966918945312, 61.98797607421875, 0.18874359130859375, -28.855743408203125, 6.270328521728516, 160.44989013671875, 193.58657836914062, 103.33526611328125, 16.905006408691406, 147.3623504638672, 66.00839233398438, 225.75344848632812, 98.38037109375, 128.17596435546875, 36.44427490234375, 129.5875244140625, 67.22103118896484, 3.5463428497314453, 122.09646606445312, 59.69891357421875, -172.93283081054688, 130.33494567871094, 99.63693237304688, -30.180908203125, -111.0833740234375, -52.786895751953125, -74.31800842285156, 11.990623474121094, 27.181640625, 58.036376953125, 194.9149169921875, -7.897789001464844, 162.0181884765625, 220.72238159179688, 51.31182861328125, 116.81015014648438, 125.53140258789062, 115.87506103515625, -3.8243331909179688, 26.119125366210938, -4.757360458374023, 109.950927734375, 293.5340881347656, -75.21774291992188, 150.51455688476562, 60.41729736328125, -2.434053421020508, -33.45562744140625, 131.4696807861328, 125.12621307373047, -14.188217163085938, 13.100349426269531, 152.43711853027344, 85.39553833007812, 246.58175659179688, 66.68429565429688, 44.78782272338867, -100.76646423339844, 0.5853271484375, 6.9940032958984375, 12.325820922851562, 20.1304931640625, 169.12875366210938, 140.40081787109375, 30.702903747558594, 47.69480895996094, -151.12603759765625, -13.727447509765625, 50.171348571777344, 1.075408935546875, 214.4337158203125, 91.26612854003906, 61.31907653808594, 123.91802978515625, 190.20712280273438, 148.4857940673828, 76.81951904296875, -58.23646545410156, 275.18768310546875, 98.5958251953125, 111.72686767578125, 264.01812744140625, 102.6579818725586, 50.391754150390625, 71.96070861816406, 85.48185729980469, -114.28482055664062, 154.46136474609375, 52.370689392089844, 158.20611572265625, 3.207550048828125, 14.142837524414062, 121.29763793945312, -95.45700073242188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000258.npy"}
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 65.73184204101562, "std": 99.16793823242188, "min": -206.57037353515625, "p10": -41.76592407226561, "median": 56.6458740234375, "p90": 188.6759521484375, "max": 481.2120361328125, "pos_frac": 0.7109375, "sample": [134.343994140625, -2.1817474365234375, 120.75027465820312, 19.67392349243164, -52.16404724121094, 43.626007080078125, 141.86131286621094, 181.53768920898438, 187.3045654296875, -16.70367431640625, 71.6685791015625, 136.94326782226562, 96.86334228515625, 183.30731201171875, 49.329673767089844, -34.592437744140625, 59.5286865234375, -84.98910522460938, -22.94189453125, 153.5273895263672, -7.685821533203125, 149.46755981445312, 191.8758544921875, 22.19861602783203, 240.48886108398438, 151.51547241210938, -1.8835525512695312, 140.10159301757812, -88.15975952148438, -37.38746643066406, -0.12232208251953125, -2.071117401123047, -35.311004638671875, 126.07061767578125, 137.74853515625, 3.317169189453125, 10.43927001953125, -2.0958633422851562, 44.858428955078125, 49.75285339355469, -54.318359375, 65.8397216796875, 125.59771728515625, 12.992616653442383, -23.295654296875, 5.16290283203125, 9.98260498046875, 203.14581298828125, -7.829681396484375, 56.49932861328125, 125.74609375, 120.61935424804688, 211.4273681640625, -109.3980712890625, 38.10113525390625, -87.2330322265625, 80.06170654296875, 200.26052856445312, 149.6288604736328, 18.43242645263672, 107.70538330078125, 30.769317626953125, -4.90484619140625, 54.943885803222656, -13.483200073242188, -23.444927215576172, 148.09457397460938, -106.23880004882812, -32.77779006958008, 147.8641357421875, 195.9599609375, 72.1348876953125, -9.775070190429688, 173.2595977783203, 217.48890686035156, 0.686492919921875, 68.8243408203125, 0.0, -2.73162841796875, 173.91009521484375, 85.04118347167969, 75.393798828125, -7.17486572265625, 134.20883178710938, 104.7080078125, 101.8792724609375, 151.15966796875, -106.6138916015625, 14.028297424316406, 105.9979248046875, -206.57037353515625, -26.178192138671875, 216.69976806640625, 481.2120361328125, 37.48553466796875, 52.822021484375, 11.585250854492188, -38.405059814453125, 159.625732421875, 121.13690185546875, 63.49396514892578, 110.20732116699219, -112.84671020507812, 276.4515075683594, 6.2955322265625, 51.408203125, 160.44876098632812, -109.5362548828125, 87.64324951171875, 146.005859375, 70.91131591796875, 205.32846069335938, 85.4981689453125, 10.0008544921875, 4.130090713500977, 5.392425537109375, 76.27691650390625, 56.79241943359375, 98.7451171875, 27.624237060546875, 183.7273406982422, -54.70880126953125, -49.607940673828125, 210.22503662109375, -18.918014526367188, 252.5999755859375, 151.916259765625, 150.60934448242188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000259.npy"}
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 29.77170181274414, "std": 95.39249420166016, "min": -213.35650634765625, "p10": -83.58156661987304, "median": 6.2844953536987305, "p90": 164.2564239501953, "max": 251.7119140625, "pos_frac": 0.6015625, "sample": [134.90997314453125, -63.375457763671875, 155.3606414794922, 17.568817138671875, -59.908203125, 23.707969665527344, -45.69122314453125, 196.15029907226562, -4.680633544921875, 78.95477294921875, 5.10736083984375, 14.713226318359375, 0.0, -27.0771484375, 109.26788330078125, 32.10389709472656, 110.89970397949219, 240.80062866210938, -26.836570739746094, 12.786308288574219, 227.5729217529297, -213.35650634765625, 5.065410614013672, 114.36517333984375, 214.43804931640625, -99.48489379882812, 23.065052032470703, 6.761119842529297, 62.041351318359375, 75.7989273071289, 108.91177368164062, -93.76123046875, 126.90501403808594, 6.089269638061523, -2.1873703002929688, 82.51820373535156, -80.08255004882812, -90.91033935546875, -157.5575408935547, 6.4797210693359375, -11.680938720703125, 87.458740234375, 57.3492431640625, -49.48028564453125, -173.2202911376953, 166.99261474609375, 0.3519096374511719, -89.26251220703125, -88.29632568359375, 5.946355819702148, -37.67559814453125, -77.4765625, 188.068115234375, 148.12655639648438, 10.13473129272461, 13.793609619140625, 251.7119140625, -54.54937744140625, 204.3292236328125, 111.57376098632812, -55.966552734375, 4.500255584716797, 176.6169891357422, -5.250762939453125, 21.203857421875, -95.94156646728516, 1.2020530700683594, 163.08377075195312, -10.4796142578125, 232.2708740234375, 85.68659973144531, 182.9615478515625, -40.821380615234375, 6.054473876953125, 19.223831176757812, -19.949371337890625, -12.52260971069336, 162.93576049804688, -1.485595703125, -38.56903076171875, -144.89739990234375, 48.11344909667969, 246.59274291992188, 5.2864227294921875, -21.220611572265625, 62.60772705078125, -72.48390197753906, -7.75897216796875, -12.718429565429688, -17.772994995117188, -53.527374267578125, 4.993865966796875, -55.9522705078125, 212.86529541015625, 34.3614501953125, 7.329193115234375, -28.463912963867188, 120.8778076171875, -10.4764404296875, 69.6348876953125, 136.87298583984375, 132.80752563476562, 83.80325317382812, -20.045700073242188, 70.79849243164062, -83.03863525390625, 4.440155029296875, 129.90267944335938, 70.9808120727539, 115.86941528320312, -24.83258056640625, 99.830810546875, 106.69662475585938, -8.91668701171875, -84.6749267578125, 52.570343017578125, -118.21279907226562, -120.77774047851562, 7.1895751953125, 1.408416748046875, 129.24676513671875, -52.9754638671875, 30.04937744140625, -0.06973648071289062, 137.77194213867188, -83.11298370361328, 41.082305908203125, 2.34063720703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000260.npy"}
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 53.38719177246094, "std": 107.8483657836914, "min": -197.46607971191406, "p10": -85.55784301757812, "median": 38.56006622314453, "p90": 203.22004089355468, "max": 307.9528503417969, "pos_frac": 0.65625, "sample": [-3.53363037109375, -197.46607971191406, -42.41688537597656, -71.19148254394531, 34.627410888671875, 55.73980712890625, 167.52899169921875, -30.866958618164062, 57.947174072265625, 81.90293884277344, 48.76866149902344, 0.0, -9.300296783447266, -131.524169921875, 22.81744956970215, 84.03054809570312, -29.72576141357422, 54.2822265625, 170.29452514648438, 60.66778564453125, 181.6189727783203, -146.291259765625, 179.06446838378906, -35.8887939453125, 105.23196411132812, 23.628646850585938, -92.50569152832031, 249.64881896972656, -63.55731201171875, -117.10673522949219, 60.830413818359375, 181.2687225341797, 293.68975830078125, 13.15231704711914, 153.7445068359375, 230.719970703125, 22.23217010498047, 5.18853759765625, -138.64102172851562, -20.355484008789062, 24.43729019165039, 39.870025634765625, 199.07919311523438, -92.3756103515625, 97.73529052734375, -29.637008666992188, -0.7163181304931641, 227.9227752685547, -19.036331176757812, 28.45907974243164, 94.84976196289062, -44.280303955078125, 109.9959716796875, 163.82534790039062, 194.5577392578125, 48.55438232421875, 6.327262878417969, -85.09872436523438, 248.281494140625, 151.78256225585938, 243.85198974609375, 19.782821655273438, 75.72117614746094, -22.436431884765625, 88.91168212890625, -13.154335021972656, 34.98127746582031, -51.569480895996094, 169.55177307128906, 25.150238037109375, 95.80120849609375, 288.16265869140625, 100.7218017578125, 212.92007446289062, 114.33378601074219, 97.46932983398438, -101.256591796875, 52.695438385009766, -6.81793212890625, 22.47809600830078, 127.877685546875, -15.982833862304688, 37.25010681152344, -124.15472412109375, 203.24227905273438, 125.73721313476562, 82.48370361328125, 94.29736328125, 176.43722534179688, -98.32745361328125, -14.931503295898438, 54.548828125, -86.62911987304688, -33.992706298828125, -159.24609375, 160.69052124023438, 149.15975952148438, 201.36495971679688, 120.89031982421875, -71.86640930175781, -52.32850646972656, 56.31827926635742, 20.671802520751953, 130.87066650390625, -0.3483734130859375, 255.56640625, -44.76177978515625, 220.17214965820312, 203.44854736328125, 21.960426330566406, 203.21051025390625, -28.290420532226562, 32.50050354003906, -161.19337463378906, 60.836151123046875, -0.2631072998046875, -25.317291259765625, 150.3616943359375, 1.11541748046875, 122.10774993896484, 112.57952880859375, 307.9528503417969, 48.286651611328125, -40.0181884765625, 28.05097770690918, -42.28816223144531, 75.19383239746094, 28.22907257080078], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000261.npy"}
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 68.55213165283203, "std": 98.93804168701172, "min": -165.29119873046875, "p10": -37.818200683593744, "median": 57.76837158203125, "p90": 214.40743103027341, "max": 285.8017883300781, "pos_frac": 0.734375, "sample": [219.18038940429688, 98.8809814453125, 105.1778564453125, 23.275711059570312, -0.95758056640625, 63.72044372558594, 171.36883544921875, 264.05023193359375, 30.641876220703125, 110.77157592773438, 191.13275146484375, -20.9437255859375, 113.1219711303711, 20.456649780273438, -22.910980224609375, 1.6971549987792969, 227.20523071289062, 92.90304565429688, -31.42657470703125, 35.60138702392578, 74.76470947265625, -47.329315185546875, -132.1485595703125, 200.37396240234375, 33.10064697265625, 193.92544555664062, 20.040603637695312, 29.826202392578125, 75.01214599609375, 226.82373046875, 185.79266357421875, -10.457626342773438, 66.79597473144531, -26.6441650390625, 8.523147583007812, -95.16130065917969, 65.37548828125, 18.3828125, -34.36625671386719, 85.512939453125, 116.93807983398438, -10.500419616699219, 56.35455322265625, -121.46981811523438, 144.03704833984375, -34.68711853027344, 24.75628662109375, -0.825714111328125, 13.10565185546875, -77.75167846679688, 157.9677734375, 227.0087890625, 254.40106201171875, 175.06900024414062, -73.67523193359375, -3.3013458251953125, 14.834381103515625, 94.0018310546875, -42.95062255859375, 36.11891174316406, 169.20880126953125, -10.621185302734375, 40.843505859375, 107.53799438476562, 14.728141784667969, 1.6873779296875, 174.43374633789062, 73.11749267578125, 198.13424682617188, 133.85635375976562, 11.8101806640625, -88.10728454589844, 179.57913208007812, 42.31793212890625, 113.92471313476562, -22.677825927734375, 26.416519165039062, 212.36187744140625, 8.0792236328125, 13.638298034667969, -126.01608276367188, -20.215858459472656, 0.0, -14.247413635253906, 243.09341430664062, 32.429901123046875, 73.10848999023438, 155.39666748046875, -4.372158050537109, -16.896377563476562, 194.2700653076172, 61.800018310546875, 199.02517700195312, 30.085052490234375, 228.8902587890625, 247.07861328125, -35.61859130859375, 35.78950500488281, 10.222274780273438, 251.69412231445312, -6.921344757080078, -56.832855224609375, 75.04376220703125, 83.63821411132812, -50.59393310546875, 74.16168212890625, 59.18218994140625, 61.4693603515625, 206.51754760742188, 134.77369689941406, 33.99000549316406, 35.31327819824219, 285.8017883300781, 167.6020050048828, 100.23175048828125, 100.19757080078125, 260.34967041015625, -68.28099060058594, 211.42909240722656, 136.76478576660156, 114.69256591796875, -165.29119873046875, 73.83877563476562, 71.11465454101562, 62.11680603027344, 246.85745239257812, 14.733993530273438, -13.5357666015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000262.npy"}
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 63.323760986328125, "std": 103.21694946289062, "min": -209.11068725585938, "p10": -57.19773254394531, "median": 52.120811462402344, "p90": 188.17222900390624, "max": 295.935546875, "pos_frac": 0.71875, "sample": [264.9248046875, 181.69845581054688, 193.32083129882812, 49.98876953125, 38.79867172241211, 83.83895874023438, -28.641571044921875, 20.50042724609375, 117.22408294677734, 145.523681640625, 4.098836898803711, 43.83024597167969, 54.97100830078125, 23.49169921875, -43.7965087890625, 9.411430358886719, -0.7425899505615234, -19.758941650390625, 59.879608154296875, 18.5655517578125, 190.72955322265625, 23.999404907226562, 25.11957550048828, 247.692138671875, -13.744293212890625, 123.18377685546875, 116.49990844726562, 112.66205596923828, -41.50701904296875, -143.88290405273438, -185.430419921875, 83.56521606445312, 183.85772705078125, 204.50564575195312, 162.747314453125, 51.48387145996094, 216.60601806640625, 117.17195129394531, -56.5706787109375, 182.0259246826172, 45.45951843261719, 0.605255126953125, 40.022857666015625, 295.935546875, 242.90533447265625, 6.293575286865234, 99.70132446289062, -74.05220031738281, 223.9688720703125, 147.647705078125, 128.9793701171875, -67.80899047851562, -77.54463195800781, 135.3653564453125, 161.85690307617188, 141.093017578125, -108.5511245727539, 57.0863037109375, 236.54672241210938, 207.5131072998047, 0.0, 95.4600830078125, -191.79837036132812, -0.40494728088378906, -9.725341796875, 21.80438232421875, 8.280517578125, -43.05213928222656, 45.433006286621094, 59.03507995605469, 97.86412048339844, -13.573799133300781, -19.92115020751953, 119.42996978759766, -7.3494415283203125, 59.58837890625, 99.33090209960938, 177.2845458984375, 94.6663818359375, -8.307235717773438, -58.722686767578125, 43.11073303222656, -107.12957763671875, 40.01788330078125, -19.018417358398438, 4.667675018310547, -124.27508544921875, 163.9461669921875, 180.58734130859375, 13.111801147460938, -16.90087890625, 161.71319580078125, 15.030744552612305, -209.11068725585938, 1.0708503723144531, -47.203857421875, -58.660858154296875, 161.7337646484375, 35.80451965332031, -177.413818359375, 57.58509826660156, -36.227447509765625, 0.0, 34.767845153808594, 131.4083251953125, 62.2227783203125, 294.291748046875, 101.94331359863281, 186.146728515625, 141.55628967285156, 174.15237426757812, 186.83843994140625, 169.36708068847656, 0.0, 52.75775146484375, 161.78981018066406, 97.61933898925781, 46.41096496582031, 133.3441162109375, 36.452247619628906, 77.15718078613281, 187.07623291015625, 132.59352111816406, -13.50299072265625, 224.29000854492188, -1.27899169921875, 84.6864013671875, 134.72415161132812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000263.npy"}
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 73.27791595458984, "std": 100.36682891845703, "min": -197.7974395751953, "p10": -34.46376800537109, "median": 56.811126708984375, "p90": 204.59370117187495, "max": 369.5586242675781, "pos_frac": 0.7421875, "sample": [10.531225204467773, 89.25091552734375, 166.55706787109375, 31.169952392578125, 180.96026611328125, 200.95541381835938, 70.5287857055664, 99.29588317871094, -3.571409225463867, -29.096710205078125, -126.63395690917969, 101.57977294921875, 73.90322875976562, 43.221954345703125, 74.5076904296875, 123.38279724121094, 148.93545532226562, 146.0480499267578, -27.430017471313477, -7.653648376464844, 98.14208984375, 157.072021484375, -8.186553955078125, 17.04043960571289, -103.56471252441406, 160.24920654296875, 163.37844848632812, 118.76002502441406, 39.09906005859375, 213.08303833007812, 12.1873779296875, -45.41020202636719, 239.07435607910156, 369.5586242675781, 264.301025390625, 0.0, 59.4591064453125, 299.7869873046875, 0.890472412109375, 24.011390686035156, -14.20574951171875, 170.625732421875, -37.908599853515625, 109.617431640625, 22.517333984375, 264.1708984375, -197.7974395751953, 235.21902465820312, -96.8892822265625, 102.898681640625, 26.29486083984375, 4.017822265625, 191.98089599609375, 69.67129516601562, 100.61550903320312, -136.10079956054688, 171.4389190673828, 166.050537109375, 9.089323043823242, 38.143165588378906, 98.17031860351562, 70.39234161376953, 101.8184585571289, 44.265663146972656, 53.077850341796875, 31.42535400390625, -22.17547607421875, 19.112648010253906, 142.22821044921875, 180.29949951171875, 63.36277770996094, -0.7751922607421875, -95.9744873046875, 3.852947235107422, 28.133590698242188, -32.13665771484375, -43.64043045043945, 43.22257995605469, 158.70703125, 155.84112548828125, 171.96124267578125, -32.98741149902344, 28.72842025756836, 125.07597351074219, -0.629302978515625, 237.07666015625, -41.03944396972656, 0.0, 37.02972412109375, 124.98384094238281, -76.512451171875, 53.098907470703125, 53.43562316894531, 176.8887939453125, 43.9244384765625, 85.49282836914062, -6.549900054931641, 48.22076416015625, 44.90220642089844, 38.7904052734375, 228.93983459472656, 11.81884765625, -50.21168518066406, -25.6165771484375, -23.18731689453125, 130.08920288085938, 4.20159912109375, 83.8739013671875, -21.374603271484375, 235.00564575195312, 119.76972961425781, 239.626953125, 90.98260498046875, 155.78919982910156, 87.73171997070312, 286.42535400390625, -26.636993408203125, 54.16314697265625, 197.38723754882812, -29.977508544921875, -39.8074951171875, 146.20562744140625, 63.63038635253906, 77.89816284179688, 258.0711669921875, -7.676271438598633, 186.84085083007812, 189.68685913085938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000264.npy"}
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 74.02152252197266, "std": 111.37838745117188, "min": -218.7161865234375, "p10": -57.064538574218744, "median": 63.281211853027344, "p90": 211.4256408691406, "max": 375.29730224609375, "pos_frac": 0.6953125, "sample": [3.2988338470458984, 165.93222045898438, 300.99609375, -78.34283447265625, 64.90811157226562, 1.9437255859375, 179.78001403808594, 110.55294799804688, 87.06672668457031, -2.71978759765625, 137.48133850097656, 167.5111846923828, 158.6339111328125, -12.65338134765625, 181.91015625, -16.8583984375, 166.00482177734375, 375.29730224609375, 132.68222045898438, 142.6280975341797, 281.0098876953125, 261.6208801269531, 159.63858032226562, 124.40998077392578, 196.20513916015625, 8.927001953125, 215.61773681640625, 183.76077270507812, 178.58566284179688, 17.299560546875, 5.493766784667969, 83.6812744140625, 155.67774963378906, 104.8201904296875, 224.98684692382812, 141.9449462890625, 108.51123046875, 15.7906494140625, 95.509033203125, 191.36306762695312, -39.014007568359375, 38.809661865234375, -50.236083984375, 140.833251953125, -1.482208251953125, 61.65431213378906, 21.72314453125, -39.61492919921875, 3.2509613037109375, 126.00810241699219, 1.581787109375, -63.556396484375, -78.48883056640625, -3.2510986328125, -8.360794067382812, 235.60662841796875, 163.7547607421875, 83.95346069335938, -45.479888916015625, -78.6531982421875, 47.41510009765625, 159.63214111328125, -56.43231201171875, 190.31472778320312, -46.483734130859375, -10.1405029296875, 198.5555419921875, 106.34848022460938, 141.8689422607422, -25.423194885253906, -35.923858642578125, 200.1540985107422, 286.49969482421875, 25.80035400390625, 0.0, -19.503421783447266, -166.76333618164062, -124.78204345703125, 280.36029052734375, 270.7422790527344, 229.15167236328125, 49.54986572265625, 94.08445739746094, 14.677932739257812, 189.5897979736328, 247.88510131835938, 0.0, -96.15593719482422, -218.7161865234375, 147.77328491210938, 56.3785400390625, -58.53973388671875, -39.43267822265625, -32.820396423339844, 184.48434448242188, 12.069374084472656, 72.399658203125, 239.11605834960938, -19.63812255859375, -49.261566162109375, -0.3688812255859375, 67.74349975585938, 55.89025115966797, 36.830936431884766, -46.622650146484375, -34.35682678222656, -97.68623352050781, 206.33914184570312, 152.75857543945312, -67.18191528320312, -12.866256713867188, 113.35609436035156, 128.2691650390625, 8.432586669921875, 31.172286987304688, 131.36627197265625, 26.036731719970703, 203.94566345214844, 120.66015625, 132.8640899658203, 209.6290283203125, -62.8466796875, 48.630859375, 49.91448974609375, 29.425569534301758, -144.0992431640625, 107.57618713378906, 145.16134643554688], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000265.npy"}
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 74.36134338378906, "std": 98.36548614501953, "min": -195.89572143554688, "p10": -24.715281867980956, "median": 57.44938659667969, "p90": 200.12748718261716, "max": 370.03369140625, "pos_frac": 0.7890625, "sample": [62.03729248046875, 140.46603393554688, 223.10311889648438, 148.33631896972656, -12.212432861328125, -64.90643310546875, 128.2525177001953, -17.019378662109375, 3.8935508728027344, 14.419403076171875, 15.222564697265625, 91.84963989257812, 7.5346527099609375, 36.44329833984375, 122.53807067871094, -5.475860595703125, -50.207763671875, 36.5391731262207, 112.30876159667969, -195.89572143554688, 234.6368865966797, 35.897300720214844, 108.14474487304688, 42.84521484375, 195.5240478515625, 36.686004638671875, -20.91778564453125, 183.1304931640625, 370.03369140625, 139.90948486328125, 64.12548065185547, 269.87493896484375, 54.882476806640625, 63.604644775390625, 16.408218383789062, 245.83837890625, -93.1995849609375, 29.403770446777344, 70.16261291503906, 180.28329467773438, 33.05500793457031, 26.783935546875, 50.46150207519531, 63.421142578125, 139.79229736328125, 6.740020751953125, 99.09603881835938, 180.12762451171875, 13.318809509277344, 90.5428466796875, 93.61758422851562, 138.2149658203125, -124.02430725097656, -76.30705261230469, 1.3994483947753906, 0.06060791015625, -11.840421676635742, 53.00682067871094, 33.05657958984375, 122.88900756835938, 159.86712646484375, 111.23320007324219, 30.650344848632812, 26.871673583984375, -12.873611450195312, 271.6934814453125, 99.16021728515625, 95.5174560546875, -41.10980224609375, 60.671661376953125, 204.03240966796875, 231.8226776123047, 115.4339599609375, 53.829681396484375, 162.16455078125, 280.24603271484375, -0.11199951171875, -7.264404296875, -55.2196044921875, -20.845277786254883, -186.85919189453125, 29.6900634765625, 43.561065673828125, 264.30487060546875, 104.80352783203125, 174.57662963867188, 60.01629638671875, 115.91098022460938, 245.03265380859375, 66.32522583007812, 162.78790283203125, 44.65069580078125, -25.89119529724121, -2.5919189453125, 82.10043334960938, 263.3182373046875, 118.20108032226562, 177.04061889648438, -40.81524658203125, -87.55149841308594, -30.85479736328125, 90.54043579101562, 74.7923583984375, 134.73699951171875, 30.29088020324707, -6.4520263671875, -2.326061248779297, 16.093589782714844, 35.094940185546875, -4.762481689453125, 148.64413452148438, 198.45394897460938, 294.701416015625, 14.461349487304688, 181.59024047851562, 8.15435791015625, 192.23446655273438, 23.705612182617188, 0.7437782287597656, 131.67593383789062, 159.95849609375, -24.211318969726562, 25.509445190429688, 52.36145782470703, 106.81126403808594, 107.23092651367188, 32.8189697265625, 193.9598388671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000266.npy"}
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 59.47343063354492, "std": 101.44120025634766, "min": -258.16705322265625, "p10": -49.81722869873047, "median": 56.4345817565918, "p90": 174.41387939453125, "max": 370.9993896484375, "pos_frac": 0.703125, "sample": [-4.84272575378418, 100.2584228515625, 162.7991485595703, 86.48980712890625, 370.9993896484375, 139.5605010986328, -23.689308166503906, 85.89857482910156, 67.556640625, -25.7100830078125, -7.736606597900391, 59.35601806640625, -49.592041015625, 152.86325073242188, -0.4958362579345703, 4.60675048828125, 31.843833923339844, 176.42578125, 78.09027099609375, 59.844085693359375, -50.34266662597656, 230.001953125, 42.192657470703125, -18.780868530273438, 51.00483703613281, -37.831024169921875, 62.163330078125, 118.33746337890625, 110.3966064453125, -258.16705322265625, 19.269012451171875, -36.30067443847656, 67.54583740234375, -30.538604736328125, 44.33897399902344, 98.54911804199219, 10.966480255126953, 211.5460205078125, 169.10452270507812, 106.31275939941406, -31.638565063476562, 226.2735595703125, 2.732574462890625, 145.47445678710938, 55.49870300292969, 306.4921569824219, 110.87530517578125, 56.67425537109375, 231.2150421142578, 139.8118438720703, -42.31764602661133, -124.58892822265625, 50.191650390625, 97.0924072265625, 28.8701171875, 275.27996826171875, 85.51588439941406, 120.03018188476562, 154.42068481445312, 122.33810424804688, 221.1654510498047, 107.87496948242188, -197.84613037109375, 17.431854248046875, 115.20967102050781, 34.39997863769531, 76.73077392578125, -52.36355972290039, 114.30654907226562, 25.221771240234375, 39.054534912109375, 137.00531005859375, 160.68328857421875, -168.55926513671875, -76.083251953125, 125.99453735351562, 84.48043823242188, 221.15414428710938, -25.298709869384766, 8.183584213256836, -1.7464752197265625, -32.31964111328125, -25.780670166015625, 112.13592529296875, 10.487014770507812, 132.109619140625, -58.8284912109375, 153.4314422607422, 105.85443115234375, 80.06015014648438, 86.51756286621094, -21.107208251953125, 60.92613220214844, -45.956512451171875, 28.041946411132812, 173.5516357421875, 146.88897705078125, 120.8614273071289, 186.9667205810547, -23.974517822265625, 123.12889862060547, 76.96951293945312, -34.631378173828125, 13.025146484375, -8.647361755371094, 37.61865234375, 131.29647827148438, 95.90693664550781, -1.3269424438476562, 356.3731689453125, -76.1628646850586, -63.73735046386719, 95.13255310058594, 16.662567138671875, -19.69482421875, -5.7513427734375, 184.5887451171875, 149.66021728515625, -47.343536376953125, -70.555419921875, 155.05419921875, 51.806671142578125, 20.421142578125, 0.5300674438476562, 56.194908142089844, -133.56195068359375, -66.70538330078125, 34.945556640625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000267.npy"}
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 70.96157836914062, "std": 102.05659484863281, "min": -172.86215209960938, "p10": -43.249377441406246, "median": 63.075408935546875, "p90": 206.37672424316403, "max": 336.104736328125, "pos_frac": 0.7421875, "sample": [105.64588928222656, 8.854278564453125, 34.90092468261719, 142.35150146484375, -4.9072265625, -22.393218994140625, 81.98524475097656, -11.38592529296875, -6.6140289306640625, -70.7845230102539, -28.136024475097656, 272.0079040527344, 156.56546020507812, 88.96894836425781, 21.186279296875, 64.0257568359375, 336.104736328125, -6.180580139160156, -2.72747802734375, 136.70913696289062, 186.38575744628906, 119.12408447265625, 154.14340209960938, -166.1485595703125, 212.2489013671875, 64.96826171875, 4.105188369750977, 78.07830810546875, 78.86474609375, 87.90262603759766, 29.93951416015625, 135.968994140625, 152.83135986328125, 68.84153747558594, -72.66358947753906, -7.173069000244141, -62.987709045410156, -119.57675170898438, 18.259780883789062, 225.07763671875, 104.31936645507812, 60.30694580078125, 103.20880126953125, 128.474609375, 99.14886474609375, 15.770004272460938, 161.3330078125, -123.63088989257812, -10.88299560546875, 15.835006713867188, 191.18386840820312, 53.156036376953125, 231.38156127929688, 252.61358642578125, 1.5572357177734375, 24.587615966796875, 269.25384521484375, -52.28265380859375, 32.42218017578125, -172.86215209960938, 203.86007690429688, 38.28315734863281, 197.3845977783203, -42.390594482421875, -4.89520263671875, 156.76593017578125, 175.50729370117188, 112.6722412109375, -16.474288940429688, -19.50750732421875, 228.12161254882812, 122.10932922363281, 232.72003173828125, 121.67547607421875, 102.18521118164062, -21.757980346679688, 67.72698974609375, 189.36660766601562, 200.4053955078125, 138.0916748046875, 16.835479736328125, 148.9599609375, 154.1368408203125, -47.28549575805664, 81.06829833984375, 110.18726348876953, -157.7388916015625, 188.3804931640625, 96.12313842773438, 77.61456298828125, 173.68487548828125, 199.47158813476562, 236.61001586914062, 69.39068603515625, 9.88031005859375, 90.903076171875, 54.09159851074219, 7.9974822998046875, 63.05792236328125, 41.14495849609375, 31.55462646484375, 31.646949768066406, 63.0928955078125, 11.450942993164062, 6.705680847167969, 248.9677276611328, -12.4605712890625, -45.253204345703125, -2.070648193359375, -4.035186767578125, 221.89605712890625, 5.406917572021484, 127.66696166992188, 145.75965881347656, 14.34417724609375, 13.337112426757812, -129.6890411376953, -16.96136474609375, 307.206787109375, 60.81585693359375, -16.81240463256836, -90.28973388671875, 35.905601501464844, 171.32212829589844, 36.1575927734375, 194.13662719726562, 7.654541015625, 0.0], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000268.npy"}
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 54.28797912597656, "std": 103.16228485107422, "min": -306.93438720703125, "p10": -55.77085266113281, "median": 36.92753601074219, "p90": 186.82765502929684, "max": 275.130615234375, "pos_frac": 0.6953125, "sample": [-118.683349609375, 115.99121856689453, 110.66311645507812, -98.45155334472656, 136.74171447753906, 14.664398193359375, -46.16722106933594, 180.2783660888672, 83.23965454101562, 19.00146484375, 85.75459289550781, 33.033660888671875, 28.557144165039062, 161.4072723388672, 33.03509521484375, 158.49774169921875, 18.937515258789062, 4.893096923828125, 212.65496826171875, 34.74915313720703, 221.39181518554688, 140.20208740234375, 40.97895812988281, 201.52294921875, 115.535400390625, -17.872474670410156, 173.5721435546875, 139.361328125, 15.16424560546875, 136.812255859375, 22.099273681640625, 235.678466796875, 128.4072265625, -102.98945617675781, 116.56634521484375, 0.0, 159.00018310546875, -21.95916175842285, 27.51354217529297, 195.3704833984375, -40.81255340576172, 41.091796875, 239.6463623046875, 244.4582061767578, 29.24932861328125, -134.5841522216797, 131.11846923828125, 0.0, 8.02798843383789, 38.28887939453125, 176.18524169921875, -17.115264892578125, 233.18405151367188, -185.85012817382812, -52.64239501953125, 246.79896545410156, 163.00906372070312, 81.1000747680664, -120.01046752929688, -2.664379119873047, 131.66912841796875, 16.274871826171875, 59.944915771484375, -39.59808349609375, 155.93994140625, 16.355010986328125, -7.69378662109375, -14.195846557617188, 116.5657958984375, 15.589767456054688, 64.46319580078125, -13.1390380859375, 0.0, 275.130615234375, 27.764389038085938, 63.546478271484375, -190.126953125, -41.6674690246582, 226.78839111328125, 145.25296020507812, 9.937774658203125, -14.939453125, 174.62713623046875, 92.54786682128906, 120.14131164550781, 48.371612548828125, 59.317047119140625, 70.08319091796875, 38.58001708984375, 51.57305908203125, 22.819320678710938, 102.92367553710938, 9.655609130859375, -73.320556640625, 136.63754272460938, -57.9447021484375, -59.657012939453125, 176.9774932861328, -42.64863586425781, 26.446258544921875, 97.47314453125, 69.06354522705078, -5.056694030761719, -44.16349792480469, 0.7615547180175781, 42.28837585449219, -61.82032775878906, 151.29751586914062, -54.839202880859375, 183.16644287109375, -37.41876220703125, 137.69818115234375, 81.49273681640625, 33.09693908691406, 59.33229064941406, -31.58502960205078, -133.90701293945312, -25.462936401367188, 35.566192626953125, -3.2793655395507812, 159.084716796875, 208.50531005859375, -35.427032470703125, -306.93438720703125, -52.15966796875, 28.007835388183594, 218.13958740234375, 161.31527709960938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000269.npy"}
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 63.22173309326172, "std": 110.65121459960938, "min": -221.94805908203125, "p10": -69.72678375244139, "median": 62.85469055175781, "p90": 188.82734374999998, "max": 408.47381591796875, "pos_frac": 0.75, "sample": [-180.2667236328125, 93.624267578125, 157.24737548828125, 39.643089294433594, -52.981285095214844, 282.8054504394531, -65.93199157714844, 99.14288330078125, 134.4073486328125, -111.05548095703125, 176.17230224609375, 176.23663330078125, 62.066650390625, 35.08985900878906, 93.23637390136719, 59.60406494140625, 34.043731689453125, 227.97158813476562, 5.922172546386719, -61.31768798828125, -51.63934326171875, -10.031303405761719, 128.68223571777344, 102.65962219238281, -19.488723754882812, -0.971710205078125, 167.46206665039062, 145.3365478515625, -31.91925048828125, 2.616659164428711, 19.32769775390625, 194.84068298339844, 66.61294555664062, 9.967761993408203, 51.87437438964844, 27.985305786132812, -139.8916015625, 52.322174072265625, -170.42205810546875, 186.2501983642578, 55.55853271484375, 123.33200073242188, 108.88862609863281, 21.401565551757812, 152.43084716796875, -60.774993896484375, -16.901050567626953, 289.3905029296875, 119.37049102783203, -83.17776489257812, -179.37762451171875, 214.02886962890625, 147.6319580078125, 165.12156677246094, -49.24591064453125, 50.64823913574219, 109.51239013671875, -78.581298828125, 75.30244445800781, 5.8892974853515625, 4.4495849609375, 12.164093017578125, 24.202392578125, 22.015113830566406, 246.75924682617188, 120.69088745117188, 226.09564208984375, 144.3293914794922, 171.6551513671875, -59.99809265136719, 63.642730712890625, 78.313232421875, -166.39920043945312, 157.23849487304688, 3.703125, 253.10342407226562, 124.26036071777344, 3.7037220001220703, 99.61607360839844, -150.39700317382812, 26.043609619140625, 21.927825927734375, 8.990478515625, 65.93173217773438, -3.8110427856445312, 70.70164489746094, 64.71218872070312, 88.14848327636719, 25.80840301513672, -29.258140563964844, -20.40185546875, -85.32363891601562, 202.5926513671875, 252.35272216796875, -85.73942565917969, 12.435049057006836, 185.24752807617188, 139.8321990966797, 118.767822265625, -12.54876708984375, 110.28192138671875, 152.4588623046875, -12.94183349609375, 8.121322631835938, 38.458251953125, -221.94805908203125, 73.22442626953125, -16.817138671875, 162.10592651367188, 65.39889526367188, 11.869384765625, 178.6376953125, 52.650238037109375, 122.4136962890625, 168.29876708984375, 185.6741943359375, 210.74517822265625, 84.61709594726562, 17.27276611328125, 320.271240234375, -46.07220458984375, 408.47381591796875, 150.2943115234375, 178.9620819091797, 74.3145751953125, 104.33038330078125, 83.0164794921875, -104.9722900390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000270.npy"}
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 58.87036895751953, "std": 101.8916015625, "min": -197.0692138671875, "p10": -51.02376251220703, "median": 46.97040557861328, "p90": 193.8305908203125, "max": 352.8604431152344, "pos_frac": 0.71875, "sample": [57.48150634765625, -93.3863525390625, -8.1474609375, 37.40301513671875, 20.186279296875, -14.256637573242188, 19.866539001464844, -24.388595581054688, 56.40960693359375, 5.5804595947265625, 38.24609375, 46.074005126953125, 145.21353149414062, 70.78239440917969, -109.289794921875, 142.95034790039062, 121.72955322265625, 142.49740600585938, 133.3300018310547, -22.4459228515625, 25.54047203063965, 21.8306884765625, 200.67874145507812, 114.10453796386719, 11.164535522460938, 11.4122314453125, -197.0692138671875, 100.489013671875, 35.42413330078125, 22.419525146484375, 211.25778198242188, 198.44688415527344, 1.70428466796875, 98.17718505859375, 91.33739471435547, 16.763336181640625, 124.4371337890625, 63.87212371826172, 0.435028076171875, -82.44313049316406, 246.57138061523438, 62.802947998046875, -13.244110107421875, 172.57070922851562, 170.16253662109375, -0.549713134765625, 73.21856689453125, 1.203887939453125, 63.05223083496094, 65.36495971679688, 47.47877502441406, 184.30352783203125, 130.05128479003906, 121.19023132324219, 352.8604431152344, 274.1904602050781, -7.0759124755859375, 177.84774780273438, 20.877044677734375, 178.38890075683594, 97.63105773925781, -9.130462646484375, -4.66607666015625, 301.7366638183594, 62.980064392089844, -54.034576416015625, -37.758941650390625, 282.2947998046875, 48.171173095703125, 18.0640869140625, 179.30694580078125, 194.50021362304688, -49.39959716796875, 35.85269546508789, -0.38482666015625, 42.247802734375, -58.852386474609375, 50.38232421875, 46.4620361328125, 17.415462493896484, -37.7967529296875, 51.279388427734375, 66.10238647460938, -137.7172088623047, 271.205810546875, 90.86203002929688, -103.12555694580078, -14.272918701171875, 27.62560272216797, -90.15347290039062, -12.126808166503906, 215.69119262695312, 161.29415893554688, -35.56329345703125, -6.7990875244140625, 89.31198120117188, 96.54936218261719, 169.15957641601562, 44.12318420410156, 7.365478515625, 48.453887939453125, -27.832748413085938, -170.02182006835938, 187.6441650390625, 193.54360961914062, 47.99418640136719, -101.45797729492188, 179.79141235351562, 49.41969299316406, 100.806640625, 54.3624267578125, 76.79928588867188, 141.202392578125, -49.73341369628906, -22.19464111328125, -3.664794921875, 219.63790893554688, 193.367431640625, -156.06663513183594, -64.68154907226562, 31.766475677490234, -42.04785919189453, 286.16314697265625, 113.14166259765625, 0.21681594848632812, 62.691192626953125, 17.110671997070312, -3.92279052734375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000271.npy"}
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 59.87489700317383, "std": 110.55757904052734, "min": -243.0888671875, "p10": -59.85226135253905, "median": 42.550804138183594, "p90": 210.37552795410156, "max": 389.47314453125, "pos_frac": 0.6953125, "sample": [-43.16035461425781, 0.0, 36.83416748046875, 25.708145141601562, -226.8749542236328, 88.31875610351562, 344.67425537109375, -9.3760986328125, 172.43157958984375, -28.9853515625, -5.0302886962890625, 4.879167556762695, 144.86849975585938, 173.57601928710938, 73.94366455078125, 242.4835205078125, 222.21148681640625, 63.2939453125, 71.25936889648438, -8.471138000488281, 250.3504638671875, 22.300785064697266, 2.63201904296875, -68.57855224609375, 125.78236389160156, -1.991912841796875, 5.708560943603516, 208.98776245117188, 0.2174072265625, 12.930450439453125, 134.22732543945312, 343.817138671875, 112.30322265625, -10.856147766113281, 389.47314453125, 102.36285400390625, 213.6136474609375, 126.76776123046875, 61.270263671875, -11.243370056152344, 99.6744613647461, -110.08026123046875, -33.994964599609375, 175.32815551757812, 50.65374755859375, 42.98052978515625, 13.417037963867188, -57.10345458984375, 145.7146759033203, -0.5581035614013672, 236.9221954345703, 186.24240112304688, -22.81448745727539, 42.12107849121094, 134.7223358154297, 195.9249267578125, -103.85292053222656, 80.68414306640625, 139.708740234375, 10.891773223876953, -6.98396110534668, 57.57403564453125, 45.19712829589844, -37.4534912109375, 26.38116455078125, -14.711260795593262, 248.6673583984375, 143.82379150390625, 97.13833618164062, 27.59228515625, 249.5030517578125, -66.26614379882812, 121.52731323242188, 88.36636352539062, -243.0888671875, -67.3592529296875, 51.6383056640625, 5.081512451171875, -180.2042236328125, 185.26580810546875, 177.69943237304688, 131.04757690429688, 33.24311065673828, 218.69085693359375, 5.254547119140625, -197.30526733398438, 18.985469818115234, 84.8541259765625, 68.52590942382812, 0.349761962890625, -14.446083068847656, 117.0113525390625, 61.423492431640625, -94.28732299804688, -70.6360855102539, 134.1668701171875, 89.06356811523438, 167.58995056152344, -22.88818359375, 11.166498184204102, 0.0, -37.61865234375, 29.97685432434082, -47.511444091796875, -12.759429931640625, -8.769775390625, 137.97512817382812, -12.694282531738281, 102.54647064208984, 91.47219848632812, 109.80355072021484, 230.1124267578125, -157.421630859375, -14.5072021484375, 92.725830078125, -72.68203735351562, 77.24644470214844, 19.137235641479492, -7.923208236694336, 6.0279541015625, 231.81423950195312, 128.14625549316406, 19.197555541992188, 22.8848876953125, 81.45480346679688, 199.05010986328125, 182.8665771484375, 32.99571228027344], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000272.npy"}
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 65.82425689697266, "std": 107.24958801269531, "min": -173.66534423828125, "p10": -59.317649078369136, "median": 60.1727180480957, "p90": 199.4786148071289, "max": 352.304931640625, "pos_frac": 0.7265625, "sample": [90.47921752929688, 159.13665771484375, 98.37921142578125, 141.0597381591797, 198.53001403808594, -17.379417419433594, 337.13201904296875, 88.885498046875, 156.23043823242188, 52.36450958251953, -50.0723876953125, 4.9395751953125, -126.78138732910156, 104.72994995117188, 226.53021240234375, -173.66534423828125, 97.046875, 205.10284423828125, 14.01226806640625, 60.74041748046875, -11.062744140625, -39.34325408935547, 185.68511962890625, 23.43408203125, 159.75103759765625, 33.6695556640625, -134.76516723632812, 88.22657775878906, 155.8858642578125, 31.273483276367188, 15.813858032226562, -58.323150634765625, -33.3037109375, 194.02650451660156, -61.638145446777344, -51.144805908203125, 25.494354248046875, -97.32208251953125, -29.899673461914062, 75.18353271484375, -87.53153991699219, -47.752105712890625, -25.884117126464844, 78.6611328125, 88.68310546875, 56.66827392578125, 59.605018615722656, -14.101766586303711, -64.31340789794922, 14.046112060546875, 81.9869384765625, -85.82192993164062, 275.2298278808594, 32.317626953125, 18.283920288085938, 7.429107666015625, -33.29510498046875, 55.01763916015625, 111.07754516601562, 129.58340454101562, -121.965087890625, 201.6920166015625, 87.35440826416016, 166.0546875, -14.837127685546875, 44.067047119140625, 114.1524658203125, 28.14275360107422, 117.08479309082031, 7.637725830078125, 229.72756958007812, -9.989486694335938, -20.956939697265625, 18.585174560546875, 152.30145263671875, 238.17575073242188, -11.214385986328125, 178.20584106445312, 70.77960205078125, 123.79425048828125, 84.13491821289062, 149.52316284179688, 125.55780792236328, -44.79156494140625, 89.43353271484375, 82.4994125366211, 109.67367553710938, 62.726173400878906, 258.8000793457031, 86.5967025756836, 73.78692626953125, -24.979881286621094, 17.764205932617188, 187.59197998046875, -170.7213134765625, 62.665924072265625, -78.6429443359375, 125.96931457519531, 264.7132568359375, 3.7032623291015625, 307.9769592285156, 107.92762756347656, 179.5264892578125, 78.01057434082031, 193.95770263671875, -11.084728240966797, -49.4130859375, 40.247161865234375, 166.11434936523438, 1.1634674072265625, 47.67280578613281, 43.198486328125, 2.6774139404296875, 352.304931640625, 45.817718505859375, 130.21063232421875, 15.3775634765625, 260.2481994628906, 93.7520751953125, -12.917861938476562, 32.211753845214844, 187.96307373046875, -3.537017822265625, 192.88156127929688, 244.47726440429688, -91.70993041992188, 185.1566162109375, -168.436279296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000273.npy"}
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 77.7999496459961, "std": 101.07211303710938, "min": -179.47128295898438, "p10": -35.59246215820312, "median": 75.30226135253906, "p90": 204.81398010253906, "max": 332.1785888671875, "pos_frac": 0.796875, "sample": [-16.125011444091797, 168.6690673828125, -88.23068237304688, 33.89439392089844, 2.719898223876953, 80.1107177734375, 150.55218505859375, 184.52027893066406, 200.60536193847656, 4.390724182128906, 115.50479125976562, 142.57737731933594, 288.0226135253906, 39.05908203125, -8.34466552734375, 125.78623962402344, 259.8797607421875, 275.15875244140625, -29.909149169921875, -8.868370056152344, 36.500244140625, 204.12637329101562, 242.48397827148438, 180.28256225585938, 89.85194396972656, 18.82977294921875, -121.12380981445312, 6.664031982421875, 10.812515258789062, 32.231536865234375, 30.03277587890625, 138.67984008789062, 178.11659240722656, 166.45361328125, 164.94882202148438, 169.1744384765625, -40.959228515625, 109.86787414550781, 13.073486328125, 177.642578125, 94.39767456054688, 9.981882095336914, -33.29241943359375, -11.899604797363281, -19.015716552734375, 94.33230590820312, 153.50189208984375, -22.649742126464844, 18.9195556640625, 193.2096710205078, 4.250244140625, 94.55419921875, 113.34037780761719, -73.12858581542969, -102.73119354248047, -43.85829162597656, 31.19793701171875, 108.02133178710938, 1.5090141296386719, 206.41839599609375, 5.203411102294922, 27.005863189697266, 54.84819030761719, 46.630615234375, -176.17755126953125, 21.47962188720703, -3.973297119140625, -1.5547428131103516, 310.8221435546875, 108.13539123535156, 42.35844421386719, -179.47128295898438, 180.3289794921875, 136.8153076171875, 98.59088134765625, 54.57684326171875, -43.415252685546875, 142.63168334960938, 332.1785888671875, 161.9041748046875, 261.6676025390625, 121.57931518554688, 44.8736572265625, 213.18438720703125, -59.950164794921875, 130.8819580078125, 211.200927734375, 148.56787109375, 157.19204711914062, 196.2974853515625, 149.09918212890625, 46.49778747558594, -123.95997619628906, 111.62933349609375, 133.28030395507812, 141.24188232421875, 92.3721923828125, 255.77206420898438, 209.68441772460938, 49.33428955078125, 0.46185302734375, 20.1424560546875, 92.14971923828125, -16.55950927734375, -6.835968017578125, 31.27999496459961, 30.517227172851562, 194.78981018066406, 40.622406005859375, 116.217041015625, 52.83465576171875, 129.94996643066406, 23.12066650390625, 86.82533264160156, 70.49380493164062, -12.61688232421875, 86.57064819335938, 23.603713989257812, -62.537506103515625, 110.55752563476562, 6.969135284423828, -130.01226806640625, 136.53024291992188, 7.351991653442383, 49.820098876953125, 163.92828369140625, 141.27703857421875, 216.85531616210938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000274.npy"}
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 76.83008575439453, "std": 98.50930786132812, "min": -119.44052124023438, "p10": -28.144404983520506, "median": 64.47290802001953, "p90": 212.50235290527343, "max": 342.0455017089844, "pos_frac": 0.7421875, "sample": [211.5025634765625, -10.5419921875, -18.310733795166016, -7.20733642578125, 149.89120483398438, 19.6317138671875, 333.3010559082031, 84.447509765625, 74.65823364257812, 10.155517578125, 15.403106689453125, 342.0455017089844, -5.389434814453125, 218.09286499023438, 0.0, 36.4627685546875, 120.73780822753906, -28.028518676757812, 3.1612606048583984, 16.58587646484375, 0.0, 0.0, 134.63426208496094, 131.0059356689453, -0.7589645385742188, 3.9561004638671875, 150.08123779296875, 137.04541015625, 13.48980712890625, 88.02427673339844, 235.42160034179688, 22.61968994140625, 68.0781478881836, 143.4754638671875, 224.1527099609375, 155.50473022460938, 18.372987747192383, 14.568252563476562, -86.6494140625, 82.0738296508789, 333.48187255859375, 45.219207763671875, 177.74305725097656, 69.20242309570312, -46.522491455078125, -6.254604339599609, -24.133453369140625, 182.59317016601562, 205.33705139160156, -92.92413330078125, 149.65170288085938, -119.44052124023438, 3.6639785766601562, -15.829071044921875, 28.73723602294922, 38.128387451171875, 65.24000549316406, 12.3433837890625, 272.35504150390625, 0.0, 44.13287353515625, -39.23536682128906, 98.60104370117188, 312.8792419433594, 15.458511352539062, 175.11691284179688, 231.8987579345703, 85.3363037109375, -16.240737915039062, 0.0, -15.286693572998047, 185.99142456054688, 21.773666381835938, 178.23379516601562, -31.872093200683594, 67.042236328125, 148.9862060546875, 220.39227294921875, 91.57180786132812, 10.523963928222656, 133.29464721679688, 187.40390014648438, 91.546142578125, 212.85626220703125, 94.63513946533203, 5.137331008911133, 63.705810546875, 212.35067749023438, -13.77008056640625, 121.0694580078125, -34.8626708984375, 43.74664306640625, 24.66558837890625, 87.12615966796875, 241.44900512695312, 12.9093017578125, -107.3570556640625, -38.04173278808594, 253.816650390625, -42.8348388671875, 27.848190307617188, 127.19705200195312, 70.02105712890625, 10.295791625976562, 175.869384765625, 67.75199127197266, -3.1942291259765625, 110.27555084228516, 22.524261474609375, -54.634033203125, 154.691162109375, -52.47654724121094, 172.77383422851562, 31.4609375, 77.39935302734375, -0.6896629333496094, 127.85345458984375, 105.44287109375, 209.96395874023438, 76.79367065429688, 7.075286865234375, 208.41012573242188, 24.509368896484375, -28.414806365966797, 145.715087890625, -12.4232177734375, 159.39154052734375, 158.38198852539062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000275.npy"}
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 70.29570007324219, "std": 107.15750122070312, "min": -304.7744445800781, "p10": -31.868655395507812, "median": 49.46142578125, "p90": 219.28907165527343, "max": 345.2376708984375, "pos_frac": 0.765625, "sample": [345.2376708984375, 74.6611328125, 137.48257446289062, -13.007034301757812, 118.40740966796875, 29.355133056640625, 66.7606201171875, 18.846435546875, 110.315185546875, 5.907796859741211, 74.36211395263672, -43.917633056640625, 50.099456787109375, -8.919479370117188, -102.71737670898438, 115.53630065917969, 161.39144897460938, 17.80462646484375, -32.5654296875, 199.62393188476562, 173.56295776367188, -19.018814086914062, 4.934101104736328, 196.25729370117188, 135.67697143554688, -97.4168701171875, -242.334716796875, -24.590087890625, -42.2783203125, 131.23199462890625, 61.583831787109375, -4.409210205078125, 306.581298828125, 81.5482177734375, 33.324127197265625, 36.00189208984375, 81.1044921875, 1.1113948822021484, 67.467041015625, 82.12890625, 41.6689453125, 23.3800048828125, 155.10983276367188, 16.075042724609375, 35.22313690185547, 264.9223327636719, 5.370849609375, 238.32693481445312, 82.36920166015625, 31.029388427734375, 18.416847229003906, 204.26065063476562, 152.123779296875, 39.282257080078125, 186.0515594482422, 19.597251892089844, -31.570037841796875, 96.28372192382812, 198.96859741210938, -31.5303955078125, 0.0, -89.58779907226562, 61.86241149902344, -5.902229309082031, -3.0178146362304688, -12.794342041015625, 26.541168212890625, 183.19775390625, -26.162734985351562, 65.07431030273438, -6.309192657470703, 14.828460693359375, 73.30267333984375, 239.68472290039062, 136.2161865234375, 221.12432861328125, 197.84140014648438, -16.671798706054688, 223.64389038085938, 38.943084716796875, 108.02301025390625, 218.50253295898438, 98.49151611328125, 170.41714477539062, 20.289520263671875, 48.823394775390625, -304.7744445800781, 71.3320541381836, 6.880165100097656, 39.47601318359375, 2.9276161193847656, 235.77130126953125, 198.86883544921875, 18.554229736328125, 41.948486328125, 90.46795654296875, -54.17681884765625, 256.14129638671875, 27.708541870117188, 274.5863037109375, 218.10174560546875, 110.54399108886719, 247.62738037109375, 26.261474609375, -30.402618408203125, 129.52774047851562, -14.686729431152344, -8.612342834472656, 44.53564453125, 171.8333740234375, 11.130775451660156, 156.88168334960938, 50.98323059082031, 121.882080078125, 127.48556518554688, -135.61459350585938, 109.59710693359375, 30.667755126953125, 60.750701904296875, -136.28707885742188, 158.48794555664062, 20.2669677734375, 257.1340637207031, -101.12637329101562, -37.118499755859375, 243.911376953125, 197.09600830078125, 42.426368713378906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000276.npy"}
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 49.63103485107422, "std": 106.20406341552734, "min": -243.32196044921875, "p10": -72.01137084960936, "median": 39.59300231933594, "p90": 171.7243927001953, "max": 370.3343505859375, "pos_frac": 0.65625, "sample": [85.14689636230469, -22.36090087890625, 17.274517059326172, 135.4447021484375, 348.036865234375, -102.85015869140625, -37.46441650390625, 8.40765380859375, 66.11572265625, -12.369705200195312, 117.0892333984375, 92.82435607910156, 37.68565368652344, 128.39044189453125, -46.893272399902344, 118.94027709960938, 55.333763122558594, -138.8316650390625, 115.21566772460938, 66.6849365234375, 41.50035095214844, -5.140729904174805, 31.11717987060547, 90.68133544921875, 56.772125244140625, 147.71221923828125, -10.568771362304688, -50.405120849609375, 106.87802124023438, 80.67105102539062, 14.68414306640625, -162.606689453125, -9.583454132080078, -113.0323486328125, -12.559661865234375, 43.97686767578125, 114.73927307128906, 25.58490753173828, 143.52621459960938, 84.51080322265625, -243.32196044921875, -51.64666748046875, -97.73514556884766, -37.01361083984375, 25.263885498046875, 62.478759765625, 68.7059326171875, 181.92105102539062, 86.04635620117188, 81.96282958984375, 134.750244140625, -24.95411491394043, -1.3393478393554688, 370.3343505859375, 148.78750610351562, 0.0, -87.75982666015625, 149.75518798828125, 83.61940002441406, 62.131927490234375, 295.78704833984375, 155.57144165039062, -92.11870574951172, -6.014167785644531, -30.444366455078125, 19.502685546875, 113.83539581298828, -121.12739562988281, 268.630615234375, 297.3189697265625, -18.536029815673828, -62.067352294921875, 153.1298370361328, 75.76113891601562, 12.203857421875, -184.39132690429688, 107.96011352539062, -21.562255859375, -76.69734191894531, 5.323413848876953, 32.20489501953125, 34.574241638183594, -117.26515197753906, -34.522491455078125, 104.70458984375, 187.34866333007812, 109.17821502685547, 77.2896728515625, -70.00309753417969, 296.18853759765625, 8.849372863769531, 36.1290283203125, -27.399185180664062, -14.214004516601562, 177.38302612304688, 22.34088134765625, -19.994873046875, 147.78924560546875, 93.20062255859375, 255.45858764648438, 158.40969848632812, 68.14894104003906, 70.52993774414062, 169.66131591796875, 97.26785278320312, 176.53823852539062, 59.855316162109375, 279.219482421875, 16.694061279296875, 0.0, 169.10153198242188, 180.11407470703125, 10.607063293457031, -13.07476806640625, -69.15162658691406, -0.1054840087890625, -42.974609375, 89.17929077148438, 74.45132446289062, 14.061046600341797, 0.459320068359375, -148.15139770507812, -5.6358642578125, 3.138713836669922, 57.732177734375, 92.69786071777344, 95.83302307128906, -5.404977798461914], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000277.npy"}
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 48.51531982421875, "std": 122.65709686279297, "min": -170.54653930664062, "p10": -81.72252044677734, "median": 10.668787002563477, "p90": 200.90907897949216, "max": 553.5645141601562, "pos_frac": 0.6015625, "sample": [318.172119140625, 7.926862716674805, -158.818359375, 10.385250091552734, -27.460662841796875, 71.58340454101562, 101.9002685546875, -6.7702178955078125, -21.940105438232422, 228.80352783203125, 75.87408447265625, 16.78058624267578, 125.36698150634766, 163.57176208496094, 153.63055419921875, 89.50875854492188, -123.9599609375, -61.154327392578125, 95.74140930175781, -27.6829833984375, -50.818809509277344, 89.98333740234375, 5.672027587890625, 0.0290069580078125, -22.326446533203125, 304.6844482421875, 249.14544677734375, 10.05764389038086, 3.972381591796875, -133.4901123046875, -170.54653930664062, 553.5645141601562, -2.2433624267578125, -25.505126953125, 163.75067138671875, 43.75086975097656, 14.015363693237305, -91.52987670898438, 254.22039794921875, -7.512504577636719, 168.82598876953125, 118.4364013671875, 157.55709838867188, 8.0609130859375, 21.847209930419922, 88.16064453125, -53.35003662109375, 189.24713134765625, 155.07302856445312, -41.45014953613281, 7.116512298583984, 190.40496826171875, -2.9527931213378906, -43.28717803955078, -69.75192260742188, -66.95352172851562, -2.9227657318115234, 333.09588623046875, -107.17149353027344, -38.413490295410156, 22.185623168945312, 142.34161376953125, -23.60357666015625, 9.0064697265625, -23.74176025390625, 199.27716064453125, -131.92453002929688, 153.20425415039062, 171.76475524902344, 92.741943359375, 229.977294921875, 253.3951416015625, 15.284143447875977, -58.852874755859375, 121.03033447265625, -168.48831176757812, 13.382352828979492, 10.952323913574219, 97.89752197265625, 143.50579833984375, -159.53070068359375, -80.69123840332031, -0.1838245391845703, 137.99041748046875, 169.32896423339844, 41.544883728027344, -28.79827880859375, 194.27532958984375, -3.56182861328125, 47.52564239501953, -7.09967041015625, -28.16705322265625, 110.83740997314453, 369.4639892578125, 2.545166015625, 248.06936645507812, -116.65188598632812, -1.18255615234375, 50.409217834472656, 78.22991943359375, -19.91888427734375, 31.90350341796875, 35.20220947265625, -140.93295288085938, -124.11135864257812, 51.57221984863281, 132.51242065429688, -27.01482391357422, -16.904571533203125, 114.18035125732422, 152.8791046142578, 11.895763397216797, -52.35296630859375, 305.78997802734375, 3.4415321350097656, -39.314964294433594, 4.326444625854492, -76.13385009765625, -3.57855224609375, -41.48942565917969, 6.457405090332031, -5.566680908203125, 21.33782958984375, 197.32296752929688, -84.12884521484375, 81.3717041015625, -39.09473419189453, 204.71688842773438], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000278.npy"}
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 74.13917541503906, "std": 113.85585021972656, "min": -188.17440795898438, "p10": -55.61590423583984, "median": 56.57050132751465, "p90": 206.958251953125, "max": 430.6695556640625, "pos_frac": 0.796875, "sample": [77.33404541015625, 212.66221618652344, 198.08929443359375, 61.304443359375, 16.450820922851562, 11.015716552734375, 41.60760498046875, 160.16226196289062, 21.24428939819336, 151.39404296875, -3.774078369140625, 192.66973876953125, 115.19805908203125, -4.797210693359375, 0.57659912109375, 104.9217529296875, 206.1214599609375, 136.63433837890625, 118.335205078125, 168.05389404296875, 28.302261352539062, -2.96661376953125, 19.38861083984375, 5.0791015625, 38.199432373046875, 12.35455322265625, -24.622169494628906, 24.815460205078125, -172.07159423828125, 0.4984588623046875, 161.0012969970703, 232.23297119140625, 162.33917236328125, 85.15191650390625, 92.51455688476562, -23.778060913085938, 20.373397827148438, 55.99021530151367, -188.17440795898438, 12.750762939453125, 36.21039962768555, 36.77494812011719, 60.7562255859375, 169.73623657226562, 36.65257263183594, 195.29066467285156, 28.638153076171875, 19.183059692382812, 41.48878479003906, -5.1252593994140625, 199.9864501953125, 198.45071411132812, -87.11029052734375, 29.208072662353516, 106.14248657226562, -183.22244262695312, 248.65875244140625, -138.12513732910156, 156.35653686523438, 205.11410522460938, -67.09996795654297, 59.84857177734375, 100.13368225097656, 109.6588134765625, -49.36065673828125, -78.67361450195312, -55.412322998046875, 65.60353088378906, 9.171024322509766, 177.34600830078125, 1.212310791015625, 397.426513671875, 40.90277099609375, -12.196731567382812, 3.330476760864258, -162.82464599609375, -4.8238525390625, 430.6695556640625, 237.44415283203125, 23.73492431640625, -89.00772094726562, 67.90011596679688, 237.93072509765625, 410.195556640625, 198.02294921875, 98.9976806640625, 49.840606689453125, -5.8275604248046875, 137.47030639648438, 65.13311767578125, 46.99299621582031, -56.09092712402344, 62.24627685546875, 146.78253173828125, 4.874446868896484, -74.4451904296875, 208.9107666015625, 20.100217819213867, 25.598236083984375, 150.17453002929688, 173.347412109375, 218.30722045898438, -44.992576599121094, 118.168701171875, 124.90069580078125, 166.84506225585938, 128.86788940429688, 35.365570068359375, 3.44036865234375, 133.1772003173828, -47.852386474609375, 98.7471923828125, 213.43978881835938, 197.42919921875, 3.7696533203125, 71.391845703125, -58.165985107421875, 134.94180297851562, -146.17208862304688, 0.31316375732421875, 346.966064453125, 94.52688598632812, 187.5819091796875, 38.956947326660156, 153.43475341796875, 15.348175048828125, 57.150787353515625, 259.0394287109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000279.npy"}
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 58.03026580810547, "std": 102.695068359375, "min": -263.82012939453125, "p10": -30.67902908325194, "median": 41.58980369567871, "p90": 199.4222900390625, "max": 324.13739013671875, "pos_frac": 0.7265625, "sample": [211.80557250976562, 24.958152770996094, -9.690277099609375, 4.074462890625, -8.015615463256836, -154.82122802734375, 170.39051818847656, 188.29644775390625, 137.18182373046875, 65.1256103515625, -14.385009765625, 39.78364562988281, 4.670677185058594, 185.7161865234375, 2.6297607421875, -39.39313507080078, 101.54217529296875, 58.215118408203125, 173.68756103515625, -13.40338134765625, 70.023193359375, 202.62176513671875, -228.69764709472656, 124.69752502441406, -140.63848876953125, 228.83937072753906, 8.803131103515625, 47.96697998046875, 107.69097900390625, 250.2221221923828, 0.0, 93.68824768066406, 188.96145629882812, -63.2745361328125, 49.795989990234375, 175.01611328125, 21.997512817382812, 128.71328735351562, 19.764238357543945, -9.02227783203125, 55.5028076171875, 50.93439483642578, 24.035755157470703, 23.32525634765625, -17.908523559570312, 23.131561279296875, 54.588623046875, -26.944412231445312, 138.7947235107422, -11.6612548828125, 164.84796142578125, 127.58023834228516, 9.75152587890625, 29.02872085571289, 34.87025451660156, 214.0863037109375, 87.0828857421875, 3.95831298828125, -10.755645751953125, 198.05108642578125, 242.17559814453125, 65.58908081054688, 153.4046173095703, -3.6899795532226562, -17.492191314697266, -17.092254638671875, -70.54534912109375, 179.37240600585938, 30.208999633789062, 99.6573486328125, 269.37957763671875, -40.516090393066406, 84.627685546875, 123.12973022460938, 203.27557373046875, -263.82012939453125, 148.81190490722656, 12.501617431640625, -5.8009033203125, 110.66143798828125, 62.55128479003906, 212.13198852539062, 25.98651123046875, 66.42141723632812, 249.81875610351562, 22.250396728515625, 34.5274658203125, -96.21907043457031, 40.94207763671875, -16.64483642578125, 172.45574951171875, 25.1895751953125, 2.9851341247558594, -187.78643798828125, 73.05303955078125, 186.43563842773438, -41.49430847167969, 59.641624450683594, 34.26092529296875, 79.6275634765625, 101.24111938476562, 39.92970275878906, 289.35626220703125, -25.791290283203125, 99.93954467773438, -19.209457397460938, -3.3088531494140625, 17.668106079101562, 324.13739013671875, 24.003738403320312, 77.03244018554688, 49.79376220703125, 4.019710540771484, 149.99166870117188, -96.2337646484375, 42.23752975463867, 66.207763671875, 169.18414306640625, -1.5938720703125, -175.12989807128906, 0.9410400390625, 50.213294982910156, 113.72528076171875, -18.475894927978516, 51.68504333496094, -7.956531524658203, 222.1220703125, -5.667724609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000280.npy"}
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 69.6479721069336, "std": 105.00955200195312, "min": -223.68423461914062, "p10": -39.51864929199219, "median": 51.18708038330078, "p90": 202.92733154296874, "max": 546.86767578125, "pos_frac": 0.7109375, "sample": [135.55661010742188, 70.07579803466797, -39.574554443359375, 30.220123291015625, 113.23880004882812, 44.37396240234375, 89.35360717773438, 28.713760375976562, 100.588623046875, 119.53253173828125, 199.3804931640625, -2.572174072265625, -64.0787353515625, 233.34713745117188, -62.33308410644531, 134.68475341796875, 41.69130325317383, 167.0954132080078, 35.82078552246094, -23.565200805664062, 80.234375, 6.3515472412109375, 37.24638366699219, -36.06903076171875, 101.01895141601562, 79.09644317626953, -75.074462890625, 252.93380737304688, 40.98748779296875, 26.933746337890625, 0.0, 89.387939453125, 41.263763427734375, 161.809326171875, -1.265380859375, 173.6811981201172, 203.71902465820312, 2.293731689453125, 102.3621826171875, 93.5408935546875, -223.68423461914062, 5.155364990234375, 142.97271728515625, -15.637001037597656, 0.0, -39.49468994140625, -0.023590087890625, 88.5765380859375, 9.8077392578125, 51.38841247558594, 184.61090087890625, 50.985748291015625, -18.44757080078125, 84.05252075195312, -0.8017578125, 231.34439086914062, -35.42376708984375, 58.57099914550781, 18.55859375, 59.88440704345703, 157.19662475585938, 115.44891357421875, -45.97998046875, 3.2147560119628906, 60.68251037597656, 73.7459716796875, 120.80043029785156, 6.24993896484375, -6.9820556640625, -16.71612548828125, -29.93206787109375, 125.9937744140625, -3.34454345703125, 203.89585876464844, 214.21768188476562, 17.943023681640625, 163.73202514648438, 132.06777954101562, 161.8555908203125, 54.433349609375, -4.096290588378906, 128.37423706054688, 546.86767578125, 372.3894958496094, 310.20556640625, 0.09322357177734375, -8.218795776367188, 106.65738677978516, 71.73249816894531, 12.719154357910156, 216.02301025390625, 27.010009765625, 10.50885009765625, 16.038475036621094, 175.35977172851562, -35.06622314453125, 8.923126220703125, 203.0135498046875, 304.81787109375, -9.379150390625, -67.71749877929688, 87.41973876953125, 0.0, 202.890380859375, -51.923187255859375, -64.34474182128906, 33.94203186035156, -76.549560546875, 220.18496704101562, 154.37054443359375, 127.88418579101562, 71.37466430664062, 171.27410888671875, 29.677001953125, 142.33377075195312, -109.21722412109375, -7.409820556640625, 188.81875610351562, -4.985725402832031, 173.53619384765625, 94.27944946289062, -57.3349609375, -30.757064819335938, 202.379150390625, 122.02328491210938, -71.755126953125, 79.4339599609375, 8.218795776367188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000281.npy"}
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 62.281951904296875, "std": 108.77175903320312, "min": -277.94244384765625, "p10": -60.700865173339835, "median": 49.62113952636719, "p90": 210.0685363769531, "max": 361.91949462890625, "pos_frac": 0.734375, "sample": [190.31680297851562, -37.10536193847656, 49.92169189453125, 286.2061767578125, -48.27447509765625, 13.451606750488281, -179.56344604492188, 234.3126220703125, 73.43411254882812, -51.240745544433594, -13.026863098144531, 124.14098358154297, 39.438995361328125, -15.38616943359375, 95.0172119140625, 24.873558044433594, 121.97201538085938, -129.94619750976562, 104.89816284179688, 140.15750122070312, 11.45855712890625, 18.464080810546875, -29.628360748291016, 231.144287109375, 96.954345703125, 239.3623046875, 181.41835021972656, 130.70062255859375, -98.4417724609375, 8.8697509765625, -29.7998046875, 100.87197875976562, 160.4967041015625, 185.791748046875, 3.7911376953125, -10.340911865234375, 146.00990295410156, -2.2303409576416016, 66.18417358398438, 119.167236328125, 102.18638610839844, 8.571502685546875, 107.9180908203125, -37.601898193359375, 236.77874755859375, -72.41691589355469, -138.29513549804688, 93.91963195800781, 66.3084716796875, 33.989501953125, 65.595947265625, 34.462677001953125, -23.681243896484375, 21.236526489257812, -66.75540161132812, 217.81613159179688, 65.9925537109375, 55.96221923828125, -29.489990234375, -108.7174072265625, -7.560272216796875, 35.16302490234375, -19.066802978515625, 133.90277099609375, 180.58709716796875, 128.18255615234375, 114.36304473876953, -69.14098358154297, 43.295204162597656, 135.629638671875, -277.94244384765625, -170.40298461914062, 159.77853393554688, 61.91851806640625, 49.320587158203125, 236.41961669921875, 235.18646240234375, 4.5489044189453125, 253.232421875, -174.7974853515625, -58.10606384277344, 15.653690338134766, 38.52557373046875, 204.42108154296875, -2.013397216796875, 30.108505249023438, 6.497528076171875, 15.35772705078125, 152.05685424804688, 186.7701416015625, 204.77603149414062, 32.230377197265625, 42.626617431640625, 106.384033203125, -28.6075439453125, 53.51116943359375, 119.8328857421875, 56.78028869628906, 0.7006797790527344, 15.508575439453125, 206.74813842773438, 361.91949462890625, 172.80198669433594, -4.148826599121094, 120.86585998535156, 12.974884033203125, -118.84814453125, 47.849029541015625, 259.2166748046875, 63.278778076171875, 73.7618637084961, 24.47271728515625, 54.11712646484375, 22.601524353027344, -13.231040954589844, 147.61328125, 225.57862854003906, 181.44732666015625, 292.31622314453125, 46.1673583984375, 91.77386474609375, -11.061866760253906, 9.362968444824219, -97.58164978027344, 157.46209716796875, -12.685684204101562, 131.45663452148438, 92.60589599609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000282.npy"}
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 60.91192626953125, "std": 105.17179870605469, "min": -199.9971923828125, "p10": -49.857620239257805, "median": 37.06289291381836, "p90": 185.26731567382814, "max": 418.265869140625, "pos_frac": 0.703125, "sample": [-72.7647705078125, 66.18814086914062, 76.94596862792969, -2.58026123046875, 7.280736923217773, -79.98849487304688, 129.72433471679688, -31.37004852294922, 275.7105712890625, -63.97328186035156, -52.3455810546875, 162.37567138671875, 158.88209533691406, 10.613019943237305, -15.972034454345703, -26.661514282226562, -4.663177490234375, 170.30316162109375, 76.75927734375, -42.0174560546875, 418.265869140625, 267.4384765625, 79.42218017578125, -58.59918212890625, 263.04022216796875, 201.7374267578125, 179.67391967773438, -24.934722900390625, 128.27081298828125, 93.86734771728516, 156.87527465820312, -28.799606323242188, -37.36634826660156, 164.52597045898438, 55.053070068359375, 185.33349609375, 15.778121948242188, 3.385953903198242, 126.72616577148438, 9.230819702148438, -40.245361328125, 188.90383911132812, -41.375640869140625, 14.774650573730469, 21.61054229736328, 142.703857421875, 34.2137451171875, -56.260894775390625, 54.90022277832031, 184.0811767578125, 0.869964599609375, 158.46157836914062, 120.19219207763672, -15.406379699707031, 320.7393798828125, 0.6479949951171875, -9.485301971435547, 105.82073974609375, 30.126220703125, 135.28012084960938, 144.1597900390625, 247.301513671875, 194.22802734375, 113.44967651367188, 149.1612548828125, 1.409423828125, 144.84042358398438, 49.737884521484375, 97.61107635498047, 49.73985290527344, -33.15861511230469, 10.048843383789062, -55.979408264160156, 123.24148559570312, -48.791351318359375, 8.665771484375, 41.72663879394531, -10.681755065917969, 0.0, 10.51580810546875, -19.507652282714844, 22.089645385742188, 150.07070922851562, -3.71038818359375, 116.79165649414062, -7.24029541015625, 36.52284240722656, -45.23808288574219, -75.53985595703125, 50.240570068359375, 5.77447509765625, -199.9971923828125, 93.4395751953125, -43.0277099609375, 61.9766845703125, 13.31610107421875, 146.5736083984375, 105.21759033203125, -167.4180450439453, 97.15078735351562, 0.8470668792724609, 27.0181884765625, 416.24346923828125, 37.602943420410156, 114.36767578125, 2.9399032592773438, 81.64520263671875, -105.99462890625, 58.053070068359375, -60.877349853515625, 121.61410522460938, 20.37939453125, -10.759941101074219, 3.385953903198242, 0.0, 185.23895263671875, 194.9659881591797, 334.39361572265625, 150.08370971679688, 14.82497787475586, -78.17570495605469, 109.80302429199219, 98.79348754882812, 93.45651245117188, -23.771331787109375, 1.50726318359375, 99.5758056640625, 42.929779052734375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000283.npy"}
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 65.26899719238281, "std": 115.0948486328125, "min": -216.740478515625, "p10": -65.7642822265625, "median": 57.827850341796875, "p90": 222.20047607421873, "max": 379.7127380371094, "pos_frac": 0.71875, "sample": [-11.253921508789062, 36.00201416015625, -43.16380310058594, -192.33285522460938, 45.42791748046875, 40.222808837890625, 235.48681640625, 251.72039794921875, 3.38836669921875, 34.42924499511719, 17.565107345581055, 113.99725341796875, 82.8271255493164, -27.414817810058594, 216.37632751464844, 117.18011474609375, 89.353515625, 196.0523681640625, 248.80364990234375, 9.113622665405273, 127.93603515625, 379.7127380371094, -42.764312744140625, 116.37283325195312, -19.098491668701172, 87.1357650756836, 14.423934936523438, -144.54595947265625, 41.09871292114258, -43.718780517578125, -50.83604431152344, 208.5887451171875, 95.6046142578125, -113.9248046875, 49.073333740234375, 70.71817016601562, 152.071044921875, 163.677734375, 56.961151123046875, -109.65226745605469, -3.2242584228515625, 38.928466796875, 17.920867919921875, 69.228759765625, -51.059967041015625, 93.25222778320312, 5.3060150146484375, 337.211181640625, 142.59469604492188, 236.54190063476562, 58.694549560546875, -46.239044189453125, 205.27349853515625, 168.0601806640625, 198.19573974609375, 106.25637817382812, -64.93243408203125, -14.596317291259766, 225.22503662109375, 335.68597412109375, 95.21923828125, 148.34693908691406, 68.43035888671875, -28.268516540527344, 121.65228271484375, 96.3736572265625, 242.9049072265625, 54.91546630859375, 109.37971496582031, 15.447555541992188, 230.52780151367188, 18.216392517089844, -1.0388412475585938, -8.29443359375, 14.206672668457031, 296.17962646484375, 24.864044189453125, 12.436927795410156, 108.29348754882812, 58.916290283203125, 258.20758056640625, 138.54571533203125, 109.73755645751953, 193.9208984375, -151.45639038085938, -216.740478515625, 3.149974822998047, -71.54852294921875, 7.834747314453125, 23.885284423828125, 124.54191589355469, 29.01263427734375, 99.28314208984375, 146.71389770507812, -7.112499237060547, -19.499893188476562, 104.23870849609375, -73.46664428710938, 132.30831909179688, 62.91941833496094, 226.353759765625, -56.4383544921875, 135.023193359375, 108.06515502929688, 17.241893768310547, 101.06918334960938, -176.63943481445312, 35.22743225097656, -67.70526123046875, 133.2073974609375, 220.90423583984375, 171.65887451171875, 46.92314147949219, -23.065765380859375, -168.4302520751953, -54.45025634765625, 8.032333374023438, -114.2450180053711, -5.051242828369141, -24.16619873046875, 179.974853515625, 105.58253479003906, 171.27099609375, 198.46334838867188, 149.30233764648438, -36.603233337402344, -158.03262329101562, 96.83479309082031], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000284.npy"}
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 42.88224411010742, "std": 109.73712158203125, "min": -211.0850830078125, "p10": -78.31904296875, "median": 30.14611053466797, "p90": 196.88404083251953, "max": 358.7254638671875, "pos_frac": 0.6484375, "sample": [38.22918701171875, 69.02532958984375, 93.01034545898438, 22.88189697265625, 194.32699584960938, 147.12667846679688, 129.42108154296875, -64.17616271972656, 43.737823486328125, 48.387115478515625, 15.060394287109375, -148.2724609375, 52.93498229980469, -67.79928588867188, -50.9869384765625, 56.910186767578125, 14.715744018554688, 10.9842529296875, -98.9222412109375, 148.9805908203125, -35.046234130859375, 36.06201171875, -100.79954528808594, 155.7726287841797, 25.033447265625, 47.45426940917969, 293.4103088378906, 159.00643920898438, 65.63954162597656, 358.7254638671875, 249.9229736328125, -83.31527709960938, 21.5611572265625, -67.91412353515625, 240.57679748535156, 232.21788024902344, 109.49139404296875, -165.98455810546875, -74.9962158203125, -60.968994140625, -63.772735595703125, -81.7987060546875, -66.68984985351562, 17.797637939453125, -4.350671768188477, 67.51580810546875, 103.08056640625, 52.03753662109375, 25.079360961914062, 130.8165283203125, 229.29330444335938, -5.492740631103516, -18.229248046875, -30.243003845214844, 320.3113708496094, -5.62646484375, -52.219810485839844, -15.611248016357422, 195.7621307373047, 3.6121978759765625, -68.81980895996094, 267.21380615234375, 73.95309448242188, 75.15916442871094, 73.02618408203125, -111.86273193359375, 32.1068115234375, -9.986940383911133, -6.2076568603515625, 156.40338134765625, -53.843719482421875, 145.9134521484375, 55.89692687988281, 9.396026611328125, 99.3030776977539, 45.54624938964844, 38.0235595703125, -3.6215667724609375, 95.4326171875, 28.97998046875, 20.303146362304688, 120.20883178710938, 41.80609130859375, 54.22984313964844, 16.770782470703125, 123.57952880859375, 126.33944702148438, 168.8878173828125, 296.3947448730469, 97.61958312988281, 49.12146759033203, 75.16387939453125, 3.3820419311523438, -176.99969482421875, -59.930389404296875, 52.668304443359375, 21.806396484375, 226.1014404296875, -28.899871826171875, 22.15887451171875, 141.3982696533203, -9.635765075683594, 31.312240600585938, -54.97406005859375, -211.0850830078125, -124.37936401367188, 0.4544677734375, -22.221965789794922, 199.5018310546875, 53.119384765625, -125.34002685546875, -39.61539077758789, 191.4530029296875, -137.27357482910156, -23.9248046875, -172.77914428710938, 19.243053436279297, 41.07939147949219, 248.638671875, -24.02264404296875, 24.2762451171875, -14.126861572265625, -16.829071044921875, 86.02459716796875, -76.8277587890625, 128.67547607421875, 105.19866943359375, 216.16632080078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000285.npy"}
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 63.524391174316406, "std": 115.18657684326172, "min": -211.97705078125, "p10": -69.25048828124999, "median": 46.82298278808594, "p90": 192.5287017822265, "max": 495.22772216796875, "pos_frac": 0.71875, "sample": [-12.838134765625, 70.6236572265625, 69.03399658203125, 54.222625732421875, 37.22697448730469, 118.82830047607422, 29.56195068359375, 10.087348937988281, 23.915771484375, 109.51358032226562, 149.4390869140625, 62.560546875, 49.274742126464844, 175.20611572265625, -114.22163391113281, 54.8975830078125, 68.64759063720703, 163.99185180664062, 150.3179931640625, 169.77374267578125, -50.75494384765625, 104.83255004882812, 89.83084106445312, 42.282867431640625, -81.15587615966797, 59.765777587890625, 46.155975341796875, -76.2923583984375, 83.5464096069336, 9.482536315917969, -168.77099609375, 47.489990234375, 226.87747192382812, 206.04837036132812, -15.23309326171875, 110.42760467529297, 115.75540161132812, -7.29327392578125, 28.333534240722656, 140.6028594970703, -24.315643310546875, 47.76585388183594, 173.984619140625, -132.6059112548828, 122.3275146484375, -66.2325439453125, 90.39450073242188, 438.92388916015625, 7.41595458984375, -6.72515869140625, 100.12551879882812, -93.27828979492188, 129.74880981445312, 118.9418716430664, 32.73028564453125, 150.6060791015625, -40.93376159667969, 366.9538879394531, -40.561912536621094, 223.33531188964844, -80.00921630859375, 0.0, -13.24072265625, -94.01591491699219, 7.979133605957031, 98.34280395507812, -16.317031860351562, -19.293212890625, 20.784622192382812, 31.23809051513672, -115.21356201171875, 85.34368896484375, -21.452392578125, 121.01528930664062, 1.2406654357910156, 0.806427001953125, 238.47906494140625, 51.82550811767578, -29.64185905456543, 30.606781005859375, 112.5975341796875, -6.961559295654297, 2.2901039123535156, 235.55523681640625, 237.73597717285156, 83.71891784667969, -3.0629425048828125, 16.349594116210938, 163.93507385253906, -123.6495361328125, 86.56095123291016, 170.4481201171875, 115.4898681640625, 246.9844207763672, 186.73455810546875, 167.32501220703125, 44.826927185058594, 3.127910614013672, 82.21377563476562, 85.32516479492188, 495.22772216796875, 40.384552001953125, 166.0858154296875, 6.79998779296875, -5.946624755859375, -108.42425537109375, 152.02737426757812, 116.08261108398438, 13.95880126953125, 19.810657501220703, -211.97705078125, 281.47882080078125, -1.010833740234375, -18.1534481048584, -30.345672607421875, 414.00372314453125, -161.339599609375, 31.76519775390625, 58.330108642578125, 15.620849609375, 167.2572479248047, 214.2097930908203, -6.385915756225586, 6.4917755126953125, 175.7969970703125, -21.754531860351562, 128.91738891601562, 35.591102600097656], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000286.npy"}
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 61.376522064208984, "std": 112.17217254638672, "min": -228.95277404785156, "p10": -71.17091369628906, "median": 46.75318908691406, "p90": 214.01882324218747, "max": 319.10809326171875, "pos_frac": 0.6875, "sample": [141.40969848632812, -3.479206085205078, -70.82296752929688, -1.3506355285644531, -77.06025695800781, 21.897048950195312, 200.51223754882812, 180.28060913085938, 207.03636169433594, 144.41275024414062, 99.30169677734375, 7.1955718994140625, 18.186599731445312, 111.41190338134766, 89.45697021484375, 16.561279296875, -146.23782348632812, -11.43988037109375, 73.51643371582031, 47.16839599609375, -19.16693115234375, 231.61239624023438, 301.9088134765625, -48.215179443359375, 211.28189086914062, -228.95277404785156, -6.21600341796875, 150.309326171875, 77.02410888671875, 66.06243896484375, 25.40985870361328, 319.10809326171875, 10.161476135253906, 227.26513671875, -55.684417724609375, -46.98976135253906, -38.125152587890625, 165.13726806640625, 10.863203048706055, 171.1488494873047, 149.812744140625, 101.59555053710938, -16.206268310546875, 132.09747314453125, 251.10214233398438, -12.274463653564453, 182.85025024414062, 118.45906066894531, 185.89453125, 119.09185791015625, -11.4666748046875, 238.81797790527344, 243.80404663085938, 123.59056854248047, 74.6456298828125, 65.00543212890625, 42.801910400390625, 59.234405517578125, 220.40499877929688, -71.9827880859375, 170.63584899902344, -49.5826416015625, 201.47799682617188, 11.654916763305664, 17.873390197753906, 70.92193603515625, 162.60763549804688, 272.7462463378906, 126.1522216796875, 151.84872436523438, -150.01394653320312, 2.9501419067382812, 144.30474853515625, 162.06219482421875, 63.58489990234375, 41.562591552734375, 26.4197998046875, 197.40579223632812, -22.2342529296875, 29.68082046508789, 147.11178588867188, -34.69189453125, -17.208145141601562, 24.76696014404297, -103.22608947753906, 15.130636215209961, -81.79521179199219, 164.15277099609375, -124.93972778320312, 123.10533142089844, 259.1929931640625, 150.28744506835938, -199.1680908203125, -65.32528686523438, -39.892486572265625, -5.348907470703125, 25.418750762939453, 5.573619842529297, -12.979827880859375, 99.06219482421875, -45.439453125, 65.66552734375, 46.337982177734375, 73.8995361328125, -162.80401611328125, -35.16888427734375, 165.07322692871094, -89.5596923828125, 249.0419921875, 7.896400451660156, 166.35794067382812, -2.06982421875, 118.835205078125, 222.39370727539062, 240.2252655029297, -19.45330810546875, 34.98710632324219, 119.28350830078125, 107.47052001953125, 6.965953826904297, 2.6100635528564453, 24.671329498291016, -149.79547119140625, -135.58099365234375, -38.05413818359375, 106.29747009277344, -45.73860168457031, 91.38288879394531], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000287.npy"}
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 54.61041259765625, "std": 110.98861694335938, "min": -283.0112609863281, "p10": -58.76816864013671, "median": 34.1453742980957, "p90": 210.79364013671875, "max": 343.25262451171875, "pos_frac": 0.703125, "sample": [107.50958251953125, -110.67912292480469, -51.247406005859375, 27.465606689453125, 115.4239501953125, 27.89874267578125, 90.34379577636719, -166.1087646484375, 24.187828063964844, 129.25497436523438, 155.840087890625, 228.49835205078125, -13.98345947265625, 127.77951049804688, 135.09408569335938, -67.76905822753906, 18.440460205078125, 127.49468994140625, 5.755126953125, 0.0, -2.9783935546875, -9.940673828125, 13.253936767578125, 36.79380798339844, 21.03655242919922, -6.6334228515625, 10.89862060546875, -94.0767822265625, -219.06912231445312, 127.8792724609375, 259.8336486816406, 192.04446411132812, 29.29767608642578, 44.228271484375, 25.832733154296875, 0.0, -163.2789306640625, 12.092720031738281, 136.80517578125, 57.291351318359375, -42.58024597167969, 194.45443725585938, -95.30838012695312, -77.18574523925781, 1.2900218963623047, 189.4010467529297, 211.3785400390625, 114.74456787109375, -205.22467041015625, -1.0274810791015625, 208.96615600585938, -51.93419647216797, 124.25042724609375, -7.202392578125, 93.18135070800781, -113.102294921875, -1.8879966735839844, 43.366455078125, 67.65983581542969, -53.078369140625, 343.25262451171875, 40.43690490722656, 57.69287109375, -47.752166748046875, 214.14544677734375, 20.800323486328125, -21.576499938964844, -283.0112609863281, 242.31948852539062, 32.643402099609375, 253.5609130859375, -18.574188232421875, -19.50592041015625, 82.60075378417969, -54.91064453125, -16.81488037109375, 23.3565673828125, 21.210494995117188, -76.75, 35.64734649658203, -52.038818359375, 240.0625, 7.159759521484375, -23.279083251953125, -35.45536804199219, 4.4794158935546875, 65.70925903320312, 322.70123291015625, 198.5354766845703, 90.84219360351562, 160.368408203125, 46.359283447265625, 142.97637939453125, -2.977153778076172, 156.46908569335938, 39.709999084472656, 24.112876892089844, -96.12879943847656, 328.123779296875, 20.1697998046875, 44.47355651855469, 28.1790771484375, 16.57861328125, 281.6796569824219, 32.29357147216797, 51.326416015625, 23.789396286010742, -28.827911376953125, 59.856292724609375, 7.7349853515625, -21.516876220703125, 172.0933837890625, 103.83990478515625, 39.59808349609375, 105.89456176757812, 76.11752319335938, 55.585205078125, 46.82012939453125, 147.38934326171875, 159.89456176757812, 19.62139129638672, 210.54296875, 238.29090881347656, 153.19546508789062, 221.7506866455078, 100.30722045898438, 145.34844970703125, 48.93382263183594], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000288.npy"}
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 40.09748458862305, "std": 102.90096282958984, "min": -210.30117797851562, "p10": -88.52638549804686, "median": 29.27341079711914, "p90": 185.46477813720702, "max": 302.7764892578125, "pos_frac": 0.6640625, "sample": [-24.814453125, -8.97772216796875, 19.07208251953125, 35.88892364501953, -56.6322021484375, 13.638946533203125, 36.91839599609375, 17.184356689453125, 83.70208740234375, -87.343505859375, 207.37353515625, -64.38909912109375, 9.442655563354492, 39.64227294921875, 57.261016845703125, 26.96368408203125, -210.30117797851562, 140.362060546875, 45.418060302734375, 22.28131103515625, 50.5198974609375, 162.13916015625, 137.44544982910156, 43.9134521484375, 97.9556884765625, -60.9310302734375, -139.65762329101562, 5.617912292480469, -3.18487548828125, 12.8515625, -81.21749877929688, 195.3826904296875, -57.616607666015625, 108.45527648925781, 13.448463439941406, 0.0, 27.44892120361328, 117.102294921875, 94.01358032226562, 62.54010009765625, 50.05797576904297, -68.3519287109375, -154.51480102539062, -75.845947265625, 117.10723876953125, -66.27157592773438, 185.2855224609375, 31.097900390625, 67.83595275878906, 15.532928466796875, -121.14767456054688, 205.57904052734375, -205.49496459960938, -139.94989013671875, 34.33235168457031, 1.3051490783691406, -32.51142120361328, 20.20291519165039, -16.655393600463867, 180.99758911132812, -91.28643798828125, 185.9233856201172, 22.056739807128906, 82.65678405761719, 70.6556396484375, -6.9910430908203125, -11.457626342773438, -52.42888641357422, -66.50636291503906, 10.625358581542969, 253.1866912841797, -92.1236572265625, 69.78298950195312, 152.2582550048828, 43.80541229248047, 54.91534423828125, 8.584449768066406, 302.7764892578125, 45.411407470703125, 124.458251953125, 7.548736572265625, -48.9381103515625, -93.03652954101562, 220.94406127929688, -58.17881774902344, 24.120742797851562, -39.77610778808594, 219.1819305419922, -43.292236328125, 164.67669677734375, -33.24798583984375, 185.88304138183594, 199.04417419433594, -5.639892578125, 90.7239990234375, 84.97848510742188, 108.623779296875, 1.568817138671875, 147.83230590820312, 86.93241882324219, 165.85958862304688, 191.66366577148438, 63.52679443359375, 139.75335693359375, -105.06622314453125, 115.13557434082031, 43.4615478515625, -92.58050537109375, 14.10418701171875, 288.81878662109375, -164.4661865234375, 155.4608154296875, 91.02630615234375, -15.70196533203125, 230.02035522460938, -114.68524169921875, -51.56451416015625, 128.32896423339844, -2.8278121948242188, 155.14645385742188, 41.71820068359375, 61.42266845703125, 168.9825439453125, -53.85295104980469, 42.87944030761719, 25.10938262939453, -15.8861083984375, 182.92864990234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000289.npy"}
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 60.12238311767578, "std": 117.53836059570312, "min": -293.0514831542969, "p10": -83.02414093017576, "median": 56.69583702087402, "p90": 213.10021362304687, "max": 404.9365234375, "pos_frac": 0.734375, "sample": [287.19512939453125, 11.14764404296875, 6.12823486328125, 250.00901794433594, 75.29296875, -92.89926147460938, 14.703384399414062, 151.5784912109375, -10.18609619140625, 58.153350830078125, 312.25616455078125, 167.9261474609375, 101.0841064453125, -52.04034423828125, 186.96072387695312, 93.52920532226562, 68.75401306152344, 58.900146484375, 270.65045166015625, 118.765869140625, 141.31283569335938, 3.4095840454101562, 35.57476806640625, 99.59957885742188, 281.76776123046875, -131.47271728515625, -97.17628479003906, -188.7720947265625, -9.083755493164062, 4.078948974609375, 404.9365234375, 8.161529541015625, 108.363525390625, 102.333740234375, 213.25885009765625, 355.041748046875, 170.23733520507812, 16.803295135498047, -18.4019775390625, -182.5833740234375, 137.49505615234375, 56.87046813964844, 3.65185546875, 73.08694458007812, 57.22998046875, 4.107536315917969, 6.13623046875, -56.671844482421875, -13.373636245727539, 68.9578857421875, -124.2677001953125, 129.41522216796875, 101.15142822265625, 5.292755126953125, 144.34768676757812, -49.96978759765625, -68.57291412353516, 132.94622802734375, 42.296142578125, 46.04194641113281, 82.73019409179688, 24.00421905517578, -94.43769836425781, 159.15399169921875, 133.67620849609375, 164.081298828125, 16.720848083496094, 66.1220703125, 75.5654296875, 258.7080078125, -72.91790771484375, 41.561187744140625, -19.521331787109375, 68.50701904296875, -16.266082763671875, 243.1717529296875, 10.357805252075195, -112.86627197265625, 7.14996337890625, 177.20120239257812, -14.95977783203125, -13.790077209472656, -78.79194641113281, 194.41639709472656, 55.83681869506836, 129.71347045898438, 116.65222930908203, 61.28448486328125, -205.3220672607422, 61.61763000488281, 149.1427459716797, -55.70896911621094, 45.90045166015625, 26.074981689453125, -42.819793701171875, 4.674140930175781, -113.41229248046875, -16.690536499023438, 29.057220458984375, 189.06948852539062, 56.52120590209961, -113.6313247680664, 105.30740356445312, 229.1181640625, 296.255126953125, 72.94351196289062, 278.23504638671875, 157.311279296875, 213.0322265625, 53.24925231933594, -70.24578857421875, 142.80282592773438, 192.0941619873047, 84.71780395507812, -114.25845336914062, 1.202606201171875, -11.662895202636719, 111.83868408203125, -293.0514831542969, 156.5247344970703, -24.45074462890625, 51.40531921386719, 65.95751953125, 22.629079818725586, -10.628463745117188, 35.83587646484375, 110.71124267578125, 69.78411865234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000290.npy"}
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 79.79718017578125, "std": 110.80087280273438, "min": -252.3477783203125, "p10": -57.897125244140625, "median": 69.41683959960938, "p90": 234.4188690185547, "max": 393.0818176269531, "pos_frac": 0.75, "sample": [262.4884033203125, -12.090648651123047, 267.6871643066406, 43.01422119140625, 185.35186767578125, 79.18531799316406, 166.88302612304688, 241.86668395996094, 38.5657958984375, 69.61407470703125, 46.246063232421875, 21.377334594726562, 46.72758483886719, -57.775177001953125, 233.18624877929688, 20.214828491210938, 359.9037170410156, -5.337432861328125, 136.84979248046875, 181.65689086914062, 116.8060302734375, 103.91070556640625, 87.15045166015625, -21.888397216796875, -95.45341491699219, 156.67056274414062, 13.500091552734375, -11.773567199707031, 128.63534545898438, -64.0872802734375, -86.71766662597656, 24.472496032714844, -58.181671142578125, 20.8446044921875, -252.3477783203125, 173.8824462890625, 117.98651123046875, 32.27166748046875, 57.62408447265625, 59.74137878417969, 197.255126953125, -42.61680603027344, 0.13507080078125, -56.49275207519531, 291.0910949707031, 111.34957885742188, 1.5180740356445312, 156.53378295898438, 71.29461669921875, -29.991195678710938, 131.47915649414062, 125.02210998535156, 136.4336700439453, -42.868133544921875, 175.35626220703125, 251.29246520996094, -68.0377197265625, 17.304931640625, -34.701141357421875, 122.41268920898438, 183.8895263671875, 88.224365234375, 68.18218994140625, -73.74700927734375, 189.22145080566406, -81.0313720703125, 48.45460510253906, 51.14610290527344, -80.31320190429688, 68.54627990722656, -71.98171997070312, 179.87051391601562, 9.052347183227539, 266.3245849609375, 69.2196044921875, 181.12646484375, 30.255165100097656, 216.607666015625, 132.0600128173828, -36.93450927734375, 149.09483337402344, 4.494804382324219, 102.51068115234375, -44.48297119140625, 51.25103759765625, 246.24716186523438, 81.52738189697266, -42.26666259765625, 0.0, -0.9906463623046875, 60.90565490722656, 166.51495361328125, 99.98516845703125, -10.158096313476562, 239.54098510742188, 32.35114288330078, 171.25921630859375, 114.43344116210938, 177.2684326171875, -73.92190551757812, 71.76812744140625, 393.0818176269531, 70.39013671875, 251.27001953125, 11.49261474609375, 149.87347412109375, 22.513702392578125, 73.0455322265625, 342.284912109375, -19.676025390625, -115.42425537109375, 150.360107421875, 143.27279663085938, 23.29104232788086, 163.53135681152344, -0.355438232421875, 62.84295654296875, 0.7375049591064453, 219.9827880859375, 199.66326904296875, 102.82156372070312, 108.84532165527344, -105.22647094726562, -6.621004104614258, 204.21983337402344, 147.42007446289062, 5.1669921875, 237.29498291015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000291.npy"}
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 58.60597229003906, "std": 111.57447052001953, "min": -274.4296875, "p10": -50.74673156738281, "median": 47.01422882080078, "p90": 195.58793182373046, "max": 411.031982421875, "pos_frac": 0.7421875, "sample": [26.095703125, 37.189422607421875, 36.68791198730469, -7.246826171875, 27.969736099243164, 29.227798461914062, 130.58673095703125, 307.35687255859375, 58.03889465332031, 187.59979248046875, 372.53802490234375, 93.2210693359375, -42.414306640625, -51.953338623046875, 97.38937377929688, 69.3533935546875, -8.131834030151367, 17.19703483581543, 23.964996337890625, -2.3560791015625, 116.93515014648438, -19.731094360351562, 62.032501220703125, 86.93057250976562, 181.96368408203125, 93.26873779296875, 93.70846557617188, 149.89712524414062, 53.211456298828125, 4.484851837158203, 256.2976379394531, -0.21170806884765625, 13.576257705688477, 47.10536193847656, 172.54415893554688, 38.30567169189453, 64.89764404296875, 23.55474853515625, -250.00640869140625, 50.88298797607422, 113.55130004882812, -89.72706604003906, 177.86239624023438, 35.223907470703125, -27.6788330078125, 170.32383728027344, 7.211952209472656, -165.54476928710938, 45.43120574951172, 205.36968994140625, -92.1624755859375, 215.7547607421875, -11.0048828125, -4.94110107421875, -31.227142333984375, 47.84684753417969, 26.143310546875, 32.955413818359375, 55.909423828125, -2.9893951416015625, 148.6583251953125, 64.21221923828125, 60.453948974609375, -73.89678955078125, 48.08111572265625, 2.1881561279296875, 141.37542724609375, 96.23388671875, 411.031982421875, 163.09381103515625, 21.775390625, 75.18536376953125, -50.016021728515625, 51.2822265625, -4.639713287353516, 360.2213134765625, 18.401885986328125, 155.57369995117188, 56.54077911376953, -28.663766860961914, -96.73895263671875, -216.23394775390625, 119.206787109375, 208.58074951171875, 195.991455078125, 39.97584533691406, 72.76220703125, 36.874542236328125, 23.092649459838867, 198.774169921875, -50.2296142578125, -18.87451171875, 40.082149505615234, 51.03741455078125, -114.23599243164062, -56.50872039794922, -5.671283721923828, -23.942901611328125, 16.557188034057617, 74.37066650390625, 84.65798950195312, 39.671630859375, 42.78996276855469, 19.06266212463379, 140.96240234375, 96.35614013671875, 162.56759643554688, 4.4290008544921875, 60.67549133300781, 46.923095703125, 22.404998779296875, 300.9205322265625, 99.88632202148438, -16.061492919921875, 191.50408935546875, 147.7261962890625, 204.91204833984375, -178.328369140625, 93.23764038085938, 13.689590454101562, 238.43307495117188, -274.4296875, -158.74862670898438, 119.78805541992188, 195.4149932861328, 158.1925811767578, 95.75347900390625, -13.060480117797852], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000292.npy"}
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 71.85774230957031, "std": 105.71854400634766, "min": -173.03338623046875, "p10": -57.92851181030274, "median": 60.01862335205078, "p90": 227.58115234374998, "max": 341.2639465332031, "pos_frac": 0.7265625, "sample": [189.7992401123047, 38.618141174316406, 236.57794189453125, 125.0487060546875, 59.435577392578125, -173.03338623046875, 41.969818115234375, 111.5909423828125, 86.08839416503906, 71.23126220703125, 217.08554077148438, 246.67630004882812, 169.9764862060547, -49.336883544921875, 9.61920166015625, 60.03480529785156, 73.04484558105469, -2.62628173828125, 16.417861938476562, 115.17752075195312, 27.11370849609375, -71.007568359375, 332.51800537109375, 117.79168701171875, 0.6569061279296875, 0.0, 234.51031494140625, 79.78044128417969, 125.498779296875, 61.693138122558594, 109.62948608398438, -44.9478759765625, -59.03205871582031, -28.311065673828125, 232.3232421875, -63.93049621582031, -27.500244140625, 197.3173065185547, 110.72866821289062, 23.57415771484375, 105.7343978881836, -7.069591522216797, 143.39031982421875, 117.71990966796875, 17.1707763671875, -142.4937744140625, 225.548828125, 92.78682708740234, 341.2639465332031, 19.64916229248047, 37.60353088378906, 117.19061279296875, 107.11485290527344, 96.77447509765625, -68.97048950195312, 233.61578369140625, 14.634521484375, -7.27294921875, -18.824256896972656, 60.9488525390625, -17.56664276123047, 256.6635437011719, 130.60507202148438, -29.930076599121094, -95.20986938476562, -78.32818603515625, 13.13125228881836, -138.9866180419922, 128.5645751953125, -0.648956298828125, -84.73072814941406, 39.89390563964844, 198.6809539794922, 161.76043701171875, -18.434555053710938, 50.92924499511719, 80.985107421875, 159.182861328125, -13.929779052734375, 49.4659423828125, 33.154083251953125, 124.25164794921875, 126.17465209960938, 73.78231048583984, 118.75848388671875, 47.012451171875, 28.630279541015625, 175.6220703125, 197.64434814453125, -58.04041290283203, -2.7265625, 292.3731689453125, 60.00244140625, 68.3353271484375, 116.37533569335938, 148.6062774658203, 131.20457458496094, -10.712043762207031, 36.187713623046875, -5.483865737915039, 179.7257537841797, 324.85955810546875, 249.9141387939453, 224.819580078125, 157.38818359375, 204.84878540039062, 24.528411865234375, 99.42192077636719, -76.68333435058594, 0.8701362609863281, -55.42304992675781, -57.88055419921875, 0.184814453125, 159.99481201171875, 287.865234375, 234.07440185546875, 29.07220458984375, 28.576187133789062, 130.29078674316406, -111.29669952392578, -11.495208740234375, -44.04766845703125, 25.886940002441406, 19.681900024414062, -1.2421760559082031, 12.250762939453125, 103.17768859863281, 76.8590087890625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000293.npy"}
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 64.56130981445312, "std": 116.92916870117188, "min": -202.20501708984375, "p10": -55.634848022460936, "median": 47.09197998046875, "p90": 221.05806884765624, "max": 533.618408203125, "pos_frac": 0.703125, "sample": [163.50643920898438, 140.00732421875, 95.52200317382812, -16.180511474609375, 25.20196533203125, 25.33094024658203, 0.7785911560058594, -5.87030029296875, 35.84235382080078, 230.36387634277344, 75.61944580078125, 103.61260986328125, 94.08990478515625, -100.0316162109375, 49.40057373046875, 13.037841796875, 155.62257385253906, 119.23300170898438, 216.8524169921875, 60.0218505859375, 81.50064086914062, 148.77499389648438, 23.720306396484375, -26.665252685546875, -6.399993896484375, 220.6937255859375, -9.893386840820312, -58.396331787109375, 57.65582275390625, 9.457740783691406, 284.6385498046875, -33.753631591796875, 22.79754638671875, -81.0926513671875, -12.318313598632812, 159.37838745117188, -162.0369873046875, -169.97152709960938, 192.72933959960938, -28.5106201171875, 40.752471923828125, 215.97122192382812, -29.571380615234375, 90.97211456298828, 225.09481811523438, 255.90252685546875, 145.90501403808594, 51.171173095703125, 9.412422180175781, 83.28851318359375, 216.79595947265625, 3.8461227416992188, 256.7022705078125, 533.618408203125, 52.28907012939453, 362.89813232421875, 77.01934814453125, -48.139129638671875, 37.9285888671875, 2.3534774780273438, 27.49383544921875, 69.45693969726562, 24.363067626953125, 234.30807495117188, -202.20501708984375, 94.78117370605469, 161.31948852539062, 107.93830871582031, 23.689682006835938, -12.70965576171875, 28.75958251953125, 309.57830810546875, 139.5841064453125, 0.0, 200.9105224609375, 229.00054931640625, -24.570953369140625, 1.015350341796875, 51.35567855834961, 90.66522216796875, -5.751605987548828, 80.09307861328125, -3.743692398071289, -195.70228576660156, -37.127197265625, 221.908203125, -18.877105712890625, 39.23816680908203, 94.70294189453125, -54.45135498046875, 44.78338623046875, 50.82720947265625, -63.46990966796875, -98.03976440429688, -73.13634490966797, 120.7244873046875, 7.5936279296875, 5.13360595703125, 143.49066162109375, -71.10395812988281, 59.849090576171875, 99.95291900634766, 213.44873046875, 13.481477737426758, 65.63975524902344, 167.3598175048828, -187.862548828125, -44.70697021484375, 281.13604736328125, -3.594146728515625, 19.698402404785156, 278.3001403808594, 153.2503662109375, 164.531494140625, 126.30422973632812, 49.73594665527344, -9.499588012695312, -0.06258392333984375, 128.28823852539062, -33.12665557861328, 93.96234130859375, -38.321044921875, 13.643661499023438, -7.957845687866211, 164.0693359375, -128.87509155273438, 16.10761260986328, 188.7866973876953], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000294.npy"}
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 60.28120803833008, "std": 104.30779266357422, "min": -194.4112548828125, "p10": -57.99956130981445, "median": 38.31829833984375, "p90": 199.2947296142578, "max": 378.27783203125, "pos_frac": 0.7109375, "sample": [197.9681396484375, 96.05435180664062, -149.41839599609375, 16.081787109375, -8.01513671875, 35.351104736328125, -5.737205505371094, 142.86764526367188, -91.25265502929688, -101.67123413085938, 5.491630554199219, -9.486061096191406, 204.81988525390625, 19.4053955078125, -24.48345947265625, 65.67720794677734, -175.78555297851562, 45.56219482421875, -98.95358276367188, 84.7107925415039, 51.343017578125, 207.19769287109375, -86.28759765625, -194.4112548828125, 11.405609130859375, 37.897674560546875, 39.052978515625, -69.11244201660156, -55.667633056640625, 60.31697082519531, 23.05157470703125, 87.23744201660156, 175.255615234375, 184.1847686767578, 164.8126220703125, -0.247711181640625, 38.738922119140625, 53.227447509765625, 191.73593139648438, 96.33529663085938, 240.783935546875, -23.205917358398438, 133.69525146484375, 114.32916259765625, -16.20947265625, 75.70675659179688, 9.719947814941406, 214.01528930664062, 186.05426025390625, 105.5850830078125, -78.62577819824219, -57.57666778564453, 20.9892578125, 378.27783203125, 94.1993408203125, 145.59600830078125, -86.88841247558594, 115.37055206298828, 96.9461669921875, 171.25433349609375, 70.42779541015625, -10.549591064453125, -23.52606201171875, 27.25597381591797, 202.39010620117188, -36.78766632080078, 15.898208618164062, 69.50070190429688, 12.190765380859375, 22.065948486328125, 14.259574890136719, 12.22161865234375, 2.0806350708007812, 208.0784912109375, -42.020233154296875, 140.97865295410156, 164.72079467773438, 71.8122787475586, 17.9678955078125, 137.09326171875, 43.82276916503906, 150.0723114013672, -41.026893615722656, 5.336284637451172, 114.04689025878906, 52.43707275390625, 0.0, 137.93087768554688, 33.01637268066406, 13.387664794921875, 305.1361083984375, 210.26593017578125, 143.614990234375, -45.072265625, 90.65264892578125, -22.722518920898438, -20.49395751953125, 211.796875, 11.208168029785156, 186.14576721191406, 113.98432922363281, 10.539321899414062, 5.152717590332031, -25.1207275390625, 154.52261352539062, -52.467796325683594, 125.61579132080078, 14.4178466796875, -29.391372680664062, -58.98631286621094, -38.614959716796875, 282.7199401855469, 197.17718505859375, 107.90692901611328, 79.53184509277344, 285.3507080078125, 21.7474365234375, -14.345001220703125, 32.049560546875, -111.8985595703125, -70.6063232421875, -17.67401123046875, 166.55490112304688, 164.77093505859375, 130.34768676757812, 284.7310791015625, 30.998748779296875, 166.09564208984375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000295.npy"}
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 81.817138671875, "std": 120.91117095947266, "min": -256.9273376464844, "p10": -66.41215515136719, "median": 79.8573989868164, "p90": 222.56273193359374, "max": 366.706787109375, "pos_frac": 0.7578125, "sample": [31.0562744140625, 17.128870010375977, 63.12274169921875, 17.48114013671875, 175.91705322265625, -11.143218994140625, -19.35064697265625, -15.287981033325195, 21.760467529296875, 6.888481140136719, 131.608154296875, -135.44244384765625, 160.26400756835938, 162.4630126953125, 9.74578857421875, 123.96017456054688, 77.29263305664062, 158.47216796875, 188.3787384033203, -82.58758544921875, 366.706787109375, 74.251220703125, 53.51513671875, 182.7582244873047, 72.1715087890625, 145.74929809570312, 3.422710418701172, 58.612220764160156, 221.19052124023438, -67.54147338867188, 25.048110961914062, -4.190971374511719, 233.69061279296875, -11.899169921875, -105.84686279296875, 190.50054931640625, 139.83978271484375, -65.34440612792969, 180.1025848388672, 123.34467315673828, -155.463623046875, 80.685302734375, 209.49505615234375, 153.66616821289062, 23.5772705078125, -70.89495849609375, 167.60614013671875, -152.65235900878906, 213.86602783203125, 52.617401123046875, 356.6983642578125, 119.47090148925781, -65.92816162109375, 165.09661865234375, 336.9473876953125, 176.10116577148438, 10.768806457519531, 246.55581665039062, -39.333038330078125, 216.58355712890625, -206.76727294921875, 188.9379425048828, 83.26739501953125, -135.9356689453125, 259.84722900390625, 82.19798278808594, 135.88812255859375, 219.34765625, -164.99075317382812, 137.6475830078125, 209.73123168945312, 43.86235046386719, -161.31283569335938, 91.69645690917969, 128.15069580078125, 89.02659606933594, 249.40322875976562, 288.000244140625, -25.963973999023438, 71.2552490234375, 160.6376953125, 264.656005859375, 79.02949523925781, -20.176666259765625, 238.4254150390625, 73.80126953125, -36.84144592285156, -256.9273376464844, 77.449462890625, -20.338836669921875, 150.44052124023438, 127.20291137695312, 188.64794921875, -55.859222412109375, 189.30056762695312, 0.0, 57.048126220703125, 53.845550537109375, 131.82998657226562, 193.981689453125, 67.92645263671875, 72.70391845703125, 260.17315673828125, 214.32968139648438, 90.3127670288086, 31.75823974609375, 39.337066650390625, 9.0423583984375, -40.6251220703125, 158.19873046875, 185.18362426757812, 86.4295425415039, 23.627382278442383, -52.400054931640625, -10.421646118164062, 169.61026000976562, 143.24267578125, 25.615493774414062, 24.97161865234375, 222.77340698242188, 10.462776184082031, 103.1661376953125, 88.05712890625, 222.47244262695312, 114.26737976074219, 281.47625732421875, -143.252197265625, -48.56184387207031], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000296.npy"}
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 61.278297424316406, "std": 105.45077514648438, "min": -181.80264282226562, "p10": -50.94264526367187, "median": 38.3953800201416, "p90": 205.8287780761719, "max": 345.2022399902344, "pos_frac": 0.6953125, "sample": [305.0715637207031, 7.9718475341796875, 173.73654174804688, 174.892822265625, 37.7066650390625, 128.63201904296875, 248.0799560546875, 233.62069702148438, 203.07867431640625, -60.87994384765625, -44.39839172363281, -11.643463134765625, -31.97193145751953, 9.064117431640625, 222.086181640625, -124.86396026611328, -12.138925552368164, 42.03106689453125, 301.07965087890625, 168.84823608398438, 12.3349609375, 65.46235656738281, -70.76549530029297, 88.8524169921875, 38.37969970703125, 11.19830322265625, -54.6217041015625, -181.80264282226562, 10.6207275390625, -3.1695480346679688, -18.4793701171875, 144.27952575683594, 136.30685424804688, -7.772743225097656, -114.78451538085938, -5.28851318359375, 166.34469604492188, 69.31991577148438, 194.93804931640625, -88.8536376953125, 98.92804718017578, 13.811813354492188, 105.71516418457031, 25.460052490234375, 4.0377197265625, -23.451690673828125, -106.37173461914062, 85.73098754882812, -138.95843505859375, 182.24554443359375, -35.93269348144531, -29.73846435546875, -3.087371826171875, 78.34783935546875, 15.263427734375, 214.82550048828125, 38.41106033325195, 23.79680633544922, -15.920143127441406, -169.64334106445312, 28.96593475341797, 59.688995361328125, -143.266845703125, -6.2866668701171875, 55.977325439453125, 93.6749267578125, -7.941009521484375, 120.80029296875, 267.82916259765625, 31.098281860351562, 197.01214599609375, 156.11715698242188, 164.66879272460938, 179.19189453125, 231.85955810546875, -96.7994384765625, 31.140453338623047, 21.951675415039062, 41.63427734375, -31.039976119995117, -98.26876831054688, 79.95761108398438, 131.31503295898438, -12.301239013671875, 61.64117431640625, 72.73049926757812, 49.47418212890625, -0.49273681640625, 181.59786987304688, 69.26028442382812, 130.35211181640625, 25.721343994140625, -43.42670440673828, 135.9426727294922, 15.42364501953125, 156.3630828857422, 42.32157897949219, 142.433837890625, 57.111045837402344, 178.30084228515625, 205.56576538085938, 128.88250732421875, 37.6605224609375, 94.58772277832031, -23.997947692871094, 45.03546142578125, 32.088958740234375, 3.58984375, 61.166038513183594, 206.44247436523438, 237.0556640625, 264.2056884765625, -35.16670227050781, -29.255706787109375, 2.2659912109375, -28.12377166748047, -49.36590576171875, 131.592041015625, 168.6544952392578, 28.45947265625, 32.21661376953125, 35.625213623046875, 226.04083251953125, -11.007476806640625, -6.6571044921875, 114.58810424804688, 128.56370544433594, 345.2022399902344], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000297.npy"}
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 69.17764282226562, "std": 120.33387756347656, "min": -267.3484802246094, "p10": -63.147476196289034, "median": 44.81526184082031, "p90": 230.0312530517578, "max": 352.009033203125, "pos_frac": 0.703125, "sample": [141.08985900878906, 174.12063598632812, -25.271820068359375, 21.665557861328125, -177.03558349609375, -107.011474609375, -84.44906616210938, -108.8740234375, -28.974365234375, 231.25930786132812, 112.94720458984375, -21.414657592773438, 0.920318603515625, 5.465972900390625, 10.456161499023438, -27.932159423828125, 146.896484375, 157.77371215820312, 90.89698028564453, 123.16376495361328, 183.70394897460938, 229.50494384765625, 41.5479736328125, 313.015625, -85.8468017578125, -48.7491455078125, 198.19747924804688, -114.672119140625, 153.33474731445312, 187.44020080566406, 223.47283935546875, -0.23822021484375, 148.93800354003906, 37.18522644042969, 61.8204345703125, 30.754058837890625, 35.94679260253906, 210.82772827148438, 111.47666931152344, -5.253700256347656, 101.00088500976562, 64.65646362304688, 155.19271850585938, -215.68536376953125, 198.19125366210938, 261.400634765625, 64.52874755859375, 116.7530517578125, -226.18057250976562, -55.437957763671875, 37.68120574951172, 269.97100830078125, 84.98870849609375, 155.92108154296875, -114.81939697265625, -40.21125793457031, 123.93080139160156, 146.95068359375, 74.3702392578125, 199.4599609375, 302.8441467285156, -267.3484802246094, 48.082550048828125, -46.3468017578125, -2.1970081329345703, 202.4420166015625, 188.54373168945312, 22.9114990234375, -7.93487548828125, 0.0, 23.033748626708984, 14.389602661132812, 352.009033203125, -4.5531005859375, 57.42383575439453, -25.988319396972656, 139.36819458007812, 33.14830780029297, 61.28451919555664, 9.71917724609375, -41.93843078613281, -6.611413955688477, 19.38172721862793, 184.0210418701172, 173.85394287109375, 35.33042907714844, 239.66799926757812, 258.37109375, 231.39376831054688, 29.1107177734375, 282.44110107421875, 28.71942138671875, -43.91523742675781, 6.3175048828125, -35.86626434326172, 83.4239501953125, 73.27767944335938, 5.540496826171875, -21.30704689025879, 31.37408447265625, -33.67388916015625, 101.62078857421875, 118.520263671875, 132.7867431640625, 222.367919921875, 156.4735107421875, 29.074737548828125, 141.23922729492188, 301.5360107421875, 169.02291870117188, 269.0101013183594, 25.67840576171875, 15.323741912841797, 71.66094207763672, 168.74761962890625, -17.53607177734375, -32.79730224609375, 219.95980834960938, -110.68817138671875, 302.66619873046875, -15.926433563232422, 134.44619750976562, 72.0458984375, 38.494049072265625, 8.04254150390625, -81.1363525390625, -27.846473693847656, -108.55470275878906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000298.npy"}
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 57.785362243652344, "std": 106.04495239257812, "min": -258.7022705078125, "p10": -45.991172790527344, "median": 36.882171630859375, "p90": 190.20761260986328, "max": 356.0594177246094, "pos_frac": 0.7421875, "sample": [128.3523712158203, 304.66595458984375, 26.12104034423828, 133.19749450683594, 20.18133544921875, -9.0892333984375, 165.7852020263672, -52.5296630859375, 81.2669677734375, -219.26727294921875, 345.34515380859375, 211.4984893798828, 160.39501953125, 126.94586181640625, -45.125244140625, 94.13522338867188, -33.7099609375, 61.32456970214844, 10.035125732421875, -258.7022705078125, 225.57235717773438, 15.5911865234375, 203.43182373046875, -30.848419189453125, 140.171630859375, -31.249755859375, -124.31977844238281, -37.727508544921875, 31.213905334472656, 9.08404541015625, 170.90728759765625, 1.163238525390625, 164.10391235351562, 19.164695739746094, -214.54788208007812, 104.1046371459961, 92.04756164550781, -22.6507568359375, -81.11861419677734, 90.58648681640625, 151.41928100585938, 356.0594177246094, 24.95738983154297, -7.993648529052734, -4.7201690673828125, 11.65423583984375, 132.45394897460938, -14.031822204589844, 3.1914215087890625, -30.98480987548828, -143.44732666015625, 37.08447265625, 53.04804992675781, 36.67987060546875, 11.02899169921875, -29.439117431640625, 194.020751953125, -80.93862915039062, 179.4378662109375, 23.663436889648438, 126.00192260742188, 22.796417236328125, 6.8209228515625, 128.9022216796875, 94.7379150390625, 256.1447448730469, -18.087371826171875, 8.172109603881836, 40.3641357421875, 114.583984375, 131.29815673828125, 18.183895111083984, 81.03790283203125, 14.91156005859375, 138.21450805664062, 219.17626953125, 72.13641357421875, 11.224594116210938, 13.77374267578125, 157.55499267578125, 44.676361083984375, 125.88644409179688, 29.03260040283203, 25.8001708984375, -22.160751342773438, 43.05816650390625, 260.5904846191406, 120.99725341796875, -17.4766845703125, -33.030426025390625, -44.716583251953125, 97.78280639648438, 188.5734100341797, -72.94796752929688, -3.9942893981933594, -2.6859588623046875, 38.88542938232422, 42.693504333496094, 181.606201171875, -20.88678741455078, -90.4989013671875, 12.642608642578125, 19.972503662109375, -48.01167297363281, 12.956329345703125, 316.8110046386719, 64.05294799804688, 205.88348388671875, 66.11880493164062, 4.67578125, 159.83099365234375, 23.147457122802734, 77.24600219726562, -94.2156982421875, 22.735107421875, 104.68878173828125, 44.198577880859375, 46.037322998046875, 139.9464111328125, 216.23812866210938, -57.13927459716797, 6.786651611328125, 187.40158081054688, 154.10919189453125, 54.99676513671875, 169.97589111328125, 20.7618408203125, 56.895263671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000299.npy"}
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 57.6943359375, "std": 113.90972137451172, "min": -262.2806091308594, "p10": -82.92764053344726, "median": 51.131446838378906, "p90": 206.01240234375, "max": 342.20440673828125, "pos_frac": 0.6796875, "sample": [177.6287841796875, 100.43157958984375, 24.85468292236328, -172.75839233398438, 21.94141387939453, -82.35488891601562, 191.88623046875, -25.815887451171875, 25.231481552124023, 6.22515869140625, 213.8341064453125, 229.4885711669922, 93.34625244140625, -3.157562255859375, 89.79739379882812, 342.20440673828125, -197.26651000976562, 64.96176147460938, -66.94696044921875, -24.658782958984375, 291.6580505371094, 94.50982666015625, 75.0113525390625, 79.01432800292969, 52.79356384277344, 119.64358520507812, 82.27682495117188, -84.2640609741211, 41.01800537109375, -67.22428894042969, 168.95364379882812, 117.3013916015625, 162.5145263671875, -20.92138671875, 177.866455078125, -53.67683410644531, -262.2806091308594, 118.90350341796875, 8.445245742797852, 213.61849975585938, 174.58377075195312, -36.764495849609375, -69.59686279296875, 14.71600341796875, 245.76698303222656, 167.82083129882812, -103.63356018066406, 59.216407775878906, 122.53532409667969, 276.5697021484375, 19.098735809326172, -33.40660095214844, -132.95126342773438, 205.84918212890625, 37.01856994628906, -3.193378448486328, 113.8590087890625, 33.27178192138672, 79.99404907226562, -30.95026397705078, 67.23562622070312, 80.95559692382812, -163.50323486328125, 184.10739135742188, 47.047760009765625, -4.49444580078125, -4.21441650390625, 102.42681884765625, 49.469329833984375, 215.33999633789062, 86.92877197265625, 14.163938522338867, 3.8276233673095703, -12.414169311523438, -11.349296569824219, -21.8245849609375, 266.32855224609375, 67.07904052734375, -9.061126708984375, 154.04705810546875, 240.97885131835938, 124.25835418701172, 89.905517578125, 100.28753662109375, -3.7515106201171875, 120.09625244140625, -31.67626953125, 0.0, 94.6707763671875, 35.54649353027344, 159.60162353515625, 66.006591796875, -157.58074951171875, 160.95425415039062, 32.1673583984375, 268.2379150390625, 31.768218994140625, 8.827545166015625, 158.58587646484375, 104.3895263671875, 114.29495239257812, 152.54931640625, 3.1860198974609375, -10.9859619140625, -138.969970703125, 187.25723266601562, 203.94183349609375, -109.9393310546875, -98.96807861328125, -192.66189575195312, 105.51397705078125, 8.608419418334961, -11.329818725585938, -31.273727416992188, 109.416259765625, 135.88047790527344, 102.94369506835938, -10.888290405273438, -29.008071899414062, -95.74055480957031, 206.39324951171875, 78.40983581542969, 46.00531005859375, -34.10432434082031, 169.7030029296875, 336.3548583984375, 9.069343566894531, 0.008636474609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000300.npy"}
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 62.59714126586914, "std": 107.79400634765625, "min": -261.6910400390625, "p10": -65.85172424316406, "median": 47.70820999145508, "p90": 205.1068359375, "max": 312.28021240234375, "pos_frac": 0.6875, "sample": [-2.5797119140625, 267.1474304199219, 99.26239013671875, 138.45452880859375, 12.156196594238281, 204.96359252929688, 4.501132965087891, -67.34133911132812, 56.122406005859375, 14.667766571044922, 4.698219299316406, -25.76513671875, 81.0147705078125, 28.623565673828125, 17.190673828125, 19.311195373535156, -126.45491027832031, -57.352325439453125, 222.0264892578125, 205.44107055664062, 15.781654357910156, 43.999847412109375, 121.46588134765625, 34.8172607421875, 74.0601806640625, 28.47625732421875, 8.347320556640625, 265.0216064453125, 154.5612030029297, 84.22848510742188, 76.87091064453125, -0.454071044921875, 0.0, 119.448974609375, 114.88298034667969, 133.12229919433594, -87.91168212890625, 312.28021240234375, 0.34918975830078125, 195.45547485351562, -8.554071426391602, 182.04861450195312, 225.6051483154297, 75.25271606445312, -154.01904296875, 105.64178466796875, -96.98226928710938, 141.73095703125, -84.21331787109375, 32.3433837890625, 162.37832641601562, -26.83447265625, -11.5521240234375, 141.4364013671875, 181.3013916015625, 44.05396270751953, 78.4337158203125, 207.05636596679688, -3.2470703125, 306.23309326171875, -36.490333557128906, 94.67578125, -4.986602783203125, 120.25289916992188, -65.21331787109375, -5.868701934814453, 116.90927124023438, -261.6910400390625, -110.48748779296875, -61.95398712158203, 2.9020767211914062, 167.25535583496094, -14.7962646484375, 11.496562957763672, 144.87518310546875, 42.23114013671875, 173.81475830078125, 7.63433837890625, 153.310791015625, -127.02279663085938, 23.87896728515625, -29.227745056152344, 310.5639953613281, 154.99957275390625, 141.82598876953125, 24.346923828125, -24.78487777709961, -26.4017333984375, 104.41879272460938, -41.518890380859375, 250.5572967529297, 200.3951416015625, -6.555908203125, -20.337173461914062, 169.33673095703125, -117.63790893554688, -75.25951385498047, 116.64910888671875, 62.171112060546875, 56.2298583984375, 198.47021484375, 157.520751953125, 234.07867431640625, 51.362457275390625, 87.37286376953125, 151.164306640625, 110.9554443359375, 24.6737060546875, 222.43673706054688, -32.84611511230469, 84.40248107910156, 63.33386993408203, 222.25848388671875, -48.230255126953125, -10.030227661132812, 146.89707946777344, -106.84161376953125, -0.7677860260009766, 5.460487365722656, 131.42410278320312, 176.95140075683594, -137.966796875, -24.45697021484375, 28.253822326660156, 135.23123168945312, 175.51266479492188, 67.77195739746094, -11.462005615234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000301.npy"}
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 74.50103759765625, "std": 111.5886459350586, "min": -185.27890014648438, "p10": -37.291178894042964, "median": 58.88941192626953, "p90": 230.4385498046875, "max": 347.080322265625, "pos_frac": 0.7265625, "sample": [126.54446411132812, 191.13677978515625, 78.26361846923828, 138.1775665283203, 42.581607818603516, 66.253662109375, 0.0, 97.57156372070312, 14.95330810546875, 71.02536010742188, -185.27890014648438, 250.50929260253906, 206.85662841796875, 169.3424835205078, -140.16110229492188, 29.70551300048828, -108.88909912109375, 118.99359130859375, -31.972747802734375, -32.44496154785156, 292.2193603515625, 165.4384765625, 273.54254150390625, -6.271484375, 126.70599365234375, 26.174774169921875, 234.9691162109375, 14.981117248535156, -75.2576904296875, 27.826614379882812, 120.14604949951172, 226.11187744140625, 154.26849365234375, 120.50131225585938, 61.43336486816406, -48.1236572265625, 313.45166015625, 95.31085205078125, 158.43434143066406, 49.066436767578125, 30.644287109375, -18.639602661132812, -30.01092529296875, 66.73455810546875, 100.29442596435547, 100.56953430175781, 230.2264404296875, 137.6409149169922, 224.11920166015625, 230.9334716796875, -34.76947021484375, 93.08082580566406, 80.50003814697266, 24.882110595703125, 137.34207153320312, -128.53231811523438, 54.36793518066406, 106.70526123046875, 177.67288208007812, -41.92497253417969, 84.06832885742188, 115.89441680908203, 154.8026123046875, 115.29275512695312, 115.53497314453125, -4.6717529296875, -126.57693481445312, 8.435382843017578, -8.311634063720703, -8.337329864501953, 180.85037231445312, 123.04013061523438, 53.5223388671875, -86.73040771484375, 278.6527099609375, 148.1263427734375, -12.647987365722656, 116.66421508789062, 178.0423583984375, 106.06684875488281, 26.42803955078125, -2.505706787109375, 230.0964813232422, -172.7510986328125, 347.080322265625, 27.155731201171875, -98.72723388671875, 11.176017761230469, -35.305267333984375, -2.3680877685546875, -10.537328720092773, 55.65814208984375, -138.44241333007812, 25.48321533203125, -11.425472259521484, 308.41015625, 15.714996337890625, 241.08514404296875, 209.50433349609375, -28.104774475097656, 15.058563232421875, -24.7724609375, -118.10438537597656, 34.723838806152344, 219.74411010742188, -3.4187660217285156, 56.345458984375, 160.21786499023438, 36.099403381347656, 149.8776092529297, -34.8292236328125, 6.5998382568359375, 68.2969970703125, -30.56414794921875, 137.89151000976562, -13.268817901611328, 267.3936767578125, 4.7099761962890625, 203.11160278320312, 5.4073486328125, 30.515365600585938, 288.9075927734375, 112.65422058105469, 121.10227966308594, 46.86793518066406, 20.06329345703125, 256.20184326171875, 14.022857666015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000302.npy"}
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 67.70027160644531, "std": 112.6309585571289, "min": -273.15972900390625, "p10": -70.78825149536132, "median": 64.15045166015625, "p90": 200.8623092651367, "max": 353.2171630859375, "pos_frac": 0.7578125, "sample": [35.089813232421875, 17.624893188476562, -158.48614501953125, -125.18495178222656, 173.54476928710938, 12.201622009277344, 164.838623046875, 267.6344909667969, 135.0465087890625, 163.25067138671875, -70.46392822265625, 4.159553527832031, 199.61636352539062, 202.08937072753906, 100.34356689453125, -71.69949340820312, -146.28680419921875, 73.237060546875, 2.4820919036865234, 170.11956787109375, 63.036773681640625, 74.7796630859375, 18.441680908203125, 27.012603759765625, 95.78033447265625, 65.26412963867188, 69.17813873291016, 168.96694946289062, 199.61322021484375, 99.4665756225586, 164.07139587402344, 162.4300994873047, -154.32862854003906, 232.1193389892578, -39.079559326171875, 47.769439697265625, 10.313568115234375, 71.24472045898438, 0.0, 28.37195587158203, 15.9681396484375, -1.6967010498046875, -9.373397827148438, 5.7395477294921875, 19.707813262939453, 13.83538818359375, 175.35714721679688, 31.75225830078125, 158.509033203125, 161.60255432128906, 138.5131378173828, -17.27777099609375, 148.109130859375, -211.20571899414062, 200.33642578125, 263.15789794921875, 112.0565185546875, 3.1814041137695312, 133.92111206054688, 104.15925598144531, 2.0932388305664062, -37.27778625488281, -56.39111328125, -20.0810546875, -83.87078857421875, 43.33526611328125, 117.37997436523438, 15.121139526367188, 229.69381713867188, 25.564655303955078, 184.06402587890625, -26.080703735351562, -22.28472900390625, -49.12132263183594, 81.97257995605469, 174.03634643554688, -61.832794189453125, 186.20758056640625, 197.75831604003906, 284.8497314453125, 171.80226135253906, -273.15972900390625, 184.41416931152344, -71.54500579833984, 115.17910766601562, 38.54998779296875, 230.7574462890625, 22.56427001953125, -19.335205078125, 101.061767578125, 106.54776000976562, -123.325927734375, 178.4039306640625, -7.8604278564453125, 138.0909423828125, -2.803497314453125, 10.630340576171875, 2.0360336303710938, -25.833892822265625, 101.52272033691406, -118.89276123046875, -166.7947998046875, 68.96856689453125, 2.7918853759765625, -86.83977508544922, 44.56318664550781, 176.84857177734375, 211.29818725585938, 243.30120849609375, 37.669349670410156, 58.639984130859375, 175.00531005859375, 23.449443817138672, 98.61441040039062, 83.80979919433594, 211.57684326171875, 151.2759552001953, -20.45281982421875, 134.35995483398438, 71.92277526855469, 353.2171630859375, 8.8084716796875, 319.39923095703125, 8.009536743164062, 94.43560791015625, 55.943695068359375, 139.14840698242188, 222.76010131835938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000303.npy"}
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 73.56123352050781, "std": 116.89955139160156, "min": -185.90525817871094, "p10": -57.67705192565918, "median": 53.22618103027344, "p90": 219.83766479492186, "max": 420.8973388671875, "pos_frac": 0.71875, "sample": [0.0, 63.848358154296875, -37.751373291015625, 23.689796447753906, -107.23663330078125, 8.35333251953125, -59.763954162597656, 179.8394775390625, 125.67413330078125, -46.51654052734375, 141.56288146972656, 81.91436767578125, 124.66947937011719, -6.6587066650390625, -153.02301025390625, 124.82418823242188, 14.830169677734375, 129.18084716796875, 160.44143676757812, 17.982666015625, 119.97378540039062, 28.94992446899414, 10.6229248046875, 11.983749389648438, 75.93746948242188, 195.50572204589844, 99.27493286132812, 121.70930480957031, 69.81114196777344, 48.298583984375, -8.756072998046875, 186.7609100341797, 158.71893310546875, -2.71795654296875, 261.5544738769531, 107.75082397460938, 180.4938507080078, -16.283912658691406, 263.1723937988281, 136.9415283203125, 19.408721923828125, 66.68082427978516, 281.46820068359375, -47.045257568359375, 200.56585693359375, 51.90431213378906, -61.73942565917969, 41.29718017578125, -75.53643798828125, 125.55682373046875, 259.0814208984375, -52.00218200683594, 129.35577392578125, -138.39569091796875, 0.0, 173.40943908691406, -112.02432250976562, 13.340835571289062, -1.4185791015625, 90.30120849609375, -36.287628173828125, 11.089218139648438, -35.427093505859375, 161.93399047851562, 93.52107238769531, 23.448936462402344, -185.90525817871094, 19.291259765625, 283.7511901855469, 303.3197021484375, -36.62164306640625, 113.20037841796875, 111.092041015625, -8.09771728515625, -63.67877197265625, 358.47552490234375, 198.79168701171875, -56.78266525268555, 176.61630249023438, -70.44155883789062, 236.54075622558594, 136.91490173339844, 50.93486785888672, -128.45877075195312, 387.77392578125, 54.54804992675781, 420.8973388671875, 3.583251953125, 185.19622802734375, -136.867431640625, -26.375518798828125, -5.667333602905273, 14.747467041015625, 112.20703125, 352.5748291015625, 28.909650802612305, 24.123779296875, 142.24575805664062, 160.2728729248047, 132.1900634765625, 26.55572509765625, 98.438720703125, 77.50386047363281, 13.509628295898438, 26.33452606201172, 130.54391479492188, 17.58056640625, 276.6872253417969, 88.44076538085938, 222.8228759765625, 147.87733459472656, 51.842529296875, -45.02439880371094, -47.77557373046875, 183.15707397460938, 63.935089111328125, -42.5723876953125, -17.526275634765625, 10.360092163085938, 202.39883422851562, 40.06719970703125, 153.8358154296875, -108.33676147460938, 161.40274047851562, -28.514923095703125, 163.04624938964844, 23.336669921875, 218.55828857421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000304.npy"}
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 107.00007629394531, "std": 122.18753814697266, "min": -281.548828125, "p10": -11.137113952636712, "median": 103.86210632324219, "p90": 267.9576599121093, "max": 504.67724609375, "pos_frac": 0.84375, "sample": [40.38066482543945, 137.0421142578125, 19.97283172607422, 251.1082763671875, 24.39642333984375, -0.79925537109375, 42.63018035888672, -59.753135681152344, 128.8170623779297, 41.416473388671875, 124.91482543945312, 50.0958251953125, -35.158538818359375, 102.99786376953125, 1.4409942626953125, -15.28997802734375, 35.39178466796875, 61.68914794921875, 130.8529052734375, 15.817684173583984, 160.39154052734375, 290.3794250488281, 131.3809814453125, 124.45950317382812, 17.600845336914062, 19.407196044921875, 42.200714111328125, -6.6454620361328125, 33.5560302734375, 70.632568359375, 53.19183349609375, 195.7425537109375, 150.3419189453125, 164.2662353515625, 235.38204956054688, 86.267822265625, 9.875564575195312, -6.465660095214844, 199.2079315185547, 166.81390380859375, -136.09730529785156, 84.04638671875, 4.778938293457031, 163.98318481445312, 380.4287414550781, 183.2122802734375, 343.80609130859375, 195.61257934570312, 104.72634887695312, 68.46556091308594, -35.50774002075195, -189.92083740234375, 186.51205444335938, 321.0468444824219, 162.86685180664062, 121.40682983398438, 504.67724609375, 162.76344299316406, 240.99029541015625, 145.531005859375, 43.427886962890625, 112.745361328125, 52.613525390625, 263.0833740234375, 56.938507080078125, -9.357315063476562, 183.9158935546875, -6.399986267089844, 308.25396728515625, 209.84446716308594, -1.081268310546875, 142.574951171875, -137.88217163085938, 257.7267150878906, 152.1605224609375, -281.548828125, 29.837966918945312, 138.30137634277344, -81.27914428710938, 124.04757690429688, 13.663818359375, 279.33099365234375, 233.2440948486328, 39.85533905029297, 64.76264190673828, 122.99287414550781, -47.1444091796875, 362.4481506347656, 17.669540405273438, 26.79547119140625, 284.2501220703125, 140.7342529296875, 79.21868896484375, 154.54595947265625, 181.34234619140625, 68.93848419189453, 252.66017150878906, 234.1563720703125, 344.866455078125, 139.19598388671875, 144.63223266601562, 22.54290771484375, 194.5520782470703, 91.9747314453125, 174.136474609375, 78.52481079101562, 78.5787353515625, 39.729248046875, 69.05508422851562, 157.514892578125, 181.3797607421875, -101.28611755371094, 119.32950592041016, 19.60638427734375, 114.75401306152344, 26.903587341308594, 184.80007934570312, -83.56368255615234, 87.8201904296875, 53.15168762207031, 324.31353759765625, 124.43487548828125, 327.3902587890625, 225.54043579101562, 294.56585693359375, -76.04814147949219, -1.723876953125, 22.668930053710938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000305.npy"}
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 65.02589416503906, "std": 110.6356201171875, "min": -181.58328247070312, "p10": -64.46298370361328, "median": 46.988922119140625, "p90": 222.84918212890625, "max": 313.7558898925781, "pos_frac": 0.6796875, "sample": [39.3126220703125, 64.52178955078125, 33.37294006347656, 222.50421142578125, -128.12136840820312, 313.7558898925781, 11.679035186767578, -65.48556518554688, -130.36651611328125, 232.37091064453125, -20.261404037475586, 23.266693115234375, 221.82281494140625, 96.60592651367188, -45.95109558105469, -54.408843994140625, -76.04254150390625, 223.65411376953125, -18.43926239013672, 40.7156982421875, 91.454345703125, 79.69468688964844, -125.945556640625, -32.8592529296875, -65.08438110351562, 233.21572875976562, 204.73956298828125, 154.31472778320312, 135.78219604492188, 6.787261962890625, 74.670654296875, -1.0439453125, -12.256744384765625, 168.40850830078125, 8.36480712890625, 148.61370849609375, 102.78936767578125, 234.04391479492188, 140.03851318359375, 33.47216796875, -22.044342041015625, 15.590412139892578, -2.18621826171875, 18.094619750976562, 53.26214599609375, -2.6114349365234375, 65.49078369140625, 104.57736206054688, 53.330352783203125, -151.5762939453125, -101.394287109375, 196.66738891601562, -0.21099090576171875, 190.7867431640625, 79.12823486328125, -181.58328247070312, 12.448722839355469, 147.89056396484375, -30.723350524902344, 184.7596435546875, -55.0030517578125, 1.23638916015625, 216.00543212890625, -0.807403564453125, 119.25288391113281, 27.736059188842773, 124.18418884277344, 15.353687286376953, 181.4999542236328, -64.19667053222656, 88.0586166381836, -95.81768798828125, 95.722412109375, 121.1866455078125, 253.70379638671875, 9.801948547363281, 75.3297119140625, 35.52204513549805, -1.57647705078125, -45.6083984375, 63.166015625, 65.21493530273438, 100.40631103515625, 145.16949462890625, 220.5003662109375, -52.99357604980469, -9.9261474609375, 112.74688720703125, 36.80572509765625, 241.26754760742188, 117.22927856445312, 250.46630859375, 76.34304809570312, -37.0369873046875, 62.6309814453125, 153.2919158935547, 152.2738037109375, -20.148712158203125, 198.6649169921875, -5.436185836791992, 272.0396728515625, 137.87005615234375, 290.8395080566406, 240.13946533203125, 20.642799377441406, 36.40840148925781, 108.24488830566406, -119.6376953125, 7.1835784912109375, -133.04791259765625, -40.783077239990234, 4.872314453125, 162.57479858398438, 200.06024169921875, 37.44610595703125, 278.490966796875, -140.53286743164062, -17.66705322265625, -13.7342529296875, -41.55377197265625, 11.186431884765625, 178.12680053710938, -34.16230773925781, -14.436729431152344, 118.70723724365234, 201.08734130859375, 257.9857177734375, 149.3404541015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000306.npy"}
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 69.71328735351562, "std": 109.98136901855469, "min": -198.365966796875, "p10": -67.51798095703124, "median": 68.17425537109375, "p90": 205.8122360229492, "max": 349.9957275390625, "pos_frac": 0.765625, "sample": [165.03732299804688, 257.424560546875, 81.50149536132812, 238.71649169921875, -1.87017822265625, -10.73088264465332, 37.16189193725586, 152.73806762695312, 125.51132202148438, 201.89125061035156, -195.99591064453125, 21.58374786376953, 164.34915161132812, -133.57936096191406, -30.7864990234375, 20.197036743164062, 302.9921569824219, 224.36495971679688, 104.50819396972656, 68.10125732421875, 217.17855834960938, -141.30654907226562, 68.24725341796875, 7.778907775878906, 181.261962890625, -25.34600830078125, 41.88043212890625, 55.055450439453125, 86.741455078125, 89.67218017578125, 88.94673156738281, -41.7080078125, 169.62838745117188, 113.4113540649414, 42.9822998046875, -16.33935546875, 146.85104370117188, 95.13916015625, 39.15185546875, 104.38248443603516, -41.53466796875, 15.550519943237305, 24.96485137939453, 17.11773681640625, 110.85403442382812, -138.90060424804688, 128.62242126464844, -21.656055450439453, 37.12158203125, 124.0831298828125, 28.61187744140625, 55.3325080871582, 189.5384521484375, 31.94281005859375, 33.17417907714844, 157.6438751220703, 140.6846160888672, 206.01548767089844, 49.33372497558594, -11.851875305175781, 33.86041259765625, 138.20123291015625, -198.365966796875, 102.35223388671875, 14.402584075927734, -152.4013671875, 205.72512817382812, 240.98220825195312, 100.21756744384766, 10.997978210449219, 12.111814498901367, 76.4816665649414, 25.78948974609375, -25.0302734375, -67.084716796875, -13.192962646484375, 120.43438720703125, 227.22561645507812, 70.441162109375, 49.908355712890625, 146.8203887939453, -148.87808227539062, 6.135219573974609, 160.15216064453125, 171.942626953125, 222.24365234375, 21.95849609375, 194.8109588623047, 21.479400634765625, 31.411834716796875, 192.1114959716797, -91.26864624023438, 77.98204040527344, 253.08578491210938, 113.07461547851562, 21.188507080078125, 153.21212768554688, -90.3840560913086, -115.51024627685547, 182.67498779296875, -107.11724853515625, 286.94134521484375, 47.588958740234375, 153.37905883789062, -192.44482421875, -32.49835205078125, -25.80706787109375, 99.95893859863281, 188.41482543945312, -11.423973083496094, -68.5289306640625, 23.5528564453125, 349.9957275390625, 70.93008422851562, 59.81983947753906, 151.81967163085938, 99.17353820800781, -51.488311767578125, 81.4873046875, 189.72158813476562, -13.294036865234375, 13.676109313964844, 178.78317260742188, 83.00180053710938, 190.18792724609375, 0.5225448608398438, 257.2078857421875, 149.0662841796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000307.npy"}
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 50.35519027709961, "std": 117.71638488769531, "min": -322.1843566894531, "p10": -111.08927154541016, "median": 43.61216735839844, "p90": 187.63970336914062, "max": 482.46612548828125, "pos_frac": 0.6875, "sample": [212.09410095214844, 52.46263885498047, -41.5877685546875, 59.20746612548828, 77.2237548828125, -113.37652587890625, -39.1583251953125, 117.59429931640625, 22.746612548828125, 17.401206970214844, -112.78582763671875, 130.72982788085938, -110.73182678222656, -154.47842407226562, -27.76226806640625, -16.01715087890625, 106.98002624511719, 84.77799987792969, 182.80850219726562, 188.29010009765625, 191.88922119140625, 54.746673583984375, 83.51773071289062, 3.7091445922851562, 27.434478759765625, 160.60037231445312, -322.1843566894531, -29.70551300048828, 7.624267578125, 223.87893676757812, 157.52354431152344, -24.641448974609375, -2.40728759765625, 87.98173522949219, -157.66851806640625, -52.715789794921875, 133.90167236328125, 19.88292694091797, 100.50616455078125, 50.30462646484375, -3.602935791015625, 40.130332946777344, 278.5557861328125, -102.8758544921875, -121.502197265625, 56.419464111328125, 5.15228271484375, 154.3917236328125, -67.75841522216797, 176.932373046875, 14.783073425292969, 96.3495101928711, 9.2227783203125, 187.3609619140625, -111.92330932617188, 104.656005859375, 218.81024169921875, -13.325645446777344, -172.77816772460938, 128.83145141601562, -18.13842010498047, -43.901092529296875, -32.9765625, -117.584716796875, 365.1433410644531, 32.111968994140625, 0.11134147644042969, -140.83474731445312, 144.6372528076172, 99.38092803955078, 118.15997314453125, 168.8956298828125, 162.55503845214844, 103.90875244140625, 65.02743530273438, 96.38861083984375, 309.409423828125, 191.36004638671875, 482.46612548828125, -14.002700805664062, 96.12481689453125, 41.8980712890625, -20.44470977783203, -131.27084350585938, 98.83480834960938, -108.24057006835938, 140.1927490234375, 140.874755859375, -70.3565673828125, 126.2257308959961, 33.6068115234375, 42.27688980102539, -12.5540771484375, 43.57275390625, 130.5350341796875, 59.203636169433594, 141.35748291015625, 248.90176391601562, 31.236663818359375, 154.17031860351562, 72.8690185546875, 158.5523681640625, 261.2046203613281, -118.37605285644531, 75.06583404541016, 25.835952758789062, 44.04486083984375, -9.91604995727539, 120.68074035644531, 19.195648193359375, -201.220947265625, 43.651580810546875, -0.8482666015625, 27.184959411621094, 15.116340637207031, -46.34942626953125, -18.497299194335938, 75.20935821533203, -95.11257934570312, 12.751708984375, 27.18017578125, 14.770071029663086, 211.7266845703125, 51.997955322265625, 120.16854095458984, -79.7164306640625, 173.95538330078125, 77.65029907226562], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000308.npy"}
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 82.73484802246094, "std": 121.43352508544922, "min": -258.45062255859375, "p10": -46.18614120483398, "median": 69.43290710449219, "p90": 241.5038879394531, "max": 422.8736572265625, "pos_frac": 0.75, "sample": [-13.115585327148438, 96.23484802246094, 247.954833984375, 107.68623352050781, -15.748191833496094, -65.84196472167969, 158.8753662109375, -7.766277313232422, 142.84658813476562, 74.541748046875, 162.77603149414062, 12.920166015625, -63.10093688964844, 76.91790771484375, -41.23839569091797, 119.969482421875, 13.1103515625, 7.887401580810547, -4.310340881347656, 144.70565795898438, 10.925567626953125, 137.77590942382812, -4.837066650390625, -151.93106079101562, 17.16632080078125, 21.241851806640625, -151.26513671875, 265.5785217285156, 118.92605590820312, 95.83428955078125, -27.232452392578125, 211.5372314453125, 60.630889892578125, 7.7624664306640625, 16.016830444335938, -29.11822509765625, 126.55064392089844, -16.068710327148438, 273.1353759765625, -232.38668823242188, 248.79067993164062, 103.53164672851562, 23.609947204589844, 31.083694458007812, 225.1411895751953, -50.23479461669922, -31.091445922851562, 16.44464111328125, -44.45100402832031, 1.0147705078125, 199.90052795410156, -0.05246162414550781, 93.27828979492188, -112.03033447265625, 134.905029296875, 112.51171875, 37.37361145019531, 21.94976806640625, -6.6143798828125, 125.75872802734375, 238.73919677734375, -35.340576171875, 125.37444305419922, 338.8231201171875, -35.02137756347656, 392.26708984375, -1.0165557861328125, 57.053131103515625, 248.9869842529297, 108.03506469726562, -94.32957458496094, 185.9830322265625, -3.4263458251953125, 422.8736572265625, 52.35504150390625, 45.19300842285156, 92.26675415039062, 151.65782165527344, 31.809566497802734, 298.297119140625, 165.7738037109375, 191.66265869140625, 30.509567260742188, 109.35260009765625, 130.4726104736328, 185.492431640625, 64.32406616210938, 50.53697204589844, 280.83148193359375, 17.43683624267578, 166.63330078125, 24.330474853515625, -19.574478149414062, -129.94065856933594, 139.03579711914062, 168.4959259033203, 92.08506774902344, 48.60968780517578, 140.8944091796875, 11.235511779785156, -58.527923583984375, 56.941925048828125, 167.22064208984375, 198.02377319335938, 180.00244140625, -60.94439697265625, 218.22402954101562, 78.68841552734375, -16.320770263671875, -258.45062255859375, 131.58712768554688, 315.69464111328125, 28.423141479492188, 211.50674438476562, 20.814010620117188, 222.29168701171875, 102.42474365234375, 409.4725646972656, 10.097412109375, 218.83741760253906, 253.08953857421875, 127.64424896240234, 187.62179565429688, -54.39447021484375, 108.12997436523438, 47.512237548828125, 3.908754348754883, 215.3956298828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000309.npy"}
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 77.39212036132812, "std": 113.7319107055664, "min": -163.37435913085938, "p10": -55.774482727050774, "median": 67.9891586303711, "p90": 222.78987884521484, "max": 428.0374755859375, "pos_frac": 0.7421875, "sample": [293.9129638671875, 152.45877075195312, -13.419635772705078, 311.849609375, -53.3665771484375, 136.50051879882812, 47.579864501953125, 156.07669067382812, 173.3658447265625, 79.29519653320312, 170.17544555664062, 365.8345947265625, 154.4467315673828, 169.9727325439453, 83.9814453125, 230.23086547851562, 83.7393798828125, 24.77638053894043, 221.45172119140625, -147.6114501953125, 37.88523864746094, 68.26858520507812, 9.315078735351562, 100.2110595703125, 120.03822326660156, 56.06117248535156, 83.59579467773438, -93.01922607421875, 127.74829864501953, -10.800445556640625, 162.11903381347656, 0.22063827514648438, -68.5697021484375, 219.4322509765625, -31.1834716796875, -61.39292907714844, 39.46751403808594, 174.90606689453125, 28.17718505859375, -106.6981201171875, 14.758796691894531, 16.813568115234375, 40.66874694824219, 114.55487060546875, 1.1433181762695312, -61.7125244140625, 153.46401977539062, 142.02606201171875, 52.36909484863281, 141.12843322753906, 5.7442626953125, 39.569488525390625, 179.25790405273438, 27.527557373046875, 67.70973205566406, 2.166961669921875, 428.0374755859375, 176.52581787109375, -11.600135803222656, -163.37435913085938, -2.5427169799804688, 252.74295043945312, 113.57601928710938, 29.81719970703125, -143.13156127929688, -30.506454467773438, -39.37982177734375, 156.0950927734375, 119.84048461914062, 8.25335693359375, -48.731285095214844, -62.745086669921875, -23.446853637695312, 117.8253173828125, 193.9423370361328, 174.78436279296875, 172.46026611328125, 292.390869140625, 233.93466186523438, 43.03059387207031, -136.80624389648438, 141.29324340820312, 159.63690185546875, 113.58056640625, -45.13072204589844, 74.09078979492188, 134.01425170898438, -4.132720947265625, 11.751443862915039, 60.267364501953125, 15.677764892578125, 37.18402099609375, 180.73426818847656, 8.197784423828125, -23.227096557617188, -8.137527465820312, 76.97650146484375, 211.0204620361328, -77.9886474609375, 69.47700500488281, 237.84457397460938, 225.91224670410156, -37.570037841796875, -109.74041748046875, 198.82009887695312, 78.30291748046875, -27.352127075195312, -108.71685791015625, -45.06413269042969, 48.63676452636719, -28.39354705810547, 71.65826416015625, 28.706893920898438, 141.84671020507812, 44.290496826171875, -21.496337890625, 214.71685791015625, 73.49638366699219, 44.95758819580078, -33.4385986328125, 131.96456909179688, 7.331958770751953, 279.1024169921875, 244.66891479492188, 123.4180908203125, 165.81016540527344, 218.44525146484375, 317.5311279296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000310.npy"}
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 78.53181457519531, "std": 120.2965087890625, "min": -280.884765625, "p10": -67.20005645751952, "median": 76.96509170532227, "p90": 223.66952667236328, "max": 415.25335693359375, "pos_frac": 0.75, "sample": [110.7176513671875, -45.564422607421875, 180.47415161132812, 44.64765930175781, 221.26864624023438, -13.612409591674805, 183.722900390625, 42.08589172363281, 239.20501708984375, 163.17755126953125, 167.898193359375, 144.81654357910156, 13.808435440063477, -17.68732452392578, -69.19775390625, -203.83929443359375, 97.59503173828125, 85.64797973632812, -0.6844825744628906, 59.98089599609375, 210.34304809570312, 175.2215576171875, 58.179443359375, 76.90824890136719, 8.642173767089844, -48.49345397949219, 181.64404296875, -63.96867370605469, 37.53684997558594, 70.9373779296875, -69.5560531616211, -267.4251708984375, -280.884765625, 99.72102355957031, 105.6112060546875, 248.637939453125, -138.96533203125, 3.91082763671875, 134.15301513671875, 97.53317260742188, -130.10208129882812, 193.076416015625, 62.930137634277344, 240.50299072265625, 4.024394989013672, 160.11302185058594, -15.538436889648438, 74.05294799804688, -167.96820068359375, 116.41282653808594, 142.87989807128906, 56.74566650390625, 67.10458374023438, 186.7742156982422, 158.61587524414062, 48.270538330078125, 77.02193450927734, 46.6904296875, -31.0806884765625, 195.34478759765625, 354.23291015625, 181.85955810546875, 224.06277465820312, 129.2364501953125, 239.14093017578125, -59.5511474609375, 74.46063232421875, -3.0533294677734375, 121.20482635498047, 32.2569580078125, 132.75701904296875, -66.37748718261719, 211.8448486328125, 116.93231201171875, 83.74679565429688, 49.831581115722656, 200.7760009765625, 149.59954833984375, -6.548583984375, 269.6979675292969, 271.7376708984375, 15.657958984375, -69.119384765625, 4.305946350097656, -18.673805236816406, 9.964996337890625, 223.50099182128906, 302.85235595703125, 285.2063293457031, 172.1109619140625, 58.940338134765625, 84.09967041015625, 79.28195190429688, 51.434326171875, 172.51824951171875, -85.01313781738281, 415.25335693359375, -20.00933837890625, -145.9464874267578, 187.96778869628906, 17.763065338134766, 123.38543701171875, -21.673465728759766, 65.43368530273438, 113.95831298828125, -6.124515533447266, 175.62144470214844, 252.15570068359375, -8.466043472290039, 198.067138671875, 62.17106628417969, 82.17337036132812, 4.46673583984375, 151.6234130859375, -87.08221435546875, -27.244277954101562, 168.60440063476562, 182.7491912841797, 134.7763214111328, 7.1590423583984375, -122.53926086425781, 15.883125305175781, 96.7990493774414, 239.52398681640625, 152.69253540039062, 44.755489349365234, 125.40646362304688, -52.16648864746094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000311.npy"}
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 68.55606079101562, "std": 138.89002990722656, "min": -334.0816650390625, "p10": -94.27965393066405, "median": 39.596649169921875, "p90": 234.01960144042965, "max": 461.17169189453125, "pos_frac": 0.71875, "sample": [174.54478454589844, 0.09625244140625, 79.96417236328125, 231.2025146484375, 70.17457580566406, -330.3914489746094, 265.5328369140625, 69.8074951171875, -1.1512680053710938, 7.9475555419921875, 88.90029907226562, -123.52000427246094, 16.818817138671875, 93.11041259765625, 183.78363037109375, 420.0189208984375, 34.615962982177734, 156.91859436035156, -109.57235717773438, -36.3414306640625, -7.09716796875, 26.525230407714844, 199.02850341796875, 16.8345947265625, -31.493553161621094, -25.1912841796875, 46.13848876953125, 211.7816162109375, -48.862396240234375, -26.688156127929688, 322.8902282714844, 23.91381072998047, 18.849365234375, 461.17169189453125, 122.8980712890625, -106.95433044433594, 195.06988525390625, -7.956489562988281, 10.779752731323242, 219.03265380859375, 205.89425659179688, 411.49188232421875, -34.66326141357422, 155.84344482421875, 54.55731964111328, 144.24197387695312, 81.83673095703125, 240.59280395507812, 155.53736877441406, 42.23297119140625, 230.467529296875, 217.65972900390625, -93.06356811523438, 5.065574645996094, 14.52630615234375, 55.997161865234375, 71.05253601074219, 64.35357666015625, 5.2482452392578125, 335.4289245605469, 9.229843139648438, -16.4088134765625, 263.7626647949219, -97.1171875, 13.538848876953125, -334.0816650390625, 68.74845886230469, -32.878143310546875, -114.35223388671875, 77.87225341796875, 4.3565673828125, 98.05966186523438, -0.9582405090332031, 3.2403526306152344, 21.69024658203125, 146.2257080078125, -3.5840072631835938, 242.951416015625, -3.0936813354492188, 174.18701171875, 185.3468017578125, 179.06124877929688, -83.33587646484375, 28.492477416992188, -22.900306701660156, 0.0, 189.9306640625, 56.697296142578125, -45.39544677734375, -177.07720947265625, 4.5791015625, 36.9603271484375, 168.27554321289062, 120.51065063476562, 127.14935302734375, 1.5208358764648438, -235.60877990722656, 35.51093292236328, 298.173583984375, 16.221969604492188, 30.53167724609375, 10.290725708007812, 117.191650390625, 184.03114318847656, 90.464111328125, -149.60760498046875, 294.7928466796875, 122.9166259765625, 304.9750061035156, 399.34222412109375, 169.50668334960938, -28.824188232421875, 197.07659912109375, -8.576335906982422, -35.79676818847656, 33.237518310546875, 186.162841796875, -105.91487121582031, -180.9097900390625, 145.63351440429688, -168.35592651367188, 200.595458984375, -38.056304931640625, 81.3753662109375, 156.61070251464844, 65.14639282226562, 10.537103652954102, 7.86444091796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000312.npy"}
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 70.84982299804688, "std": 134.65269470214844, "min": -224.14625549316406, "p10": -95.59059753417968, "median": 49.65448760986328, "p90": 231.29346923828118, "max": 754.385009765625, "pos_frac": 0.6796875, "sample": [307.7290954589844, 293.0137939453125, -33.794189453125, -52.41752624511719, -4.5675048828125, -0.8715133666992188, 18.58453369140625, 118.99154663085938, 293.95379638671875, -37.54052734375, -56.44171142578125, -77.25115966796875, -93.24237060546875, 281.9474182128906, -46.391014099121094, 217.8617401123047, 61.161102294921875, 102.40725708007812, 34.108123779296875, -105.95135498046875, -61.53448486328125, 50.1793212890625, -1.9608039855957031, 155.51187133789062, 128.20147705078125, -9.6126708984375, -145.5655517578125, 56.1097412109375, 120.20303344726562, 256.8411865234375, 159.050537109375, 254.59039306640625, 96.80947875976562, 142.87493896484375, -87.882568359375, 192.6029052734375, 45.869728088378906, 43.50665283203125, 259.1672058105469, 47.56098937988281, -150.8812255859375, 295.259033203125, 288.8810729980469, 0.5561981201171875, -136.86184692382812, 31.932907104492188, 66.4962158203125, -105.25607299804688, 15.249191284179688, 224.6435546875, 147.8018035888672, 23.338701248168945, -152.14593505859375, 23.211694717407227, 23.1663818359375, -16.79916000366211, 179.89820861816406, 51.638755798339844, 29.246002197265625, -1.666595458984375, 158.4215087890625, -13.63336181640625, 169.9521026611328, 26.03112030029297, 100.7400894165039, -8.049240112304688, -18.997482299804688, 161.82571411132812, -35.97697448730469, -45.813385009765625, 154.0448760986328, -10.250778198242188, 223.62423706054688, 169.90322875976562, -10.3050537109375, 24.3775634765625, 69.92732238769531, 218.84100341796875, -38.64649963378906, 122.98564147949219, 15.74200439453125, -101.06979370117188, 62.715850830078125, -111.95773315429688, 52.6265869140625, 179.28097534179688, 37.579925537109375, -9.723602294921875, 23.488601684570312, 49.12965393066406, 60.7633056640625, 754.385009765625, 115.2596206665039, 191.5115966796875, 122.3515625, 198.1531524658203, 51.32314682006836, 246.8099365234375, -127.23464965820312, -126.69613647460938, 57.1644287109375, 151.63909912109375, 32.76871109008789, 223.49969482421875, 44.92071533203125, 12.526763916015625, 34.38734436035156, -51.38928985595703, 282.9886169433594, 298.31787109375, 171.88992309570312, 142.45297241210938, 141.251708984375, 34.753265380859375, 137.6001434326172, -27.982284545898438, 177.3138885498047, 123.72059631347656, -30.82464599609375, -224.14625549316406, -17.124099731445312, -194.4779052734375, -153.10098266601562, 159.58566284179688, 144.5277099609375, 184.79653930664062, 153.0238037109375, 193.73043823242188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000313.npy"}
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 72.81683349609375, "std": 112.10604095458984, "min": -220.54507446289062, "p10": -60.50263519287109, "median": 64.9961929321289, "p90": 205.55146484374998, "max": 399.4545593261719, "pos_frac": 0.7421875, "sample": [107.05610656738281, 200.20681762695312, 190.2145233154297, 78.15753173828125, 145.84393310546875, 55.46405029296875, -58.95045471191406, 108.81118774414062, -68.08460998535156, -29.845687866210938, -220.54507446289062, 184.09213256835938, 311.1433410644531, 116.86859130859375, 180.8197021484375, -168.1107177734375, -136.865478515625, 194.61245727539062, 9.273834228515625, 195.81036376953125, 184.95285034179688, -69.97509765625, 19.849849700927734, 53.397247314453125, -94.22444152832031, 100.35105895996094, 136.24664306640625, 69.15740966796875, 35.653411865234375, 286.0260009765625, 215.662353515625, -50.17268371582031, -38.63494873046875, -48.96043395996094, 136.79823303222656, 39.9752197265625, 187.10440063476562, 109.53620910644531, 73.38104248046875, 281.6669921875, 21.22760009765625, -11.350250244140625, 47.058685302734375, 155.15606689453125, 19.762575149536133, 88.9034423828125, 219.80557250976562, 80.7580795288086, 133.65185546875, -23.78509521484375, 150.2835693359375, 17.07274627685547, 100.16670227050781, 156.15081787109375, 24.299415588378906, 144.95748901367188, 63.566497802734375, 43.68255615234375, 57.52516174316406, 12.3712158203125, 170.79270935058594, 199.29327392578125, 165.25570678710938, 0.0, -107.3848876953125, -151.91427612304688, -89.47381591796875, 152.52655029296875, 29.31341552734375, 87.41763305664062, 47.53460693359375, 15.66717529296875, 149.17709350585938, 206.77377319335938, -47.39581298828125, 22.34906005859375, 2.139801025390625, 235.03707885742188, -42.946990966796875, -185.40081787109375, 119.14109802246094, 110.14163208007812, 54.34645080566406, 194.4404296875, 0.0, 214.07571411132812, 34.738922119140625, 159.36849975585938, 189.66693115234375, 66.42588806152344, 143.31936645507812, 43.07806396484375, 215.5167236328125, 42.87286376953125, -41.626068115234375, 278.3468017578125, 178.34661865234375, 92.99580383300781, 32.30975341796875, 11.8651123046875, 93.26199340820312, 16.058372497558594, -30.541473388671875, -190.18264770507812, 46.733489990234375, -10.865936279296875, 228.16531372070312, 154.09115600585938, 82.29165649414062, 135.9443817138672, 252.89190673828125, -30.00408172607422, -64.1243896484375, 32.731109619140625, -1.240570068359375, 399.4545593261719, 179.43002319335938, -43.54383087158203, 20.605918884277344, 128.53140258789062, -2.731475830078125, 21.36467742919922, -17.162322998046875, 129.92401123046875, -90.2646484375, 205.02761840820312, 167.5273895263672, -15.97915267944336], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000314.npy"}
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 81.02194213867188, "std": 116.6976089477539, "min": -186.4298095703125, "p10": -53.27767944335937, "median": 63.36716651916504, "p90": 233.5232360839844, "max": 344.65557861328125, "pos_frac": 0.75, "sample": [-33.954315185546875, 109.19426727294922, 4.285068511962891, -48.86820983886719, -134.5964813232422, -34.56658935546875, -72.71002197265625, 47.367462158203125, 228.14630126953125, -64.18011474609375, 168.54425048828125, -11.342437744140625, 198.2127227783203, 59.72509765625, 34.666717529296875, 16.009368896484375, 61.692535400390625, 221.5888671875, 82.67706298828125, -82.48439025878906, -6.540975570678711, -9.036651611328125, 6.396331787109375, -95.01228332519531, 245.43218994140625, 6.24237060546875, 120.64674377441406, 192.60693359375, 174.55145263671875, 39.999114990234375, 9.660541534423828, 74.84893798828125, 64.0294189453125, 270.69769287109375, 146.22853088378906, 217.05996704101562, 168.74273681640625, 193.92550659179688, 42.37837219238281, -21.34967041015625, -15.650360107421875, -50.830230712890625, 56.86102294921875, 302.6401062011719, 16.809463500976562, -5.10284423828125, 44.89794921875, -99.53857421875, -57.142333984375, 123.03952026367188, 138.5848846435547, -21.662052154541016, 291.7061462402344, 200.58700561523438, 232.68707275390625, 263.599853515625, 39.513031005859375, 29.917587280273438, 174.56756591796875, 138.3103790283203, 91.76394653320312, -19.404052734375, -69.62184143066406, 9.947189331054688, 222.74246215820312, 233.43902587890625, 118.76434326171875, 213.65394592285156, -20.1005859375, 240.2322235107422, -186.4298095703125, 168.197265625, 344.65557861328125, 52.288421630859375, 157.79953002929688, -51.62139892578125, -163.83193969726562, 146.975341796875, 9.605157852172852, 255.12152099609375, 280.19671630859375, 28.247467041015625, 141.52230834960938, 220.4326171875, 187.0565185546875, 17.530593872070312, 97.07118225097656, 126.27265930175781, 23.9278564453125, 198.50213623046875, 16.910934448242188, -31.63720703125, -170.06466674804688, -41.690521240234375, 118.54013061523438, 306.789794921875, 156.61383056640625, 2.365846633911133, -180.58877563476562, 196.6409912109375, 2.999786376953125, 29.65643310546875, 163.58486938476562, 44.194679260253906, -35.65428161621094, -112.61231994628906, -39.565185546875, 245.4464111328125, 7.051061630249023, 132.43621826171875, 87.78140258789062, 62.70491409301758, 68.83145141601562, 301.3632507324219, 38.93499755859375, 108.15924072265625, 233.7197265625, -16.5518798828125, 110.310302734375, 116.41920471191406, 121.29463195800781, 20.077110290527344, 185.9510955810547, 143.0523681640625, 176.8528289794922, 201.03427124023438, 104.99078369140625, 24.818866729736328], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000315.npy"}
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 76.04742431640625, "std": 116.98184204101562, "min": -159.1656951904297, "p10": -61.065740203857416, "median": 54.860260009765625, "p90": 240.48591156005858, "max": 390.423095703125, "pos_frac": 0.7421875, "sample": [246.83526611328125, 7.668682098388672, 79.7945556640625, 134.532958984375, 27.22265625, -3.30206298828125, -5.107452392578125, 316.2249450683594, 7.0888671875, 226.96145629882812, -59.25677490234375, -91.18417358398438, 82.87493896484375, -54.362327575683594, -49.4488525390625, 145.91567993164062, 33.37590026855469, -159.1656951904297, 55.05401611328125, 133.86764526367188, 301.6327209472656, 225.3857421875, -40.28033447265625, -67.12776947021484, 156.73260498046875, 310.8720703125, 61.98736572265625, 189.96954345703125, 196.67300415039062, 345.12725830078125, -57.38182830810547, 181.1478271484375, -30.20306396484375, 117.71219635009766, 139.1859130859375, 26.469406127929688, 106.13428497314453, 244.34478759765625, 130.013427734375, 25.376907348632812, -24.894012451171875, 96.35658264160156, 51.711669921875, 161.5345458984375, 8.417427062988281, 247.90869140625, 2.3462295532226562, 121.28314208984375, -3.66571044921875, 15.929611206054688, 24.632537841796875, 0.809417724609375, -149.854736328125, -114.4490966796875, -35.63551330566406, 124.50732421875, 20.836181640625, 37.25663757324219, 57.19696044921875, 5.1835174560546875, -15.751754760742188, 183.85910034179688, 203.7237548828125, 273.2044982910156, 153.21456909179688, -3.0023193359375, -128.34637451171875, 390.423095703125, 81.30723571777344, 224.8485107421875, 74.36932373046875, -41.64149475097656, -15.98516845703125, 46.13214111328125, 50.65948486328125, 78.25701904296875, 44.35174560546875, 238.8321075439453, 12.356498718261719, 71.50503540039062, 167.01144409179688, 143.2642822265625, 235.78369140625, -65.28665924072266, 10.535600662231445, 107.23065185546875, 111.62655639648438, 268.77392578125, -124.34429931640625, 205.1448974609375, -128.41201782226562, 25.818603515625, 0.0, -139.99612426757812, 116.343017578125, 208.59149169921875, 168.89151000976562, 95.709716796875, -20.945205688476562, 232.30691528320312, 54.66650390625, 29.818618774414062, -2.8288612365722656, 10.451919555664062, -101.60076904296875, 94.04525756835938, 266.3522644042969, 12.701507568359375, 128.80221557617188, 64.84329223632812, 305.89581298828125, 19.17364501953125, 257.8057861328125, -66.15020751953125, 149.74656677246094, 99.806884765625, -56.15620040893555, -89.2518081665039, 126.79901123046875, 28.761978149414062, 103.83074951171875, 3.3096847534179688, 5.7232513427734375, 46.4884033203125, 168.91648864746094, 47.461395263671875, 216.79849243164062, -19.280792236328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000316.npy"}
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 74.09947204589844, "std": 99.27224731445312, "min": -208.31900024414062, "p10": -35.41374969482421, "median": 63.377357482910156, "p90": 207.68578491210937, "max": 312.94317626953125, "pos_frac": 0.7421875, "sample": [86.51138305664062, -90.4100341796875, 0.0, 10.223838806152344, 239.129150390625, -49.930076599121094, 249.47772216796875, 188.36724853515625, 40.626243591308594, 139.51556396484375, 25.702911376953125, 65.86457824707031, 133.84722900390625, 6.2839202880859375, 161.01266479492188, -89.71380615234375, -10.198776245117188, 126.16415405273438, 126.35641479492188, 53.16405487060547, 174.85369873046875, 23.31121826171875, 184.53892517089844, 38.94753646850586, 6.171043395996094, 104.28369140625, 16.67998504638672, 0.3909473419189453, -8.96746826171875, 4.467437744140625, -18.530649185180664, 84.4769287109375, 282.9656677246094, -22.682159423828125, -3.8229827880859375, 70.76571655273438, 154.0068359375, 197.76869201660156, 91.78883361816406, -114.70184326171875, 89.28761291503906, -34.02665710449219, 306.22637939453125, -47.489044189453125, -38.650299072265625, 238.86419677734375, 180.01300048828125, 193.74545288085938, 58.04259490966797, 0.0, 10.373405456542969, 85.48724365234375, 175.18914794921875, 253.06585693359375, -78.85417175292969, -25.99169921875, 60.89013671875, 76.90115356445312, 81.03257751464844, 6.50140380859375, 39.625030517578125, 171.61990356445312, 211.35264587402344, 7.567584991455078, 0.0, 28.02215576171875, 59.43963623046875, 69.52241516113281, 103.50715637207031, 215.014404296875, 87.93289184570312, -18.357666015625, -20.496963500976562, -208.31900024414062, 270.5833740234375, 312.94317626953125, 271.6671142578125, -59.50518798828125, 38.984130859375, -47.80940628051758, 78.19644165039062, 174.10353088378906, 136.7642822265625, 88.1393051147461, 36.16796875, 30.49578857421875, 73.85948944091797, -26.02471923828125, 153.93301391601562, 114.30143737792969, -2.4229583740234375, 159.76992797851562, 148.2777099609375, 46.92645263671875, 101.75051879882812, -0.4027385711669922, -0.99151611328125, 97.92691040039062, 135.6575469970703, 177.4322509765625, -51.594818115234375, 149.0037841796875, 214.16888427734375, 157.93118286132812, -178.77883911132812, 60.762725830078125, 58.78118896484375, 113.72076416015625, -0.4010906219482422, -5.53558349609375, 115.74967956542969, 25.962265014648438, 162.662353515625, 14.591323852539062, 103.42660522460938, 135.726318359375, 24.999595642089844, -1.2957401275634766, 245.79983520507812, 23.888099670410156, 11.562089920043945, 71.61671447753906, -54.210968017578125, 158.08197021484375, 18.96851348876953, 177.73622131347656, -1.1739349365234375, 206.11427307128906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000317.npy"}
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 88.67483520507812, "std": 108.14126586914062, "min": -139.36953735351562, "p10": -46.68524169921875, "median": 89.23200988769531, "p90": 253.33691864013673, "max": 353.51483154296875, "pos_frac": 0.7578125, "sample": [114.57005310058594, 115.72576904296875, 55.095909118652344, 165.26600646972656, 12.973602294921875, 176.23287963867188, 25.28802490234375, 135.15634155273438, 353.51483154296875, 270.1286926269531, 23.882293701171875, -59.55092239379883, 283.4648742675781, 60.86648178100586, 253.16424560546875, -28.32269287109375, 140.36026000976562, 15.431976318359375, 75.23545837402344, 36.489288330078125, -26.16082763671875, -91.96759033203125, 102.83676147460938, 144.127197265625, -31.124603271484375, 63.5875244140625, 189.77484130859375, 127.63790893554688, -3.4344635009765625, 187.36749267578125, 106.47808837890625, 20.984024047851562, -6.219482421875, 32.71458435058594, -43.29954528808594, -36.55596923828125, 6.554126739501953, 144.82241821289062, 297.76202392578125, 171.7813720703125, 81.02389526367188, 183.91094970703125, 269.6770324707031, 137.2230224609375, 13.977127075195312, 8.68310546875, -138.61419677734375, -4.5026092529296875, -0.356231689453125, 114.492431640625, -64.7889404296875, 127.24586486816406, 114.84306335449219, 156.631591796875, 146.83978271484375, 187.64419555664062, -0.5241317749023438, 189.93983459472656, 253.7398223876953, -8.2623291015625, 84.06646728515625, -7.55950927734375, 281.31072998046875, 73.4735107421875, -75.50154113769531, 50.312744140625, 85.2435302734375, 62.580902099609375, 298.5324401855469, -105.13427734375, 120.13873291015625, 104.23446655273438, -17.61004638671875, 56.36695098876953, 172.38546752929688, 184.44451904296875, -16.76983642578125, 103.67652893066406, 110.58221435546875, 15.06109619140625, 93.22048950195312, 34.448272705078125, 260.26116943359375, 138.8760223388672, 93.76899719238281, -22.1712646484375, 150.92822265625, 250.28375244140625, -60.548484802246094, -23.6865234375, 6.40740966796875, 170.31179809570312, 75.56060791015625, 116.81761932373047, 151.13507080078125, 82.34832763671875, 199.42608642578125, 38.302947998046875, 282.4877624511719, -46.556610107421875, 127.59626007080078, 229.46795654296875, 306.65966796875, 100.05410766601562, -82.69607543945312, 247.0177001953125, 100.0084228515625, 39.22418212890625, 10.735427856445312, -57.473602294921875, 256.4057312011719, 71.01483154296875, 145.341064453125, 114.537841796875, 9.015350341796875, 248.9809112548828, 129.47860717773438, 267.7750244140625, -46.985382080078125, 175.4354248046875, -139.36953735351562, -50.239990234375, 30.940780639648438, 120.81927490234375, 158.01197814941406, 46.080596923828125, -130.37741088867188, 0.0], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000318.npy"}
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 78.45974731445312, "std": 101.38890838623047, "min": -204.8463134765625, "p10": -35.702758789062486, "median": 73.78475189208984, "p90": 201.95439910888672, "max": 360.8902893066406, "pos_frac": 0.7578125, "sample": [154.5484619140625, -20.754053115844727, 187.6219482421875, -4.164520263671875, 60.546775817871094, 265.2326354980469, 208.417724609375, -49.39935302734375, 319.8324279785156, 153.31170654296875, -18.12701416015625, -5.472381591796875, -15.00347900390625, 151.49331665039062, 172.2415771484375, 218.16970825195312, -21.9853515625, 0.0, -0.17645263671875, 103.09750366210938, 24.284000396728516, 107.13314819335938, 33.43910217285156, 9.798995971679688, 94.96389770507812, 185.01437377929688, 12.641754150390625, 11.537643432617188, 15.052886962890625, 92.29386138916016, 40.59039306640625, -22.04737091064453, 119.78659057617188, 47.63654327392578, 116.78823852539062, 195.64511108398438, -2.4784698486328125, 360.8902893066406, 147.57595825195312, 2.7403488159179688, 37.978515625, 93.19085693359375, 172.748779296875, -125.5513916015625, 34.57551574707031, 61.223236083984375, 105.52880859375, 98.828369140625, -23.53668212890625, 184.1669921875, 89.44915771484375, 219.25677490234375, 144.53150939941406, -75.81900024414062, 50.02099609375, 60.680877685546875, 40.140106201171875, -31.71991729736328, 126.88993835449219, 164.7364501953125, -186.72882080078125, 131.51878356933594, 68.9112548828125, -21.375946044921875, 109.96749114990234, -77.09425354003906, 219.07130432128906, -46.74250793457031, 147.89761352539062, 136.90231323242188, -43.76409149169922, 159.25799560546875, 200.87258911132812, 44.15850830078125, -9.53094482421875, -42.970001220703125, 80.5438232421875, 182.99282836914062, 271.5835266113281, -65.09356689453125, 103.30445861816406, 161.00515747070312, 199.8922119140625, 287.05108642578125, 68.75396728515625, 5.0064849853515625, 227.33123779296875, 289.64862060546875, -32.588226318359375, -46.238677978515625, 5.3154296875, 122.41201782226562, 13.002197265625, 29.17303466796875, 203.97361755371094, 13.845138549804688, 78.65824890136719, 201.08901977539062, -50.151397705078125, 168.933837890625, 49.37567138671875, -7.62359619140625, 163.5099334716797, 124.82534790039062, 43.846038818359375, 111.2552490234375, 152.7997589111328, 31.175186157226562, 163.91664123535156, -24.539981842041016, 106.98513793945312, -7.251958847045898, -132.9119873046875, 211.03363037109375, -204.8463134765625, 174.81103515625, 95.36927032470703, 129.8380126953125, 29.01055145263672, 25.179107666015625, 184.5401611328125, 37.832763671875, 105.31423950195312, 1.8129730224609375, 47.530487060546875, 87.821044921875, 52.08189392089844, 100.32320404052734], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000319.npy"}
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 73.4688720703125, "std": 126.57585906982422, "min": -241.19296264648438, "p10": -52.31007232666015, "median": 47.065101623535156, "p90": 233.5869171142578, "max": 407.49871826171875, "pos_frac": 0.765625, "sample": [52.591552734375, 13.951135635375977, -40.753631591796875, -241.19296264648438, 180.93218994140625, 196.564697265625, 6.40899658203125, 33.635894775390625, 278.1164245605469, 175.471923828125, 35.01483154296875, 130.541259765625, -32.15228271484375, 42.42633056640625, 233.09603881835938, 373.18634033203125, -39.2821044921875, 10.845829010009766, 10.15301513671875, 38.30602264404297, 6.133308410644531, 187.02438354492188, 234.7322998046875, -20.78582763671875, 37.22332763671875, -154.02490234375, 173.25628662109375, 174.23997497558594, 11.237945556640625, -28.61516761779785, 154.87841796875, 137.939208984375, 241.6568603515625, 84.8917236328125, 29.74078369140625, 174.59014892578125, 10.383697509765625, 99.02207946777344, 76.966796875, 3.7993087768554688, -0.6743221282958984, 145.92616271972656, 161.20114135742188, -188.48251342773438, 230.54046630859375, 292.65106201171875, 45.61077880859375, 97.28189086914062, 66.99468994140625, 180.04220581054688, -58.659942626953125, -3.793609619140625, -83.30816650390625, 2.740234375, 197.2977752685547, -45.092376708984375, 162.36004638671875, 29.709732055664062, 138.0171356201172, 154.04876708984375, 118.5693359375, 106.89141845703125, 36.905517578125, 233.01417541503906, 193.3961181640625, -32.8326416015625, 253.20751953125, -27.5269775390625, 254.0989990234375, 245.4852752685547, 34.442771911621094, 0.2786865234375, 166.40557861328125, 2.287628173828125, -220.39251708984375, 100.15156555175781, 27.033966064453125, 183.6483154296875, 65.843017578125, 197.2888946533203, 238.13937377929688, 9.478355407714844, -5.442216873168945, 45.01123046875, -41.91496276855469, 75.21815490722656, 34.65632629394531, 54.40948486328125, 141.74447631835938, -20.226951599121094, 86.60990905761719, 2.551513671875, 149.603271484375, -27.5567626953125, 26.687713623046875, -49.58869934082031, 61.3154296875, 23.150550842285156, -64.94497680664062, 189.834716796875, 371.3447265625, -77.17919921875, 225.4932861328125, 149.30911254882812, 19.3665771484375, 279.1553955078125, 185.01608276367188, -126.89289093017578, 18.949432373046875, -164.64389038085938, 0.9227752685546875, 185.91226196289062, -195.2865753173828, 33.54737091064453, 391.2248229980469, 72.37094116210938, 219.64157104492188, -33.2877197265625, 84.08552551269531, 76.03569030761719, -33.17264175415039, 46.91229248046875, 47.21791076660156, -184.6930389404297, 407.49871826171875, 6.180168151855469, -175.35272216796875, 90.84336853027344], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000320.npy"}
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 56.423927307128906, "std": 115.16559600830078, "min": -200.94114685058594, "p10": -65.92676239013672, "median": 26.677871704101562, "p90": 214.15672302246094, "max": 448.87109375, "pos_frac": 0.640625, "sample": [231.87869262695312, -132.90139770507812, 108.83757019042969, -186.282470703125, 11.2049560546875, 215.51162719726562, -21.477783203125, -171.7403564453125, -69.92440795898438, 203.96957397460938, 179.88632202148438, 35.56901550292969, -17.417984008789062, -27.927764892578125, 210.12144470214844, 100.52520751953125, -0.10822296142578125, 212.75120544433594, 280.6414794921875, 20.4603271484375, 29.1425724029541, -128.06155395507812, -10.428443908691406, -34.7674560546875, 255.8834228515625, 54.989898681640625, -11.460708618164062, 122.54263305664062, 80.06864929199219, -114.71109008789062, 149.58880615234375, 161.95611572265625, -7.997749328613281, -43.69921112060547, 138.53533935546875, -104.41262817382812, -52.849609375, 44.244285583496094, 73.47906494140625, 60.9285888671875, 149.4171600341797, 23.13916015625, -29.13275146484375, 27.412445068359375, -118.67657470703125, 244.67547607421875, 13.27508544921875, 102.72978210449219, 213.5760498046875, 3.8296661376953125, 87.07110595703125, 132.81637573242188, -24.090957641601562, 116.68038940429688, 12.093475341796875, -22.259490966796875, 205.90130615234375, -9.665828704833984, 104.32765197753906, -33.002777099609375, -11.693016052246094, -3.5465774536132812, 78.79450988769531, 160.2551727294922, 162.9788055419922, 218.96263122558594, 58.47515869140625, 25.94329833984375, 4.411956787109375, -101.63152313232422, -1.900787353515625, 98.1412353515625, -48.89292907714844, 154.03228759765625, -44.42643737792969, 62.76850891113281, 232.99130249023438, 448.87109375, 53.948699951171875, 85.91677856445312, 9.4345703125, -106.36846923828125, 72.63723754882812, 195.2086944580078, 2.877410888671875, 143.56173706054688, -19.911453247070312, 245.16455078125, -56.29059600830078, 151.5333251953125, -18.908672332763672, 122.01506805419922, -22.289588928222656, 204.46212768554688, -158.1558074951172, 354.69476318359375, -10.48388671875, 93.92239379882812, 236.26193237304688, 161.36065673828125, 18.571258544921875, -15.1138916015625, 13.877105712890625, 46.717864990234375, 2.16375732421875, 121.1488037109375, -22.404308319091797, 13.920583724975586, 17.215152740478516, -64.21348571777344, -93.75840759277344, -39.2042236328125, 21.02027702331543, -39.32627868652344, -47.35545349121094, 226.45303344726562, -200.94114685058594, 179.94735717773438, 61.197906494140625, 76.312255859375, 16.035400390625, 125.22604370117188, 157.1424560546875, -11.918495178222656, 273.072021484375, 94.354248046875, 0.0, 10.33197021484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000321.npy"}
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 70.656982421875, "std": 118.39958953857422, "min": -287.868408203125, "p10": -51.344992065429686, "median": 43.331695556640625, "p90": 228.68241271972656, "max": 420.912841796875, "pos_frac": 0.703125, "sample": [-287.868408203125, 35.167236328125, -20.928985595703125, 163.5545654296875, 32.638824462890625, 14.632080078125, 102.72370910644531, -93.3853759765625, 74.62425231933594, 362.6290283203125, -65.64193725585938, 24.395599365234375, 209.088623046875, 24.816650390625, 149.56307983398438, 186.39476013183594, 9.99169921875, 29.15643310546875, -1.2919235229492188, 177.25650024414062, 162.71453857421875, -80.05722045898438, 172.28768920898438, -39.369964599609375, -23.509506225585938, 192.1168212890625, -11.942596435546875, 106.05366516113281, 80.04020690917969, 122.43258666992188, 48.26158142089844, 160.8480224609375, 12.600799560546875, -39.664093017578125, 35.634986877441406, 51.024169921875, 108.18351745605469, 188.40847778320312, -88.78488159179688, 181.52798461914062, 143.11074829101562, 179.87652587890625, 37.261871337890625, -51.327789306640625, 37.7838134765625, -52.76702880859375, 68.46408081054688, -39.752899169921875, 256.4990539550781, -21.559484481811523, 55.286346435546875, 48.46759033203125, 220.27870178222656, 343.51690673828125, -39.63055419921875, 191.61138916015625, 39.775352478027344, -31.5072021484375, 80.68527221679688, 8.851272583007812, 182.50201416015625, -33.884552001953125, 230.88424682617188, 238.0277099609375, 21.626800537109375, 110.90776062011719, -51.3851318359375, 168.1337890625, 68.67161560058594, -6.5480499267578125, 233.425537109375, 302.60235595703125, 56.73114013671875, 97.02825927734375, 7.242462158203125, 0.0, -92.05096435546875, 85.347412109375, 40.74896240234375, -65.51577758789062, 302.7015075683594, -4.299442291259766, 25.97930908203125, 277.51202392578125, -44.22386169433594, 112.75103759765625, 173.1053466796875, -19.9879150390625, 147.407470703125, 101.00736999511719, 227.73876953125, -101.20053100585938, 3.7070465087890625, -100.3394775390625, -13.04254150390625, 14.656646728515625, -8.200042724609375, 45.9144287109375, -271.9463195800781, 140.3934326171875, 24.48248291015625, -20.0023193359375, 64.8507080078125, 152.3069610595703, 111.509033203125, -13.310256958007812, -15.359710693359375, 359.5490417480469, 272.20806884765625, 11.004020690917969, -29.17767333984375, 9.705673217773438, 76.84619140625, 79.55654907226562, -90.70034790039062, -4.66357421875, -12.248931884765625, 18.444091796875, 143.0162353515625, 27.576194763183594, 91.11911010742188, 282.2272033691406, 162.22572326660156, 38.17822265625, 35.99932861328125, 420.912841796875, 132.01608276367188, 170.44549560546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000322.npy"}
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 49.38531494140625, "std": 116.84378814697266, "min": -208.416015625, "p10": -91.08151550292969, "median": 37.195655822753906, "p90": 191.03461608886718, "max": 524.0745849609375, "pos_frac": 0.6640625, "sample": [32.72815704345703, 40.21710205078125, -189.9241943359375, -36.445465087890625, 158.0855712890625, 77.7525634765625, 38.95751953125, -54.82292938232422, 132.6396484375, 242.08258056640625, -126.75674438476562, 175.82518005371094, 61.80517578125, -43.1441650390625, -39.7318115234375, 116.23413848876953, 195.73228454589844, -65.63961791992188, -89.53683471679688, -90.98321533203125, 188.02554321289062, 162.0205078125, -91.31088256835938, 119.51513671875, 73.68135070800781, 281.1305236816406, 51.60009765625, -29.526531219482422, -174.27940368652344, 121.15927124023438, -38.96630859375, -123.3653564453125, -16.05803680419922, 123.47611236572266, 2.9281539916992188, 115.5032958984375, 55.33002471923828, -0.7954826354980469, 111.545166015625, 133.6042938232422, 54.623565673828125, 151.82846069335938, 26.683746337890625, 297.3876953125, 201.2183837890625, 5.4734344482421875, -39.28138732910156, -115.99462890625, 80.51361083984375, -6.715850830078125, 9.106142044067383, 32.498863220214844, -134.37696838378906, 22.537433624267578, 83.9569091796875, 59.232940673828125, 72.794921875, 208.72242736816406, 182.28045654296875, -17.817581176757812, 152.19464111328125, 72.20159912109375, 166.61619567871094, -33.859100341796875, 191.02197265625, 14.541427612304688, 92.60824584960938, -57.598880767822266, -106.06060791015625, 141.2593994140625, 58.120025634765625, 19.532983779907227, 115.72161865234375, 20.83782958984375, 198.1517791748047, 96.69525146484375, 181.71197509765625, 0.8688201904296875, 79.94973754882812, -20.01654052734375, 191.06411743164062, -165.9365234375, 73.36015319824219, -53.90875244140625, 111.26809692382812, 297.8313293457031, -208.416015625, 134.1195068359375, -101.49530029296875, 27.8523006439209, 153.36732482910156, 126.01779174804688, 35.43379211425781, 50.37379455566406, 75.46926879882812, -153.07444763183594, 39.306854248046875, 3.7407684326171875, 121.0201416015625, -54.890472412109375, -89.50048828125, 18.26092529296875, -58.790283203125, 83.813232421875, -34.9764404296875, -30.471206665039062, 10.733993530273438, -23.848159790039062, 95.0389404296875, 524.0745849609375, 7.854423522949219, -34.61372375488281, 141.37416076660156, 331.28485107421875, -66.00786590576172, -102.79861450195312, 29.461410522460938, 20.426605224609375, 43.70075225830078, -12.6617431640625, -9.078161239624023, 18.68878173828125, -28.57659912109375, 242.59817504882812, 174.39334106445312, -73.57553100585938, 31.62468719482422, 250.92282104492188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000323.npy"}
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 82.51322937011719, "std": 104.56196594238281, "min": -246.04135131835938, "p10": -26.267771911621093, "median": 75.24686431884766, "p90": 225.38316192626948, "max": 324.71746826171875, "pos_frac": 0.796875, "sample": [324.71746826171875, 293.088623046875, 21.02679443359375, 140.99281311035156, 40.45135498046875, 143.962158203125, 20.2169189453125, 140.34840393066406, 207.2079620361328, 92.7918701171875, 28.47308349609375, 0.960784912109375, -12.962493896484375, 242.724609375, 189.15353393554688, 131.06991577148438, 97.64419555664062, 26.037090301513672, -246.04135131835938, 0.0, 35.7177734375, -3.94464111328125, 136.06854248046875, 12.430423736572266, 24.531803131103516, 147.15493774414062, 69.78746032714844, -173.7364959716797, 89.21282958984375, -57.72224426269531, 0.7021560668945312, 112.85516357421875, 262.830810546875, 256.3094482421875, 31.82263946533203, 214.7621307373047, -0.3385467529296875, 111.894287109375, 115.87310791015625, 0.0, -22.36058807373047, 118.7855224609375, 283.8788146972656, 152.93344116210938, 41.062774658203125, 143.986572265625, 0.0, -18.625411987304688, 149.76251220703125, -122.18930053710938, -26.327682495117188, 109.3913345336914, 15.676498413085938, 7.44268798828125, 54.84893798828125, 41.32441711425781, 81.79812622070312, 145.20779418945312, 212.33221435546875, 155.07281494140625, -37.09196853637695, 297.83935546875, 19.76495361328125, 199.14430236816406, -3.4783554077148438, -17.465635299682617, 189.003662109375, 120.26986694335938, 43.323394775390625, -66.79987335205078, 187.53729248046875, 24.735336303710938, -55.30830383300781, 161.36990356445312, -26.242095947265625, 50.667205810546875, 28.724536895751953, 50.75506591796875, 79.41185760498047, 276.62493896484375, 298.135986328125, 1.9414253234863281, 85.3995361328125, 52.610504150390625, 220.921142578125, 267.47314453125, 133.03533935546875, 188.11785888671875, 21.3480224609375, -37.18951416015625, 202.5906219482422, 146.21389770507812, -84.17789459228516, 69.5008544921875, 8.958885192871094, 101.69439697265625, 179.66448974609375, 95.857421875, 97.5531005859375, -52.57781982421875, -81.72335815429688, -17.543399810791016, 31.517532348632812, 7.196723937988281, -66.243408203125, 163.47247314453125, 235.79454040527344, 187.46343994140625, 3.2257080078125, 18.98394775390625, 272.47589111328125, 53.356353759765625, 30.396804809570312, 82.0479736328125, -7.916473388671875, 252.8750762939453, 191.171142578125, 187.91018676757812, 3.051572799682617, 96.68862915039062, 78.50401306152344, 1.0542755126953125, 22.533935546875, 177.37832641601562, 80.71916198730469, 71.98971557617188, 82.4056396484375, 90.97286987304688], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000324.npy"}
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 72.10031127929688, "std": 127.26569366455078, "min": -333.7734375, "p10": -54.158413696289045, "median": 43.25399398803711, "p90": 228.87807006835936, "max": 517.71728515625, "pos_frac": 0.71875, "sample": [98.58819580078125, 103.4549560546875, -45.02162170410156, -93.6378173828125, 95.24591064453125, 32.71356201171875, 15.189849853515625, 246.7691650390625, 97.76617431640625, 33.337066650390625, 30.05975341796875, -2.506561279296875, 200.1009063720703, 149.50106811523438, 83.78201293945312, 28.00656509399414, 178.98585510253906, 211.83416748046875, 117.78656005859375, 410.0615234375, 149.3845672607422, 316.8751525878906, -3.6320953369140625, 3.8672542572021484, 166.48477172851562, 108.13653564453125, 18.090591430664062, 98.69183349609375, 275.8744812011719, 163.10818481445312, 43.72148895263672, -50.213470458984375, -5.447307586669922, 73.02760314941406, 100.93157958984375, 104.06687927246094, 152.9940643310547, 273.00482177734375, 35.61692810058594, 0.3126220703125, 19.881790161132812, -2.025390625, 218.1338653564453, 17.745079040527344, -333.7734375, 1.587188720703125, 305.25640869140625, -232.0767059326172, 216.69537353515625, -9.8211669921875, 15.115066528320312, 21.869537353515625, 171.425537109375, 38.589141845703125, 33.64002990722656, 70.36167907714844, -36.90220642089844, -81.26385498046875, -27.64837646484375, -117.30291748046875, 8.879791259765625, 218.6458740234375, -3.3487548828125, -31.248199462890625, 93.0862045288086, 69.0452880859375, 46.912628173828125, -38.03973388671875, 228.0760498046875, -63.36328125, 42.7864990234375, 138.64324951171875, 517.71728515625, 47.32948303222656, -33.118408203125, 24.543304443359375, -6.0405426025390625, 5.520927429199219, 18.81768798828125, 83.11880493164062, -64.12716674804688, -97.19140625, 15.92193603515625, 368.1219482421875, 24.586990356445312, 120.7767333984375, 132.60968017578125, 230.74945068359375, 164.9769287109375, -74.80410766601562, -6.8680267333984375, 81.80110168457031, 268.565673828125, 21.07220458984375, 2.5079479217529297, -219.75186157226562, -38.24701690673828, 220.39691162109375, 26.616912841796875, -33.70996856689453, 210.9263916015625, 47.11334228515625, 111.73220825195312, 199.80697631835938, 178.21578979492188, -19.852691650390625, -20.233659744262695, 147.23715209960938, 44.159820556640625, -6.1680908203125, -109.08224487304688, 213.87530517578125, 156.96957397460938, -6.16693115234375, 3.2864646911621094, 197.01608276367188, 193.57586669921875, 289.57763671875, -154.68557739257812, 61.74889373779297, 27.91217041015625, -32.843231201171875, 235.80343627929688, -23.844696044921875, -82.47091674804688, 263.5330505371094, 84.728759765625, 198.6007537841797], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000325.npy"}
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 72.38761901855469, "std": 112.9830093383789, "min": -219.32687377929688, "p10": -67.72174987792968, "median": 73.45848846435547, "p90": 213.73300933837888, "max": 423.07867431640625, "pos_frac": 0.7578125, "sample": [70.45040893554688, 196.68421936035156, 22.991729736328125, 423.07867431640625, 35.92372131347656, 179.94442749023438, -69.67086791992188, 17.148834228515625, 164.784423828125, -143.22979736328125, 225.6904296875, 45.906158447265625, 43.86627197265625, 156.04873657226562, 81.55530548095703, -28.490516662597656, 28.398998260498047, 297.1148681640625, 304.86212158203125, 104.92240905761719, 104.62518310546875, 225.54791259765625, 70.80760192871094, -101.27792358398438, 167.85275268554688, 11.823745727539062, 220.29759216308594, 183.54107666015625, 20.062850952148438, 70.91505432128906, -146.8824005126953, 27.61077880859375, 76.00192260742188, 179.03262329101562, -66.88641357421875, -82.66635131835938, 31.251049041748047, 48.96312713623047, -16.98602294921875, 142.88250732421875, 79.12518310546875, 11.598121643066406, -0.23687744140625, 188.58206176757812, 43.070831298828125, 101.19149017333984, -157.98829650878906, 168.18841552734375, -26.86181640625, 289.6169128417969, 138.613525390625, 94.40951538085938, 321.3141174316406, 142.8109893798828, 142.99127197265625, -55.596397399902344, -60.34918212890625, 19.041748046875, 157.85321044921875, -16.131011962890625, 101.00323486328125, 104.92799377441406, 47.5565185546875, 150.5388641357422, 0.0, -114.21572875976562, 25.407028198242188, 2.94659423828125, 18.56066131591797, 2.16754150390625, -12.336380004882812, 124.86459350585938, 152.62628173828125, 181.86062622070312, 76.09529113769531, -36.787750244140625, 44.67680358886719, -17.954421997070312, 110.79448699951172, 128.13723754882812, 3.880828857421875, 264.30120849609375, -27.52532958984375, 80.00381469726562, 116.60819244384766, 118.01626586914062, -117.94827270507812, -27.32830810546875, 177.19873046875, 1.302255630493164, 88.17476654052734, 35.34912109375, 148.6072998046875, 7.541748046875, 61.410255432128906, 63.3133544921875, 106.14790344238281, 35.53436279296875, 6.880859375, 94.69589233398438, 91.53111267089844, 0.0, 133.5986328125, 93.75291442871094, 113.32199096679688, 152.34165954589844, 32.604095458984375, 138.57289123535156, 129.77552795410156, 0.0, 147.09259033203125, -219.32687377929688, -80.17266845703125, -50.01488494873047, 253.07427978515625, -139.4754638671875, 194.25918579101562, -178.8876953125, 210.91961669921875, 152.84234619140625, 123.6463394165039, 22.48046875, -28.29681396484375, -134.83973693847656, 227.24537658691406, 266.3243103027344, 264.3441162109375, 116.1265869140625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000326.npy"}
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 80.97723388671875, "std": 113.66191864013672, "min": -190.62860107421875, "p10": -48.513616943359374, "median": 65.51426696777344, "p90": 220.89046936035155, "max": 529.80615234375, "pos_frac": 0.7890625, "sample": [38.007606506347656, 155.88565063476562, 40.3522834777832, 26.370895385742188, 220.56515502929688, -107.72097778320312, 188.07763671875, 45.327667236328125, 54.1376953125, 148.53018188476562, 56.4415283203125, 198.37179565429688, 10.90606689453125, -190.62860107421875, 58.92945098876953, 126.00933837890625, -30.45648193359375, 16.44287109375, 222.3829345703125, 255.3887481689453, 52.979713439941406, 159.87461853027344, 225.3601531982422, 43.43388366699219, 19.2064208984375, -49.660003662109375, 102.03302001953125, 193.12783813476562, 18.828277587890625, 31.73865509033203, 53.6221923828125, 158.89642333984375, -130.3909912109375, 125.41705322265625, 116.23944091796875, 205.4483642578125, 148.04452514648438, 48.99517822265625, 352.8396911621094, 12.318939208984375, -7.0479736328125, 9.264484405517578, -52.211822509765625, 187.30833435058594, 68.6830062866211, 49.765777587890625, 200.18914794921875, 73.78726196289062, 187.34136962890625, 24.875701904296875, 199.822998046875, -2.873443603515625, -20.579910278320312, -129.7120361328125, 34.97272491455078, 22.134552001953125, 233.6597900390625, 141.5995330810547, 174.69284057617188, 122.1510009765625, -39.85997772216797, 137.25979614257812, 104.64108276367188, 119.62384033203125, 248.57235717773438, 214.80535888671875, 265.2337341308594, 36.4093017578125, 104.95562744140625, -60.433868408203125, 63.312347412109375, -27.68365478515625, -3.850494384765625, -104.74485778808594, -46.14485168457031, 13.867324829101562, 103.95695495605469, 131.9188232421875, 266.1856689453125, 66.55645751953125, -20.558380126953125, 29.6170654296875, 94.39605712890625, 125.31094360351562, 72.33952331542969, 200.36160278320312, -24.4521484375, 0.0, 68.49786376953125, 28.884109497070312, 17.233306884765625, 529.80615234375, 271.6855163574219, 61.519073486328125, 92.20382690429688, 172.26898193359375, 139.66690063476562, 122.16818237304688, 312.0631103515625, 25.523086547851562, 110.13812255859375, 52.3489990234375, -124.30479431152344, -36.281349182128906, 69.5679931640625, 128.72206115722656, 96.24911499023438, 18.814720153808594, 70.27838134765625, 102.46762084960938, -187.29840087890625, 159.4661865234375, 221.6495361328125, 27.007293701171875, -114.65008544921875, 188.84921264648438, -59.53001403808594, 214.82742309570312, 118.72838592529297, -33.42478942871094, 347.06536865234375, 113.55998229980469, 59.40318298339844, -48.022308349609375, -93.84780883789062, 6.264251708984375, 39.951629638671875, 64.47207641601562], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000327.npy"}
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 73.3765869140625, "std": 106.4517593383789, "min": -185.59222412109375, "p10": -49.57692108154297, "median": 60.317291259765625, "p90": 205.85649719238282, "max": 378.7660827636719, "pos_frac": 0.7421875, "sample": [15.04595947265625, 56.840545654296875, 150.86593627929688, 100.25128173828125, 215.24114990234375, 47.796630859375, 1.9862060546875, 220.80787658691406, 184.1325225830078, 73.6404800415039, 17.6414794921875, 6.6178741455078125, 44.6756477355957, 105.2752685546875, 110.7816162109375, -4.393566131591797, 199.9488525390625, 15.466796875, 48.42230224609375, 119.24620819091797, 105.56523132324219, 24.482585906982422, -4.714263916015625, -14.90716552734375, 147.95565795898438, 122.31060791015625, -50.04643249511719, 96.9605712890625, 112.88494873046875, 124.43059539794922, -46.1077880859375, 195.4996337890625, 40.76275634765625, 205.18899536132812, 316.91595458984375, 109.70018768310547, 81.34895324707031, 14.596847534179688, 378.7660827636719, -3.9208221435546875, 55.26239013671875, 179.8481903076172, 120.62158203125, -35.93170166015625, -12.744918823242188, -93.63517761230469, -33.60076904296875, 97.35110473632812, -49.375701904296875, 101.66461181640625, 25.493896484375, 57.91552734375, -90.63296508789062, 277.17327880859375, 76.85635375976562, 84.255126953125, 9.5567626953125, 32.94708251953125, 52.215904235839844, -31.571365356445312, -13.932281494140625, -35.671348571777344, 126.18118286132812, 161.45556640625, 192.01495361328125, 281.16949462890625, 196.84161376953125, -185.59222412109375, 79.67132568359375, 16.84484100341797, 12.848068237304688, 54.03070068359375, 178.83248901367188, 164.69256591796875, 160.01861572265625, 317.2319641113281, 234.02316284179688, -25.092437744140625, -85.0709228515625, 62.851318359375, -64.05487060546875, 6.27264404296875, 117.06475830078125, 81.57403564453125, -74.07830810546875, 26.43585205078125, 105.93504333496094, -0.1499481201171875, 150.5386199951172, -174.77847290039062, -42.350616455078125, 165.25927734375, 27.79782485961914, 42.81024932861328, 49.8721923828125, -68.98330688476562, 162.39968872070312, 153.65965270996094, -146.19090270996094, 152.336181640625, 147.1534881591797, 69.82992553710938, 206.19635009765625, 15.826171875, -79.39556121826172, 338.80255126953125, 260.1856689453125, -50.94940185546875, 15.076095581054688, -22.517288208007812, 76.05880737304688, 62.71905517578125, 167.48916625976562, -18.38720703125, 46.3330078125, -35.24079895019531, 42.66664123535156, 198.61959838867188, -15.92197036743164, -78.27841186523438, 17.905914306640625, 80.2171859741211, 245.37869262695312, -3.652740478515625, 82.13946533203125, 241.35272216796875, 205.71084594726562, 100.5339584350586], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000328.npy"}
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 81.00325012207031, "std": 135.79881286621094, "min": -223.0411376953125, "p10": -76.89131164550781, "median": 57.7718505859375, "p90": 279.8712890625, "max": 471.0093994140625, "pos_frac": 0.7265625, "sample": [-132.89385986328125, 16.958847045898438, -117.83245849609375, 221.2529296875, 34.44464874267578, 97.63710021972656, 41.31773376464844, 105.95938110351562, -22.07817840576172, 128.65899658203125, 91.75289916992188, 33.661277770996094, -184.1566162109375, -28.448516845703125, 28.817398071289062, 205.24111938476562, -8.3985595703125, -76.92303466796875, 113.58439636230469, 58.264862060546875, -87.26065063476562, 445.2135314941406, 97.2108154296875, 96.57574462890625, 221.0880126953125, 202.45999145507812, -87.02375793457031, 83.17315673828125, 296.21343994140625, 59.13311767578125, 77.25929260253906, 32.3587646484375, -72.183349609375, 176.70687866210938, -71.85723876953125, 37.4423828125, 167.5887451171875, 118.1446533203125, 57.278839111328125, 119.91983032226562, 228.072021484375, 471.0093994140625, 141.703369140625, -58.31512451171875, 72.11370849609375, 25.5887451171875, -22.675308227539062, 282.16790771484375, -4.932098388671875, -223.0411376953125, -83.4494857788086, 310.95477294921875, -83.284912109375, 28.635040283203125, 302.96197509765625, 123.92431640625, 68.04145812988281, 201.5150146484375, 66.65868377685547, -76.87771606445312, 0.5567207336425781, 206.2838134765625, 66.880126953125, -107.70677185058594, 14.379905700683594, -14.582275390625, 15.367889404296875, 47.428009033203125, 204.902587890625, 136.68565368652344, 36.933349609375, -6.76995849609375, 111.66641235351562, 433.7532958984375, 19.637527465820312, 95.39761352539062, 183.0850830078125, -52.908233642578125, -1.7627849578857422, 120.778564453125, 186.71994018554688, 111.8157958984375, -181.88397216796875, 128.75540161132812, 6.93157958984375, 324.4866943359375, 224.93698120117188, 3.229217529296875, 155.75082397460938, 165.72686767578125, 241.9854736328125, -6.933929443359375, 15.637161254882812, 457.67791748046875, 161.34815979003906, 5.14208984375, 319.2967224121094, 278.88702392578125, -2.0270233154296875, 54.9228515625, 22.475875854492188, 32.66636657714844, 145.9112548828125, 29.722259521484375, -26.70635986328125, -17.03166961669922, 318.77117919921875, 131.42343139648438, 238.89988708496094, 377.735107421875, -30.650772094726562, 105.55859375, 153.60853576660156, -125.33645629882812, 312.10382080078125, 18.854446411132812, 75.38131713867188, -36.128662109375, 20.444046020507812, -1.9644775390625, -60.104957580566406, 14.401748657226562, 17.337596893310547, 38.85002136230469, 185.5931396484375, 77.81303405761719, -138.4466552734375, -22.21172332763672], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000329.npy"}
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 72.27506256103516, "std": 112.5998306274414, "min": -220.94631958007812, "p10": -45.47237091064453, "median": 63.882225036621094, "p90": 205.07160339355468, "max": 463.14068603515625, "pos_frac": 0.734375, "sample": [-23.805999755859375, 140.95968627929688, 23.064773559570312, -220.94631958007812, 142.56033325195312, 98.12228393554688, -32.619049072265625, 372.01409912109375, 190.45718383789062, 83.66384887695312, 26.751708984375, 65.819091796875, -62.114288330078125, -177.46453857421875, 114.6708984375, 55.65055847167969, 388.72332763671875, 124.45306396484375, 150.5552978515625, 43.101776123046875, 140.43309020996094, -34.91313171386719, -129.7139129638672, -10.17462158203125, -43.81306457519531, 176.931396484375, 33.5191650390625, 21.529630661010742, 112.0680160522461, -7.9587554931640625, 95.49949645996094, 109.19337463378906, 155.2567138671875, 166.48171997070312, 49.059288024902344, 44.91064453125, -9.20458984375, 41.93072509765625, -32.16961669921875, 179.74569702148438, 210.766357421875, 190.5299072265625, 134.01483154296875, 92.30859375, -17.781539916992188, 73.81201171875, 11.532331466674805, 276.28985595703125, 48.242210388183594, 155.73974609375, -5.90203857421875, 113.6739730834961, 147.19400024414062, -151.73641967773438, 130.35623168945312, 38.50166320800781, 90.69229125976562, 55.993194580078125, 172.88241577148438, 49.54658508300781, 73.94720458984375, 200.69300842285156, 24.421218872070312, 205.52664184570312, -129.19537353515625, 55.683685302734375, 105.39286804199219, -11.04730224609375, 11.72113037109375, -14.55279541015625, 66.126953125, -213.0225830078125, 230.15567016601562, 104.23944091796875, 0.83331298828125, 118.70758056640625, -53.52454376220703, 200.42327880859375, -69.21536254882812, -37.01361083984375, 76.041748046875, 222.72933959960938, 61.94535827636719, 31.723052978515625, 6.5712890625, 133.06044006347656, 68.38531494140625, 276.6984558105469, 287.33135986328125, 40.77837371826172, 179.03836059570312, 119.7685546875, 42.34107971191406, 73.6820068359375, 32.318359375, -5.3130035400390625, 110.443359375, 8.119733810424805, -1.00384521484375, 163.3806610107422, 204.8765869140625, 68.09613037109375, -80.79660034179688, -1.0342025756835938, -50.938385009765625, 255.24557495117188, 131.5406494140625, 94.68832397460938, -9.131610870361328, 43.037933349609375, 146.09188842773438, -16.11074447631836, 129.475341796875, -36.75177001953125, 20.18896484375, 233.28076171875, 115.81399536132812, 28.123817443847656, 0.446502685546875, 283.6502380371094, -15.014617919921875, 463.14068603515625, -74.23341369628906, -41.161376953125, -49.344085693359375, 162.79010009765625, 0.4601764678955078, 67.55145263671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000330.npy"}
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 73.2829360961914, "std": 130.75218200683594, "min": -290.9646301269531, "p10": -85.14270782470703, "median": 66.22596740722656, "p90": 244.84868927001952, "max": 599.9236450195312, "pos_frac": 0.7578125, "sample": [29.160888671875, 230.2099609375, 187.72398376464844, -186.54771423339844, 248.53399658203125, 70.32920837402344, 0.488525390625, 69.50225830078125, 84.20956420898438, 43.190826416015625, 65.20870971679688, -29.468353271484375, 143.2823486328125, -18.021392822265625, 599.9236450195312, -95.73350524902344, 97.7259521484375, 26.59050178527832, -126.46078491210938, -81.69204711914062, 45.07097625732422, 74.39923095703125, -162.24928283691406, 39.45030212402344, -34.91252899169922, 147.80462646484375, 85.71656799316406, 197.13348388671875, 68.4715576171875, 139.9287567138672, 18.58306884765625, 102.85662841796875, -94.00421142578125, 34.504669189453125, 81.24803161621094, 214.18118286132812, 134.63726806640625, -7.9901123046875, 198.10195922851562, 243.26927185058594, -72.74620819091797, 35.369842529296875, 104.98040771484375, -160.38034057617188, 71.33380126953125, 4.1973419189453125, 283.80267333984375, 82.12838745117188, -0.2872314453125, 0.0, 70.72625732421875, 5.7899322509765625, 83.783447265625, -25.31597900390625, 6.21759033203125, 133.9705047607422, 10.269927978515625, 176.27122497558594, 65.46487426757812, 74.4453125, 71.76806640625, 49.3831787109375, -151.03048706054688, -89.83734130859375, 249.8065185546875, 66.987060546875, 254.19802856445312, 104.3255386352539, 13.09466552734375, -30.241836547851562, 11.15484619140625, -11.7037353515625, 353.50494384765625, 92.17904663085938, 157.96836853027344, 5.79277229309082, -142.31773376464844, 63.18896484375, 28.208694458007812, -83.13072204589844, 2.3145751953125, 53.87554931640625, 234.10205078125, 81.00888061523438, -198.46224975585938, 188.7713623046875, 21.048431396484375, 272.045166015625, -91.06315612792969, 320.48681640625, 234.51309204101562, 121.7703857421875, 164.84141540527344, 377.146484375, 45.0267333984375, 172.36221313476562, 253.92230224609375, -4.553169250488281, -5.54046630859375, 144.8864288330078, 85.1116943359375, -49.5013427734375, 217.5477752685547, 161.17050170898438, -165.43545532226562, 108.23091125488281, 414.2651672363281, 0.9567966461181641, -11.670120239257812, 44.433712005615234, 41.66722106933594, 30.49169921875, 123.1328125, 128.97451782226562, 132.1295166015625, 152.3060302734375, -20.1514892578125, 278.2408447265625, 29.783870697021484, 155.62757873535156, -290.9646301269531, 110.49972534179688, 264.82098388671875, 133.74542236328125, 23.2327880859375, -19.51214599609375, 58.1759033203125, 6.7240142822265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000331.npy"}
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 74.4039306640625, "std": 128.8959503173828, "min": -289.17633056640625, "p10": -59.779273986816406, "median": 61.34946060180664, "p90": 239.0409393310547, "max": 479.3482360839844, "pos_frac": 0.7109375, "sample": [-10.641054153442383, 116.69384765625, 5.0597686767578125, 438.4559326171875, 40.878082275390625, 116.35673522949219, -58.561859130859375, 261.91864013671875, 116.831787109375, 260.7584533691406, 96.96670532226562, 20.052732467651367, 9.093963623046875, 122.20052337646484, 168.6566619873047, 26.6416015625, -88.05244445800781, -44.259033203125, 151.47943115234375, 38.56896209716797, 122.3038330078125, 51.76557922363281, 349.6333923339844, -1.1482658386230469, 126.2608871459961, 74.8616943359375, 8.581024169921875, 27.936859130859375, 20.219501495361328, 68.23746490478516, 29.1744384765625, 119.10693359375, -86.44281005859375, 11.4254150390625, 127.88009643554688, 136.25486755371094, 179.20343017578125, 250.54086303710938, 161.10861206054688, 197.5298614501953, 129.37173461914062, 238.11080932617188, 202.84808349609375, -6.985939025878906, 103.64810180664062, 108.8978271484375, 166.85879516601562, 241.21124267578125, -9.837173461914062, 49.802215576171875, 302.547607421875, 479.3482360839844, -84.65632629394531, -113.15716552734375, 191.322509765625, 117.0885238647461, -60.10552978515625, 60.411476135253906, 275.471435546875, -181.5537109375, 131.84658813476562, 172.71783447265625, 236.09828186035156, 3.4304733276367188, -59.26008605957031, 38.884246826171875, 106.16132354736328, 16.13519287109375, 4.11149787902832, -82.56063842773438, 4.7667236328125, 237.27947998046875, -59.63945007324219, 66.29051208496094, 60.186431884765625, 357.01513671875, 84.02517700195312, 170.15811157226562, -110.55931854248047, 38.501434326171875, -28.076187133789062, -19.265335083007812, -18.890335083007812, -29.773834228515625, 187.9248046875, -179.49195861816406, 97.07168579101562, 0.561279296875, 6.672119140625, 155.27748107910156, 89.90167236328125, 99.72527313232422, -42.87196350097656, 86.79376220703125, 183.719970703125, -45.97845458984375, 14.078628540039062, -147.59243774414062, -32.203826904296875, 82.0709457397461, 282.4941711425781, 228.50289916992188, -289.17633056640625, -131.06179809570312, 154.49560546875, 211.1357421875, 54.637420654296875, -27.573837280273438, 32.68980407714844, -12.278533935546875, 103.39193725585938, 62.287445068359375, 165.23583984375, 322.53564453125, 134.28976440429688, 106.32376098632812, -40.604827880859375, 140.00393676757812, -9.931447982788086, 160.8736572265625, 36.64068603515625, 335.97222900390625, -29.866867065429688, -52.620025634765625, 0.0, -17.44921875, -189.0255889892578, -57.636962890625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000332.npy"}
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 67.19111633300781, "std": 111.65773010253906, "min": -177.30047607421875, "p10": -47.352750396728496, "median": 44.17915344238281, "p90": 232.74513854980466, "max": 340.55316162109375, "pos_frac": 0.71875, "sample": [228.02951049804688, -149.82546997070312, 21.376720428466797, 325.0659484863281, 156.91207885742188, -177.30047607421875, 102.44366455078125, 140.3363037109375, -2.35601806640625, 248.27276611328125, -9.644241333007812, 25.46484375, -106.26336669921875, 199.72735595703125, 229.98776245117188, 47.6900634765625, 81.51795196533203, 34.955169677734375, 165.90850830078125, 115.55853271484375, -133.0501251220703, -37.247764587402344, -23.863128662109375, 150.1937255859375, 119.59805297851562, 85.66486358642578, 117.91494750976562, 94.40875244140625, 290.525634765625, 9.234283447265625, 25.990875244140625, 227.36170959472656, 28.287322998046875, 173.5120849609375, 55.51387023925781, 18.161449432373047, 255.1226806640625, -42.17327117919922, 221.78170776367188, 5.1541900634765625, 86.60488891601562, 33.813819885253906, 58.937652587890625, 8.58721923828125, 146.94586181640625, -14.42425537109375, 6.341848373413086, 43.429168701171875, 57.9140625, -58.660133361816406, -67.72021484375, 111.63534545898438, 0.0, 10.00946044921875, 2.28021240234375, 180.40634155273438, 25.9993896484375, 313.60382080078125, 70.22550964355469, 115.166015625, 14.8800048828125, 190.22132873535156, -75.15289306640625, 309.6726379394531, 58.830322265625, -42.50672912597656, 105.49786376953125, 206.713134765625, 41.94928741455078, 79.7830810546875, -164.65806579589844, 286.34906005859375, 181.97335815429688, -157.05850219726562, -8.965255737304688, 195.66793823242188, -31.030059814453125, 4.8482666015625, -37.2474365234375, 45.405120849609375, 59.09649658203125, 5.008491516113281, 23.199981689453125, 203.58941650390625, -59.506195068359375, 340.55316162109375, 44.92913818359375, -25.49371337890625, -13.48687744140625, 105.03567504882812, -15.357803344726562, 212.622802734375, -17.483352661132812, 132.5283966064453, 26.58929443359375, -27.495758056640625, 43.221099853515625, 48.480865478515625, 1.543497085571289, 57.366844177246094, 103.37783813476562, 139.11651611328125, -67.27093505859375, -30.785659790039062, 259.572998046875, 132.17080688476562, 32.50592041015625, 258.7756042480469, 95.77944946289062, 239.17901611328125, 34.974884033203125, -6.520259857177734, 245.14056396484375, -123.69287109375, 22.646888732910156, -5.32818603515625, -79.65565490722656, -5.138282775878906, -29.4737548828125, 86.1328125, 107.50323486328125, -39.21923828125, -28.449417114257812, 41.797889709472656, 3.187225341796875, 248.64898681640625, 114.88467407226562, 49.442405700683594], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000333.npy"}
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 75.2254409790039, "std": 119.54232025146484, "min": -243.01791381835938, "p10": -62.14653778076172, "median": 67.69905090332031, "p90": 231.56400451660153, "max": 356.3311767578125, "pos_frac": 0.7578125, "sample": [210.46774291992188, 127.56259155273438, 31.96826171875, -92.74102783203125, 0.0, -48.77392578125, 24.173492431640625, -48.03181457519531, 30.428386688232422, -17.49615478515625, 66.72262573242188, 114.65455627441406, 256.1641845703125, 204.74615478515625, 51.086944580078125, 166.05279541015625, -54.13563537597656, 146.29315185546875, 62.218475341796875, -2.3267822265625, 143.1733856201172, 139.0733642578125, 237.205810546875, 143.83633422851562, -70.86492919921875, 258.7593994140625, 71.87161254882812, 136.7474365234375, -45.58977508544922, 15.360061645507812, 51.36480712890625, -34.419349670410156, 5.352550506591797, 114.0277099609375, 146.61952209472656, 70.16204833984375, 34.00353240966797, 97.16397094726562, -138.89889526367188, 229.14608764648438, -14.267829895019531, 31.596160888671875, 15.7860107421875, 1.140869140625, 59.03923034667969, 70.67095947265625, 23.91119384765625, -98.6895751953125, 160.65057373046875, -12.76641845703125, 114.74014282226562, -18.475860595703125, 324.2574462890625, 8.011962890625, 275.97344970703125, 154.37271118164062, -1.4195537567138672, 103.60968017578125, 23.2347412109375, 115.27529907226562, 61.559078216552734, 33.7747802734375, 252.77264404296875, 95.329833984375, -193.89598083496094, -61.78828430175781, 259.299560546875, -106.97174072265625, 29.729583740234375, 95.2362060546875, 68.67547607421875, 5.077423095703125, -86.40473937988281, 189.15512084960938, 214.3760986328125, 140.13804626464844, 72.98715209960938, 161.72174072265625, 9.403076171875, 71.1408920288086, 175.61297607421875, 85.0714111328125, -3.758056640625, 118.56375122070312, 161.0068359375, 335.6737060546875, 9.511062622070312, 163.85470581054688, 179.15869140625, -4.509552001953125, 228.65997314453125, 176.71929931640625, 356.3311767578125, -30.500228881835938, 53.79768371582031, 15.327102661132812, 89.80139923095703, 61.85600280761719, 305.77874755859375, 331.16949462890625, 185.00949096679688, 3.50628662109375, 32.486209869384766, -243.01791381835938, 129.99447631835938, -144.01535034179688, -62.09767150878906, 198.80886840820312, 75.18345642089844, 197.74700927734375, -185.54283142089844, 253.22512817382812, 181.19577026367188, -7.248939514160156, 14.60113525390625, -227.0059814453125, 28.8817138671875, 60.4166259765625, -145.22940063476562, 146.59783935546875, 52.079803466796875, 209.97653198242188, 117.331298828125, 121.9981689453125, 114.0718994140625, 4.117401123046875, -62.26055908203125, 287.8232421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000334.npy"}
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 75.95929718017578, "std": 126.23539733886719, "min": -209.75314331054688, "p10": -84.44908065795897, "median": 81.09173202514648, "p90": 216.61668548583975, "max": 435.9850158691406, "pos_frac": 0.7265625, "sample": [-58.1376953125, -209.75314331054688, 168.26258850097656, -149.74652099609375, 115.39764404296875, -8.756195068359375, 206.88510131835938, -0.31201171875, 163.91104125976562, 1.26055908203125, 113.82861328125, -162.80880737304688, 146.50262451171875, 29.12843132019043, 0.3623046875, 57.054931640625, 33.52745056152344, 0.5644912719726562, -45.37054443359375, -200.78404235839844, -52.5318603515625, 150.27130126953125, 4.0079345703125, 122.63298034667969, -18.294921875, -24.757904052734375, 83.28929138183594, 82.80816650390625, -127.94317626953125, 357.9503173828125, -149.2999267578125, 196.41098022460938, -56.875579833984375, 183.565673828125, 80.796142578125, 354.1923828125, 15.162490844726562, 288.09747314453125, 192.29214477539062, 103.7938003540039, 435.9850158691406, 165.60108947753906, 151.21844482421875, 34.60639953613281, 201.66278076171875, 247.45635986328125, -13.592674255371094, 187.6055145263672, 135.10696411132812, 239.32371520996094, -29.410003662109375, 142.88674926757812, 42.078590393066406, 142.92413330078125, 164.8993377685547, 184.6549530029297, 271.135498046875, 35.31280517578125, 41.37031555175781, 17.46612548828125, -4.606719970703125, 24.135101318359375, 5.8137054443359375, 99.79641723632812, 27.059722900390625, 145.91412353515625, -33.705238342285156, 253.1063232421875, 177.2818603515625, 132.07814025878906, 152.2373504638672, 24.777557373046875, 174.16470336914062, 26.05633544921875, 254.00595092773438, 185.82958984375, 25.988327026367188, -142.74664306640625, 54.67823028564453, 39.15734100341797, 11.960601806640625, 142.1243133544922, -133.17398071289062, 190.5260772705078, 386.52392578125, -80.65100860595703, -104.490234375, 32.25941467285156, 282.3524169921875, 79.11186981201172, -93.31124877929688, 23.444961547851562, 147.9922637939453, 27.191421508789062, -100.47260284423828, -28.232345581054688, 97.75731658935547, -6.4444580078125, 197.24066162109375, -80.25758361816406, 185.13458251953125, 89.23727416992188, 133.43161010742188, 190.93701171875, 239.74867248535156, 49.936553955078125, 0.0, 194.2371826171875, 201.9178466796875, 150.19149780273438, -69.7799072265625, 91.83103942871094, 124.20286560058594, -149.7996826171875, -36.571434020996094, 363.1278076171875, -5.24244499206543, -24.90857696533203, 81.38732147216797, 21.4085693359375, 115.26307678222656, 97.52580261230469, 100.65896606445312, -33.94708251953125, -128.47189331054688, 122.35671997070312, 96.4185791015625, 125.20698547363281], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000335.npy"}
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 75.75871276855469, "std": 135.4607391357422, "min": -224.9076690673828, "p10": -60.70408020019531, "median": 48.822364807128906, "p90": 245.42398681640617, "max": 713.7245483398438, "pos_frac": 0.7265625, "sample": [342.4547119140625, 47.731292724609375, 151.166259765625, 2.2984619140625, 28.308746337890625, 138.15478515625, -28.963653564453125, 141.50119018554688, 203.98922729492188, 163.2626495361328, -29.204986572265625, 100.12950134277344, 162.8511199951172, 35.46506881713867, 28.316326141357422, 73.50860595703125, 76.27684020996094, 167.87200927734375, 105.78995513916016, 331.74310302734375, 346.82318115234375, -34.205101013183594, 30.30310821533203, 174.63856506347656, -107.32559204101562, 20.227783203125, 24.470733642578125, 30.09722900390625, -0.293701171875, 89.3017578125, 77.02322387695312, 157.5918731689453, 16.314292907714844, 296.04345703125, 116.15902709960938, 276.0735778808594, 92.89871215820312, 344.17449951171875, 70.28528594970703, -10.22564697265625, -224.9076690673828, 171.52178955078125, -85.3516845703125, 138.59527587890625, 46.423095703125, -45.01603698730469, -53.724403381347656, 237.22265625, -145.37939453125, 11.637714385986328, 46.511253356933594, 36.07490539550781, 2.691162109375, 199.2530517578125, 202.19680786132812, -68.65155029296875, -7.00701904296875, 94.75630187988281, 66.79344940185547, -58.188926696777344, 4.814788818359375, 24.8299560546875, -203.6700439453125, 168.14016723632812, -30.175506591796875, 97.26812744140625, 85.95089721679688, -80.48678588867188, 66.14877319335938, 143.8776397705078, 88.62777709960938, 9.53274917602539, 83.29202270507812, 89.99337768554688, -7.8095855712890625, 13.90338134765625, -6.6776123046875, -7.437896728515625, 22.110824584960938, 9.036651611328125, 8.796934127807617, 67.68438720703125, 88.2239990234375, 7.9583740234375, 231.16650390625, 72.38157653808594, 126.61126708984375, 281.62445068359375, -26.259170532226562, 116.95843505859375, -51.67645263671875, -63.41876220703125, 192.47164916992188, 264.5604248046875, 572.6121215820312, 278.1058349609375, 316.27801513671875, -18.040603637695312, 64.07574462890625, 195.11358642578125, 63.82456970214844, 288.47406005859375, 40.9075927734375, -87.8656005859375, -62.6939697265625, 713.7245483398438, 45.204925537109375, 155.2308349609375, 218.11090087890625, -165.73025512695312, 78.86962127685547, 84.29976654052734, -55.48004913330078, 47.311302185058594, 49.91343688964844, -32.5355110168457, 228.07666015625, -59.979248046875, -136.28460693359375, 0.0, -3.23516845703125, 23.436676025390625, 53.375885009765625, 137.2696533203125, -62.395355224609375, 2.270233154296875, 23.474578857421875, -5.435920715332031], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000336.npy"}
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 78.75270080566406, "std": 115.86016845703125, "min": -202.15260314941406, "p10": -57.61214141845703, "median": 69.93878936767578, "p90": 224.49175567626952, "max": 416.694580078125, "pos_frac": 0.7734375, "sample": [73.694091796875, 43.87257385253906, 69.85060119628906, 161.57760620117188, 23.555156707763672, -57.53326416015625, 22.24720001220703, 224.50103759765625, 62.619873046875, 364.67205810546875, 49.394256591796875, 18.596826553344727, 30.97216796875, 129.21160888671875, 393.73388671875, 119.1339111328125, 84.31622314453125, 60.10920715332031, -143.59234619140625, 60.85322570800781, -141.32943725585938, 76.92000579833984, 43.68077087402344, -64.87791442871094, 163.28599548339844, 6.4134979248046875, -30.966064453125, 137.3843994140625, -2.1598739624023438, 227.09445190429688, 121.7667236328125, 132.67510986328125, 77.36669921875, 295.47503662109375, 201.159423828125, -30.725662231445312, 143.05264282226562, 28.39099884033203, 47.51153564453125, 100.70208740234375, -52.65217208862305, -111.64778137207031, 55.20863342285156, 127.35968017578125, 80.61343383789062, 291.9754943847656, 34.191925048828125, -46.1728515625, 5.5831298828125, 85.0303955078125, 26.18304443359375, 176.19151306152344, 92.33343505859375, 211.73716735839844, 101.66400146484375, -57.79618835449219, 60.8502197265625, 313.7718505859375, 29.41699981689453, 70.0269775390625, -59.723358154296875, 105.97557067871094, 81.18135070800781, 276.7110595703125, 173.29637145996094, -30.249542236328125, 205.34219360351562, 78.80716705322266, -97.89494323730469, -33.41094970703125, 236.755859375, 118.4588623046875, 71.05303955078125, 63.43017578125, 31.8055419921875, 16.068084716796875, -52.0462646484375, -202.15260314941406, 242.4105224609375, 37.13489532470703, -18.455167770385742, 204.0926513671875, 131.11904907226562, 64.33999633789062, 176.55868530273438, 213.18240356445312, 1.7308902740478516, 126.3795166015625, 22.725906372070312, 9.208856582641602, 73.29202270507812, 224.48777770996094, 55.87629699707031, 205.551513671875, -45.525386810302734, 162.04998779296875, 24.857635498046875, -61.835601806640625, 27.72848892211914, 111.05572509765625, 119.83732604980469, -51.01239013671875, -46.02759552001953, 214.92544555664062, -94.79672241210938, -35.49007797241211, 416.694580078125, 73.40536499023438, 15.499979019165039, 15.244277954101562, -45.3303108215332, -13.901100158691406, 384.32611083984375, 169.49148559570312, -62.16455078125, 187.18789672851562, -109.61810302734375, 44.72022247314453, -129.54718017578125, 90.37882232666016, 125.28067016601562, 194.78028869628906, 9.323394775390625, 146.0316619873047, 117.05856323242188, 148.7276611328125, 239.00868225097656, 166.46209716796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000337.npy"}
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 82.44564819335938, "std": 113.41641235351562, "min": -170.31878662109375, "p10": -40.476660156249984, "median": 67.5185775756836, "p90": 234.1526138305664, "max": 438.5716857910156, "pos_frac": 0.7421875, "sample": [235.1331329345703, -151.9232940673828, 7.4970703125, -22.248321533203125, 341.0672607421875, 395.5369567871094, 113.50456237792969, -18.298553466796875, -2.9285049438476562, -110.82162475585938, 107.12368774414062, 221.89932250976562, 35.739990234375, 144.95693969726562, 438.5716857910156, 212.81947326660156, -29.83642578125, 58.530364990234375, -50.99041748046875, -89.82980346679688, 181.73843383789062, 80.71127319335938, -2.151956558227539, 100.367431640625, 45.893463134765625, 125.02310180664062, 233.73239135742188, 110.82959747314453, -35.566070556640625, 13.629524230957031, 70.98736572265625, 28.686279296875, 41.287841796875, 120.28431701660156, 20.772933959960938, 151.7293243408203, 208.63714599609375, 121.08004760742188, 154.34263610839844, 80.42225646972656, -2.5411376953125, 63.3587646484375, 110.48471069335938, 134.22824096679688, 153.73260498046875, 21.432205200195312, -33.460205078125, 169.3426513671875, 211.51678466796875, 125.00335693359375, 155.56130981445312, -59.131011962890625, 281.4019775390625, -27.145164489746094, 62.87200927734375, 6.870933532714844, 85.29327392578125, 120.3740234375, -114.95401000976562, 268.5597839355469, 247.77249145507812, 249.7161865234375, 177.85617065429688, 60.294891357421875, 110.53179931640625, 5.2549896240234375, -30.014514923095703, 6.529815673828125, -57.140716552734375, -10.000328063964844, 113.28600311279297, 115.27201843261719, 272.96014404296875, 5.989593505859375, -15.308441162109375, 14.55276107788086, -2.0529518127441406, 180.91224670410156, 228.9103240966797, 64.04978942871094, 43.90228271484375, 1.9753303527832031, 134.11256408691406, 43.70868682861328, 41.647239685058594, -51.7855224609375, 132.55715942382812, 8.9215087890625, -59.703704833984375, 196.42193603515625, 80.01812744140625, 276.9684143066406, 61.7772216796875, 17.419326782226562, -13.847457885742188, 270.50616455078125, 108.60009765625, 158.28253173828125, -104.46478271484375, 44.20850372314453, -23.452682495117188, -30.392135620117188, -2.3884048461914062, 140.37509155273438, -24.278953552246094, 341.2499084472656, 183.5823974609375, 74.73008728027344, -35.97076416015625, 208.9649658203125, 63.320343017578125, 88.32624816894531, 7.099458694458008, 109.02703857421875, -17.5926513671875, 224.15167236328125, 79.79769897460938, 86.62522888183594, 90.42379760742188, 31.4315185546875, -170.31878662109375, -53.81866455078125, 22.47265625, 26.559249877929688, 226.88092041015625, 155.05514526367188, -67.59384155273438, 267.4370422363281], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000338.npy"}
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 86.39607238769531, "std": 127.89737701416016, "min": -206.72198486328125, "p10": -91.37068176269531, "median": 89.8990478515625, "p90": 240.79803466796872, "max": 379.93548583984375, "pos_frac": 0.71875, "sample": [17.159820556640625, -105.82940673828125, 219.078125, 167.86181640625, 30.551513671875, -26.73345947265625, 67.38461303710938, -92.46176147460938, 58.45635986328125, -63.693634033203125, 328.6823425292969, 73.05126953125, 15.150634765625, 64.84127807617188, 219.79330444335938, 1.6129150390625, 212.97618103027344, 183.65859985351562, 105.06640625, -72.31912231445312, 223.6845703125, -51.709075927734375, 208.77171325683594, 147.17547607421875, 0.0, 328.2578125, 2.7049102783203125, 149.0715789794922, 0.0, 163.85031127929688, 146.36097717285156, -117.428955078125, 142.30392456054688, 186.96401977539062, -130.9227294921875, -32.81964874267578, 81.9876708984375, 207.19381713867188, 47.61289978027344, -46.822959899902344, 139.88784790039062, 186.90577697753906, -37.40985107421875, 214.43576049804688, 117.180419921875, 57.22926330566406, 93.87939453125, -55.49382019042969, 84.2701644897461, 237.2122802734375, 59.005218505859375, -90.903076171875, 249.92691040039062, 121.05757141113281, 215.21719360351562, 207.81326293945312, 36.04456329345703, 163.09725952148438, 85.918701171875, 145.76348876953125, 227.87506103515625, -6.821311950683594, 3.71356201171875, -93.12603759765625, 0.6049385070800781, 193.26580810546875, -101.06858825683594, 134.53054809570312, 262.46728515625, 50.93365478515625, 108.29965209960938, -0.854217529296875, -67.40039825439453, 283.06591796875, -9.0740966796875, 175.81622314453125, 207.78411865234375, 192.22933959960938, -65.38824462890625, -16.29454803466797, -28.601715087890625, 144.96212768554688, 185.28762817382812, -42.4940185546875, 345.5245056152344, 255.13031005859375, 143.64813232421875, 0.0, 107.9688720703125, 93.9971923828125, 31.54339599609375, 152.544189453125, 80.46621704101562, 189.58218383789062, 249.164794921875, -35.761871337890625, 82.1937255859375, 59.066558837890625, 163.73440551757812, 120.38360595703125, 197.58251953125, -121.78317260742188, 64.31015014648438, 109.27908325195312, 49.0238037109375, 141.0223388671875, 138.88967895507812, 29.36846923828125, -6.3661041259765625, -189.1706085205078, 249.78370666503906, 378.68975830078125, -181.02481079101562, -7.167816162109375, 113.4998779296875, 14.172760009765625, 136.2850341796875, -106.02969360351562, 303.82281494140625, -174.33358764648438, -206.72198486328125, 379.93548583984375, 166.09811401367188, 95.567626953125, 184.6040802001953, 7.37982177734375, -114.34175872802734, 335.8602294921875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000339.npy"}
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 77.30957794189453, "std": 146.6755828857422, "min": -379.81793212890625, "p10": -82.3828369140625, "median": 62.235904693603516, "p90": 280.7218048095703, "max": 434.27996826171875, "pos_frac": 0.7109375, "sample": [434.27996826171875, 40.0506591796875, -341.327880859375, 162.44422912597656, 15.25018310546875, 322.369873046875, -13.239990234375, -49.54103088378906, -190.815185546875, 271.6844177246094, -64.59765625, 317.7040710449219, -59.65155029296875, 37.3770751953125, 0.0, 192.33018493652344, 3.4700927734375, -379.81793212890625, 100.73001098632812, 198.69493103027344, 138.54000854492188, 212.3697967529297, -111.47589111328125, 83.39453125, -113.30479431152344, 44.514984130859375, -38.12603759765625, 111.68567657470703, 93.73890686035156, 61.6575927734375, 348.176025390625, -10.057357788085938, 102.29302978515625, 290.2132568359375, 42.81427001953125, -54.314208984375, 0.5451736450195312, -86.04766845703125, 195.89463806152344, 201.78814697265625, 266.908447265625, 169.385009765625, -65.89181518554688, 124.31143188476562, -70.65145874023438, 259.30572509765625, -35.12794494628906, 119.69764709472656, 36.855072021484375, -4.2454833984375, 178.9691162109375, 398.1783447265625, -10.454984664916992, -142.1170654296875, 8.873241424560547, 6.24615478515625, 244.12710571289062, 279.3051452636719, 382.31280517578125, -58.81007385253906, 128.97109985351562, -128.2645263671875, 267.27752685546875, 0.0, 291.9580078125, 50.61102294921875, 234.82919311523438, 68.21566009521484, -7.72198486328125, 59.1263427734375, -156.50262451171875, -26.079221725463867, 184.9076385498047, -154.2723388671875, 275.3096618652344, 81.68484497070312, 78.40518188476562, -25.08221435546875, 135.5435791015625, -38.813720703125, 303.01025390625, 346.89447021484375, -286.82012939453125, -99.44243621826172, 23.471038818359375, 39.3116455078125, 74.36572265625, 31.207763671875, 226.769775390625, 48.920501708984375, 81.36175537109375, -26.04010772705078, 139.19183349609375, 20.5711669921875, 92.26456451416016, 186.20245361328125, 138.72808837890625, 1.844818115234375, 90.06976318359375, 150.06857299804688, 228.4906005859375, 17.184974670410156, 284.3219299316406, 114.40048217773438, 313.07476806640625, -65.08297729492188, 65.62171936035156, -129.81410217285156, 203.25425720214844, 107.30841064453125, 180.20046997070312, 74.88107299804688, -0.8236503601074219, 62.81421661376953, 284.02734375, 172.38140869140625, 64.88360595703125, 47.119773864746094, 43.34733581542969, 19.370330810546875, 30.20806884765625, 48.10772705078125, 11.9957275390625, -15.232999801635742, -80.81219482421875, 69.22329711914062, 233.7589569091797, 10.489595413208008], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000340.npy"}
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 63.04977798461914, "std": 147.22543334960938, "min": -370.8626708984375, "p10": -141.2131576538086, "median": 56.7041015625, "p90": 251.03160705566404, "max": 484.02740478515625, "pos_frac": 0.703125, "sample": [17.535072326660156, 96.64498901367188, -13.161651611328125, 141.129150390625, 106.39547729492188, 104.04039001464844, 74.01322937011719, 53.89385986328125, 198.02003479003906, -15.391510009765625, 484.02740478515625, 109.19169616699219, 166.87210083007812, 11.615875244140625, 302.81292724609375, 286.095947265625, -15.4766845703125, 133.121337890625, 61.8494873046875, 6.062662124633789, 229.3289794921875, 136.66363525390625, 164.23324584960938, -370.8626708984375, 372.1702880859375, 131.1057891845703, 56.000762939453125, 75.19317626953125, -241.064697265625, 22.458984375, 289.05780029296875, 272.26171875, -102.451904296875, 20.692123413085938, -173.64593505859375, 53.798095703125, 55.6876220703125, -82.1690673828125, -3.3791046142578125, -7.283363342285156, 136.00827026367188, 153.8007354736328, 95.6009521484375, 8.535675048828125, 148.6748809814453, -35.19476318359375, 111.08267211914062, 128.87521362304688, 1.0837249755859375, 69.70463562011719, 108.0174560546875, 169.27645874023438, 260.10009765625, -16.66143798828125, 211.79287719726562, 79.53616333007812, 101.75111389160156, 26.35272216796875, 30.313812255859375, -237.17266845703125, 99.94488525390625, -144.46939086914062, 402.5986328125, 39.84192657470703, 125.593017578125, -166.01596069335938, -108.46099853515625, -8.055755615234375, 189.23507690429688, 15.17608642578125, 94.13192749023438, 241.54286193847656, 14.723480224609375, 47.39250183105469, 29.70996856689453, 101.7569351196289, 199.69818115234375, 10.7841796875, 93.34930419921875, -162.89547729492188, 135.63084411621094, 57.407440185546875, 190.1871337890625, -51.456512451171875, 288.2250061035156, 16.3759765625, 75.97113037109375, 156.926025390625, -0.4996337890625, 247.14511108398438, -79.29815673828125, 49.946807861328125, 199.78781127929688, 17.569087982177734, -183.03546142578125, -1.29791259765625, -141.07997131347656, -24.020366668701172, 0.850189208984375, 120.62039184570312, -15.266510009765625, 190.49066162109375, 57.4364013671875, 174.27685546875, -46.278533935546875, 183.23388671875, 71.648681640625, -65.45994567871094, -198.67758178710938, -187.38832092285156, 94.67973327636719, -90.1563720703125, -141.52392578125, 300.0355224609375, -90.60870361328125, -70.46817016601562, 167.00807189941406, 316.68438720703125, 110.02755737304688, 410.52606201171875, 12.70062255859375, 30.154373168945312, -15.957260131835938, -297.87249755859375, 378.29547119140625, 44.06072998046875, -196.07559204101562, -5.256397247314453], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000341.npy"}
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 76.46883392333984, "std": 112.53009033203125, "min": -228.79061889648438, "p10": -61.18943481445312, "median": 63.14045715332031, "p90": 228.76891174316404, "max": 368.2289733886719, "pos_frac": 0.7890625, "sample": [28.3173828125, 10.459884643554688, 21.84197998046875, 75.8963623046875, 136.62435913085938, 180.5430908203125, 49.685279846191406, -163.25936889648438, -3.745330810546875, 81.21124267578125, 180.87924194335938, 58.9703369140625, -94.14157104492188, 133.6209716796875, 129.20907592773438, 225.37496948242188, -99.9604263305664, 366.24395751953125, -66.96817016601562, -53.56977844238281, 257.9505920410156, 77.65399169921875, 46.6087646484375, 124.24547576904297, -63.330535888671875, 178.90164184570312, 125.60697937011719, -121.28857421875, 83.49923706054688, 95.73031616210938, 58.27014923095703, 245.01263427734375, 43.42915344238281, 122.80990600585938, 4.955169677734375, 47.7830810546875, 96.59355926513672, 120.53629302978516, -79.85514831542969, 99.69671630859375, 11.69967269897461, 157.1339111328125, 5.03460693359375, -10.687408447265625, -109.05718994140625, -44.32121276855469, -155.37664794921875, 64.52487182617188, 324.25054931640625, 13.614105224609375, 44.453025817871094, 21.2681884765625, 9.21600341796875, 211.15109252929688, 21.793838500976562, 201.4641876220703, 108.65484619140625, 61.75604248046875, 272.2151794433594, 299.18255615234375, 39.579925537109375, 368.2289733886719, 61.71038818359375, 211.93016052246094, 248.2325897216797, -15.536327362060547, 14.375274658203125, 0.0, 65.27471923828125, 31.705078125, 112.98342895507812, -10.51681900024414, 170.69149780273438, 10.19023323059082, -57.82738494873047, 78.25057983398438, 182.4656982421875, 52.198387145996094, 198.48165893554688, 16.58390235900879, 20.17120361328125, 107.0006103515625, 67.29090881347656, 86.9698486328125, -70.39373779296875, 44.62407684326172, -228.79061889648438, -27.39862060546875, 207.0692138671875, 18.43766975402832, 152.8164520263672, 84.74942016601562, 37.95349884033203, 44.032135009765625, 17.642921447753906, 172.076904296875, 143.05450439453125, 123.48480987548828, -18.296836853027344, -60.271820068359375, 78.74246215820312, 127.79397583007812, 9.875534057617188, 278.9808044433594, 125.18673706054688, 129.1295166015625, -12.53106689453125, 71.31741333007812, 134.79412841796875, -87.46599578857422, 311.5269775390625, 163.545166015625, 13.4725341796875, 205.17459106445312, 53.241424560546875, 238.2755126953125, 87.8946533203125, 8.041656494140625, 192.9165496826172, 161.65850830078125, 37.845703125, -19.794593811035156, -155.60516357421875, 269.30279541015625, 175.66964721679688, -20.674612045288086, 236.6881103515625, 1.76971435546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000342.npy"}
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 67.89884948730469, "std": 113.26075744628906, "min": -214.5242462158203, "p10": -80.38582839965821, "median": 66.50140380859375, "p90": 217.12852172851564, "max": 346.82501220703125, "pos_frac": 0.6875, "sample": [-138.48086547851562, 124.21707153320312, 197.69677734375, 1.6779346466064453, 1.911865234375, 89.0892333984375, 73.372802734375, -20.226806640625, -39.040496826171875, 70.4666748046875, 217.81292724609375, -82.81741333007812, -37.889373779296875, -31.08697509765625, 21.037506103515625, 275.355224609375, 45.38208770751953, 62.037353515625, -39.009544372558594, -36.63127136230469, 145.26513671875, 42.582000732421875, 182.37112426757812, 51.515380859375, 121.14822387695312, 138.7495880126953, 103.27049255371094, 35.893646240234375, 2.43951416015625, 195.05697631835938, 63.70660400390625, -214.5242462158203, 18.40185546875, -27.739341735839844, -80.18074798583984, 186.73182678222656, -17.977752685546875, 245.85647583007812, 6.3935546875, 187.23992919921875, 63.162200927734375, 263.8438415527344, 69.84161376953125, 77.40692138671875, 0.0, -86.95162963867188, 273.8338928222656, 181.79942321777344, 167.4281768798828, 300.4613037109375, 52.123382568359375, 109.46118927001953, -5.26324462890625, 9.146955490112305, 216.835205078125, -98.0755615234375, 76.28871154785156, 104.35653686523438, 22.652923583984375, 103.66572570800781, 151.86502075195312, 81.99298095703125, 39.00715637207031, -12.088104248046875, 94.62474060058594, 208.22433471679688, -98.37850952148438, 346.82501220703125, 31.4775390625, -125.01806640625, -62.34069061279297, -75.60543823242188, 123.71124267578125, -36.161224365234375, -87.92620086669922, 117.5462646484375, -21.38165283203125, 248.402099609375, 136.17861938476562, 69.29620361328125, 83.08015441894531, 120.16905212402344, 46.94493103027344, 76.32498168945312, 156.76124572753906, 148.55120849609375, 36.90032958984375, 88.48788452148438, -80.86434936523438, -66.14788818359375, 7.142864227294922, 325.00872802734375, 313.2368469238281, 218.8155517578125, 37.7669677734375, 130.526123046875, 112.31741333007812, 196.55599975585938, 121.5491943359375, 98.56419372558594, 187.9591064453125, 106.2938232421875, 58.80072021484375, 207.3846435546875, -159.35577392578125, 176.63232421875, 34.499603271484375, 131.01254272460938, -36.339324951171875, -119.04251098632812, 84.98652648925781, 195.71412658691406, 171.87396240234375, -18.333770751953125, -12.423248291015625, -32.916015625, -18.192276000976562, -38.44854736328125, 165.51707458496094, 262.3725891113281, -120.39054870605469, -23.025352478027344, 77.49081420898438, -23.068222045898438, -19.77886962890625, 230.52096557617188, -40.14031982421875, -83.58637237548828], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000343.npy"}
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 71.52964782714844, "std": 122.47824096679688, "min": -256.1362609863281, "p10": -72.01531677246092, "median": 75.73274230957031, "p90": 236.24721679687497, "max": 326.8836364746094, "pos_frac": 0.6875, "sample": [126.47927856445312, -23.016845703125, 194.74835205078125, 62.227020263671875, -30.408203125, 258.34820556640625, 299.8932800292969, 118.24899291992188, 171.2799835205078, -2.12762451171875, 131.77162170410156, -13.939727783203125, -14.302742004394531, -41.58348083496094, 165.95135498046875, 247.91143798828125, 116.04379272460938, 118.45619201660156, -256.1362609863281, 191.90017700195312, -61.984649658203125, 126.17267608642578, -65.16815185546875, 52.182342529296875, 310.4832763671875, 175.72525024414062, -36.59654235839844, 223.59539794921875, 140.63595581054688, 35.790435791015625, 182.791015625, -19.064208984375, 35.942108154296875, 36.23295593261719, -202.390625, 60.12654113769531, 178.89459228515625, 115.59424591064453, 170.84222412109375, -3.5445556640625, -222.94850158691406, 110.12545776367188, 30.03765869140625, -158.01593017578125, -25.30937957763672, 128.5714569091797, -60.9713134765625, 20.572166442871094, 121.62478637695312, 136.36424255371094, -131.1708984375, -111.73436737060547, 108.0767822265625, -139.64212036132812, 120.94061279296875, -53.370574951171875, 167.4281768798828, -34.6671142578125, 150.78176879882812, 184.43109130859375, 201.8326873779297, 88.38478088378906, 303.20172119140625, 107.94638061523438, 258.8978271484375, -25.153701782226562, -8.8326416015625, 300.0986328125, -51.57415771484375, -116.1307373046875, -22.761199951171875, 52.384918212890625, 77.134033203125, 297.4471740722656, 36.166595458984375, 156.3428955078125, 16.7745361328125, 159.78237915039062, -55.4864501953125, 48.94706726074219, 156.5171661376953, 47.680511474609375, 27.9906005859375, 159.2262420654297, 10.469146728515625, -10.396728515625, -87.99203491210938, 209.19349670410156, 24.97314453125, 13.145126342773438, 261.14251708984375, 74.33145141601562, 117.3168716430664, -42.563079833984375, 61.143104553222656, 130.78355407714844, 232.51849365234375, 4.091423034667969, 71.963623046875, -154.29852294921875, -27.089557647705078, 146.44845581054688, -88.05877685546875, 263.85223388671875, -2.439208984375, 41.79473876953125, 176.26821899414062, 96.3304443359375, 84.18769836425781, -28.19049072265625, 92.16317749023438, 13.918106079101562, -56.76971435546875, 308.3572998046875, 168.48681640625, 35.070526123046875, 119.86322021484375, -88.71135711669922, 128.6571044921875, 204.82937622070312, 326.8836364746094, 83.14202880859375, 115.83258056640625, 134.9123992919922, 94.81594848632812, -19.29302215576172, 244.94757080078125, -166.18246459960938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000344.npy"}
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 61.69767761230469, "std": 117.13240814208984, "min": -227.12466430664062, "p10": -82.3407989501953, "median": 46.71414756774902, "p90": 208.1032958984375, "max": 408.1368408203125, "pos_frac": 0.7265625, "sample": [32.78131103515625, 203.8761749267578, 315.38665771484375, -192.71595764160156, 259.4049072265625, 211.87109375, 89.40869140625, -39.966766357421875, 7.33319091796875, 0.3538665771484375, -24.49712371826172, -13.96270751953125, 21.504119873046875, 33.425262451171875, -10.220855712890625, -117.32546997070312, 408.1368408203125, 60.15147399902344, 24.771453857421875, -22.293121337890625, 80.80581665039062, -54.792236328125, 225.56748962402344, 127.43107604980469, -7.964338302612305, 74.17782592773438, 178.31085205078125, 206.488525390625, -176.83584594726562, 12.916206359863281, 17.407012939453125, 5.770233154296875, -7.39141845703125, 15.007156372070312, 16.09527587890625, 102.68487548828125, 59.3065185546875, 188.1800537109375, 17.581939697265625, 115.97134399414062, 51.65203857421875, 5.927263259887695, -31.38861083984375, 66.09515380859375, 102.38088989257812, 12.54010009765625, 168.12811279296875, 12.28333854675293, 268.8160705566406, 185.9605712890625, 143.0248260498047, 192.393798828125, 141.45721435546875, 103.25898742675781, 62.37385559082031, 165.31585693359375, 26.696044921875, -113.49496459960938, 29.947662353515625, 25.131080627441406, 92.98818969726562, 50.10064697265625, -122.57614135742188, 42.5283203125, 34.12672424316406, 144.77435302734375, -227.12466430664062, 46.24345016479492, -108.83303833007812, -98.9964599609375, 136.55999755859375, 164.94546508789062, 112.87875366210938, 73.89286804199219, 57.20005798339844, -68.75746154785156, 269.0208435058594, -116.5569076538086, 84.66830444335938, -106.63512420654297, 279.66912841796875, 336.04608154296875, 91.12541198730469, -217.16259765625, 54.682891845703125, 146.7582244873047, -67.19869995117188, 276.9141845703125, 176.12432861328125, 83.10897064208984, 344.57757568359375, 137.75067138671875, 94.93927001953125, -77.79360961914062, 21.80938720703125, -52.383766174316406, 218.15286254882812, 228.40771484375, 42.27079772949219, -10.902969360351562, 160.1307373046875, -54.29425048828125, 122.55325317382812, -81.12606811523438, 24.365203857421875, 139.80599975585938, -92.55754089355469, 64.38870239257812, 39.756988525390625, -85.1751708984375, -0.9893798828125, 160.31158447265625, -21.836273193359375, 38.52192687988281, 205.77490234375, 167.19879150390625, 47.184844970703125, 178.41641235351562, 0.0, 19.183002471923828, 33.26847839355469, 64.22195434570312, -20.097129821777344, 2.945707321166992, 118.48512268066406, -1.00909423828125, 117.74520111083984, -79.85589599609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000345.npy"}
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 80.96846008300781, "std": 127.41040802001953, "min": -292.71148681640625, "p10": -69.39771347045898, "median": 74.25240707397461, "p90": 224.07213134765624, "max": 399.23553466796875, "pos_frac": 0.7734375, "sample": [-90.52496337890625, 1.9068183898925781, 26.796031951904297, -112.70663452148438, 31.67498779296875, 109.6107177734375, 313.0382080078125, 191.5616455078125, -0.24938201904296875, -36.32568359375, 235.029541015625, -60.68269348144531, 94.117919921875, -20.610260009765625, 157.3470458984375, -79.5576171875, -13.020217895507812, 160.90101623535156, 124.2095947265625, 253.57763671875, 13.4267578125, 135.32872009277344, 71.59695434570312, 26.14703369140625, 13.240558624267578, 209.82843017578125, 116.51507568359375, 218.51885986328125, -32.594207763671875, -140.35108947753906, 45.68316650390625, 86.25617218017578, 18.702077865600586, 49.167388916015625, -72.96087646484375, 285.484375, 182.3465576171875, 105.27120971679688, 2.7801074981689453, 130.52444458007812, 100.5433349609375, 21.290786743164062, -214.8546142578125, -30.8612060546875, -266.9285888671875, 33.39215087890625, 20.997833251953125, -52.151092529296875, 28.062347412109375, 90.43231201171875, 114.63623046875, 65.72915649414062, 160.85877990722656, -67.87064361572266, 91.1282958984375, 115.58058166503906, 377.7037658691406, 148.6490020751953, 14.461883544921875, 203.18402099609375, 186.75332641601562, 171.1207275390625, 342.8607177734375, 23.601730346679688, 18.57098388671875, 72.38729095458984, 233.06979370117188, -38.20989990234375, 144.146240234375, 195.03988647460938, 185.14016723632812, 112.09478759765625, 27.144397735595703, 297.39361572265625, 378.70257568359375, 221.947265625, -57.44818115234375, 163.1876220703125, 93.89627075195312, 39.99188232421875, 24.4598388671875, -230.28591918945312, 215.41058349609375, 56.4166259765625, -4.213371276855469, 3.7401351928710938, -44.01573181152344, 33.462623596191406, -121.28407287597656, 112.715087890625, 187.480224609375, 172.232666015625, -77.96459197998047, 191.35910034179688, 28.944351196289062, -80.49440002441406, 135.06690979003906, -292.71148681640625, 56.751991271972656, 252.55841064453125, -15.283124923706055, 32.410797119140625, 12.98406982421875, 217.10360717773438, 154.61807250976562, 166.9257354736328, 190.04469299316406, 308.0145263671875, 9.656906127929688, 118.7984619140625, 0.0, 84.75242614746094, -105.22282409667969, 32.350830078125, 221.3934326171875, 399.23553466796875, 57.8515625, 129.09884643554688, 15.847381591796875, 55.87812805175781, 178.10110473632812, 219.04641723632812, -44.404754638671875, 182.29751586914062, 229.0301513671875, 76.11752319335938, 95.77420043945312, 205.53179931640625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000346.npy"}
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 67.21463012695312, "std": 113.25947570800781, "min": -211.152099609375, "p10": -77.92013244628906, "median": 59.78499984741211, "p90": 205.02430114746093, "max": 405.99462890625, "pos_frac": 0.7265625, "sample": [85.7184066772461, 187.718505859375, -7.837432861328125, 74.18559265136719, -24.69140625, 57.185951232910156, 24.595909118652344, 206.05206298828125, 163.52670288085938, 238.7047119140625, -211.152099609375, 107.39849853515625, 127.980712890625, 45.412078857421875, 189.1162109375, -169.8511199951172, -204.97787475585938, -159.60205078125, 188.6395263671875, 315.54119873046875, 109.05830383300781, -0.04815673828125, 177.26278686523438, -2.6073532104492188, -33.34203338623047, 60.85028076171875, 282.1615905761719, 99.22967529296875, 14.195175170898438, 30.8565673828125, -46.13629150390625, 59.1478271484375, 113.25418090820312, 59.395477294921875, 243.85684204101562, 32.14179229736328, 133.3277587890625, 21.35106658935547, 74.34561157226562, -26.483047485351562, 2.251434326171875, 264.277587890625, 255.27151489257812, 49.458740234375, 100.8172607421875, 2.46697998046875, 60.174522399902344, 5.291259765625, 111.08206176757812, -5.091888427734375, -6.98516845703125, 9.840988159179688, 122.16352844238281, 73.95645141601562, -96.83328247070312, -12.049240112304688, 21.78924560546875, 188.89532470703125, 248.5478057861328, -56.56988525390625, 148.75180053710938, 31.974395751953125, -13.10318374633789, 22.482336044311523, 101.79164123535156, 108.65785217285156, 204.58383178710938, 120.54022216796875, 209.43560791015625, 129.40859985351562, 88.03875732421875, -74.72271728515625, -204.95216369628906, 138.602294921875, 189.22451782226562, -5.308834075927734, -130.44882202148438, 37.71509552001953, -80.85922241210938, -11.33978271484375, -1.5391616821289062, 153.94061279296875, 405.99462890625, 27.24981689453125, 183.74880981445312, -108.62033081054688, 153.52392578125, 200.12933349609375, 130.8031005859375, 110.86508178710938, -116.88775634765625, 82.71821594238281, -97.30224609375, 109.21581268310547, 54.465240478515625, -82.71099853515625, -5.876739501953125, 7.6770477294921875, 144.64938354492188, 44.507080078125, -76.6605224609375, -153.74819946289062, 263.3726501464844, 188.08392333984375, 0.0, 61.71075439453125, 3.433319091796875, 96.2435302734375, 190.84716796875, 160.07876586914062, 135.35787963867188, 27.883140563964844, 88.1744155883789, 240.62892150878906, -15.6630859375, 100.04818725585938, 1.7686309814453125, 17.436614990234375, 54.74720764160156, -3.63031005859375, 157.66903686523438, 225.49673461914062, 188.41427612304688, 137.75735473632812, 49.488525390625, -63.82464599609375, 110.70693969726562, 36.391510009765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000347.npy"}
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 57.19635772705078, "std": 125.07121276855469, "min": -256.28509521484375, "p10": -75.21012725830077, "median": 32.630252838134766, "p90": 247.52609100341795, "max": 513.87890625, "pos_frac": 0.640625, "sample": [63.9227294921875, 100.2498779296875, 13.015726089477539, 3.64215087890625, -7.353515625, -25.617584228515625, 244.1482391357422, -21.498950958251953, 189.80323791503906, 199.04815673828125, -148.6055908203125, 140.99766540527344, 47.80926513671875, 71.63604736328125, 57.00779724121094, 20.09368896484375, 255.40774536132812, 90.2779541015625, 60.9503173828125, 0.0, 37.01708984375, 260.55047607421875, 32.28443145751953, 1.6369132995605469, -57.536163330078125, 316.28546142578125, -109.38150024414062, -10.863494873046875, -7.99334716796875, 214.8289794921875, 181.45004272460938, -6.214817047119141, -0.807586669921875, 9.326896667480469, 124.877197265625, 34.877197265625, 169.58399963378906, 27.95635223388672, 184.84091186523438, 163.02392578125, 4.405466079711914, 14.4444580078125, -164.48284912109375, 26.68902587890625, -15.85711669921875, 200.9061279296875, 43.21014404296875, -55.754364013671875, 56.92475891113281, 177.70994567871094, -7.455986022949219, 91.11944580078125, 109.39328002929688, 159.25779724121094, -18.990814208984375, 55.32470703125, 132.93017578125, 258.8313293457031, 236.40057373046875, -132.3510284423828, -41.061187744140625, -77.24737548828125, -87.95962524414062, 45.12553405761719, 14.196229934692383, -53.366912841796875, 282.3823547363281, 1.5029067993164062, 50.246246337890625, -22.59075927734375, 266.55291748046875, -88.33999633789062, -150.43704223632812, -9.60760498046875, 58.10118103027344, -61.81500244140625, 57.26805114746094, -74.33702087402344, 19.72613525390625, 41.26959228515625, 256.56646728515625, 39.33349609375, 183.16629028320312, 351.76690673828125, -5.85076904296875, 283.18670654296875, -191.7373046875, -56.304107666015625, -93.4478759765625, -43.85986328125, 24.597976684570312, 90.82771301269531, 180.50994873046875, -36.979026794433594, 286.04302978515625, -21.601470947265625, 127.37458801269531, -7.105690002441406, -44.890380859375, 46.82293701171875, 145.57479858398438, 15.357269287109375, -30.322189331054688, -21.55499839782715, 111.84953308105469, 8.78765869140625, 126.71246337890625, 132.9749755859375, -0.712127685546875, 307.19671630859375, 63.19050598144531, -66.721923828125, 4.836273193359375, -28.9097900390625, -101.472412109375, 64.45091247558594, 346.8057861328125, 53.18902587890625, 35.13688659667969, 32.97607421875, -44.38482666015625, -13.2872314453125, -105.9197998046875, 19.17584228515625, 212.63650512695312, 164.61326599121094, -256.28509521484375, 513.87890625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000348.npy"}
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 87.79043579101562, "std": 131.3859100341797, "min": -376.89544677734375, "p10": -43.41656494140624, "median": 65.27171325683594, "p90": 245.85278320312497, "max": 483.399658203125, "pos_frac": 0.78125, "sample": [133.54574584960938, 82.36393737792969, 66.28372192382812, -217.1748046875, -85.31689453125, 12.151336669921875, 457.4976806640625, 187.04885864257812, 29.065589904785156, 243.1827392578125, 83.68917846679688, -137.8670196533203, 115.01744079589844, 186.271240234375, 56.45794677734375, 143.3975830078125, 196.08917236328125, 26.792327880859375, 269.4775390625, 237.63861083984375, 136.01934814453125, 58.59051513671875, -41.392333984375, -27.9373779296875, -10.496002197265625, 11.848785400390625, -108.7489013671875, 30.83154296875, 256.25872802734375, 219.74539184570312, 39.78590393066406, 169.17706298828125, 154.48727416992188, -116.642578125, 58.241912841796875, 17.432815551757812, 194.55130004882812, -49.6949462890625, 22.559715270996094, 152.03753662109375, 33.174198150634766, 35.45903015136719, 172.79318237304688, 136.4267120361328, 18.233543395996094, 181.3310546875, 213.48211669921875, 305.8810119628906, 180.20135498046875, 103.31204223632812, -19.270111083984375, -6.353303909301758, -36.890777587890625, 323.2366638183594, 217.35629272460938, 37.18919372558594, 13.191421508789062, 58.716522216796875, -108.61477661132812, 130.30908203125, 0.0, 227.48345947265625, -15.2161865234375, 216.24609375, 73.4532470703125, 52.02008056640625, 30.298187255859375, 315.61553955078125, 58.1556396484375, -1.2536163330078125, 15.7147216796875, -178.37451171875, -174.22506713867188, 304.3411865234375, 39.63226318359375, 144.33824157714844, -7.8160400390625, 176.61822509765625, -28.70623016357422, 53.80028533935547, 9.742523193359375, 204.61712646484375, 220.798095703125, 139.65740966796875, 216.2794189453125, 43.688507080078125, 188.3129425048828, 86.67462158203125, 368.7835998535156, 21.224151611328125, 87.16476440429688, 116.799560546875, 252.57736206054688, 40.937461853027344, -62.346405029296875, 41.960784912109375, 252.0828857421875, 98.71926879882812, 25.462432861328125, -15.725448608398438, 64.25970458984375, 98.19117736816406, 51.2471923828125, 157.14837646484375, -376.89544677734375, 69.65582275390625, 7.5824737548828125, 3.808238983154297, 20.286102294921875, 41.82171630859375, 34.05657958984375, 185.02658081054688, -29.84967041015625, 277.01861572265625, 234.63232421875, 241.66018676757812, 147.072021484375, -48.1397705078125, 349.312744140625, 483.399658203125, -20.238128662109375, 87.24932861328125, 90.45626831054688, 74.67092895507812, -125.42730712890625, -17.256927490234375, 68.20387268066406, 187.25534057617188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000349.npy"}
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 67.88014221191406, "std": 127.94562530517578, "min": -248.47991943359375, "p10": -102.81058197021484, "median": 69.60392761230469, "p90": 230.68325653076172, "max": 378.7763977050781, "pos_frac": 0.71875, "sample": [231.47589111328125, 180.7181396484375, 4.715812683105469, -36.00105285644531, 96.093505859375, 12.478439331054688, -102.86399841308594, 184.43255615234375, -106.07635498046875, 128.8928680419922, 73.4468994140625, 11.043899536132812, 202.04763793945312, -18.0155029296875, 59.194610595703125, 294.15130615234375, 172.575927734375, 260.1136474609375, 165.01370239257812, -138.86521911621094, 195.70770263671875, 46.764434814453125, 2.586456298828125, 43.104705810546875, 75.45854187011719, 62.584930419921875, 288.8377685546875, -46.14703369140625, 226.30148315429688, 233.521240234375, 211.66107177734375, -9.19769287109375, 30.762123107910156, 7.809410095214844, -191.12619018554688, 125.8302993774414, 81.43406677246094, 31.427467346191406, 83.8548583984375, 355.50750732421875, 6.924285888671875, 230.8968048095703, 41.81781005859375, -114.439697265625, 216.17340087890625, 39.15850830078125, -27.765640258789062, 100.09095764160156, -113.8944091796875, 2.690399169921875, 176.63865661621094, 208.78863525390625, 378.7763977050781, 151.61712646484375, 79.42974853515625, 156.31619262695312, 300.7745361328125, 67.61050415039062, 144.10784912109375, 306.4715881347656, -82.2738037109375, -70.867431640625, 132.21157836914062, -36.024505615234375, -43.57310485839844, 31.74164581298828, 154.63211059570312, 49.175872802734375, -176.24688720703125, 20.561046600341797, 98.71461486816406, 71.59735107421875, 85.42558288574219, -9.7535400390625, -34.20158386230469, 100.69558715820312, -47.074363708496094, -1.0463714599609375, -3.082000732421875, 119.19853210449219, 217.2840576171875, 174.81956481933594, 19.141822814941406, 101.9200439453125, 56.6326904296875, 74.54103088378906, 21.380859375, 283.36175537109375, 16.246200561523438, -45.38092041015625, 138.59725952148438, 214.3511962890625, 72.7227783203125, 83.31904602050781, 189.56430053710938, -90.864013671875, 66.62216186523438, 124.87702941894531, 304.8554992675781, -30.2247314453125, 86.56163024902344, -248.47991943359375, 79.66458129882812, -94.23565673828125, -214.4420166015625, 153.5641326904297, -116.7103271484375, 30.238113403320312, -69.65142822265625, 144.49786376953125, -155.97613525390625, 23.056488037109375, 144.24502563476562, 230.59173583984375, -53.16725158691406, 88.92767333984375, -102.78768920898438, 170.33206176757812, 356.522705078125, -122.79278564453125, 1.5321197509765625, 13.298408508300781, -36.20928955078125, -189.57638549804688, 137.5334014892578, -43.53877258300781, 142.8159942626953, 95.82666015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000350.npy"}
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 46.9192008972168, "std": 128.6351318359375, "min": -246.032470703125, "p10": -104.60416412353516, "median": 32.41917419433594, "p90": 224.0183639526367, "max": 408.58465576171875, "pos_frac": 0.6171875, "sample": [0.0, 214.882080078125, 1.499755859375, 287.83209228515625, 20.480987548828125, 250.1133575439453, 39.30976867675781, 141.79736328125, 172.76483154296875, 17.291778564453125, 109.72042846679688, -108.490234375, 61.61553955078125, 70.35733032226562, 273.330322265625, -147.61778259277344, -10.581148147583008, 214.21266174316406, -102.93870544433594, 46.28375244140625, 29.181411743164062, 36.56109619140625, -101.31466674804688, 313.38055419921875, 284.700927734375, 42.38646697998047, -7.028226852416992, -6.85260009765625, -42.601165771484375, 53.522682189941406, 408.58465576171875, 184.830322265625, -27.275482177734375, -16.10649871826172, 141.49075317382812, -46.259063720703125, -8.01765251159668, 196.9447021484375, -26.751235961914062, -29.258026123046875, -129.06256103515625, 22.198394775390625, 53.1832160949707, 109.55316162109375, 27.8868408203125, 106.19436645507812, 50.23235321044922, 94.84681701660156, -24.711578369140625, -154.005126953125, -129.5093994140625, -80.56109619140625, -20.1195068359375, 75.370361328125, -85.43511962890625, -5.66326904296875, 165.65646362304688, -49.29046630859375, -40.397613525390625, 15.2803955078125, 203.19924926757812, 199.00436401367188, 40.135948181152344, 281.47760009765625, 190.39088439941406, -228.50360107421875, 139.23765563964844, 82.30543518066406, 111.41888427734375, -115.05299377441406, 8.898651123046875, 104.03987884521484, -205.6361083984375, -152.6756134033203, 327.9507141113281, -206.563720703125, 102.9256591796875, -28.609588623046875, -91.75341796875, 99.73834228515625, 0.0, 2.00445556640625, 96.69480895996094, 58.833099365234375, 229.37416076660156, -0.16288375854492188, 15.088302612304688, 252.08590698242188, 139.891845703125, -48.2188720703125, 80.73040771484375, 16.58282470703125, 40.199951171875, -24.507801055908203, 0.0, 128.25152587890625, 221.7230224609375, -29.1231689453125, 18.034507751464844, 75.85089111328125, -92.91397094726562, -240.79248046875, 122.5993423461914, 81.65180206298828, 31.13568115234375, 33.702667236328125, 283.76885986328125, 29.263824462890625, 254.69216918945312, -97.46339416503906, -10.674581527709961, 143.87289428710938, -13.2410888671875, 35.12101745605469, 29.792905807495117, 208.33944702148438, 65.8648452758789, -60.642242431640625, -85.437255859375, -95.7860107421875, 94.54142761230469, 198.8975830078125, -246.032470703125, 119.22015380859375, -75.1058349609375, -140.25289916992188, 265.8699035644531, 96.77452087402344], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000351.npy"}
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 90.0728759765625, "std": 123.41556549072266, "min": -201.57672119140625, "p10": -64.30436401367186, "median": 96.97893524169922, "p90": 253.85236358642575, "max": 361.2952880859375, "pos_frac": 0.75, "sample": [117.40093994140625, 141.78921508789062, 235.18109130859375, -77.25115966796875, 51.92607879638672, 325.9164123535156, 158.99856567382812, 121.73992919921875, 145.75778198242188, -8.471954345703125, -56.143402099609375, 69.02375030517578, -190.02728271484375, 60.26942443847656, 361.282470703125, 29.328781127929688, 86.0643310546875, 289.9622802734375, 242.1478729248047, 102.76638793945312, 303.4136657714844, -2.7700061798095703, -120.427978515625, 334.90399169921875, 361.2952880859375, -127.34600830078125, 7.871429443359375, 117.23641967773438, 84.4212646484375, 172.2281036376953, -36.42955017089844, -11.52362060546875, 172.17083740234375, 96.87907409667969, 166.78887939453125, 122.48924255371094, -4.0681610107421875, 37.7705078125, -4.0770263671875, 82.87608337402344, 208.64523315429688, -152.88507080078125, 8.218170166015625, 130.62673950195312, -179.00003051757812, 3.239105224609375, 0.0, 135.58367919921875, 14.545997619628906, 225.17813110351562, 68.57479858398438, 315.17913818359375, 1.580352783203125, 102.98419189453125, 270.7857360839844, 97.07879638671875, 105.98511505126953, 0.0, 122.587646484375, 151.5255126953125, 153.63475036621094, -7.582923889160156, 39.262420654296875, 185.009033203125, -201.57672119140625, -108.42160034179688, 26.851593017578125, 136.56820678710938, 67.568115234375, 74.55380249023438, 4.414588928222656, 261.86114501953125, 138.87527465820312, -12.671089172363281, -33.201904296875, -115.80435180664062, 160.9072265625, 122.51459503173828, -154.60400390625, -45.90997314453125, 246.53237915039062, 139.91116333007812, 177.35218811035156, 123.3163833618164, 67.03936767578125, 189.03326416015625, 52.2191162109375, -116.09808349609375, -4.03216552734375, 162.976318359375, -10.39501953125, -24.39649200439453, 143.97927856445312, 43.94582748413086, 189.650146484375, 162.26715087890625, 148.62103271484375, 87.18588256835938, 91.52078247070312, 277.8256530761719, 114.50633239746094, 76.29916381835938, 272.59210205078125, 126.21722412109375, 235.09967041015625, -58.7557373046875, 250.42002868652344, 120.61477661132812, 198.3482666015625, 34.72802734375, 105.03033447265625, 53.23735809326172, 343.88055419921875, 48.136474609375, -92.2373046875, -33.67585754394531, 148.3925018310547, 301.5721435546875, 175.61932373046875, 223.15121459960938, -36.1236572265625, -99.0140380859375, 141.79800415039062, 218.36886596679688, 20.143577575683594, 121.81080627441406, 34.74370574951172, 51.92425537109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000352.npy"}
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 74.90514373779297, "std": 124.37012481689453, "min": -254.79177856445312, "p10": -63.39706878662109, "median": 55.36761474609375, "p90": 227.59458007812495, "max": 420.89276123046875, "pos_frac": 0.7109375, "sample": [163.34381103515625, 100.2593994140625, 164.7137451171875, 135.56268310546875, 39.63957214355469, 200.77359008789062, -38.991363525390625, 69.08001708984375, 63.707183837890625, -4.48541259765625, -60.5633544921875, 50.64361572265625, -63.40541076660156, 189.6333465576172, -13.982177734375, 44.533203125, -115.32656860351562, -44.4974365234375, 53.926239013671875, -63.39349365234375, 9.957107543945312, 143.1468505859375, -22.859710693359375, -196.1938934326172, 113.20864868164062, -6.58251953125, 239.70623779296875, 163.91717529296875, 308.2783203125, -13.560310363769531, 420.89276123046875, 203.8028564453125, 102.49948120117188, -86.95944213867188, 124.3131103515625, -254.79177856445312, 346.4663391113281, 173.4942626953125, 167.99798583984375, 290.234130859375, 100.87820434570312, -103.07330322265625, 83.04898071289062, 262.4485778808594, 118.98091125488281, -11.026123046875, 168.42306518554688, 419.0722351074219, 307.18707275390625, 10.092849731445312, 352.00616455078125, 99.66543579101562, -220.6123504638672, 130.2831573486328, 56.808990478515625, 211.9593505859375, 12.631601333618164, 53.55406951904297, 41.47225570678711, 150.8307342529297, 159.2828369140625, 38.718109130859375, -39.75361633300781, -81.0693359375, 7.16607666015625, 32.41981506347656, -8.1546630859375, -66.1611328125, 53.01263427734375, 38.661163330078125, 48.07344055175781, 3.5303955078125, 222.40386962890625, 171.4849853515625, 245.54833984375, 119.28933715820312, -0.5634231567382812, 190.77098083496094, -19.333831787109375, -15.870597839355469, 147.73147583007812, -20.908737182617188, -22.929031372070312, 68.34567260742188, 142.3892822265625, 352.9674072265625, 222.38140869140625, 67.7474365234375, 2.9256744384765625, -109.5287857055664, -0.9873085021972656, 33.648345947265625, 171.20648193359375, 100.19862365722656, 32.63362121582031, 17.17742919921875, -8.728315353393555, -22.46532440185547, -14.371383666992188, -7.784454345703125, 137.900634765625, 12.944625854492188, 49.9036865234375, 60.33619689941406, 213.0703887939453, 101.15020751953125, 240.84332275390625, 83.15020751953125, 253.99472045898438, -158.37294006347656, 35.67023468017578, 12.90130615234375, 157.1072998046875, 85.87274169921875, 11.160484313964844, 0.0, 136.30938720703125, 133.60308837890625, -178.67868041992188, -86.6329345703125, 23.16973876953125, -45.581085205078125, 2.956390380859375, 157.72579956054688, 95.74586486816406, 98.09014892578125, 213.1795196533203, 146.46234130859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000353.npy"}
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 56.87559509277344, "std": 122.25198364257812, "min": -259.02569580078125, "p10": -90.48443145751952, "median": 41.417335510253906, "p90": 218.42377929687498, "max": 384.31781005859375, "pos_frac": 0.671875, "sample": [29.62067413330078, -68.54367065429688, -136.0247802734375, 219.64181518554688, 133.52203369140625, 363.2945556640625, 109.20748901367188, 34.841033935546875, 42.085540771484375, 34.883575439453125, 173.04525756835938, 79.79559326171875, 39.057838439941406, 35.02192687988281, 258.8664855957031, 85.57088470458984, -53.96521759033203, -20.423080444335938, -4.1861114501953125, -18.709197998046875, 36.043052673339844, 61.11933898925781, 282.01202392578125, -124.31356811523438, 98.584228515625, 38.44023513793945, -25.077415466308594, -12.645126342773438, 163.44528198242188, 334.8330078125, -47.467010498046875, -25.78668212890625, -28.40451431274414, 298.8393859863281, 124.50985717773438, 8.168182373046875, 40.74913024902344, -14.517822265625, 96.85374450683594, 246.29074096679688, 96.14810180664062, -74.115966796875, 94.38748168945312, -9.317718505859375, 73.81967163085938, 168.44686889648438, -97.38998413085938, 9.709125518798828, -182.59048461914062, 93.98001098632812, 121.32281494140625, 275.0087890625, 128.64471435546875, 306.7044982910156, 108.97076416015625, 268.9180908203125, 195.28952026367188, 75.33871459960938, 55.792236328125, 17.463348388671875, 102.29415893554688, -41.99249267578125, -126.83942413330078, -149.06591796875, 86.17587280273438, 47.70960998535156, 70.87422943115234, 384.31781005859375, 22.907150268554688, 1.5090560913085938, 199.2232666015625, -24.155776977539062, 160.38998413085938, 8.248226165771484, 217.90176391601562, 101.783447265625, 30.624954223632812, -27.73455047607422, -30.700347900390625, -2.975860595703125, 98.95849609375, -167.27670288085938, 139.762451171875, 56.022796630859375, -35.60113525390625, 54.267913818359375, 64.28271484375, 192.6676025390625, 36.46165466308594, -26.08977508544922, 119.86795043945312, -181.54388427734375, 92.72138977050781, 2.0866622924804688, 126.82843017578125, 65.81974792480469, -8.688011169433594, -11.845916748046875, -30.880203247070312, -60.53401184082031, 162.2035369873047, 7.644744873046875, 13.814849853515625, -20.101776123046875, 152.115234375, 109.60116577148438, -35.9564208984375, -58.7872314453125, 116.72915649414062, 329.186279296875, 187.1927490234375, -4.932168960571289, -161.24072265625, 233.75704956054688, -87.89518737792969, 131.53948974609375, -259.02569580078125, 170.49169921875, -204.0484619140625, 18.0621337890625, 105.4324951171875, 145.83358764648438, -96.5260009765625, 92.52820587158203, 19.8624267578125, -139.4111785888672, 178.13442993164062, 1.279083251953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000354.npy"}
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 85.92352294921875, "std": 139.39344787597656, "min": -340.61181640625, "p10": -42.167584991455065, "median": 71.97336196899414, "p90": 252.03102416992186, "max": 639.0926513671875, "pos_frac": 0.7890625, "sample": [39.54680633544922, -147.16943359375, 170.46893310546875, 57.32438659667969, -7.0863037109375, 3.5262069702148438, 66.99176025390625, 9.485931396484375, 247.42877197265625, 67.59405517578125, 330.207763671875, 17.40367889404297, 19.470855712890625, 160.6001739501953, 0.06802177429199219, 143.76425170898438, 204.4680938720703, 197.68051147460938, -7.55224609375, 19.392578125, -217.10009765625, -19.906036376953125, 41.22498321533203, 462.49200439453125, 13.65435791015625, 141.93719482421875, -5.529144287109375, 21.17340087890625, 53.8447265625, 127.46127319335938, 69.729736328125, -26.3779296875, 48.590667724609375, 639.0926513671875, 292.13555908203125, 56.986000061035156, 214.52401733398438, 100.27159118652344, 28.46234130859375, 87.87098693847656, 171.84410095214844, 142.12374877929688, 96.83551025390625, 86.6761474609375, -196.93490600585938, -18.35040283203125, 133.842529296875, 16.205955505371094, 205.73135375976562, 30.24853515625, 85.85610961914062, 145.9392547607422, 188.0841064453125, 23.104598999023438, 64.1324462890625, 65.772705078125, 32.48536682128906, 158.50424194335938, 31.09619140625, -121.72740173339844, 119.08657836914062, 26.838268280029297, -103.33639526367188, -10.586135864257812, 73.04755401611328, -193.6519317626953, 50.871498107910156, 308.99395751953125, 187.01840209960938, 192.17388916015625, 6.621450424194336, 99.76150512695312, -232.9688720703125, 163.55889892578125, -33.81327819824219, 23.007476806640625, 295.65814208984375, 128.72674560546875, 285.9792785644531, 105.64404296875, 121.091552734375, 215.7793731689453, -9.3583984375, -24.63232421875, 189.2445526123047, -340.61181640625, 149.71493530273438, -65.82730102539062, 84.99594116210938, 136.86512756347656, 205.44708251953125, -15.66290283203125, -275.63531494140625, 70.899169921875, 14.2218017578125, 34.752197265625, 116.169189453125, 152.88235473632812, -128.61053466796875, 157.9656982421875, 68.33538818359375, -3.9390029907226562, 203.3564453125, 144.36767578125, 369.7160949707031, 206.25587463378906, 257.8862609863281, 278.7841491699219, -30.909820556640625, 21.09979248046875, 73.36520385742188, 266.91802978515625, -52.10334777832031, 109.11392211914062, 199.93560791015625, 183.56491088867188, 367.72735595703125, 121.22076416015625, 76.39506530761719, 139.23977661132812, -37.909400939941406, 51.52716064453125, 18.0611572265625, 292.38043212890625, 63.38800048828125, -111.7679443359375, 249.52163696289062, 124.73553466796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000355.npy"}
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 59.14048385620117, "std": 100.6990737915039, "min": -200.3653564453125, "p10": -53.695517730712886, "median": 40.98690414428711, "p90": 194.71952209472653, "max": 358.87054443359375, "pos_frac": 0.7109375, "sample": [25.856658935546875, -52.066925048828125, 212.02955627441406, -24.9053955078125, 96.7066650390625, -51.5279541015625, -58.46875, -65.5924072265625, 3.9355220794677734, 26.794143676757812, 5.2550811767578125, 26.471710205078125, 66.99380493164062, 115.24859619140625, -10.697786331176758, 151.28280639648438, 48.744441986083984, -114.05610656738281, 48.4580078125, -14.963409423828125, -95.29150390625, 152.0863494873047, 22.775741577148438, 122.175048828125, -26.647165298461914, 358.87054443359375, -26.908187866210938, 25.966110229492188, -6.67803955078125, 123.15521240234375, 6.316253662109375, 90.43771362304688, 191.87789916992188, 39.01338195800781, 167.8223419189453, 207.55255126953125, 100.47921752929688, 191.55816650390625, 184.68577575683594, 201.3499755859375, -77.67803955078125, 93.48318481445312, 0.8077392578125, -1.2089958190917969, 61.62452697753906, 90.39273071289062, -30.5504150390625, 246.04293823242188, 252.22039794921875, 0.07799720764160156, -52.17597961425781, 32.86297607421875, 230.14443969726562, 70.352783203125, -8.621805191040039, 249.72369384765625, -73.70564270019531, 155.05990600585938, 136.66851806640625, 32.78533935546875, 136.249755859375, -139.81260681152344, 120.01539611816406, 69.0709228515625, 114.26716613769531, 129.13360595703125, 220.76739501953125, -28.379257202148438, -53.42286682128906, 26.001800537109375, 25.73694610595703, 33.387603759765625, 105.83877563476562, 62.326385498046875, 6.0059814453125, 53.701236724853516, 39.84374237060547, -43.512428283691406, 60.17427062988281, -59.09056091308594, 26.147048950195312, -54.331703186035156, 287.46087646484375, 151.2892303466797, 44.68400573730469, 96.60540771484375, 182.857421875, 81.42072296142578, -28.942113876342773, 4.0364227294921875, 21.602462768554688, -121.37633514404297, -37.93858337402344, -166.284912109375, 202.90084838867188, 73.60824584960938, 50.69000244140625, 0.0, 22.24835205078125, 137.557861328125, 131.66603088378906, 32.916175842285156, -11.989446640014648, 139.25555419921875, -6.42399787902832, 42.13006591796875, 65.05133056640625, 46.777442932128906, -5.89984130859375, 27.83336067199707, 37.888641357421875, 279.87774658203125, 163.0494384765625, 116.29168701171875, -21.240615844726562, 75.34690856933594, 182.32061767578125, 32.06916046142578, 181.85394287109375, 162.69253540039062, -111.09259033203125, 118.28144836425781, 32.09222412109375, -200.3653564453125, 189.985107421875, -13.399543762207031, -46.548248291015625, 202.6219482421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000356.npy"}
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 75.67054748535156, "std": 128.61109924316406, "min": -220.73727416992188, "p10": -68.44936141967773, "median": 50.66064453125, "p90": 237.0452911376953, "max": 425.89947509765625, "pos_frac": 0.6953125, "sample": [65.65155029296875, 274.499755859375, -33.064109802246094, 88.22128295898438, 109.53871154785156, 88.0694580078125, 0.0, 5.2806549072265625, 188.34559631347656, -29.176162719726562, 199.14501953125, -92.13446044921875, -19.65656280517578, 330.26458740234375, 348.6544189453125, 155.4837646484375, -16.848918914794922, 236.6748046875, 120.07414245605469, -23.70550537109375, 153.74447631835938, 50.847900390625, 123.26043701171875, 332.723388671875, 263.4730224609375, -105.7374267578125, 425.89947509765625, 207.51080322265625, 182.57589721679688, 237.90975952148438, 28.61614990234375, 20.422454833984375, -66.11225128173828, 16.318050384521484, 178.764404296875, 124.13123321533203, -5.6720428466796875, -40.464515686035156, 128.09909057617188, 89.5594253540039, 88.99897766113281, 305.680419921875, 59.00172424316406, 205.77005004882812, 79.8998031616211, -172.02182006835938, -56.89910888671875, -127.85194396972656, 425.06341552734375, -74.06390380859375, 11.017929077148438, 35.76830291748047, 52.582664489746094, 19.038665771484375, 22.239540100097656, 187.42123413085938, 111.94309997558594, 202.25149536132812, 125.46502685546875, -73.90261840820312, 170.1715087890625, -65.94363403320312, 37.713409423828125, -26.762115478515625, 169.5225830078125, -29.92406463623047, 158.7943572998047, 212.81553649902344, 47.996978759765625, 185.31411743164062, -102.93528747558594, -185.69827270507812, 10.984420776367188, 29.18157958984375, 23.189651489257812, 186.8620147705078, 75.15115356445312, -80.31289672851562, 41.958038330078125, -34.254150390625, -164.68601989746094, -58.773895263671875, 61.41368103027344, 30.790679931640625, 243.83319091796875, 11.281951904296875, 106.751708984375, -42.423309326171875, 40.459075927734375, -3.2913055419921875, -116.47906494140625, 50.473388671875, 144.99269104003906, -0.35515594482421875, 153.0831298828125, 171.6298065185547, -13.351470947265625, 20.390037536621094, 0.0, 293.0016784667969, -36.62446975708008, 0.0, 49.8087158203125, -78.26455688476562, 129.168212890625, 19.019193649291992, 162.69375610351562, 231.01217651367188, 106.83966064453125, 3.531585693359375, 233.150634765625, -220.73727416992188, 385.4028625488281, 153.0422821044922, 15.856109619140625, 81.2806396484375, -36.620269775390625, 137.36090087890625, 200.33375549316406, 138.12869262695312, 301.6574401855469, 4.186662673950195, -47.72399139404297, -63.598785400390625, 232.78329467773438, -49.99641418457031, 2.635894775390625, 110.34657287597656], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000357.npy"}
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 88.76396942138672, "std": 125.29612731933594, "min": -260.9207763671875, "p10": -46.30629348754882, "median": 81.36621856689453, "p90": 246.56391601562498, "max": 361.5056457519531, "pos_frac": 0.7421875, "sample": [-35.67315673828125, -28.024044036865234, 122.10379028320312, 128.37872314453125, 31.67010498046875, 31.055091857910156, -5.6895751953125, 161.80172729492188, 217.54803466796875, 283.28045654296875, 48.35429000854492, 45.039207458496094, 116.07144165039062, 61.11305236816406, 201.6336669921875, 171.0345916748047, 64.44830322265625, 292.87908935546875, 236.31068420410156, 77.1416015625, -126.57073211669922, 228.1947021484375, -28.971466064453125, -12.1361083984375, 161.83013916015625, -0.09433746337890625, 170.73532104492188, -9.88140869140625, 242.47906494140625, -9.22906494140625, 19.31964111328125, 186.27821350097656, 9.664108276367188, 169.52395629882812, 85.32391357421875, 16.774017333984375, 326.7384033203125, 12.86669921875, 295.7642822265625, 210.905029296875, -147.307861328125, 180.7010498046875, 135.99639892578125, 251.67245483398438, 1.3643798828125, 190.39138793945312, 139.655029296875, 236.45245361328125, 240.52037048339844, 152.3409423828125, -129.94790649414062, -26.894332885742188, -80.15119171142578, 71.14617156982422, 196.9395751953125, -13.774566650390625, 221.01171875, 166.846923828125, -44.90174865722656, 297.221435546875, 43.254852294921875, -163.72921752929688, 220.31494140625, -30.86609649658203, 6.0216522216796875, 236.34576416015625, 45.08160400390625, 45.53687286376953, 38.933746337890625, -30.065338134765625, 129.76312255859375, 58.562347412109375, 17.494361877441406, -93.54454040527344, -5.8597412109375, -260.9207763671875, 126.19670104980469, 108.66899108886719, -165.806884765625, 261.5384216308594, 110.79386138916016, -63.021820068359375, 81.233642578125, 93.0184326171875, 273.2915954589844, 33.852294921875, 175.55413818359375, 64.11785888671875, 147.77325439453125, 212.87286376953125, 13.279495239257812, 270.8370361328125, 361.5056457519531, 81.49879455566406, 137.60140991210938, 346.78607177734375, 23.143218994140625, 49.78668212890625, 15.1180419921875, 147.61312866210938, 259.65234375, 19.89666748046875, 125.88583374023438, 192.35955810546875, 195.2078857421875, 217.39968872070312, 271.621337890625, -230.548095703125, -50.777122497558594, -181.7027587890625, -16.99541473388672, 22.2996826171875, 182.41796875, 181.1314239501953, 120.7228012084961, -49.58356475830078, -28.824310302734375, -3.5972652435302734, 45.60990905761719, 87.1790771484375, 0.0, -18.839874267578125, 203.38232421875, 84.37603759765625, 244.37454223632812, -20.251068115234375, 82.32989501953125, 54.2100830078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000358.npy"}
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 72.80799102783203, "std": 116.60377502441406, "min": -234.0819091796875, "p10": -39.77718658447265, "median": 58.388702392578125, "p90": 230.74755859374997, "max": 461.87615966796875, "pos_frac": 0.7421875, "sample": [-38.752418518066406, -234.0819091796875, 162.30081176757812, 332.4832763671875, 235.74826049804688, -55.72459411621094, -4.4712371826171875, 33.80775451660156, 49.24517822265625, 76.87295532226562, 182.9208984375, 139.19879150390625, 63.21974563598633, -27.874347686767578, -23.949859619140625, -18.21221923828125, 143.43701171875, 171.85775756835938, 7.92620849609375, 67.59381103515625, 168.61480712890625, -52.7841796875, 211.7773895263672, 461.87615966796875, -143.433349609375, 160.92886352539062, -80.99209594726562, 170.0556640625, 258.056884765625, -212.4512939453125, 71.4415283203125, 268.13232421875, -10.623321533203125, 24.748382568359375, -7.991422653198242, 112.18942260742188, -27.657941818237305, 144.4556884765625, 146.95248413085938, 103.24261474609375, 43.236785888671875, -162.9243927001953, 116.74090576171875, -12.234039306640625, 118.27396392822266, 72.80316162109375, 18.431442260742188, 71.66110229492188, 62.940673828125, -1.6003799438476562, 94.86703491210938, 32.31327819824219, 279.9836730957031, -5.711887359619141, 59.10821533203125, 11.429946899414062, 270.1625671386719, -180.85772705078125, 147.4169464111328, 58.089111328125, 240.4166259765625, 38.55303955078125, -133.69085693359375, 201.03472900390625, 12.307373046875, -42.168312072753906, 13.47943115234375, 33.57806396484375, 195.3060302734375, 157.58355712890625, -0.546722412109375, 17.4957275390625, -67.53631591796875, 26.459793090820312, 42.31126403808594, 70.08067321777344, 21.404891967773438, 90.70571899414062, 81.67179870605469, 39.106475830078125, 55.08697509765625, 19.114456176757812, 16.807220458984375, 91.39959716796875, 252.69520568847656, 8.807586669921875, 96.681884765625, 129.57952880859375, 163.48965454101562, 20.93914794921875, 186.48951721191406, 106.99165344238281, 273.4585266113281, 173.23329162597656, 58.68829345703125, 5.897247314453125, -16.548095703125, 123.71636962890625, -12.611776351928711, -12.028936386108398, 43.083282470703125, 14.603271484375, 113.23019409179688, -91.63357543945312, 141.21087646484375, -168.46499633789062, 154.44174194335938, -16.7772216796875, 79.55499267578125, 375.7587890625, 40.20005798339844, -10.654220581054688, -6.339385986328125, -17.039358139038086, 23.9967041015625, 208.31741333007812, 312.57525634765625, 184.73611450195312, 122.4395751953125, 28.947494506835938, 12.739189147949219, 246.68270874023438, 94.25613403320312, 20.66106414794922, 204.50521850585938, 93.93058776855469, 228.60440063476562, -17.79852294921875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000359.npy"}
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 67.22767639160156, "std": 134.36245727539062, "min": -527.6532592773438, "p10": -69.2023193359375, "median": 59.10101318359375, "p90": 214.34069213867184, "max": 548.9246826171875, "pos_frac": 0.7265625, "sample": [127.52479553222656, 23.847061157226562, -130.37396240234375, 188.97824096679688, -108.47836303710938, -12.920967102050781, -168.4940185546875, 54.88897705078125, 548.9246826171875, -101.02093505859375, -7.822113037109375, 56.13206481933594, 86.95675659179688, 201.74615478515625, 14.9439697265625, 194.13348388671875, 57.28791046142578, 49.1214599609375, -83.29681396484375, 269.40380859375, 199.88925170898438, 79.87222290039062, 37.06622314453125, 197.0955352783203, -71.86004638671875, 94.55029296875, -14.914169311523438, 58.21533203125, -28.663055419921875, 189.81661987304688, 56.25634765625, 168.6885986328125, -36.89695739746094, 100.807373046875, 17.195556640625, 108.64767456054688, -18.52728271484375, 112.16876220703125, -68.06329345703125, 65.05787658691406, 225.2401123046875, 124.73165893554688, 236.9423828125, 81.43316650390625, 43.91558837890625, 61.54674530029297, -12.464691162109375, 248.02310180664062, -20.43438720703125, 48.20361328125, -52.66339111328125, 128.5653076171875, 103.82107543945312, -48.51837158203125, 485.68780517578125, -13.68267822265625, 241.19931030273438, -52.82635498046875, 137.52484130859375, 12.704418182373047, 133.12457275390625, 210.02777099609375, 158.311279296875, 152.85147094726562, 44.80535888671875, 135.47901916503906, 224.4041748046875, 62.00189208984375, 59.38230895996094, -176.4779052734375, -88.83778381347656, 178.31195068359375, -3.153371810913086, 160.8560791015625, -3.490753173828125, 70.86727142333984, 166.03590393066406, 276.1860046386719, 170.78659057617188, 92.01813507080078, 41.90824890136719, 173.44287109375, 26.55116081237793, 8.26468276977539, -527.6532592773438, 174.06910705566406, 230.9961395263672, -29.323272705078125, -75.47760009765625, 13.1981201171875, 170.45428466796875, 63.893402099609375, 288.30157470703125, 30.56220245361328, -3.7890453338623047, 75.8741455078125, 80.53860473632812, -370.5755615234375, 32.848785400390625, 4.611015319824219, 128.44424438476562, 0.0, 132.34130859375, 70.29180908203125, 236.2021484375, 58.81971740722656, 66.131103515625, 387.4130859375, -5.7938690185546875, 49.01858139038086, 81.97500610351562, -23.837188720703125, 22.224365234375, 146.87353515625, 23.960121154785156, -30.316177368164062, -78.86380004882812, -236.66583251953125, 0.0, 8.786117553710938, 36.5250244140625, 128.41343688964844, 128.8699951171875, 22.642242431640625, 5.9824371337890625, 66.80762481689453, 162.11312866210938, 97.76528930664062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000360.npy"}
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 90.28713989257812, "std": 138.626953125, "min": -288.3049011230469, "p10": -82.03281173706054, "median": 71.01022338867188, "p90": 290.5419982910156, "max": 470.5118713378906, "pos_frac": 0.7265625, "sample": [131.4347686767578, 38.37437438964844, -38.593719482421875, 97.932373046875, 39.7381591796875, 3.0911865234375, 8.363700866699219, -65.44674682617188, 134.62551879882812, -1.043670654296875, -80.43563079833984, 30.6519775390625, 3.8691864013671875, 158.3597412109375, -2.3036727905273438, 296.2628173828125, 19.290992736816406, 107.09834289550781, -79.5760498046875, 20.646240234375, 294.9061279296875, 280.7275390625, -87.65960693359375, 186.03794860839844, -14.597137451171875, 226.38217163085938, 180.88623046875, 162.03807067871094, 139.68014526367188, 290.8323974609375, 240.337646484375, 169.57066345214844, 172.9423370361328, 126.8845443725586, -3.4027328491210938, 181.17095947265625, -125.0577392578125, 200.13140869140625, 161.9127655029297, 159.01068115234375, 220.91958618164062, 375.8587951660156, -64.82138061523438, -9.474113464355469, 205.42050170898438, 61.6112060546875, -122.0777587890625, 59.27142333984375, 67.95492553710938, 7.447944641113281, -85.75956726074219, 132.80014038085938, 121.98483276367188, 0.0, 38.39997863769531, 60.77703857421875, 167.4380340576172, 377.4524230957031, 3.35980224609375, 125.59994506835938, 178.97686767578125, 177.01727294921875, 15.814935684204102, 308.86175537109375, 188.022705078125, -56.73486328125, 49.49114990234375, 92.87310791015625, 147.1875, -103.9232177734375, 143.0003662109375, -14.8818359375, -36.4530029296875, 106.23916625976562, -43.77151107788086, 0.0, 109.7794189453125, -63.91180419921875, 72.42503356933594, 68.07574462890625, 309.8377990722656, 157.01177978515625, 185.50726318359375, 96.53585815429688, -288.3049011230469, 144.56918334960938, 35.209869384765625, -90.82843017578125, 376.0975036621094, 148.11447143554688, -113.69276428222656, 69.59541320800781, -87.656982421875, 143.1962432861328, 354.6438293457031, 200.9634246826172, 36.83287811279297, 58.169464111328125, 27.247365951538086, 150.16207885742188, 290.41754150390625, -3.905813217163086, 321.672119140625, 67.89291381835938, 151.12890625, 470.5118713378906, 136.98638916015625, 0.385284423828125, 213.44561767578125, 427.69281005859375, 231.86181640625, 32.65690612792969, -170.27005004882812, 257.2780456542969, 307.9674987792969, 163.42501831054688, 3.0860595703125, 244.91802978515625, 62.13115692138672, -60.69859313964844, 157.02523803710938, -73.58683776855469, -162.02786254882812, -4.171630859375, -141.21826171875, -3.4220199584960938, 59.298492431640625, -184.26416015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000361.npy"}
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 75.93827819824219, "std": 108.02275848388672, "min": -182.09719848632812, "p10": -64.51896514892577, "median": 68.37569808959961, "p90": 211.8033233642578, "max": 415.5396728515625, "pos_frac": 0.796875, "sample": [3.826141357421875, 121.9376220703125, 53.316837310791016, 133.54083251953125, 164.20916748046875, 54.226654052734375, 222.85748291015625, 117.34489440917969, 15.63153076171875, 194.55545043945312, 70.93768310546875, -104.99099731445312, -100.70614624023438, -91.29640197753906, 160.21795654296875, -6.841907501220703, 168.18927001953125, 17.0692138671875, 183.0546417236328, 198.62396240234375, -17.307342529296875, 203.5328826904297, 317.62896728515625, 115.78900146484375, 156.63653564453125, 216.07366943359375, 55.58009338378906, -1.5777130126953125, 99.04469299316406, 108.87808227539062, -38.471282958984375, -105.52659606933594, 174.48683166503906, -54.005165100097656, 95.63831329345703, 41.709228515625, 4.1899566650390625, -8.407222747802734, 136.70751953125, 4.6691436767578125, 377.6207275390625, 8.718597412109375, 254.40869140625, 186.16915893554688, 238.19061279296875, 229.24285888671875, 172.92404174804688, 31.151641845703125, -19.1015625, 65.6502685546875, -48.8543701171875, 200.74679565429688, 72.275390625, 4.0787353515625, 13.264404296875, 22.4638671875, 76.5692138671875, 214.57424926757812, -60.54705810546875, 79.62849426269531, 50.577423095703125, 26.15203857421875, 59.79766845703125, 155.27728271484375, 83.37892150878906, 79.88735961914062, 7.842262268066406, 81.06845092773438, 46.91412353515625, 177.657470703125, 415.5396728515625, 69.22607421875, 1.1597137451171875, 11.4217529296875, 46.27691650390625, 210.61578369140625, -142.62835693359375, 67.8896713256836, 151.44549560546875, 217.08285522460938, -97.31690979003906, 73.66299438476562, -83.85660552978516, 157.14913940429688, -36.6531982421875, -22.54876708984375, -131.6204376220703, 274.2908935546875, 16.1455078125, 11.442108154296875, 55.61810302734375, 123.37018585205078, 34.10834503173828, 72.22090148925781, 141.4827880859375, 35.18858337402344, 155.59584045410156, 22.087722778320312, 102.08534240722656, 65.05239868164062, -56.320343017578125, 170.18002319335938, 233.54904174804688, 49.0982666015625, 17.132179260253906, -76.20501708984375, -68.83888244628906, 120.65328979492188, 115.13836669921875, 48.18878173828125, 125.99647521972656, 25.453908920288086, 162.0142364501953, 40.58317565917969, 36.70977783203125, -182.09719848632812, 92.67898559570312, 285.90777587890625, 68.86172485351562, 36.40462112426758, -97.62469482421875, 202.20144653320312, 146.21307373046875, 131.7855224609375, -100.73712158203125, -62.667572021484375, 166.49346923828125, 105.11041259765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000362.npy"}
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 76.81553649902344, "std": 135.2577362060547, "min": -229.63140869140625, "p10": -92.03034210205078, "median": 63.413970947265625, "p90": 231.28226623535156, "max": 583.7373046875, "pos_frac": 0.703125, "sample": [173.62506103515625, -10.968940734863281, -25.71697998046875, 276.9046630859375, 146.3123779296875, 77.7115478515625, -5.639764785766602, -118.73236083984375, -38.05145263671875, 348.0566711425781, -225.08078002929688, 38.64213562011719, 187.83151245117188, 147.69119262695312, -17.127525329589844, -127.2939453125, 74.81719970703125, 349.0902099609375, 13.920257568359375, -33.72947692871094, 27.0938720703125, 200.39822387695312, 202.34771728515625, 25.57012939453125, 62.34278869628906, 583.7373046875, 305.037109375, 148.69244384765625, 45.53340148925781, -25.605857849121094, 73.15707397460938, -3.6485137939453125, 13.0518798828125, 13.503265380859375, 169.25184631347656, 4.973783493041992, -105.01021575927734, 154.97955322265625, 253.0374755859375, -20.221435546875, 64.45903015136719, 38.00916290283203, 89.9593505859375, 171.88897705078125, 176.549560546875, 155.69635009765625, -90.61601257324219, 41.8773193359375, 75.92613220214844, 47.959259033203125, 131.8695068359375, 29.157106399536133, 418.0252685546875, 158.18115234375, -165.31524658203125, 63.022216796875, 13.71588134765625, 103.47274780273438, 58.185302734375, 377.023681640625, 47.4447021484375, 112.86962890625, 52.53852844238281, 177.79891967773438, 147.85919189453125, -58.65234375, 81.55035400390625, 107.39653015136719, 101.0518798828125, -12.506975173950195, 217.97671508789062, -9.435638427734375, 197.66571044921875, 259.85369873046875, -4.7944488525390625, 104.22320556640625, -17.09136962890625, -83.68461608886719, 10.305816650390625, 55.95599365234375, -0.6539764404296875, 274.68499755859375, -98.19842529296875, -33.682464599609375, 112.11249542236328, -201.46778869628906, -99.78378295898438, -26.943359375, -95.3304443359375, 233.82846069335938, 258.58221435546875, -156.1497802734375, 213.21310424804688, -66.00739288330078, -131.3349151611328, 198.84335327148438, 180.63421630859375, -21.427337646484375, 217.56524658203125, 160.05589294433594, 34.49894714355469, 82.87684631347656, -229.63140869140625, 209.9093017578125, 315.50189208984375, 53.027923583984375, 94.68001556396484, 171.46627807617188, 203.15179443359375, 184.84994506835938, -59.627899169921875, -157.44436645507812, -57.85272216796875, 17.532516479492188, 41.506134033203125, 117.6103515625, 15.7484130859375, 110.64060974121094, 179.69650268554688, 103.6351318359375, 171.42230224609375, 211.17123413085938, 63.80572509765625, 62.568389892578125, -56.47015380859375, 230.1910400390625, 68.44851684570312, -76.92074584960938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000363.npy"}
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 69.73213195800781, "std": 138.41310119628906, "min": -253.55984497070312, "p10": -83.73093719482422, "median": 43.24016189575195, "p90": 245.33356933593748, "max": 442.3406982421875, "pos_frac": 0.6875, "sample": [296.6170959472656, 20.29315185546875, 146.777099609375, 48.07902526855469, 196.45977783203125, 68.00547790527344, 231.1361083984375, 127.14056396484375, 100.66860961914062, -27.095489501953125, 129.59722900390625, 8.432243347167969, -14.977386474609375, 108.27130126953125, 29.1683349609375, 1.261810302734375, 418.6161804199219, -90.59686279296875, 50.27012634277344, 231.7247772216797, -222.3360595703125, 2.1529388427734375, 175.44781494140625, -164.48397827148438, 140.38275146484375, -253.55984497070312, 204.22174072265625, 150.41200256347656, -31.541015625, 140.30300903320312, 40.4931640625, 36.45375061035156, -25.167022705078125, 9.680835723876953, 36.11456298828125, 35.51240539550781, 43.682037353515625, 37.0701904296875, 49.19474792480469, 232.9729461669922, 84.69479370117188, -2.932220458984375, -230.04891967773438, 250.9759521484375, 57.446983337402344, 198.0419921875, -144.8553924560547, -25.5267333984375, -84.26925659179688, 49.638763427734375, 128.12425231933594, 168.13534545898438, -14.1407470703125, 309.645751953125, 26.73906707763672, -33.81568908691406, 242.9154052734375, -14.814836502075195, -250.47882080078125, 416.255859375, 73.27639770507812, 27.8284912109375, 19.72332763671875, -83.50022888183594, 286.8356018066406, 154.72645568847656, -23.5989990234375, 337.2672119140625, 3.410980224609375, 49.7701416015625, 62.886962890625, 215.96749877929688, 152.75396728515625, -28.673118591308594, 68.25545501708984, -18.4976806640625, 228.812255859375, 97.41445922851562, -55.313018798828125, 275.68402099609375, 157.46148681640625, -19.58319091796875, 23.85991668701172, -63.91131591796875, -8.17230224609375, 175.22756958007812, 40.14369201660156, 191.38153076171875, 19.961685180664062, 400.426513671875, 207.6773681640625, 10.764633178710938, -67.59036254882812, -99.07806396484375, 79.60279083251953, 241.26229858398438, -38.985870361328125, -131.13784790039062, 85.79058837890625, 293.05908203125, 43.461029052734375, 156.5032958984375, -183.50732421875, 234.76654052734375, 175.8192138671875, -132.3733673095703, 154.05810546875, 83.04269409179688, -5.298828125, -64.53985595703125, 38.602943420410156, -25.301788330078125, 12.439109802246094, 255.56048583984375, 43.01929473876953, 71.30354309082031, -120.01385498046875, 68.70976257324219, 442.3406982421875, -10.40499496459961, 38.67364501953125, -1.9260330200195312, 41.17121887207031, -23.4852294921875, -65.5433349609375, 219.821044921875, 310.6783752441406, -23.640960693359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000364.npy"}
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 70.25199127197266, "std": 126.69818115234375, "min": -190.57876586914062, "p10": -78.83983993530273, "median": 49.63584518432617, "p90": 245.2792724609375, "max": 379.74224853515625, "pos_frac": 0.71875, "sample": [0.0, 38.48055648803711, 70.09646606445312, 37.430885314941406, 231.50537109375, 232.67837524414062, 101.7232666015625, 66.92068481445312, 0.0, 41.8209228515625, 240.626708984375, 54.249664306640625, 237.97747802734375, 296.7852783203125, 40.05786895751953, 20.19488525390625, 14.53070068359375, 30.03192901611328, -21.79107666015625, 207.00070190429688, -57.41447448730469, 118.359375, -26.26873779296875, 12.806694030761719, 52.69940185546875, 156.9576416015625, -122.09361267089844, 132.4686279296875, 179.55421447753906, 218.39007568359375, 61.983055114746094, -84.00975799560547, -63.82781982421875, 129.38897705078125, -78.02569580078125, -67.21734619140625, 55.6671142578125, 140.55267333984375, 52.11351776123047, 83.95486450195312, -38.565879821777344, 174.00595092773438, 197.798095703125, -64.69931030273438, -70.74065399169922, 264.01788330078125, -22.95294189453125, 107.19923400878906, 6.405994415283203, 181.85723876953125, 33.36408996582031, -115.7491455078125, 23.33984375, -174.37213134765625, 16.7020263671875, 66.71795654296875, 161.70391845703125, 366.897216796875, 245.37826538085938, 89.12525939941406, 43.58624267578125, 257.070556640625, -91.10211181640625, 28.02923583984375, 11.020179748535156, -25.9312744140625, 176.5691375732422, 151.9769287109375, 134.51217651367188, 256.26348876953125, 183.10891723632812, 118.98313903808594, 66.61822509765625, -48.240753173828125, -106.48831176757812, 193.53939819335938, 177.73733520507812, -190.57876586914062, 379.74224853515625, -16.344993591308594, 308.98565673828125, 0.6042938232421875, 261.30792236328125, 35.93317413330078, 32.072654724121094, 47.158172607421875, 282.78472900390625, 321.0086669921875, 100.12210845947266, -80.73950958251953, -72.51142120361328, 20.230667114257812, 0.4327392578125, 224.95779418945312, 89.47335052490234, 74.41055297851562, -23.13953399658203, -56.98875427246094, 7.6417999267578125, 6.03460693359375, 139.11195373535156, -186.36697387695312, 3.904296875, 18.815765380859375, 235.37969970703125, 0.0, 99.84090423583984, -77.9563980102539, 38.86320495605469, 211.5673828125, -44.1590576171875, 81.54734802246094, -1.3763046264648438, -31.537734985351562, -121.29852294921875, -154.41647338867188, 267.93707275390625, 220.69729614257812, -141.159912109375, 103.14596557617188, 36.74330139160156, -190.3387451171875, 334.7039794921875, 73.98056030273438, 6.5745849609375, 245.23684692382812, 161.80650329589844, 97.36526489257812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000365.npy"}
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 81.12327575683594, "std": 123.99300384521484, "min": -322.171875, "p10": -54.48197937011719, "median": 71.85516357421875, "p90": 229.15052490234373, "max": 388.22369384765625, "pos_frac": 0.7734375, "sample": [26.9058837890625, 27.12646484375, 251.24453735351562, 83.80398559570312, 178.509033203125, 12.89447021484375, 257.7865295410156, 101.10467529296875, -32.21855163574219, 198.55593872070312, 222.8887939453125, 83.08439636230469, 177.98922729492188, -87.49896240234375, -6.196281433105469, 162.53713989257812, 110.51748657226562, 172.91421508789062, -181.59487915039062, 115.79141235351562, -208.97796630859375, 228.32940673828125, 9.497386932373047, 181.17367553710938, 82.24267578125, 335.52520751953125, 167.33932495117188, 216.40032958984375, 276.5215759277344, 204.62155151367188, 31.256561279296875, 69.71139526367188, 134.10699462890625, -75.68295288085938, 231.06646728515625, -153.0849609375, -165.32025146484375, 357.9075927734375, 65.5396728515625, -14.641265869140625, 11.211566925048828, 10.746337890625, 10.062255859375, -65.55220031738281, -20.944549560546875, 115.57186889648438, 240.29995727539062, -23.6923828125, 182.03878784179688, -27.67657470703125, 44.61907958984375, 127.77388763427734, 11.571044921875, 169.82424926757812, 10.366851806640625, 135.76715087890625, 60.318756103515625, 253.7561798095703, 204.85458374023438, 20.588531494140625, -52.2867431640625, 172.49530029296875, 58.508758544921875, 138.1792449951172, 2.7995986938476562, 214.5264892578125, 57.5725212097168, 161.23431396484375, 35.04882049560547, -322.171875, 139.11355590820312, 357.365478515625, -32.20442199707031, 0.0, 0.78729248046875, 184.07861328125, 110.58645629882812, 182.84884643554688, 78.35275268554688, -2.855609893798828, -154.2518310546875, 63.667877197265625, 243.89468383789062, 127.06039428710938, 3.8909759521484375, 73.99893188476562, 107.38961791992188, 222.75595092773438, 3.5145111083984375, 24.02837371826172, 23.657943725585938, 15.8101806640625, 121.9182357788086, 126.45314025878906, 32.28798294067383, 158.8614501953125, 120.468017578125, -27.0736083984375, 190.828125, 200.0554962158203, 171.98040771484375, 45.6416015625, 24.296157836914062, 213.27777099609375, 183.3113250732422, 43.97859191894531, 151.5290069580078, -21.09691619873047, 361.1700744628906, 49.5330810546875, -77.0672607421875, 24.070585250854492, 247.21127319335938, -23.00048828125, -53.623809814453125, -27.062957763671875, 30.364601135253906, -147.77120971679688, 88.71475982666016, 89.04364013671875, 163.088134765625, 14.65185546875, -100.64152526855469, 148.93975830078125, 8.11566162109375, -40.99267578125, 388.22369384765625, -56.484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000366.npy"}
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 82.33177185058594, "std": 118.53003692626953, "min": -191.81008911132812, "p10": -60.966116333007804, "median": 77.934814453125, "p90": 233.84882812499998, "max": 454.3050231933594, "pos_frac": 0.734375, "sample": [200.17672729492188, 157.32144165039062, -75.73722839355469, 237.08642578125, -29.114593505859375, 58.29644775390625, 12.865184783935547, -120.82070922851562, -28.11297607421875, 34.293212890625, 152.7867431640625, -19.600481033325195, 124.276123046875, 12.805587768554688, 116.15194702148438, 189.53570556640625, -57.843902587890625, 162.6239013671875, -29.348876953125, 81.98635864257812, -8.315948486328125, 184.62136840820312, 93.2386474609375, -121.9412841796875, -9.85207748413086, 171.30914306640625, -166.70262145996094, -14.663818359375, 104.22886657714844, -29.883636474609375, -110.62097930908203, 80.8687744140625, 9.931844711303711, 20.059341430664062, 230.25442504882812, -123.62603759765625, 108.70071411132812, 63.135528564453125, 68.46319580078125, -56.591217041015625, 220.00177001953125, -78.76927947998047, 71.50035858154297, 155.98486328125, 136.326171875, 18.5009765625, -126.89559936523438, 96.9403076171875, 2.9486541748046875, 105.80680847167969, 253.1551513671875, 84.86212158203125, -57.90409851074219, 58.39910888671875, 283.1322326660156, 210.74996948242188, 181.29039001464844, 85.9165267944336, -81.80757141113281, -28.3419189453125, -10.555450439453125, -53.1002197265625, 292.12200927734375, 189.07614135742188, 30.107967376708984, 180.17507934570312, 61.14054870605469, 199.07044982910156, 374.9632873535156, 170.71878051757812, 128.33648681640625, 210.76634216308594, 85.18585205078125, 131.08172607421875, 236.2091064453125, 238.26480102539062, 10.28277587890625, 128.63638305664062, 91.32716369628906, 56.96826171875, 88.67572021484375, 288.05670166015625, 216.8035430908203, 292.77862548828125, -3.90869140625, 217.53152465820312, 189.76119995117188, -2.96844482421875, -91.29910278320312, 20.025314331054688, 171.9523162841797, 177.87957763671875, 18.644073486328125, -74.31521606445312, 153.10403442382812, -9.804641723632812, -45.087318420410156, 202.4278564453125, 121.20299530029297, 312.3419189453125, 168.2689666748047, 454.3050231933594, 89.54271697998047, 120.27496337890625, 0.60101318359375, 44.195823669433594, 75.0008544921875, 232.8372802734375, 50.656005859375, 82.6748046875, 30.269271850585938, -5.976978302001953, 72.21380615234375, -191.81008911132812, 258.6407470703125, -7.866920471191406, 16.1741943359375, 158.7403106689453, 72.96136474609375, 103.98095703125, 159.5692901611328, 5.2013397216796875, 23.176513671875, 8.25616455078125, 305.8412780761719, -68.11082458496094, -15.5970458984375, 35.828453063964844], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000367.npy"}
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 98.87788391113281, "std": 111.5403060913086, "min": -180.13832092285156, "p10": -23.597720527648917, "median": 99.19108581542969, "p90": 251.119221496582, "max": 410.6182861328125, "pos_frac": 0.828125, "sample": [21.46044921875, 154.01763916015625, 72.29072570800781, 163.1025390625, 1.846282958984375, 261.9540100097656, -14.139907836914062, -97.14402770996094, 42.64996337890625, -11.781494140625, 333.93048095703125, -32.48161315917969, 175.95741271972656, 294.44390869140625, 97.34689331054688, 128.71820068359375, 255.9571533203125, 53.6890869140625, 387.3187561035156, 49.49139404296875, 6.2727813720703125, -2.132080078125, 10.0545654296875, 14.453407287597656, 18.970458984375, 48.749755859375, 49.792930603027344, 18.772567749023438, -30.204025268554688, 239.2493896484375, 25.956817626953125, 168.96453857421875, -54.96758270263672, 190.82351684570312, 38.8690185546875, -169.14208984375, -51.2198486328125, 276.76214599609375, 72.9075927734375, 160.1109619140625, 73.99940490722656, 223.26580810546875, 122.02969360351562, 193.03704833984375, 19.411407470703125, -8.77142333984375, 116.55290222167969, 103.05668640136719, 55.720733642578125, -180.13832092285156, 21.8748779296875, 130.49777221679688, -6.2724151611328125, 92.93428039550781, 3.041259765625, -5.328338623046875, 192.233642578125, 170.05642700195312, -53.24554443359375, 162.7572021484375, 153.0536346435547, 124.10943603515625, 71.82489013671875, 305.77496337890625, 244.39999389648438, 68.5453109741211, -74.46563720703125, 208.14605712890625, 168.15499877929688, 19.57568359375, -16.016128540039062, 142.83193969726562, 172.45895385742188, 48.27626037597656, 110.62644958496094, 249.0458221435547, 111.19287109375, 12.06524658203125, 193.5870361328125, 10.817703247070312, 41.07354736328125, 101.0352783203125, 190.55307006835938, 108.48138427734375, 73.77619171142578, 123.49252319335938, 10.43505859375, 197.5834503173828, 175.7341766357422, 286.3526611328125, -20.766447067260742, 29.874717712402344, 267.16558837890625, 3.6407699584960938, 141.34140014648438, 65.14366912841797, 84.24779510498047, 86.83563232421875, 55.49143981933594, 142.9004669189453, 234.37319946289062, -18.257755279541016, 410.6182861328125, 148.468994140625, -81.3756103515625, 117.495849609375, -131.94851684570312, 159.91586303710938, 281.04638671875, 223.3035125732422, 52.2451171875, 150.9979248046875, 132.0063018798828, 24.8675537109375, 52.47092819213867, 150.17613220214844, 160.1204833984375, 177.996826171875, 108.09320068359375, 132.43209838867188, 102.80517578125, -107.41216278076172, -53.347381591796875, 264.1650390625, 198.68197631835938, 274.7584228515625, 2.8963623046875, 201.9915771484375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000368.npy"}
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 73.28047943115234, "std": 108.46711730957031, "min": -227.34730529785156, "p10": -50.749114990234375, "median": 58.983367919921875, "p90": 218.01635742187494, "max": 325.47625732421875, "pos_frac": 0.75, "sample": [77.11125183105469, 159.7303466796875, 166.23297119140625, -5.7247314453125, 7.819370269775391, 47.786224365234375, 13.030235290527344, 113.869384765625, 42.81742858886719, 51.69635009765625, 53.7197265625, 73.97744750976562, -121.89009857177734, 137.05911254882812, 235.28121948242188, 179.775634765625, 129.59075927734375, 158.83358764648438, 67.255615234375, 152.8521270751953, 183.59234619140625, -5.10235595703125, 101.49720001220703, 88.13021850585938, -126.90887451171875, -125.55819702148438, 50.335113525390625, -34.380859375, 181.21014404296875, 13.865829467773438, 145.67575073242188, 81.7003173828125, 132.683349609375, 261.83819580078125, 38.06658935546875, 242.31573486328125, 15.248764038085938, 72.65191650390625, 315.47998046875, 22.304397583007812, -64.97702026367188, -51.54998779296875, 207.4775390625, -34.778900146484375, 4.054901123046875, -53.41400146484375, -15.337419509887695, 145.91265869140625, 69.32720947265625, 325.47625732421875, 8.591018676757812, 18.606063842773438, 93.24856567382812, -9.0264892578125, 156.59066772460938, 210.41787719726562, -6.607612609863281, -184.30007934570312, 288.57598876953125, 23.885833740234375, -39.518218994140625, 1.4301605224609375, -19.34282684326172, 234.72669982910156, 158.75933837890625, 3.7096214294433594, 89.57052612304688, 40.637779235839844, 45.6556396484375, -5.563751220703125, 284.6348876953125, 26.574142456054688, 122.60961151123047, 6.172119140625, 27.29449462890625, 253.19586181640625, 240.86602783203125, 0.987548828125, -227.34730529785156, 169.63430786132812, 47.127723693847656, -167.99380493164062, -26.219558715820312, -56.55755615234375, -91.47332763671875, 136.58084106445312, 205.97227478027344, 267.3249206542969, 86.2646484375, 0.0, 110.17201232910156, 240.87086486816406, -9.804794311523438, 74.88873291015625, 56.951454162597656, 4.064422607421875, 42.19483947753906, -6.0709991455078125, 122.08792114257812, 210.8547821044922, -70.76649475097656, 196.89915466308594, 88.83935546875, 113.95724487304688, 70.34524536132812, 22.23693084716797, 283.9164123535156, 58.73095703125, 204.04052734375, -59.9232177734375, -5.14569091796875, 1.170013427734375, 45.3140869140625, 144.52743530273438, 80.72279357910156, -23.9420166015625, 75.33270263671875, 159.04248046875, 80.07476806640625, -50.4058837890625, 41.23749542236328, 209.41424560546875, 59.23577880859375, -21.939559936523438, 175.2580108642578, 193.60279846191406, 161.7821044921875, -43.2213134765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000369.npy"}
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 88.60447692871094, "std": 133.69761657714844, "min": -374.1166687011719, "p10": -49.98640899658203, "median": 83.66934204101562, "p90": 245.23503112792966, "max": 488.0614318847656, "pos_frac": 0.765625, "sample": [-132.19021606445312, -129.43490600585938, 167.37615966796875, 49.79447937011719, 243.68902587890625, 115.66719818115234, -105.26986694335938, -4.826904296875, 182.11273193359375, -3.087066650390625, 209.4141845703125, 35.643035888671875, -20.540054321289062, 22.267822265625, 146.91177368164062, -6.2320556640625, 10.029571533203125, 352.013671875, -48.85625457763672, 107.48806762695312, 329.62945556640625, 142.77438354492188, 184.261962890625, 222.03570556640625, -79.40093994140625, 97.88604736328125, -1.6037635803222656, 227.56198120117188, -27.519989013671875, 48.57659912109375, 21.90667724609375, -39.86700439453125, 200.0554962158203, 32.48587417602539, 55.866485595703125, 19.77252197265625, 310.89306640625, 48.837066650390625, 10.419479370117188, 183.17840576171875, 116.66364288330078, -211.6717987060547, 261.975830078125, 14.522628784179688, 44.614383697509766, 93.59004211425781, 236.51084899902344, 129.80035400390625, 22.140716552734375, -69.22704315185547, 95.5413818359375, 0.04734039306640625, 52.082977294921875, 144.73411560058594, 155.3251953125, -69.45745849609375, 205.4486083984375, 60.41839599609375, 294.6521911621094, -374.1166687011719, 118.63526916503906, 289.97998046875, 28.323577880859375, 82.49168395996094, 24.698450088500977, 184.37696838378906, 289.305419921875, 194.08311462402344, 154.57847595214844, 3.485197067260742, 31.9864501953125, -9.783416748046875, 59.664031982421875, 220.4171142578125, 234.5025634765625, 18.511932373046875, 226.04409790039062, 448.22119140625, 96.87503051757812, 343.0573425292969, 199.731689453125, 50.85820007324219, 36.646514892578125, 259.54364013671875, 240.33843994140625, 202.10964965820312, -35.11781311035156, 36.54248046875, -176.07359313964844, 180.16079711914062, 84.84700012207031, 127.69949340820312, -29.67889404296875, 86.09808349609375, 488.0614318847656, -52.623435974121094, 212.1325225830078, -19.06884765625, 85.158203125, 52.26275634765625, 64.89891052246094, 240.61758422851562, 105.35977172851562, 168.25189208984375, -48.0921630859375, 98.70698547363281, 248.84237670898438, 138.53961181640625, 122.31143188476562, 139.11944580078125, 278.47967529296875, -206.09674072265625, -99.01455688476562, -29.507080078125, 43.69569396972656, 186.19827270507812, 198.12786865234375, 20.288692474365234, 111.90625, 39.753509521484375, 23.4354248046875, -4.519920349121094, -28.547225952148438, 107.96577453613281, -33.687095642089844, -159.8676300048828, 159.8984832763672, 1.910400390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000370.npy"}
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 65.8328628540039, "std": 118.28555297851562, "min": -366.0921630859375, "p10": -62.643922805786126, "median": 51.36765670776367, "p90": 217.1009384155273, "max": 485.8105773925781, "pos_frac": 0.7421875, "sample": [130.02639770507812, 83.0533447265625, 104.94647216796875, 98.76484680175781, -6.010951995849609, -1.8481292724609375, 248.64532470703125, 36.7568359375, -6.598785400390625, 47.36236572265625, 3.8381195068359375, 24.666748046875, -49.97137451171875, 257.73779296875, 46.07513427734375, 116.5867919921875, -90.67388916015625, 105.04896545410156, 33.91473388671875, 29.821533203125, -85.28240966796875, 209.78915405273438, -28.726045608520508, 159.05078125, 78.39186096191406, 174.77984619140625, 5.620027542114258, 63.48876953125, -104.54655456542969, 61.06329345703125, 68.95388793945312, 278.2264404296875, -21.817672729492188, -27.95819091796875, -221.73080444335938, 209.01690673828125, -197.31948852539062, 13.194778442382812, 107.37556457519531, -0.81500244140625, 191.311767578125, 51.10162353515625, 109.91017150878906, -42.82427978515625, 28.8165283203125, 55.359771728515625, 34.349700927734375, 0.0, -73.495361328125, 33.49317169189453, 245.636474609375, 165.03976440429688, 7.816650390625, -366.0921630859375, -101.18698120117188, 8.27899169921875, 61.57965087890625, 244.14617919921875, 116.073974609375, 43.63237762451172, -26.232608795166016, 54.929443359375, 83.62661743164062, 366.8379211425781, 27.194091796875, 34.63458251953125, 57.232940673828125, -8.11199951171875, 153.57855224609375, -67.55192565917969, 485.8105773925781, 128.20059204101562, 212.07754516601562, -44.125, 56.999908447265625, 14.281257629394531, 53.96331024169922, 184.53863525390625, 107.485107421875, -87.29963684082031, -82.18753051757812, 26.07220458984375, 9.596954345703125, 200.18638610839844, 33.28655242919922, 70.79269409179688, -191.76907348632812, 46.55397033691406, 57.37823486328125, -42.62870788574219, 2.3620758056640625, -15.8763427734375, 129.26251220703125, 76.89790344238281, 208.12908935546875, 24.82897186279297, 227.58843994140625, 115.60403442382812, 149.48175048828125, 225.35006713867188, 127.96952056884766, -60.54049301147461, 230.33856201171875, 147.72149658203125, 166.81399536132812, 108.776123046875, -38.160400390625, 39.640960693359375, 51.633689880371094, 40.8934326171875, 145.1896209716797, 29.00836181640625, 12.78961181640625, 98.49966430664062, -102.21609497070312, 157.5595703125, 213.5655975341797, 172.1959686279297, 331.75543212890625, 248.89474487304688, 6.145050048828125, -25.15045166015625, 6.141510009765625, -11.336830139160156, -7.8681640625, 186.97918701171875, 97.1612319946289, 227.3793487548828], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000371.npy"}
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 87.25260925292969, "std": 131.85650634765625, "min": -263.65631103515625, "p10": -68.4912612915039, "median": 80.0528335571289, "p90": 228.9023254394531, "max": 414.0303955078125, "pos_frac": 0.75, "sample": [90.717529296875, -0.7120361328125, 136.48724365234375, -68.62980651855469, -7.2865753173828125, 105.9410400390625, 209.66905212402344, -3.4171142578125, 74.85702514648438, 391.5645446777344, 78.3320541381836, 197.7733154296875, 116.70745849609375, 140.6806182861328, -100.6027603149414, 111.5538330078125, -1.084808349609375, 365.24310302734375, 23.688173294067383, 8.278923034667969, 414.0303955078125, 70.88353729248047, 158.05894470214844, -4.067230224609375, -177.64584350585938, 238.56527709960938, 226.430419921875, 43.84918212890625, 199.31944274902344, 157.50823974609375, -43.512451171875, -17.18914794921875, 205.39111328125, -173.22625732421875, 100.73004150390625, 12.973175048828125, 180.16754150390625, 7.8210296630859375, 130.9619903564453, 141.29656982421875, -25.76105499267578, 413.65887451171875, -52.52879333496094, 151.29434204101562, 282.642578125, -263.65631103515625, 171.7003173828125, 279.1488342285156, 60.799354553222656, 280.4266052246094, 2.943603515625, 106.76092529296875, -11.678009033203125, -31.64764404296875, 47.585113525390625, 28.692657470703125, 61.204986572265625, 204.15975952148438, 36.99363708496094, 196.22518920898438, 204.9327392578125, 57.75433349609375, 16.10626220703125, 80.66812133789062, 8.467113494873047, 76.38783264160156, 210.63284301757812, 23.4622802734375, 345.92041015625, -2.8418045043945312, 19.2078857421875, 168.36810302734375, 91.080322265625, 226.79718017578125, 119.29299926757812, 241.506591796875, -170.88848876953125, 206.99302673339844, 208.17214965820312, -190.27212524414062, 194.21969604492188, 11.972663879394531, 214.8132781982422, 80.3736572265625, -122.32228088378906, -166.5028533935547, 83.19525146484375, 80.292236328125, -5.431413650512695, 47.05975341796875, -49.31123733520508, 19.40576171875, 79.81343078613281, 124.293701171875, 57.0057373046875, -173.06494140625, -115.60957336425781, 149.244873046875, 140.1957550048828, -80.31317901611328, 125.37060546875, 16.87591552734375, -52.481201171875, 12.443740844726562, 160.7575225830078, 172.05947875976562, -83.06069946289062, 210.86944580078125, 384.3160400390625, 189.13299560546875, 108.43026733398438, 391.62744140625, 149.58042907714844, 12.725555419921875, 108.91738891601562, 79.53350830078125, -1.4408016204833984, 63.642333984375, -68.431884765625, 228.32025146484375, 28.0750732421875, 230.260498046875, -25.452449798583984, 177.5947265625, -7.209846496582031, 51.330322265625, 120.943603515625, 191.45440673828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000372.npy"}
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 82.415283203125, "std": 114.65169525146484, "min": -301.17620849609375, "p10": -62.065933227539055, "median": 70.54798889160156, "p90": 226.70120086669922, "max": 305.0144348144531, "pos_frac": 0.7734375, "sample": [13.899200439453125, 36.29290771484375, 229.81979370117188, 187.47134399414062, 6.410358428955078, 17.771636962890625, 41.192169189453125, 181.52249145507812, 55.37420654296875, 168.91665649414062, 97.55487060546875, -71.35992431640625, 69.63969421386719, 237.57688903808594, 50.849853515625, 227.36228942871094, 15.015506744384766, 58.1221923828125, 18.01068115234375, 112.44482421875, 137.00335693359375, 188.2566375732422, 180.0980987548828, 133.14410400390625, -52.558349609375, -0.2717742919921875, -2.735565185546875, -95.92561340332031, 45.50506591796875, 214.19070434570312, 12.542610168457031, -144.60894775390625, 145.245849609375, 34.75323486328125, 139.72357177734375, 148.595947265625, 239.53561401367188, -25.857940673828125, 125.4669189453125, 118.3768310546875, 213.86163330078125, 74.06512451171875, 80.5855712890625, 100.84332275390625, -7.31829833984375, 149.58465576171875, 20.35650634765625, -127.20391845703125, 167.83926391601562, 213.15924072265625, 17.271926879882812, 156.89169311523438, -67.67706298828125, -76.3408203125, 38.33062744140625, -68.7589111328125, 212.73492431640625, -59.661163330078125, 87.26800537109375, 47.861846923828125, 10.865850448608398, -301.17620849609375, 251.7318115234375, 7.5815277099609375, 212.06887817382812, 18.325469970703125, -52.24713897705078, 226.41787719726562, 30.95477294921875, 119.20170593261719, 56.27789306640625, 194.4605255126953, 36.97173309326172, 84.92494201660156, 142.05325317382812, 48.090606689453125, 59.96234130859375, -10.886962890625, -3.0069580078125, 26.50489044189453, -213.64756774902344, -83.48295593261719, 275.42193603515625, -96.73235321044922, 81.55848693847656, 256.16815185546875, 186.5911407470703, 189.65386962890625, 217.99420166015625, -6.382171630859375, -75.92462158203125, 200.30596923828125, 16.763427734375, 31.564437866210938, 304.87603759765625, 223.94427490234375, 71.45628356933594, 211.95965576171875, -14.013248443603516, 5.546783447265625, 154.80206298828125, 135.74876403808594, 231.20864868164062, 305.0144348144531, 37.92347717285156, 6.763702392578125, 157.2134246826172, -11.694015502929688, 265.4510803222656, 132.533203125, 6.970722198486328, -22.046791076660156, 222.97056579589844, 219.6080322265625, 20.684274673461914, -129.9840087890625, 106.78929138183594, 32.662139892578125, -20.19801902770996, 184.25877380371094, -8.14832878112793, 184.87139892578125, -40.6004638671875, 179.43399047851562, 282.0035400390625, 266.25848388671875, 71.76231384277344, 166.13438415527344], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000373.npy"}
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 71.56175994873047, "std": 119.10095977783203, "min": -248.24124145507812, "p10": -58.01807861328125, "median": 41.226741790771484, "p90": 229.80895690917967, "max": 419.3615417480469, "pos_frac": 0.6875, "sample": [258.247802734375, 0.0, 14.477142333984375, 362.0562438964844, -20.61688232421875, -61.23284912109375, -76.88394165039062, 37.70068359375, 322.3448486328125, 194.06918334960938, 25.209461212158203, 278.098876953125, 169.88076782226562, -43.477752685546875, 260.83966064453125, 34.520965576171875, 119.32083129882812, 164.35523986816406, 185.22439575195312, -38.36781311035156, 95.17208099365234, -34.325286865234375, 288.5480651855469, -74.5076904296875, 219.49179077148438, -6.228851318359375, 57.265594482421875, 2.2061920166015625, 5.232851028442383, -72.49919891357422, 18.610862731933594, 5.1583709716796875, 23.203033447265625, 32.4476318359375, -37.4984130859375, 40.00963592529297, 102.43710327148438, -41.245208740234375, 211.57595825195312, 210.5555419921875, 106.77377319335938, -22.105255126953125, 419.3615417480469, -13.036109924316406, 118.89739990234375, 3.89898681640625, -4.643009185791016, 92.23684692382812, -248.24124145507812, -30.802810668945312, 84.45500946044922, -107.5072021484375, 91.20687866210938, -33.428558349609375, 149.600830078125, 32.588134765625, 228.78916931152344, 90.35333251953125, 249.3994140625, -8.023536682128906, 69.60762023925781, 207.6180419921875, 223.6500244140625, -6.4525909423828125, 15.234298706054688, 62.300567626953125, 205.80670166015625, 24.81744384765625, -15.08135986328125, 148.21392822265625, 5.380157470703125, 46.88458251953125, 90.42628479003906, 113.28904724121094, -133.98159790039062, -17.632156372070312, -69.338134765625, 24.192626953125, 249.86114501953125, 176.29501342773438, 193.31973266601562, -86.634765625, 175.92562866210938, 163.29457092285156, 0.0, 172.44264221191406, 22.063053131103516, 42.44384765625, -88.23855590820312, -56.64031982421875, 97.57982635498047, 109.86408996582031, 232.18846130371094, 196.63613891601562, -3.24664306640625, 22.7921142578125, 303.0186767578125, 50.1881103515625, 285.1401672363281, 166.1104736328125, 5.6470947265625, -7.710090637207031, 21.013076782226562, 22.439926147460938, 9.86892318725586, -29.68470001220703, 185.74420166015625, 193.10546875, -182.39930725097656, -23.205589294433594, -26.4339599609375, -13.60113525390625, 54.817108154296875, 97.7347412109375, 73.34982299804688, 1.142974853515625, 108.91389465332031, -6.2998046875, -83.3148193359375, 121.44728088378906, 359.55352783203125, 198.36874389648438, 62.435882568359375, 102.37810516357422, 121.63482666015625, -10.30636215209961, 88.92315673828125, -136.14727783203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000374.npy"}
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 62.69805908203125, "std": 117.9583969116211, "min": -198.41702270507812, "p10": -84.03241271972657, "median": 42.39215087890625, "p90": 226.4009765625, "max": 336.651611328125, "pos_frac": 0.6953125, "sample": [100.12644958496094, 95.07215881347656, 34.40736389160156, -93.73927307128906, 76.502197265625, 50.862342834472656, 71.20339965820312, -198.41702270507812, 18.590091705322266, 37.51226806640625, 180.99755859375, 95.76783752441406, 235.9889373779297, 157.68603515625, 63.717559814453125, 0.02557373046875, 40.899658203125, 211.48097229003906, 12.89666748046875, -11.8790283203125, 193.1937713623047, 236.52230834960938, -23.443023681640625, 10.923629760742188, 51.555389404296875, 103.27465057373047, -10.878387451171875, -28.85639190673828, -73.87973022460938, 162.75706481933594, 40.77573776245117, 25.40130615234375, -31.69873046875, -98.30632019042969, -64.58688354492188, 259.6212158203125, 127.84255981445312, 123.65451049804688, 14.338737487792969, -84.0836181640625, 10.048828125, 33.4510498046875, -39.805206298828125, 20.17047119140625, -99.1934814453125, 336.651611328125, 185.44549560546875, 166.23095703125, 72.151123046875, -83.9207763671875, -165.36851501464844, -29.507232666015625, 244.15618896484375, 47.928565979003906, 123.64812469482422, -18.19561767578125, 117.7548828125, 5.525177001953125, -57.29119873046875, -113.39190673828125, 149.66131591796875, 277.46051025390625, -17.62139892578125, 326.78570556640625, -0.41309356689453125, 43.8846435546875, -44.32763671875, 15.307136535644531, 320.826171875, -17.212696075439453, -19.194000244140625, 20.146297454833984, 135.71966552734375, 120.0130386352539, 18.369293212890625, 227.96734619140625, 216.58587646484375, -103.80062866210938, 64.76580810546875, 54.91893005371094, 98.8260498046875, 50.911651611328125, 4.37835693359375, -184.534912109375, -27.587326049804688, 51.38499450683594, -47.1854248046875, 182.10235595703125, 143.6383056640625, 141.909423828125, -34.389320373535156, 212.98919677734375, 157.14276123046875, -9.45330810546875, 210.28533935546875, 280.44842529296875, -122.33676147460938, 207.0716552734375, 204.62403869628906, 291.6305847167969, -164.3789825439453, -174.6455078125, 151.6229248046875, 93.80962371826172, 167.08984375, 10.004196166992188, 212.09521484375, 246.425048828125, -9.302703857421875, 36.7994384765625, 6.08050537109375, 225.72967529296875, 157.1398162841797, 48.90025329589844, 176.0001220703125, -2.285491943359375, 16.267820358276367, 16.140098571777344, -0.7797737121582031, -124.9908447265625, 94.27896118164062, 62.908058166503906, 283.0301513671875, -5.696533203125, 13.607452392578125, 86.52944946289062, 14.966827392578125, -84.01046752929688], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000375.npy"}
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 71.82979583740234, "std": 130.48184204101562, "min": -226.91015625, "p10": -76.03076782226563, "median": 50.62485885620117, "p90": 226.13857421875, "max": 445.7083740234375, "pos_frac": 0.6953125, "sample": [-149.2076416015625, 223.38873291015625, 225.90191650390625, 226.603759765625, 163.6925048828125, 445.7083740234375, 14.723876953125, -59.21250915527344, 59.695556640625, -14.4427490234375, 154.26715087890625, -46.2891845703125, -138.52304077148438, 47.302703857421875, -36.908843994140625, 55.27247619628906, -45.335792541503906, 28.301025390625, 95.67303466796875, 276.84075927734375, 216.7134552001953, -159.65280151367188, 126.6783676147461, 38.3984375, 231.70193481445312, -76.0196533203125, -192.45999145507812, 219.20635986328125, -5.69940185546875, 235.88916015625, 81.09356689453125, 109.12559509277344, 175.00062561035156, 263.70782470703125, -51.21595764160156, -16.631256103515625, 131.23260498046875, -4.500907897949219, 368.8922119140625, -226.91015625, 0.0, 34.20836639404297, 53.405296325683594, 5.398109436035156, 12.129547119140625, 331.3671569824219, 4.558746337890625, 214.37939453125, 62.88262176513672, 19.24945068359375, -206.58270263671875, 189.2596435546875, 99.40145874023438, 20.7164306640625, 289.5302429199219, 98.65438842773438, 109.88787841796875, 150.1507568359375, -71.88018798828125, -12.8917236328125, 170.18218994140625, 19.5828857421875, 371.66455078125, 203.2030029296875, 30.7652587890625, 210.790283203125, 209.22567749023438, 26.15802001953125, 28.5732421875, -1.77935791015625, 88.3966064453125, 2.207122802734375, 7.816314697265625, -15.303268432617188, 12.652360916137695, -15.543701171875, -89.65887451171875, 19.216594696044922, -141.86065673828125, 177.1739501953125, 19.496749877929688, 182.10845947265625, 193.67889404296875, 47.84442138671875, 105.24166870117188, 116.00721740722656, 23.792490005493164, -223.27293395996094, 10.27020263671875, -55.466949462890625, 104.03118896484375, -1.4625396728515625, 87.70256042480469, -4.398223876953125, 17.336406707763672, 181.8278045654297, -126.22674560546875, -14.359130859375, -6.4886474609375, 148.65167236328125, -3.06719970703125, 114.8814697265625, -121.16046142578125, 134.71502685546875, 169.68524169921875, 86.71243286132812, 166.5263671875, 175.6136016845703, 291.373779296875, 307.95782470703125, -36.5921630859375, -76.05670166015625, -129.55581665039062, 64.84991455078125, 225.939208984375, -53.334251403808594, 66.37774658203125, -67.536865234375, 109.80197143554688, 25.33354949951172, -3.036396026611328, 148.30416870117188, 224.6667022705078, 167.0439453125, 197.86830139160156, 35.60328674316406, 298.833740234375, 156.8609619140625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000376.npy"}
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 51.40126037597656, "std": 124.66860961914062, "min": -367.82342529296875, "p10": -84.76110229492186, "median": 35.10905075073242, "p90": 213.74636993408203, "max": 365.43658447265625, "pos_frac": 0.6484375, "sample": [267.10198974609375, 40.64344787597656, 102.03485107421875, 13.373336791992188, 42.77996826171875, -2.6662445068359375, 13.60736083984375, 67.28536987304688, 171.1566162109375, -118.7923583984375, 69.59756469726562, -35.86773681640625, 172.80343627929688, -77.43939208984375, 136.03932189941406, 11.542724609375, 116.61700439453125, -64.5374755859375, -26.493637084960938, -12.66961669921875, 45.086700439453125, 0.7181549072265625, -3.7833633422851562, 10.450807571411133, -22.51398277282715, -14.916370391845703, 149.59756469726562, 30.888221740722656, 119.30133056640625, 5.609466552734375, -1.8271636962890625, 70.75286865234375, 45.341552734375, -268.834228515625, 118.78874206542969, 157.1600341796875, -2.2846202850341797, -1.48211669921875, -59.74066162109375, 49.5013427734375, 137.34637451171875, 171.59698486328125, 43.99566650390625, -367.82342529296875, 44.78858947753906, 3.53564453125, 341.8148193359375, 63.53460693359375, -105.09599304199219, -125.61724853515625, 12.3392333984375, 247.53485107421875, -1.7756729125976562, 22.727767944335938, 162.915283203125, -128.5496826171875, -197.5274658203125, -9.926795959472656, 44.93407440185547, 302.2947692871094, 198.49237060546875, 66.143798828125, 26.027252197265625, 31.1749267578125, -198.496826171875, 211.92576599121094, 126.25372314453125, 0.0, -6.3580322265625, 96.94171905517578, 96.21450805664062, -110.06501007080078, 32.85662078857422, 229.0144805908203, -11.39239501953125, 105.947998046875, 24.963729858398438, -7.751220703125, -12.609321594238281, 157.67918395996094, 217.99444580078125, 220.56890869140625, -60.568603515625, -53.23463439941406, -50.994110107421875, -101.8450927734375, 29.891929626464844, 142.47787475585938, 251.89260864257812, 161.34716796875, 187.70233154296875, 1.250640869140625, 67.68643951416016, 365.43658447265625, -40.6446533203125, 199.83352661132812, -2.6855316162109375, 8.573577880859375, 45.568206787109375, 189.41546630859375, 110.91021728515625, -190.18832397460938, -71.86778259277344, 37.361480712890625, 219.09658813476562, -19.37029266357422, 184.44210815429688, 84.07737731933594, 86.363037109375, 166.32943725585938, 61.92734146118164, 5.799842834472656, 201.76925659179688, 266.228759765625, 209.3992919921875, 167.91571044921875, 100.37161254882812, -169.10305786132812, -21.242645263671875, 219.4554443359375, -3.1644515991210938, 9.0029296875, -235.8747100830078, 269.72589111328125, -36.28404235839844, -28.957778930664062, -63.1744384765625, 204.80859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000377.npy"}
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 64.7405014038086, "std": 118.09962463378906, "min": -310.753662109375, "p10": -65.88923950195313, "median": 46.38695526123047, "p90": 214.01787109374996, "max": 453.366455078125, "pos_frac": 0.703125, "sample": [87.37847900390625, -13.54571533203125, -86.63481140136719, 176.19259643554688, -4.85906982421875, 11.597219467163086, -41.66748046875, 24.83575439453125, 127.88925170898438, 239.2415313720703, -9.251289367675781, 113.59768676757812, 58.3631591796875, 157.1257781982422, 163.316162109375, -119.77500915527344, -310.753662109375, 179.87066650390625, 363.95526123046875, 5.701934814453125, 120.12164306640625, 184.343505859375, 6.718658447265625, -85.30314636230469, 116.74801635742188, 191.6754913330078, 305.0537414550781, 134.4906005859375, 62.10186767578125, 221.00979614257812, 23.335403442382812, 111.81394958496094, 141.64486694335938, 39.991790771484375, 106.17367553710938, 114.38041687011719, -30.586132049560547, 39.743377685546875, -100.36064910888672, 158.98077392578125, 0.0, 239.38223266601562, -39.768157958984375, 259.14996337890625, 394.8392333984375, -67.71490478515625, -15.856056213378906, 256.11004638671875, -5.1376495361328125, 155.04559326171875, 91.91741943359375, 151.38198852539062, -3.9652481079101562, 60.791259765625, -2.1259765625, 42.524559020996094, -218.12783813476562, 21.896896362304688, -100.30593872070312, 453.366455078125, 198.783935546875, -1.2557220458984375, 87.67452239990234, 66.79986572265625, 208.31472778320312, 157.87765502929688, 52.48866271972656, 46.770111083984375, 261.40557861328125, 78.31327819824219, 25.4425048828125, 56.296661376953125, 5.16595458984375, 25.907882690429688, -20.71316146850586, 155.92938232421875, 23.025833129882812, -38.301544189453125, 51.76031494140625, -13.933502197265625, 58.09739685058594, -65.1068115234375, -99.10888671875, 14.805307388305664, -37.556427001953125, 91.93087768554688, 176.39205932617188, 64.11392211914062, 39.824798583984375, -70.18963623046875, 122.91558837890625, 211.02133178710938, -26.093948364257812, 44.305633544921875, 32.957122802734375, 35.188140869140625, -7.98870849609375, 94.89317321777344, 27.32763671875, -42.623291015625, -125.655029296875, 238.87628173828125, 43.056365966796875, -34.58778381347656, -53.44145202636719, -9.415740966796875, 3.94512939453125, 137.489013671875, 21.307769775390625, 101.00277709960938, 43.060791015625, 127.22819519042969, -35.559669494628906, 48.47265625, 240.1905517578125, 175.93154907226562, 269.3427734375, 3.776214599609375, 126.87399291992188, 209.64913940429688, -188.08331298828125, 46.00379943847656, 80.19338989257812, 13.82904052734375, 148.52182006835938, -94.661376953125, 55.99029541015625, -57.471832275390625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000378.npy"}
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 68.758056640625, "std": 116.11892700195312, "min": -208.30389404296875, "p10": -75.01899795532226, "median": 63.55838394165039, "p90": 219.57858581542968, "max": 310.8726806640625, "pos_frac": 0.6875, "sample": [-14.35378646850586, -13.914047241210938, 41.64868927001953, -0.1205596923828125, -2.3429031372070312, -2.281829833984375, -8.103073120117188, 167.5595703125, 105.4781494140625, 28.262405395507812, 96.94140625, 119.90985107421875, 62.38697052001953, 68.0911865234375, -8.711456298828125, -0.765625, 3.399059295654297, -52.44146728515625, 154.02847290039062, 129.6173858642578, 106.82595825195312, 181.07406616210938, 149.023193359375, -47.816650390625, 29.57311248779297, -8.300567626953125, 269.8971862792969, 45.38763427734375, -45.06683349609375, 8.952705383300781, 45.8380126953125, 156.76971435546875, 240.89105224609375, 97.269287109375, 97.55250549316406, -32.2972412109375, 175.07345581054688, 292.85235595703125, 246.9530029296875, 116.82893371582031, 150.60348510742188, -73.209716796875, 25.838836669921875, -54.448123931884766, 64.72979736328125, 222.67071533203125, -17.125953674316406, -164.94505310058594, -93.84912109375, 55.58978271484375, -18.8143310546875, -56.772178649902344, 34.5810546875, -155.33160400390625, -47.6383056640625, 218.25338745117188, 196.46826171875, 131.456787109375, 20.225242614746094, -19.94677734375, 108.25664520263672, -110.68025207519531, 31.174819946289062, 179.9943389892578, 130.69760131835938, 129.80865478515625, 76.80174255371094, 242.33425903320312, 12.418731689453125, -42.81756591796875, -34.82101821899414, 0.0, -86.57069396972656, 82.46286010742188, 93.575439453125, 202.18743896484375, 113.99392700195312, 135.93429565429688, 167.55178833007812, 280.2797546386719, 9.7701416015625, 155.00662231445312, 121.61233520507812, -56.66780471801758, 150.63284301757812, 95.93081665039062, 25.097610473632812, 154.6126708984375, -98.011962890625, 177.3758544921875, 110.43704223632812, 12.87261962890625, 274.39300537109375, 162.4369659423828, 96.628662109375, 9.933349609375, 20.72124481201172, 68.62722778320312, -79.24065399169922, 303.2164306640625, 190.34695434570312, -208.30389404296875, -42.99310302734375, 44.14419174194336, 192.56317138671875, 197.7701416015625, 296.2315673828125, -95.7398681640625, 92.05909729003906, 255.77078247070312, 187.09423828125, 165.27349853515625, 310.8726806640625, 129.40756225585938, 22.04253387451172, -172.14532470703125, 44.88761901855469, 15.186800003051758, 17.083648681640625, 167.36724853515625, -184.09837341308594, 166.1808624267578, -160.22616577148438, -20.6171875, -105.25643920898438, 172.19715881347656, -19.407470703125, 295.4669494628906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000379.npy"}
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 76.69670867919922, "std": 124.56847381591797, "min": -343.7085266113281, "p10": -67.21441192626952, "median": 59.95146179199219, "p90": 238.3290588378906, "max": 484.8651123046875, "pos_frac": 0.7421875, "sample": [147.35089111328125, 160.97296142578125, 183.41702270507812, 102.36967468261719, 7.67364501953125, 31.966766357421875, -343.7085266113281, 29.157012939453125, 1.8619918823242188, -83.1190185546875, 165.95474243164062, 124.44329833984375, 59.7183837890625, -164.89549255371094, 106.58203125, 16.123817443847656, 266.9985046386719, 6.35601806640625, 246.31698608398438, -0.33039093017578125, 0.0, 117.44752502441406, 13.201438903808594, 46.20947265625, 60.184539794921875, 143.8555908203125, -22.567398071289062, 226.36810302734375, 3.121551513671875, 43.131591796875, 70.91302490234375, 215.4234619140625, 55.348052978515625, 139.66909790039062, 114.51034545898438, -132.38197326660156, 121.02215576171875, 29.520217895507812, 74.6767578125, -27.5291748046875, 95.22274780273438, 142.5255126953125, 302.7015075683594, -37.81097412109375, -139.40921020507812, -78.4649658203125, 25.0888671875, 94.802734375, 397.94891357421875, -0.72161865234375, -100.666015625, 52.43052673339844, 127.35210418701172, -45.352203369140625, 53.24371337890625, 8.051116943359375, -1.0838699340820312, 35.84625244140625, 333.285888671875, 128.64759826660156, 81.98736572265625, 274.8961486816406, 189.8007049560547, 136.38681030273438, 127.3028564453125, 235.5458984375, -161.86785888671875, 9.939697265625, 12.62255859375, -94.74698638916016, 242.66351318359375, 342.882568359375, 9.941598892211914, 187.95330810546875, 154.50360107421875, 105.38277435302734, 163.10708618164062, -127.81582641601562, 119.93106079101562, -34.586456298828125, 75.04368591308594, 484.8651123046875, -2.4977874755859375, 216.5742950439453, 111.3603515625, -11.172874450683594, 88.45770263671875, 94.789306640625, 188.61209106445312, 166.94931030273438, 126.83709716796875, 236.471435546875, -13.66461181640625, 69.09518432617188, 42.68408966064453, -0.26244354248046875, -19.613433837890625, -140.822998046875, 397.5498962402344, 307.35284423828125, 13.781726837158203, 48.81657409667969, 27.215896606445312, 55.314544677734375, 33.09211730957031, -83.98597717285156, 164.87554931640625, 119.76614379882812, 43.81312561035156, -3.104969024658203, -1.57342529296875, -19.620071411132812, 101.69575500488281, 72.32394409179688, 59.235069274902344, 52.82501220703125, -8.7069091796875, 127.28842163085938, 113.2529296875, 259.04974365234375, 163.5819091796875, 8.7227783203125, 96.14511108398438, 106.3609619140625, -97.77534484863281, -9.389701843261719, 291.1585693359375, -62.39274597167969], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000380.npy"}
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 78.50491333007812, "std": 124.5214614868164, "min": -211.71011352539062, "p10": -42.278314208984376, "median": 56.70653533935547, "p90": 236.61844177246093, "max": 482.34259033203125, "pos_frac": 0.734375, "sample": [107.25680541992188, 333.18084716796875, 118.097900390625, -30.456932067871094, 146.3646240234375, 9.019775390625, 169.696533203125, 373.5928649902344, 171.57838439941406, 50.3256721496582, 155.49591064453125, 62.64276123046875, 156.1027374267578, 134.94166564941406, 253.15725708007812, 228.37124633789062, 0.0, 240.65435791015625, 6.730438232421875, 153.91769409179688, 482.34259033203125, 112.42724609375, -105.88005065917969, 184.76315307617188, 161.77685546875, 58.818878173828125, 32.245140075683594, -188.59124755859375, 0.0, 1.83795166015625, 97.18389892578125, 195.3298797607422, -211.71011352539062, 191.248779296875, 6.969139099121094, 27.78076171875, 15.834503173828125, 101.40084838867188, 5.36602783203125, 15.63128662109375, 250.8897705078125, 83.42459106445312, -188.32757568359375, 54.59419250488281, 327.5766906738281, 19.377716064453125, 92.01565551757812, 12.160858154296875, -39.63726043701172, 153.78775024414062, 170.61138916015625, 112.09742736816406, 113.96748352050781, 8.226058959960938, 112.946044921875, 229.40887451171875, 37.48382568359375, 234.88876342773438, 381.9918212890625, 143.8414306640625, 31.726028442382812, -31.877166748046875, 212.92715454101562, 13.414764404296875, -20.025848388671875, 107.28434753417969, 103.05292510986328, -126.88278198242188, -30.84283447265625, -53.083648681640625, 63.26934051513672, -15.484386444091797, -66.45924377441406, 205.28265380859375, -12.44873046875, -16.22119140625, -36.13959884643555, 47.95555114746094, 14.567840576171875, 1.8677253723144531, 150.16015625, -21.43614959716797, 168.41741943359375, 183.884033203125, 241.33126831054688, -10.2288818359375, -134.8785400390625, -23.721948623657227, 220.57081604003906, 21.057769775390625, 94.66162109375, -95.7655029296875, 152.87063598632812, 132.61514282226562, 81.84211730957031, -74.0054931640625, -24.938941955566406, 41.14678955078125, -17.377548217773438, -133.44097900390625, 109.70256042480469, -4.6026153564453125, 6.8221435546875, -27.494415283203125, -111.40267944335938, 284.4898681640625, -31.8043212890625, 33.388275146484375, -30.376754760742188, 46.132667541503906, 54.230987548828125, 332.2208251953125, 78.20947265625, 241.1390380859375, 75.46475219726562, 122.02589416503906, 45.72014617919922, 17.461883544921875, 234.62466430664062, 6.0340728759765625, 162.94573974609375, 387.6705322265625, 140.62673950195312, -43.377960205078125, -41.807037353515625, 186.70941162109375, 64.24163818359375, 28.21563720703125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000381.npy"}
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 90.05111694335938, "std": 117.40560913085938, "min": -225.15469360351562, "p10": -53.98852615356445, "median": 86.6272087097168, "p90": 244.0900405883789, "max": 360.94000244140625, "pos_frac": 0.75, "sample": [-97.92498779296875, 76.28497314453125, 141.77557373046875, 27.229888916015625, 202.012451171875, 174.546142578125, 312.1335754394531, -42.234771728515625, 132.69497680664062, 173.650390625, 77.48419189453125, -225.15469360351562, -18.03082275390625, 38.76216506958008, 160.78546142578125, 22.595458984375, 146.17127990722656, 35.41265869140625, 216.24072265625, -58.766483306884766, 42.8865966796875, -17.064254760742188, 243.72120666503906, -40.4720458984375, 43.38800811767578, 127.90898132324219, 46.238067626953125, -65.07614135742188, 99.51651000976562, 136.8009490966797, 90.76605224609375, 196.9656982421875, -27.049163818359375, 12.3763427734375, 141.16851806640625, 75.40997314453125, 129.22848510742188, 14.436325073242188, -34.010406494140625, 168.53375244140625, 140.38731384277344, 107.31076049804688, 189.3232421875, 71.17901611328125, 279.1688232421875, 169.58863830566406, 107.77049255371094, 112.11190795898438, 204.55502319335938, 39.285186767578125, 359.38751220703125, 69.66349792480469, 59.14891815185547, 262.32598876953125, -122.66549682617188, 116.76936340332031, 227.17779541015625, 175.9879150390625, 24.05694580078125, -13.553253173828125, -53.0035400390625, -23.95538330078125, 283.8033142089844, 79.68844604492188, 44.78736877441406, 24.698932647705078, 226.79107666015625, 167.31915283203125, 209.595703125, 22.637786865234375, 265.99835205078125, 120.93661499023438, -19.757431030273438, 33.8148193359375, 239.71380615234375, 158.15960693359375, -57.287841796875, 54.999969482421875, 82.48836517333984, 189.37286376953125, 81.96038818359375, -176.91494750976562, 105.56277465820312, 133.70291137695312, 23.550582885742188, 147.78402709960938, -82.52511596679688, 306.5726318359375, -14.851028442382812, 259.4361572265625, 254.62030029296875, 22.36248779296875, 59.36848449707031, 244.95065307617188, 261.0635986328125, -21.41021728515625, 138.13845825195312, -165.62689208984375, -52.91864013671875, -13.8470458984375, 164.53140258789062, 139.70684814453125, 173.85768127441406, 176.42808532714844, 72.43572998046875, -136.68563842773438, -21.04479217529297, 121.3636474609375, -53.78913116455078, 46.869659423828125, 44.672210693359375, -0.334869384765625, 188.06703186035156, 142.52334594726562, -18.350006103515625, -72.43746948242188, -87.35589599609375, 186.61770629882812, -51.80079650878906, 187.64234924316406, 171.30010986328125, -54.45378112792969, 360.94000244140625, 115.348388671875, 203.65272521972656, 174.89166259765625, 324.6099853515625, 5.23358154296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000382.npy"}
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 81.57150268554688, "std": 123.14766693115234, "min": -236.30032348632812, "p10": -86.31910400390625, "median": 69.01741790771484, "p90": 238.8502700805664, "max": 366.42645263671875, "pos_frac": 0.7734375, "sample": [9.571380615234375, 171.67138671875, 253.04640197753906, 175.43936157226562, -41.460052490234375, 30.20404052734375, -9.268218994140625, 1.96826171875, 1.853240966796875, 0.192291259765625, 180.3223876953125, -86.97434997558594, 178.63751220703125, 309.3246765136719, 0.0, -155.947509765625, 110.25526428222656, -100.14239501953125, 91.04611206054688, -129.85891723632812, 39.46551513671875, 340.4319763183594, 207.59002685546875, 169.89207458496094, 15.5850830078125, 235.20648193359375, -7.36773681640625, -32.227149963378906, 101.38065338134766, 7.280857086181641, 221.466064453125, 290.115478515625, 22.93553924560547, -102.32691192626953, 265.2828674316406, 71.953125, 202.88919067382812, 339.9959716796875, 81.43963623046875, 150.34280395507812, 29.40947723388672, 11.252738952636719, 197.53976440429688, 228.46664428710938, -107.91459655761719, -10.488037109375, 171.0821533203125, 199.35400390625, 87.26828002929688, 3.22576904296875, 95.06857299804688, 19.3912353515625, -27.160375595092773, 209.22393798828125, 56.28167724609375, -209.9269561767578, 7.228569030761719, 66.08171081542969, -18.266530990600586, -82.72567749023438, 59.1768798828125, 128.642578125, -9.865568161010742, 302.97711181640625, -236.30032348632812, 18.197540283203125, 151.84527587890625, 33.05372619628906, -86.03828430175781, 98.94442749023438, 15.580245971679688, 154.9122314453125, 306.088623046875, 153.29800415039062, 195.80987548828125, 118.02800750732422, 27.17486572265625, 49.183135986328125, 101.53931427001953, 215.23797607421875, 26.515899658203125, 131.57652282714844, 61.748992919921875, 191.72833251953125, 120.39189147949219, 238.1119842529297, 75.01986694335938, 6.516506195068359, 25.03466033935547, -94.44439697265625, 124.10272216796875, 2.861377716064453, 366.42645263671875, 45.32740783691406, 109.69442749023438, -1.1687393188476562, -137.4783172607422, 225.265869140625, 110.2390365600586, 218.76177978515625, 175.16259765625, 89.64315795898438, 240.57293701171875, 52.496185302734375, -7.39227294921875, 36.54505920410156, 105.71209716796875, -59.9666748046875, 0.0, 21.120969772338867, 274.8860778808594, -127.169189453125, 96.5540771484375, 90.6220703125, 107.36471557617188, -101.17015075683594, 204.43084716796875, 36.841468811035156, 306.67266845703125, 30.569610595703125, 78.71945190429688, 55.20357894897461, 309.82989501953125, -139.21124267578125, 157.9097900390625, -3.266143798828125, 180.17001342773438, 48.978302001953125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000383.npy"}
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 71.16676330566406, "std": 115.61644744873047, "min": -184.04742431640625, "p10": -70.20986022949218, "median": 56.93230438232422, "p90": 219.04450073242188, "max": 381.3563232421875, "pos_frac": 0.7265625, "sample": [258.85052490234375, -177.13555908203125, 113.92593383789062, -24.47247314453125, -34.69256591796875, 214.24044799804688, 105.2532958984375, 266.92132568359375, 53.65122985839844, 30.092384338378906, 19.03173065185547, -87.93553161621094, 127.80262756347656, -114.31710815429688, 110.5714111328125, -154.578857421875, 206.79847717285156, 32.67345428466797, 246.71490478515625, -19.337554931640625, 68.07862854003906, 219.9796142578125, 164.61233520507812, 67.39559936523438, 20.226348876953125, 260.6109619140625, 76.52835083007812, 118.43301391601562, 40.43853759765625, -84.390380859375, 114.29849243164062, -14.63677978515625, 60.21337890625, 35.435150146484375, -42.674468994140625, 111.49565887451172, 25.138595581054688, 93.94937133789062, 44.30448913574219, 147.15074157714844, -19.289276123046875, 0.553192138671875, 25.40550994873047, -15.883705139160156, 220.14659118652344, 218.64373779296875, 233.15695190429688, -163.41842651367188, -115.10739135742188, 49.71300506591797, 156.3211669921875, 217.54443359375, 10.445760726928711, 106.07843017578125, -184.04742431640625, 106.27690124511719, 247.71566772460938, 173.95343017578125, -24.02276611328125, 35.704071044921875, -18.505950927734375, -50.59759521484375, 137.84808349609375, 114.89939880371094, 337.3994140625, 178.00315856933594, 66.11962890625, 50.085693359375, 171.43258666992188, -23.2080078125, 85.23481750488281, 67.59408569335938, 248.9869384765625, 44.585693359375, 76.75241088867188, 163.43899536132812, -45.624908447265625, -5.3048553466796875, -0.2244720458984375, 177.5697021484375, 80.03570556640625, -92.435302734375, 150.13430786132812, 116.44403076171875, 211.51162719726562, 358.3132629394531, 43.43017578125, -114.57135009765625, 76.80989074707031, 76.04376220703125, 205.27224731445312, -66.39010620117188, 159.30465698242188, 381.3563232421875, -84.21359252929688, 11.124664306640625, 31.64697265625, 199.86178588867188, 122.16688537597656, -31.043853759765625, 14.037322998046875, -68.99935913085938, 145.60260009765625, -55.76971435546875, -171.92755126953125, 14.510366439819336, 11.599472045898438, 11.240007400512695, 146.65921020507812, 5.842254638671875, 204.3818359375, -58.44775390625, 146.430419921875, 33.18205261230469, -12.579423904418945, 11.13720703125, 164.37164306640625, 268.5338439941406, 38.66997528076172, 169.46607971191406, 195.62828063964844, -27.582626342773438, -12.127132415771484, 42.09490966796875, 157.383544921875, 170.18939208984375, 47.03125, -73.03436279296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000384.npy"}
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 67.46717834472656, "std": 108.15116119384766, "min": -216.70138549804688, "p10": -53.70692443847656, "median": 52.98045349121094, "p90": 226.7035369873047, "max": 415.5768127441406, "pos_frac": 0.7109375, "sample": [-59.863189697265625, 17.064983367919922, 210.9464569091797, 105.522705078125, 34.4342041015625, -53.22027587890625, 259.1623229980469, 80.16534423828125, 161.771728515625, 162.77301025390625, -8.164459228515625, -42.46209716796875, -38.01509094238281, 66.11683654785156, 18.61676025390625, 79.91903686523438, 147.61529541015625, 177.6240997314453, 54.76091003417969, -1.648529052734375, 128.839599609375, -55.54888916015625, -7.750640869140625, -173.9615020751953, 71.16555786132812, 156.0894317626953, 179.67752075195312, -0.544586181640625, 26.24090576171875, 74.84658813476562, -2.355560302734375, 255.43463134765625, 100.5008316040039, -41.980804443359375, 230.1474609375, 311.467529296875, 1.675262451171875, 14.334442138671875, 119.60196685791016, 258.00811767578125, 40.393890380859375, 15.321769714355469, 225.22756958007812, 115.69633483886719, 197.32550048828125, -119.63316345214844, 218.56443786621094, 239.17111206054688, -216.70138549804688, 90.48858642578125, 415.5768127441406, 127.24968719482422, 231.27264404296875, -17.734596252441406, 70.96940612792969, 34.18736267089844, 0.449310302734375, -4.859811782836914, -177.7081298828125, 59.04966735839844, -7.3546142578125, 274.4184875488281, -38.255126953125, 293.6004333496094, 175.55438232421875, 19.895843505859375, -10.221832275390625, 83.89132690429688, -29.568344116210938, 173.50335693359375, 125.59009552001953, 69.15625762939453, 67.93984985351562, 38.10922622680664, 29.649871826171875, -68.96261596679688, 156.97653198242188, 32.047271728515625, 171.22659301757812, -20.555450439453125, 51.19999694824219, 122.03701782226562, 171.42431640625, -25.650413513183594, 136.71652221679688, 47.69715881347656, -92.99217224121094, 112.00103759765625, -168.3192138671875, 25.91962432861328, 109.68644714355469, -9.927196502685547, 45.602874755859375, 69.47491455078125, 305.05224609375, 21.51531982421875, 0.0, 66.16191101074219, 89.42401123046875, 30.035018920898438, -35.11583709716797, 135.77163696289062, 63.266571044921875, -63.705322265625, 104.5208740234375, 28.413291931152344, 24.41552734375, -2.8553504943847656, -54.842437744140625, 237.516845703125, 135.460205078125, 0.0, -11.169830322265625, -73.07549285888672, 135.18942260742188, 3.2235107421875, 32.22393798828125, 77.99452209472656, 85.00143432617188, 138.09384155273438, 72.49043273925781, -59.616355895996094, 21.275833129882812, 45.676788330078125, -35.45851135253906, 148.46990966796875, 231.391845703125, 44.225799560546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000385.npy"}
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 93.33988189697266, "std": 123.88856506347656, "min": -200.62677001953125, "p10": -28.816723251342772, "median": 69.54554748535156, "p90": 267.8399627685547, "max": 420.6195373535156, "pos_frac": 0.796875, "sample": [35.071075439453125, -28.462520599365234, 1.4220123291015625, 97.49708557128906, 123.50286865234375, 79.00335693359375, 52.51727294921875, 271.748046875, 14.577407836914062, 174.11190795898438, 202.35218811035156, 320.8349609375, -184.22882080078125, 79.96600341796875, 246.21942138671875, 17.911209106445312, 80.265869140625, -28.59814453125, -6.1239776611328125, 311.0546875, 34.12110900878906, 201.35580444335938, 39.153106689453125, 86.19757080078125, 49.96514892578125, -140.5379180908203, 51.37162780761719, 7.58074951171875, 229.0184326171875, 29.8341064453125, 40.8270263671875, 149.1114501953125, 212.95773315429688, 113.603515625, 4.2426605224609375, 46.543853759765625, 65.88583374023438, 71.925048828125, 125.25859069824219, 273.62799072265625, 313.4205322265625, 117.3533935546875, 50.53889465332031, 88.95187377929688, 168.3514862060547, 227.75311279296875, 328.463134765625, -200.62677001953125, 55.47807312011719, 188.11163330078125, 91.02532958984375, 6.9434814453125, -124.5943603515625, -96.94127655029297, 235.67715454101562, 69.60175323486328, 69.48934173583984, 0.0, -53.451263427734375, -114.55036163330078, 21.18212890625, -1.76416015625, 138.73635864257812, 60.891998291015625, 2.9373245239257812, 23.477935791015625, 26.870086669921875, 18.5755615234375, 222.011962890625, -182.6023406982422, 249.181884765625, 118.33065795898438, 169.46795654296875, 249.40460205078125, 420.6195373535156, 178.614990234375, 10.480476379394531, -36.44627380371094, 109.64013671875, 161.35890197753906, 231.3885498046875, 193.97689819335938, -195.241455078125, 94.06097412109375, 18.290130615234375, 213.36807250976562, 188.08755493164062, -8.55682373046875, 15.5277099609375, 39.586761474609375, -15.603729248046875, -9.151214599609375, 21.4844970703125, 0.0, 299.1585693359375, 46.815704345703125, 95.061767578125, 20.612998962402344, 161.29498291015625, 82.55838012695312, 234.59056091308594, 211.60696411132812, 128.03790283203125, 343.26947021484375, 74.5007095336914, 301.4390563964844, -38.10443115234375, -16.586181640625, -16.3101806640625, 167.2286376953125, 156.73184204101562, 109.91110229492188, 320.6136474609375, -17.3636474609375, -44.52458190917969, 193.60494995117188, 35.39723205566406, 34.548583984375, 58.545387268066406, 349.15057373046875, 57.23704528808594, 198.05624389648438, 49.39308166503906, -29.326740264892578, 325.083984375, 0.0, 266.1650695800781, 67.2378921508789], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000386.npy"}
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 76.87397766113281, "std": 123.32230377197266, "min": -343.5164794921875, "p10": -63.012831878662105, "median": 64.34659576416016, "p90": 228.45024871826172, "max": 406.6053771972656, "pos_frac": 0.7734375, "sample": [62.65528869628906, -60.99372100830078, 46.472808837890625, 19.91126251220703, 15.890594482421875, 213.85443115234375, 166.43569946289062, -67.72409057617188, -332.799072265625, -81.94854736328125, 0.0, 217.61050415039062, -2.296630859375, 126.0400390625, 207.8101806640625, 406.6053771972656, 49.439170837402344, -96.296630859375, -44.33502197265625, 19.69134521484375, -21.731246948242188, 56.277618408203125, 178.00245666503906, -117.77200317382812, 79.50439453125, 221.8087158203125, -93.10470581054688, 7.5328216552734375, 233.57814025878906, 19.485116958618164, 197.71453857421875, 169.6668701171875, 198.61996459960938, 282.6399841308594, 127.57374572753906, 140.62081909179688, -8.16876220703125, 54.136756896972656, 73.88040924072266, 0.233551025390625, -79.10980224609375, 237.99481201171875, 15.105300903320312, -14.390899658203125, 228.72470092773438, 18.404685974121094, 67.52597045898438, 53.067352294921875, 83.91986846923828, 35.540679931640625, 55.166259765625, 279.6612854003906, -191.14663696289062, 3.4339599609375, -51.5806884765625, 115.77677917480469, 66.09466552734375, -46.51902770996094, 168.71627807617188, 152.86630249023438, 235.9473419189453, 115.9227294921875, 118.39501953125, 33.537841796875, 159.93820190429688, -343.5164794921875, 66.03790283203125, 169.42141723632812, 108.34725952148438, -54.12901306152344, 8.249481201171875, 3.668060302734375, 71.76025390625, 162.88970947265625, 204.28807067871094, 27.349746704101562, 182.82354736328125, 74.1291732788086, -10.90093994140625, 40.152496337890625, 150.6110076904297, -27.246856689453125, 228.33262634277344, 2.79852294921875, 103.40412902832031, 273.5321350097656, 24.656322479248047, 11.552642822265625, 112.89447021484375, 199.54173278808594, -40.44734191894531, 58.84571075439453, -72.29082489013672, 24.481689453125, 279.38873291015625, 133.4742431640625, 47.6605224609375, -3.55535888671875, 220.09353637695312, 277.5667724609375, 28.22231674194336, 80.0528564453125, -21.2208251953125, 223.24578857421875, -13.336158752441406, -110.09710693359375, -167.89913940429688, 48.76640319824219, 137.376220703125, 326.4215087890625, 6.28515625, 89.19026947021484, 52.06597900390625, 131.9854736328125, 153.43505859375, 108.67597198486328, 93.00341033935547, 184.39620971679688, 240.89605712890625, 155.4501495361328, 8.735870361328125, -139.85971069335938, 194.0589599609375, 284.7235107421875, 26.019363403320312, 206.90594482421875, 43.19291305541992, 223.796630859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000387.npy"}
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 55.634796142578125, "std": 128.0048370361328, "min": -272.5568542480469, "p10": -103.28478393554687, "median": 39.53413391113281, "p90": 207.00070190429688, "max": 383.4415588378906, "pos_frac": 0.65625, "sample": [58.116943359375, -63.31658935546875, 172.53347778320312, 0.0, 352.84228515625, 170.84112548828125, -32.9384765625, 236.38768005371094, -23.868942260742188, 207.2576904296875, -129.65606689453125, 148.86483764648438, -67.63319396972656, -4.79608154296875, -9.693099975585938, 163.67218017578125, -0.8013153076171875, -46.084930419921875, 199.40673828125, -0.54119873046875, 309.43994140625, 309.8173828125, -186.72128295898438, -23.4957275390625, 12.664459228515625, -272.5568542480469, 94.76428985595703, 178.0491943359375, 51.8299560546875, 192.07809448242188, -221.7850341796875, -58.120849609375, 0.0, 54.79071044921875, 25.658767700195312, -50.399017333984375, -51.89276123046875, 48.5811767578125, 89.16326904296875, 105.22135162353516, 139.07208251953125, 20.53447723388672, 297.2787780761719, 205.18785095214844, -147.35757446289062, 40.545654296875, 158.41244506835938, -3.93939208984375, -18.52093505859375, 4.7009429931640625, 6.935707092285156, 32.877960205078125, 230.393310546875, -149.12954711914062, 36.647857666015625, 105.61063385009766, 152.1885986328125, 179.630126953125, -29.52655029296875, 21.411651611328125, -27.68663787841797, 74.78181457519531, 154.58255004882812, -58.75987243652344, 118.68877410888672, -150.60882568359375, 90.29806518554688, -146.10714721679688, 304.3340148925781, 7.244405746459961, 109.64923095703125, 255.03079223632812, 192.153564453125, 64.79541015625, 38.522613525390625, 114.43209838867188, -15.2257080078125, 311.3284912109375, 60.886444091796875, -221.96798706054688, 32.77156066894531, 197.87680053710938, 153.20547485351562, -31.64361572265625, 14.997154235839844, 65.0277099609375, -18.741104125976562, 175.1005401611328, 21.11749267578125, 143.83724975585938, -107.10406494140625, 58.2628173828125, -22.319503784179688, -184.94805908203125, 21.137977600097656, 292.2972106933594, 383.4415588378906, 50.42115020751953, 6.012290954589844, 54.06011199951172, 8.819412231445312, 70.520751953125, 162.11978149414062, 330.22930908203125, -101.64794921875, 56.78266906738281, -2.7427978515625, 12.44549560546875, -246.78871154785156, -3.477616310119629, 148.70486450195312, 123.6011962890625, 10.57769775390625, -46.501556396484375, 20.80889892578125, -2.536602020263672, 206.89056396484375, 99.48812866210938, 106.6954345703125, 63.73095703125, -8.714759826660156, -49.793304443359375, 68.37344360351562, 90.9639892578125, 32.5350341796875, 159.20375061035156, 127.63990783691406, -118.49137878417969], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000388.npy"}
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 73.90679931640625, "std": 132.15843200683594, "min": -210.66497802734375, "p10": -76.72282257080079, "median": 67.44560623168945, "p90": 259.70482482910154, "max": 412.8236083984375, "pos_frac": 0.6796875, "sample": [-76.95762634277344, -75.3973388671875, 26.722122192382812, 90.36956787109375, 80.99462890625, 156.16741943359375, 26.055023193359375, 54.968109130859375, -70.97221374511719, 314.648681640625, 124.67596435546875, 271.294677734375, -40.82891845703125, 227.96798706054688, -26.54446029663086, 89.82351684570312, -104.87295532226562, 122.85305786132812, 68.77125549316406, 398.04376220703125, -50.95018768310547, 36.364845275878906, 78.52227783203125, 180.36602783203125, -50.89371109008789, 46.0955810546875, -63.397308349609375, 168.14141845703125, 81.91209411621094, 412.8236083984375, 38.1839599609375, -210.66497802734375, 113.78018188476562, -156.8188934326172, -38.631439208984375, -93.95440673828125, -88.61450958251953, 21.233745574951172, 276.9740295410156, 277.7545471191406, 313.5614318847656, 0.0, 39.33995056152344, 0.753448486328125, 270.0932312011719, -76.6221923828125, 282.73858642578125, 177.01036071777344, 122.76165771484375, 138.68692016601562, -3.8658599853515625, 183.738525390625, 57.895263671875, 9.644514083862305, 208.37705993652344, 23.81146240234375, 117.98623657226562, 222.55694580078125, -8.62255859375, -32.66841125488281, 22.991622924804688, -2.185302734375, 306.2449951171875, 227.0560302734375, -125.76220703125, 104.50468444824219, 151.154541015625, 182.93138122558594, 128.11773681640625, 6.19659423828125, 107.68759155273438, 79.67596435546875, 77.45791625976562, 238.02928161621094, -15.25225830078125, 260.02117919921875, 15.922805786132812, 123.16424560546875, -28.193695068359375, -8.32159423828125, 172.70693969726562, 175.77713012695312, 162.84225463867188, 412.06103515625, -26.415382385253906, 383.498291015625, -202.8580780029297, 94.95454406738281, 29.740341186523438, -24.80316162109375, 116.491455078125, 83.91229248046875, 160.47012329101562, 66.11995697021484, -99.114990234375, 4.4925689697265625, 77.44151306152344, -67.7886962890625, -68.76596069335938, 179.60406494140625, -73.13238525390625, 202.20904541015625, -25.2305908203125, -57.028221130371094, 113.08837890625, 219.30142211914062, 0.0, 155.19387817382812, 85.41897583007812, 196.5706787109375, 7.965448379516602, -86.17073059082031, -28.864593505859375, 81.61968994140625, 255.23483276367188, 237.69296264648438, 84.97967529296875, 54.91590881347656, -120.52302551269531, -68.51753997802734, 7.235870361328125, 94.02799224853516, -201.31378173828125, -91.56822204589844, 17.166061401367188, -59.708099365234375, 34.9434814453125, 259.5692443847656], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000389.npy"}
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 76.08638000488281, "std": 119.36808776855469, "min": -187.524169921875, "p10": -57.80337524414062, "median": 62.866912841796875, "p90": 226.23525848388672, "max": 410.45159912109375, "pos_frac": 0.71875, "sample": [166.477294921875, -46.145301818847656, -35.418701171875, 96.120849609375, -47.205108642578125, 29.537490844726562, 21.203195571899414, -8.47186279296875, 316.42022705078125, 139.17752075195312, -44.698974609375, 15.589630126953125, -12.380027770996094, -6.4317626953125, 166.43368530273438, 187.32235717773438, -45.10600280761719, 86.69273376464844, 227.7194366455078, 190.90957641601562, 142.87147521972656, -12.983810424804688, 245.767333984375, 25.650863647460938, 160.80908203125, 22.8082275390625, -26.46258544921875, 193.99505615234375, 145.09149169921875, 164.53543090820312, 224.334716796875, 120.59019470214844, 37.35990905761719, 161.80413818359375, 36.0650634765625, 21.8748779296875, 24.111637115478516, 44.33177185058594, -56.7608642578125, 68.36097717285156, -108.92822265625, 146.32984924316406, -81.24981689453125, 144.27655029296875, 69.46160888671875, -177.93753051757812, -18.729747772216797, -127.6279296875, 160.96270751953125, -24.964324951171875, -19.8685302734375, 132.4505615234375, -34.70442199707031, 31.36444091796875, 410.45159912109375, -133.6158905029297, 8.00634765625, 167.38107299804688, 151.1384735107422, -60.23590087890625, 124.89926147460938, 196.9545135498047, 262.96044921875, -21.57488250732422, 264.7989807128906, 36.93403625488281, 351.41796875, 161.55267333984375, -133.94845581054688, 237.2957763671875, 45.761810302734375, 206.4217529296875, -51.10162353515625, 0.0, 62.92279052734375, -14.642730712890625, 195.46954345703125, -139.47105407714844, 180.76333618164062, 155.63262939453125, -53.29602813720703, 124.53366088867188, 190.45932006835938, -187.524169921875, 16.766143798828125, 146.29345703125, 349.84423828125, 213.0648651123047, 55.8570556640625, -185.00033569335938, 99.36865234375, 189.5221405029297, 97.0245361328125, 106.28302001953125, 56.486663818359375, 17.7808837890625, -109.7264404296875, 266.27349853515625, 74.83478546142578, -78.90625, 15.243196487426758, -44.43797302246094, 269.7041015625, 184.295654296875, 23.065505981445312, 108.37213134765625, 275.85498046875, 3.3834228515625, 140.1992950439453, 3.067626953125, -23.230560302734375, 48.91668701171875, -37.949462890625, 108.11865234375, 62.81103515625, 11.569549560546875, 161.940185546875, 110.26097869873047, 95.58511352539062, 258.90203857421875, 52.43089294433594, 71.462646484375, 58.99920654296875, 225.59918212890625, 175.26348876953125, -69.970458984375, 14.644577026367188, 146.1065673828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000390.npy"}
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 80.52790832519531, "std": 124.4656982421875, "min": -209.4593505859375, "p10": -72.1929931640625, "median": 68.45375061035156, "p90": 230.25756378173827, "max": 479.823974609375, "pos_frac": 0.75, "sample": [109.7889404296875, -186.1012420654297, 28.000810623168945, 114.48486328125, 160.63600158691406, 45.677825927734375, -15.14013671875, 110.77943420410156, -52.69458770751953, -9.345022201538086, 48.7276611328125, 99.91436767578125, -53.03067398071289, 32.66963195800781, 230.03619384765625, 81.63778686523438, 49.76355743408203, 178.792724609375, -62.24769592285156, 67.47368621826172, -122.75669860839844, 129.2159423828125, 353.6337890625, 376.0737609863281, -45.96272277832031, 131.13470458984375, 187.97438049316406, 54.531951904296875, -13.101909637451172, -8.62548828125, -30.3338623046875, -144.99505615234375, 122.42919921875, 26.066701889038086, 53.291748046875, 115.44125366210938, -46.99444580078125, 2.965726852416992, -154.22532653808594, 206.1932373046875, 37.949302673339844, 199.07000732421875, 145.2494659423828, 479.823974609375, 190.7969970703125, 127.09996032714844, 190.90171813964844, 60.73353576660156, -31.33419418334961, 8.582244873046875, 162.68167114257812, 43.20744705200195, -2.8099517822265625, 186.72048950195312, -84.97454833984375, 167.3940887451172, 32.96272277832031, 341.9893798828125, 179.60523986816406, -22.596710205078125, 203.64346313476562, -95.87427520751953, 144.01576232910156, 11.327201843261719, 66.89398193359375, 215.91162109375, 260.238525390625, 13.751239776611328, 86.72859191894531, 1.2994842529296875, 27.28875732421875, 241.15447998046875, 40.93499755859375, 58.23028564453125, 57.50446319580078, -33.680145263671875, 275.42730712890625, 80.38079833984375, -70.75103759765625, 155.0569610595703, 83.32424926757812, 93.83010864257812, 144.98245239257812, 241.81463623046875, 93.969482421875, -23.569793701171875, 47.326942443847656, 166.0699462890625, 230.7740936279297, -13.421112060546875, 21.11895751953125, -1.4311084747314453, -123.79183959960938, 259.94891357421875, 212.887451171875, 94.13021850585938, 71.747802734375, 207.5889892578125, 218.4537353515625, 171.31927490234375, -209.4593505859375, -109.222412109375, 79.17684936523438, -130.24224853515625, 1.6503143310546875, 28.067188262939453, 11.674972534179688, 100.21490478515625, 239.154052734375, 220.26144409179688, 25.913482666015625, 142.0438995361328, 204.77536010742188, 68.24087524414062, 79.18734741210938, 209.292724609375, 68.6666259765625, 200.482666015625, 368.11944580078125, 94.84732055664062, -75.55755615234375, -178.0594482421875, 50.50927734375, 39.73161315917969, -8.315383911132812, 88.74868774414062, -112.38055419921875, 288.6607666015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000391.npy"}
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 97.52268981933594, "std": 123.22908782958984, "min": -401.76336669921875, "p10": -38.31715621948242, "median": 106.20073699951172, "p90": 232.16943359375, "max": 443.7958068847656, "pos_frac": 0.7578125, "sample": [136.29156494140625, 186.92405700683594, 264.0710144042969, 128.29629516601562, 106.41056823730469, 10.076187133789062, 132.3626708984375, 221.23532104492188, 119.463134765625, 219.6165771484375, -86.49388885498047, 134.42608642578125, 121.02974700927734, 47.519775390625, -10.218109130859375, -37.01776885986328, -401.76336669921875, 297.3502197265625, 58.62409973144531, -161.5889892578125, 163.44625854492188, -84.1416015625, -94.012939453125, 217.5830078125, 127.21920776367188, 166.3487548828125, 0.0, -100.1607666015625, 225.81629943847656, 375.54669189453125, -4.00640869140625, 90.46755981445312, 217.7681884765625, 114.12213134765625, 156.53868103027344, 204.89181518554688, 331.6478271484375, 147.5240020751953, 129.43655395507812, -36.69439697265625, 234.93841552734375, 197.12152099609375, 83.54654693603516, 40.7470703125, -10.629119873046875, 106.95225524902344, 1.43096923828125, -5.268310546875, 142.8401336669922, 137.39984130859375, 170.21194458007812, 201.98287963867188, 253.31808471679688, 85.8819580078125, 387.10943603515625, 162.985595703125, 6.738983154296875, 191.28555297851562, 113.8841552734375, -25.0679931640625, 88.44404602050781, -17.078086853027344, -72.08488464355469, 215.906005859375, 198.92770385742188, 47.4649658203125, -36.984771728515625, 205.94970703125, 382.5891418457031, -12.8359375, 166.486328125, 256.8868408203125, 259.06671142578125, 168.48751831054688, 82.78970336914062, 272.24273681640625, 84.51194763183594, -67.04975891113281, -5.1483917236328125, 18.261852264404297, 27.728050231933594, 115.20034790039062, -23.929412841796875, 105.99090576171875, 134.74168395996094, 65.91806030273438, 42.190460205078125, 52.752471923828125, 14.014083862304688, 9.0157470703125, -3.09930419921875, 0.0, 173.26947021484375, 25.81024169921875, 172.25933837890625, 230.98272705078125, 71.7183837890625, 107.74398803710938, 197.71383666992188, -16.975234985351562, 184.69778442382812, -10.79088020324707, 158.97193908691406, 6.917028427124023, 165.99578857421875, 112.10552978515625, 11.870269775390625, 142.34979248046875, 175.1184844970703, 29.008316040039062, -43.58831787109375, 100.42288208007812, 45.38909149169922, -10.030487060546875, 313.2408447265625, 10.72705078125, 443.7958068847656, -41.34906005859375, -83.27969360351562, 142.3497314453125, 69.09520721435547, 194.88607788085938, -118.07844543457031, 101.1993408203125, -57.983642578125, 84.14445495605469, 201.396484375, 101.0772705078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000392.npy"}
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 71.10809326171875, "std": 118.98391723632812, "min": -244.2095947265625, "p10": -54.853903198242186, "median": 55.24502944946289, "p90": 209.07058715820312, "max": 423.5979919433594, "pos_frac": 0.703125, "sample": [-123.25621032714844, 54.879615783691406, -10.411849975585938, 51.588592529296875, 125.29712677001953, -105.322265625, 152.15228271484375, 208.29833984375, 213.65719604492188, 2.32330322265625, 50.19683837890625, -55.314666748046875, 97.95027160644531, -4.7558746337890625, 133.21148681640625, 55.1859130859375, 2.949298858642578, 23.395320892333984, 103.1481704711914, 161.795166015625, 197.6204833984375, 63.34661865234375, 99.61618041992188, 111.61044311523438, -113.5921630859375, 265.7045593261719, 187.24398803710938, 161.5014190673828, 124.96212768554688, 148.03489685058594, 289.84918212890625, -8.4901123046875, -46.177032470703125, 186.64852905273438, 120.15521240234375, 101.84009552001953, 7.4465484619140625, 167.45370483398438, 25.43145751953125, 136.02488708496094, 156.40185546875, -105.26083374023438, 146.74530029296875, 70.357421875, 74.3900146484375, 186.410400390625, -66.36216735839844, -26.502532958984375, -114.99166870117188, 147.75735473632812, 383.36883544921875, 45.073585510253906, -11.713455200195312, 52.4833984375, 108.5276870727539, 45.0078125, 3.248281478881836, -42.43751525878906, 190.0272979736328, -34.30744934082031, -235.94549560546875, 53.76824951171875, 114.56582641601562, -21.715972900390625, 55.30414581298828, 132.2021484375, 47.983551025390625, 184.01455688476562, -137.1197509765625, 20.254638671875, -121.355712890625, 47.15158462524414, 24.286788940429688, 423.5979919433594, 50.45227813720703, 38.960113525390625, 360.2509765625, -53.032958984375, 246.0924072265625, -41.09640121459961, 22.49165153503418, -0.5855331420898438, 228.61993408203125, -106.70707702636719, 116.73932647705078, 33.02709197998047, 219.03997802734375, 9.844619750976562, 180.70330810546875, 163.95217895507812, -44.43260192871094, 348.99615478515625, 207.18875122070312, 221.17808532714844, 311.0946044921875, 65.51785278320312, -16.029571533203125, -54.65643310546875, 139.84048461914062, -244.2095947265625, -2.617664337158203, 128.81512451171875, 94.41192626953125, -3.6037750244140625, 209.92401123046875, 112.38667297363281, 58.44187927246094, 0.0, 70.09718322753906, 79.80032348632812, 208.704833984375, 78.89932250976562, 18.366561889648438, 3.4508056640625, -19.074615478515625, -10.139202117919922, 182.5992431640625, -11.510354995727539, -171.62844848632812, -13.069778442382812, 0.8824539184570312, -24.899200439453125, 136.80140686035156, 181.57717895507812, -18.5074462890625, -36.052734375, 104.73701477050781, 181.38925170898438], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000393.npy"}
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 69.25773620605469, "std": 133.35870361328125, "min": -198.5760498046875, "p10": -86.82018737792968, "median": 44.58347511291504, "p90": 253.73993530273438, "max": 455.16632080078125, "pos_frac": 0.6328125, "sample": [174.8115234375, -40.55218505859375, 192.27996826171875, 84.377197265625, 320.34942626953125, 231.7642822265625, -93.41253662109375, -34.89703369140625, 4.713714599609375, 156.00607299804688, 86.562744140625, -15.21331787109375, 40.1163330078125, -1.277984619140625, -6.5755615234375, -46.585166931152344, 229.14492797851562, -117.07990264892578, 298.6512451171875, -32.69183349609375, 15.63507080078125, 273.8752746582031, 281.2944641113281, -7.749744415283203, -0.7688198089599609, -38.59124755859375, 53.467376708984375, 69.09797668457031, 189.3917236328125, 95.31536865234375, -92.74508666992188, 0.0, -12.650459289550781, -36.55598449707031, 425.143310546875, 14.50628662109375, -31.633468627929688, 141.04928588867188, -31.230377197265625, 84.87820434570312, 254.00213623046875, 189.37786865234375, 83.417236328125, -116.0921630859375, 41.23820877075195, 172.163330078125, -5.3366851806640625, 28.503551483154297, 285.2362060546875, -7.974853515625, -99.28695678710938, 2.367767333984375, 115.22265625, -62.1773681640625, -27.0684814453125, 253.6275634765625, -34.97496032714844, -11.543701171875, 39.01002502441406, 14.781646728515625, 177.09231567382812, -58.59259033203125, 0.087127685546875, 14.332481384277344, 219.02056884765625, 37.10088348388672, -20.181198120117188, 119.48992919921875, -65.20995330810547, 105.65509033203125, 94.78817749023438, 328.248779296875, 318.8772277832031, 61.80500793457031, 455.16632080078125, -39.31964111328125, -198.5760498046875, -74.01690673828125, -18.959854125976562, 243.312255859375, 31.80944061279297, -108.324951171875, 133.59347534179688, -2.632904052734375, -163.23269653320312, 74.72061157226562, 67.14080810546875, 217.72125244140625, 69.8841323852539, 100.0582275390625, -14.691619873046875, -26.208126068115234, 47.928741455078125, 58.62825012207031, -76.47100830078125, 2.5048065185546875, 30.995574951171875, 59.22320556640625, 52.16023254394531, 115.3599853515625, -144.46022033691406, 242.65744018554688, 59.141387939453125, -131.68655395507812, 113.3618392944336, 355.22491455078125, -150.08139038085938, 77.55755615234375, 227.18385314941406, 188.19415283203125, 173.3387451171875, 154.3987579345703, 242.21145629882812, 294.24224853515625, 211.90670776367188, -84.28094482421875, 298.060302734375, -7.1978759765625, 34.050567626953125, 146.64581298828125, 89.17633056640625, 188.61737060546875, 9.896209716796875, 159.3979949951172, -160.19598388671875, -160.1939239501953, -72.23703002929688, 203.0596923828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000394.npy"}
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 87.89419555664062, "std": 128.0573272705078, "min": -149.54641723632812, "p10": -47.849978637695315, "median": 60.953826904296875, "p90": 254.9525695800781, "max": 511.158447265625, "pos_frac": 0.7265625, "sample": [115.49871826171875, 246.7645721435547, -8.622352600097656, -37.76551055908203, 50.74407958984375, -7.90362548828125, 13.4886474609375, 250.19598388671875, 511.158447265625, 88.29769897460938, 25.730690002441406, 94.592041015625, 59.99468994140625, -2.7270660400390625, -2.142242431640625, 113.34418487548828, 33.60630798339844, -27.40509033203125, 72.9041748046875, 94.42221069335938, 23.630889892578125, 269.02398681640625, -13.536079406738281, 13.685501098632812, 225.01309204101562, 113.7288818359375, 19.291259765625, 0.4827384948730469, 153.3218994140625, 364.6936950683594, 137.777587890625, 100.74771118164062, -89.10675048828125, -95.48411560058594, -55.93583679199219, 113.98321533203125, 155.42318725585938, 94.5594482421875, 0.09131622314453125, 47.82722091674805, 143.8084259033203, 11.951423645019531, -131.77947998046875, 78.52743530273438, 142.11196899414062, 29.937973022460938, -16.887222290039062, -30.581401824951172, 170.32835388183594, 220.6431427001953, 201.19598388671875, -149.54641723632812, 29.274505615234375, 167.8178253173828, 309.75799560546875, 23.022003173828125, 160.2987060546875, 182.75442504882812, 10.1929931640625, -48.142364501953125, -20.50115966796875, 65.39680480957031, 227.8269500732422, 25.921417236328125, 266.05126953125, -20.53509521484375, -51.02076721191406, -20.009605407714844, -1.458251953125, -29.51104736328125, -35.108428955078125, 34.648956298828125, -10.766471862792969, -140.8453369140625, 43.535186767578125, 209.59976196289062, 115.70541381835938, 59.940338134765625, -5.446929931640625, 146.818359375, 61.9129638671875, 214.37332153320312, 445.373779296875, 225.27142333984375, 193.95516967773438, 80.13018798828125, 29.325927734375, -111.92605590820312, 96.24557495117188, 207.8242645263672, 31.259601593017578, 96.63685607910156, -47.72467041015625, -13.548421859741211, -101.9720458984375, 12.406265258789062, 168.28009033203125, 273.79620361328125, 163.01583862304688, 306.1817321777344, -17.645965576171875, 145.032470703125, 43.521728515625, 143.8991241455078, 175.7698974609375, 46.05522155761719, 329.2464599609375, 37.641387939453125, 138.3668212890625, -50.012115478515625, 116.88275146484375, 214.45803833007812, 335.2857666015625, -137.46905517578125, -54.92474365234375, -38.010345458984375, 68.20361328125, 51.873817443847656, 194.96978759765625, 376.60894775390625, 484.41162109375, 291.38677978515625, 84.57974243164062, -4.424530029296875, 9.862457275390625, 157.20962524414062, 48.47389221191406, 100.06231689453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000395.npy"}
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 74.02458953857422, "std": 120.56880187988281, "min": -200.00436401367188, "p10": -96.63888854980468, "median": 62.78437423706055, "p90": 218.93621520996092, "max": 390.626953125, "pos_frac": 0.71875, "sample": [157.59735107421875, 63.778350830078125, -28.17657470703125, 384.2524719238281, 178.75244140625, -78.2520751953125, -11.250228881835938, 61.624168395996094, -200.00436401367188, 233.15267944335938, 62.01087951660156, -132.7721710205078, 7.823028564453125, 150.35223388671875, -35.98417663574219, 25.752288818359375, 238.33084106445312, -48.84857177734375, 26.163291931152344, 221.0753936767578, 43.28363800048828, -6.0824127197265625, -24.4154052734375, 368.04638671875, -151.8892364501953, 21.194915771484375, 137.18804931640625, 147.44583129882812, 104.48602294921875, 24.34302520751953, 59.86627197265625, 63.32439422607422, 123.82809448242188, 207.28311157226562, -53.174285888671875, 22.318695068359375, 59.72920227050781, -19.1365966796875, -43.773414611816406, 118.54399871826172, 88.38861083984375, 71.96247863769531, 162.67578125, 218.01942443847656, 390.626953125, 36.2850341796875, -4.034626007080078, 85.30685424804688, 6.57818603515625, 123.24366760253906, 119.40493774414062, 297.86529541015625, 160.50711059570312, 37.198394775390625, 51.149436950683594, 134.94015502929688, 140.01593017578125, -156.60232543945312, 43.88818359375, -43.105220794677734, 53.20685958862305, -34.199554443359375, -41.13868713378906, -23.502853393554688, 148.37088012695312, -108.85293579101562, 140.1422119140625, -102.50216674804688, 205.65032958984375, -48.501739501953125, 170.72047424316406, -29.738861083984375, 140.228759765625, 61.423728942871094, 168.74307250976562, 114.884033203125, 138.99493408203125, 139.43606567382812, -38.089385986328125, -136.201904296875, 44.15887451171875, 135.85968017578125, 7.48822021484375, 138.05084228515625, 81.96954345703125, 100.21826171875, 34.50592041015625, 116.38818359375, 5.591653823852539, 240.13436889648438, 106.90234375, 347.6954345703125, -100.14010620117188, -16.08355712890625, 12.15676498413086, 39.919464111328125, 213.640869140625, 137.87344360351562, -17.3133544921875, 180.41036987304688, 252.55027770996094, 194.26425170898438, 43.42987060546875, -157.2294921875, 62.244354248046875, 174.64141845703125, -115.2218246459961, 264.52850341796875, 179.59979248046875, 68.70443725585938, 0.0, 148.85342407226562, 189.3282470703125, 282.22784423828125, 97.0135498046875, 84.29006958007812, 0.0, 197.39840698242188, 164.83810424804688, -95.13836669921875, 30.560836791992188, -108.36564636230469, 59.078857421875, 270.2261962890625, 173.08505249023438, -132.56704711914062, -153.54812622070312, 129.75225830078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000396.npy"}
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 68.19586944580078, "std": 135.1341552734375, "min": -247.67971801757812, "p10": -100.44875183105468, "median": 56.91986083984375, "p90": 224.64825439453125, "max": 524.8845825195312, "pos_frac": 0.6875, "sample": [3.191253662109375, -8.70977783203125, -4.5533905029296875, 139.69625854492188, 198.6683349609375, -61.97868347167969, -37.37287902832031, 199.14425659179688, -13.344001770019531, 129.90301513671875, -22.40545654296875, 50.5943603515625, -94.75328063964844, 57.062744140625, 43.77162170410156, 156.0879364013672, 55.17994689941406, 175.06207275390625, 213.7252197265625, 213.71282958984375, -247.67971801757812, 75.60247802734375, -10.875885009765625, 82.506103515625, 69.82186889648438, 122.39273071289062, 0.0, 8.44012451171875, 252.81143188476562, 80.26766967773438, 0.0, -86.73968505859375, 249.34202575683594, -26.778594970703125, 15.62322998046875, 6.832862854003906, 89.54901123046875, 151.5786895751953, -90.23516845703125, 51.59808349609375, 181.5826416015625, 152.58944702148438, 227.15362548828125, 140.05819702148438, -34.38629150390625, -34.9100341796875, 156.355224609375, 326.05242919921875, 168.2816162109375, -97.14883422851562, 524.8845825195312, -73.58285522460938, 128.39874267578125, -203.18310546875, -123.5467529296875, -110.22268676757812, 67.75900268554688, -169.11038208007812, 36.65463638305664, -216.859130859375, -108.1485595703125, 242.90811157226562, 360.0860595703125, 130.90135192871094, 31.467254638671875, 39.47768783569336, 220.9407958984375, 70.73762512207031, -74.65849304199219, -2.2707977294921875, 259.063232421875, 240.72528076171875, 179.84970092773438, 186.78567504882812, 161.43988037109375, 97.5263671875, 205.8408203125, 63.25604248046875, -203.1988525390625, 26.248046875, 190.35116577148438, 170.56097412109375, 386.9253845214844, 299.5499267578125, 44.6878662109375, -11.630241394042969, 105.63456726074219, -12.4583740234375, 5.7593841552734375, 144.46319580078125, 132.838134765625, -37.429962158203125, 71.22682189941406, 26.907928466796875, 209.2889404296875, 132.96719360351562, 174.9144287109375, 11.187164306640625, 223.57452392578125, 54.02342224121094, 33.91943359375, -128.8843994140625, 84.56396484375, 22.233264923095703, 282.1259765625, 193.52423095703125, 87.6430435180664, -114.95272827148438, -12.800575256347656, -40.44489669799805, -171.58338928222656, 188.3912353515625, 185.3643798828125, 27.82708740234375, 63.695220947265625, 56.7769775390625, -160.55377197265625, 6.14654541015625, 124.94769287109375, 27.746307373046875, -55.21722412109375, -214.5234375, 209.33859252929688, 173.53195190429688, -8.615936279296875, -94.371337890625, 11.459716796875, 265.90374755859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000397.npy"}
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 61.36546325683594, "std": 149.21083068847656, "min": -502.41583251953125, "p10": -123.96866455078124, "median": 65.17003631591797, "p90": 241.77681274414059, "max": 363.46929931640625, "pos_frac": 0.6640625, "sample": [139.8876953125, 142.5545654296875, 164.13595581054688, 15.861907958984375, 363.46929931640625, -45.57777404785156, 256.08380126953125, 44.05633544921875, 88.42308044433594, 358.6935729980469, -108.34500122070312, -78.96630859375, -263.18548583984375, 77.46502685546875, 15.666877746582031, 62.07838439941406, 155.48556518554688, -2.087493896484375, 289.7953796386719, 259.8052673339844, 220.818115234375, -57.98480224609375, 162.64614868164062, 174.3545379638672, 77.99041748046875, 151.01998901367188, 85.75336456298828, 205.82369995117188, 130.40509033203125, 81.37254333496094, -21.800811767578125, 153.24234008789062, 179.43377685546875, -220.61981201171875, -162.65469360351562, 84.33357238769531, 150.7340087890625, 131.16871643066406, 163.9215087890625, -23.039112091064453, 4.5229644775390625, 23.65557861328125, 18.740676879882812, 220.62020874023438, -38.47663879394531, -206.2030029296875, 61.47477722167969, 117.4043960571289, 22.652252197265625, -10.02740478515625, -46.864410400390625, 77.57199096679688, 14.336845397949219, 33.733306884765625, 160.28909301757812, 103.75848388671875, -2.543577194213867, -127.3568115234375, -5.4055633544921875, -11.719675064086914, 133.80555725097656, 74.7523193359375, 284.64849853515625, -132.41390991210938, 109.35491943359375, -15.487686157226562, -8.705005645751953, -118.24755859375, 55.70289611816406, 264.80987548828125, 221.43368530273438, 252.943359375, 32.61238098144531, 229.46994018554688, -130.49761962890625, -122.5166015625, 21.373088836669922, 196.22027587890625, -299.3585205078125, -88.95783996582031, 318.9381408691406, 236.99114990234375, -34.975067138671875, 142.5325469970703, 37.80094909667969, 36.8624267578125, -145.6389617919922, 104.28431701660156, -45.10986328125, -240.1751708984375, 224.71218872070312, 221.70928955078125, -60.63587951660156, -69.95095825195312, -206.41064453125, -76.93594360351562, 204.3538818359375, -50.4508056640625, 0.948150634765625, 182.5760040283203, -83.54747009277344, -60.697052001953125, 213.06817626953125, 152.34474182128906, 68.26168823242188, -1.1161575317382812, 272.2708740234375, 323.01141357421875, 260.98040771484375, -98.95736694335938, -502.41583251953125, 337.7589111328125, 196.41058349609375, -66.96471405029297, 220.99249267578125, 69.35284423828125, -32.58128356933594, -187.50360107421875, 25.676116943359375, 191.07391357421875, 152.2012939453125, 17.2152099609375, 25.78276824951172, 186.2794952392578, 181.29446411132812, 95.06625366210938, 139.77236938476562, 4.998809814453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000398.npy"}
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 72.59346771240234, "std": 129.20462036132812, "min": -229.3101806640625, "p10": -76.28969497680663, "median": 59.09124755859375, "p90": 246.8263549804687, "max": 408.1575927734375, "pos_frac": 0.7265625, "sample": [-38.49617004394531, 297.2981872558594, 107.98287963867188, 194.376220703125, 51.30712890625, -151.5202178955078, 140.54783630371094, 105.05630493164062, 79.89337158203125, 293.9141845703125, 176.02871704101562, -31.316207885742188, -109.31112670898438, 268.848876953125, 13.748796463012695, 378.3388671875, 15.201263427734375, -35.075775146484375, 136.72215270996094, 64.2298583984375, -59.059600830078125, -66.08511352539062, -58.28521728515625, 183.22076416015625, -63.51123046875, -49.2669677734375, 22.978973388671875, -172.59909057617188, -229.3101806640625, 138.41905212402344, 185.21517944335938, 224.02435302734375, -87.64836883544922, 181.69329833984375, 57.66935729980469, 3.5194244384765625, 177.48895263671875, -160.6566619873047, -26.653701782226562, 240.978515625, 27.0433349609375, 173.871826171875, 264.6080322265625, 264.168701171875, 4.663330078125, 116.93146514892578, 203.096923828125, -106.97607421875, 185.96075439453125, 51.75266647338867, -71.42169189453125, 24.146713256835938, 51.07987976074219, 132.46347045898438, 375.59429931640625, -33.763946533203125, 35.48429870605469, 14.55206298828125, -29.89112091064453, -47.3370475769043, 7.45294189453125, 260.4713134765625, -24.111831665039062, 9.805816650390625, 85.0997543334961, 21.151477813720703, 279.3747253417969, 137.15386962890625, 140.50572204589844, 144.79144287109375, 375.4692077636719, 91.5767822265625, -37.86375427246094, -13.90716552734375, -165.223388671875, -103.97991943359375, -2.6063232421875, 63.095306396484375, 164.1773681640625, 121.0241470336914, 106.81373596191406, 187.31353759765625, 133.84619140625, 164.97201538085938, 131.4193115234375, 408.1575927734375, 91.35501098632812, 7.2680206298828125, 174.22705078125, 24.3116455078125, 172.90509033203125, 122.84201049804688, 4.2676239013671875, 143.97650146484375, 169.1361846923828, 56.33482360839844, -201.8599853515625, -19.374755859375, 266.8994140625, -16.15460968017578, 142.89019775390625, 96.8837890625, 177.02053833007812, -92.76092529296875, 34.20367431640625, 229.9061737060547, -55.19598388671875, 164.4432373046875, 110.12652587890625, 56.31455993652344, 343.61724853515625, -168.08651733398438, -108.33059692382812, 60.51313781738281, 18.995893478393555, 24.127548217773438, -64.40682983398438, 179.06094360351562, 8.467269897460938, 127.78536987304688, 20.102752685546875, 14.786529541015625, 31.935073852539062, 2.349374771118164, -70.37384033203125, 68.194091796875, 97.04414367675781, 124.30209350585938], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000399.npy"}
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 72.9753189086914, "std": 127.13419342041016, "min": -278.43408203125, "p10": -107.17769317626951, "median": 73.12876892089844, "p90": 226.81445922851557, "max": 350.25390625, "pos_frac": 0.71875, "sample": [51.270599365234375, -9.466583251953125, -48.049896240234375, 52.824951171875, 178.16973876953125, 88.18202209472656, 107.99275207519531, 57.75556945800781, 36.71665954589844, 70.40459442138672, 81.5374755859375, 175.50421142578125, -214.11380004882812, -43.986236572265625, 269.6676025390625, 103.32086181640625, 151.33370971679688, 100.40533447265625, 102.22212219238281, 117.96711730957031, 197.7601318359375, -82.96751403808594, 36.359130859375, 35.143585205078125, 155.87261962890625, 336.2618103027344, -278.43408203125, 86.4403076171875, -29.1771240234375, 164.26373291015625, 69.02207946777344, 50.0023193359375, -73.7774658203125, 118.75814819335938, 69.98686218261719, 104.50921630859375, 238.4798583984375, 221.81500244140625, -7.084228515625, 135.30685424804688, -184.1542205810547, 325.0035095214844, 189.79693603515625, 295.18731689453125, 33.00604248046875, 331.04998779296875, 85.695556640625, 65.73818969726562, 144.3671875, 15.058868408203125, 91.44512939453125, 206.47659301757812, 169.65321350097656, -126.5394287109375, 21.91864013671875, 10.588165283203125, 41.313716888427734, 74.39501953125, 252.11329650878906, -100.56806182861328, -11.031393051147461, 144.93994140625, 121.71075439453125, 184.63494873046875, 107.16378784179688, -131.95230102539062, 95.0714111328125, 62.30577087402344, 275.45355224609375, 183.22357177734375, 172.8438720703125, 28.219154357910156, -121.73109436035156, -28.609970092773438, -44.68891906738281, -200.94989013671875, 244.06781005859375, 350.25390625, 135.9700164794922, 57.98394775390625, -11.627761840820312, 220.07293701171875, 153.81265258789062, 31.58172607421875, -23.339683532714844, 198.2353515625, 142.81195068359375, -169.31405639648438, 170.60934448242188, 71.53852081298828, 42.065372467041016, -0.285888671875, -14.067276000976562, -128.05760192871094, 8.85986328125, 71.86251831054688, 15.4395751953125, 210.9140167236328, -10.043420791625977, -151.364990234375, 100.94091796875, 215.6182403564453, 142.0248260498047, 32.31610107421875, 18.210205078125, 36.00579833984375, 193.94927978515625, -1.1200599670410156, -123.84866333007812, 122.70051574707031, 161.37881469726562, -95.14900207519531, 126.023681640625, -56.0230712890625, 169.01922607421875, 83.39376831054688, 209.30580139160156, -100.94052124023438, 274.54302978515625, 267.6017761230469, 290.1561279296875, 74.75872802734375, 0.0, -122.705322265625, -123.365478515625, -18.33062744140625, -100.42044067382812, 190.46966552734375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000400.npy"}
{"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 79.57394409179688, "std": 127.80132293701172, "min": -313.96905517578125, "p10": -78.45968627929686, "median": 69.29790115356445, "p90": 241.59904479980466, "max": 432.4090881347656, "pos_frac": 0.7890625, "sample": [-127.46685028076172, -10.423736572265625, 213.64486694335938, -134.02288818359375, 81.40869140625, 255.63265991210938, 27.4281005859375, 147.26873779296875, -88.94134521484375, -33.2967529296875, 160.69171142578125, 134.58236694335938, 125.69548797607422, 85.18695068359375, 36.450531005859375, 9.335357666015625, 34.5303955078125, 109.13105773925781, 15.159271240234375, 123.80096435546875, -10.937736511230469, 91.48246765136719, -36.893638610839844, -66.22393798828125, -230.60345458984375, 338.7466735839844, 268.6353759765625, 8.705167770385742, 121.547119140625, 58.476043701171875, 158.91253662109375, 351.787109375, 44.92418670654297, 70.16650390625, 13.37896728515625, 36.26324462890625, 193.7941436767578, 207.74185180664062, 89.68412017822266, 68.4292984008789, 343.24896240234375, 112.5849609375, 250.35556030273438, 133.56817626953125, 182.35867309570312, 78.36004638671875, 14.220426559448242, 13.576614379882812, 193.96896362304688, 237.84625244140625, 19.65478515625, 172.09005737304688, 113.84298706054688, 111.01750946044922, -198.14950561523438, 185.2691650390625, 36.42962646484375, 113.271484375, 59.98448181152344, 16.609352111816406, -146.26095581054688, 203.14395141601562, 253.7900390625, 256.3431091308594, 54.470458984375, 141.08509826660156, -153.28628540039062, 126.93421936035156, 109.11544799804688, -58.0255126953125, -27.824230194091797, -74.1383056640625, 175.42388916015625, 56.96142578125, 55.01885986328125, -35.368438720703125, 262.6220703125, 418.1253662109375, 92.21795654296875, 42.02300262451172, 63.2864990234375, -88.54290771484375, 184.35675048828125, 150.473388671875, 142.30892944335938, 134.402587890625, 31.53057861328125, 10.526092529296875, 84.31585693359375, 138.9576416015625, 126.70065307617188, 98.51750183105469, -12.16619873046875, -104.1712646484375, 1.2500534057617188, 16.867446899414062, -95.39553833007812, 251.55450439453125, 370.5411682128906, 55.92169189453125, 11.27392578125, 27.0584716796875, 236.34869384765625, 85.8914566040039, 226.84632873535156, 67.14218139648438, -29.208740234375, 57.96624755859375, -26.626895904541016, 220.8489227294922, 22.235702514648438, 13.3271484375, -25.915802001953125, 4.195270538330078, 29.000808715820312, 14.329849243164062, 48.51239013671875, 72.3651351928711, 74.27468872070312, -103.37910461425781, 185.05865478515625, 191.17071533203125, 205.67550659179688, -37.868370056152344, 432.4090881347656, -313.96905517578125, -134.7685546875, 183.7425537109375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000401.npy"}
{"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 75.97816467285156, "std": 123.21009826660156, "min": -239.14198303222656, "p10": -48.168402099609374, "median": 53.75644302368164, "p90": 216.81376037597656, "max": 478.9024658203125, "pos_frac": 0.734375, "sample": [-51.035648345947266, -36.83649826049805, -47.597900390625, 115.97244262695312, 136.97842407226562, -11.09149169921875, -85.4144287109375, -215.03799438476562, 105.99652099609375, -139.1258087158203, 42.02850341796875, 370.19879150390625, 84.32717895507812, -23.059860229492188, -9.480094909667969, 280.22515869140625, 98.14387512207031, 15.953857421875, 4.2703857421875, -27.45404052734375, -26.019500732421875, 10.725173950195312, 117.84974670410156, 198.36572265625, -239.14198303222656, 6.533662796020508, 46.10968017578125, 256.11181640625, -205.02862548828125, 56.035003662109375, 104.7027587890625, -19.97308349609375, 15.184646606445312, 293.26751708984375, -20.48602294921875, 0.0, 478.9024658203125, 146.948486328125, -34.669342041015625, 217.25149536132812, 135.37713623046875, 71.3878173828125, 138.47105407714844, 2.5645751953125, 15.585250854492188, 185.91220092773438, -67.36759948730469, -62.26582336425781, 190.012939453125, 117.78860473632812, 48.369873046875, -80.05844116210938, -114.13996887207031, 150.97317504882812, 146.11474609375, 63.32258605957031, 250.64161682128906, 137.11090087890625, 3.1927490234375, 168.14605712890625, 14.565998077392578, 449.388427734375, 211.76742553710938, 54.87644958496094, 163.75344848632812, -13.652328491210938, 8.02459716796875, -9.10443115234375, 52.453392028808594, -132.76344299316406, 185.84060668945312, 190.2000274658203, 87.54957580566406, -41.59161376953125, 138.34823608398438, -7.0691070556640625, 309.3809814453125, -1.6981658935546875, 94.8939208984375, 113.90469360351562, 268.574462890625, 125.41336059570312, 130.96304321289062, 182.80978393554688, -24.012725830078125, 51.960205078125, 237.405517578125, 107.3267822265625, -49.49957275390625, 200.19549560546875, 54.16669464111328, 68.088623046875, 105.23818969726562, 9.807136535644531, 33.6553955078125, 191.31280517578125, 38.63212203979492, 216.62615966796875, 18.4765625, 158.65283203125, 178.2381591796875, 95.49418640136719, 50.324127197265625, 79.49447631835938, 437.72308349609375, -57.815093994140625, 93.92752075195312, -6.59149169921875, 31.0516357421875, 41.67266845703125, 15.84735107421875, 154.58523559570312, 53.34619140625, 299.1139831542969, -34.96112060546875, 29.4781494140625, -17.0411376953125, 216.35577392578125, 127.19038391113281, 20.598907470703125, 90.854736328125, 92.70860290527344, -1.4037208557128906, 12.032470703125, 7.85064697265625, 19.519287109375, 182.7886962890625, 6.189767837524414], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000402.npy"}
{"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 102.91226196289062, "std": 122.47110748291016, "min": -194.37078857421875, "p10": -40.809331512451166, "median": 95.51553344726562, "p90": 252.47869567871095, "max": 538.2734985351562, "pos_frac": 0.796875, "sample": [21.885299682617188, 149.67874145507812, -3.795074462890625, 278.9840393066406, 101.19668579101562, 128.291015625, 95.49447631835938, -21.824352264404297, 62.657073974609375, -42.917572021484375, 134.41036987304688, 189.86276245117188, 100.8779296875, -9.1103515625, 54.04248046875, 106.99197387695312, 43.17924499511719, 237.59552001953125, 54.84283447265625, -23.025489807128906, 245.6088104248047, -13.822206497192383, -185.08375549316406, 130.80645751953125, 203.18618774414062, 193.2156982421875, 227.3990478515625, 11.130615234375, 99.2283935546875, 231.689697265625, 212.08770751953125, -51.7838134765625, -4.752235412597656, 109.6995849609375, 9.716638565063477, 74.57772064208984, 32.370758056640625, 159.05694580078125, 0.0, -44.18109130859375, 1.8490982055664062, 237.3422088623047, 198.0540771484375, 105.25260162353516, 202.12930297851562, 241.18212890625, 218.14500427246094, 35.676971435546875, 273.4103698730469, 214.3402099609375, 60.91156005859375, 67.51223754882812, -12.8485107421875, 73.79541015625, 213.15185546875, 41.160797119140625, 33.8916015625, -80.78741455078125, 126.65380859375, -61.34539794921875, -18.512115478515625, 111.51840209960938, -143.52867126464844, 538.2734985351562, 147.4599609375, 54.451393127441406, 148.76806640625, -39.905799865722656, 48.505653381347656, 15.832000732421875, 50.035888671875, 121.90484619140625, 111.25749206542969, 332.87353515625, -7.3387451171875, 175.49073791503906, 267.00933837890625, 252.45742797851562, 116.2210693359375, 106.19775390625, 95.53659057617188, 110.71978759765625, 309.2889709472656, 230.27706909179688, 283.24932861328125, -63.740478515625, 29.671707153320312, 191.483154296875, 138.83074951171875, 158.93679809570312, 430.357421875, -4.012664794921875, 437.43060302734375, 252.5283203125, 13.610748291015625, 128.6334686279297, 26.596435546875, 224.43008422851562, 189.756591796875, 187.6771240234375, 222.1717987060547, 0.00583648681640625, 170.87596130371094, 5.59771728515625, -194.37078857421875, 265.3341064453125, -48.7608642578125, 58.30029296875, -4.698091506958008, 234.71719360351562, -72.0367431640625, 56.453834533691406, 82.60137939453125, 59.745811462402344, -122.98675537109375, 274.8836364746094, -47.10557556152344, 52.21849060058594, 37.649662017822266, 19.230884552001953, 6.59375, 163.7920379638672, 99.73138427734375, 91.16152954101562, 63.86431884765625, 91.13441467285156, 284.699462890625, 72.78448486328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000403.npy"}
{"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 76.63570404052734, "std": 133.26759338378906, "min": -296.00750732421875, "p10": -103.70298614501954, "median": 76.13488006591797, "p90": 243.7860107421875, "max": 441.9627685546875, "pos_frac": 0.75, "sample": [225.21229553222656, 191.71490478515625, 39.14959716796875, 160.60061645507812, -170.24557495117188, 243.09112548828125, 189.18020629882812, 89.09091186523438, -85.58721160888672, 0.0, -62.1409912109375, 298.03741455078125, 96.24568939208984, 29.011444091796875, -296.00750732421875, 224.0359649658203, 18.676361083984375, 245.445068359375, 7.489982604980469, 7.16571044921875, 308.18731689453125, 20.634796142578125, 21.08831787109375, 20.39874267578125, 3.797088623046875, 179.13320922851562, -231.79580688476562, 272.5845947265625, -103.89250183105469, 149.46080017089844, -8.861724853515625, -68.80615234375, 101.90667724609375, 36.9625244140625, -59.508216857910156, 8.215066909790039, -153.94229125976562, 25.533645629882812, 395.1200866699219, 82.14984130859375, 182.70123291015625, -2.2868804931640625, -0.959136962890625, -103.62176513671875, 145.87939453125, 31.031814575195312, 151.1337890625, 87.9044189453125, 38.81698989868164, 78.70965576171875, 85.6593017578125, 7.2650146484375, 205.37350463867188, -113.418701171875, 2.832275390625, 245.40740966796875, -136.36013793945312, 60.53657531738281, 42.17741394042969, 129.84548950195312, 193.01373291015625, 148.04498291015625, -68.98847961425781, 88.04244995117188, -37.946380615234375, 112.97057342529297, 264.25811767578125, 11.74591064453125, -32.96147155761719, 194.3568572998047, 121.48389434814453, 0.016805648803710938, 40.786529541015625, 24.372968673706055, 53.66009521484375, -121.56488037109375, -104.08444213867188, -152.19189453125, 178.40115356445312, 79.57130432128906, 54.10765075683594, 1.0403594970703125, 45.264739990234375, 84.3148193359375, 102.09915161132812, -9.875274658203125, -29.202255249023438, 10.23419189453125, 182.77813720703125, -217.33587646484375, 17.59058380126953, 182.70529174804688, 110.78564453125, 337.1751708984375, -142.8278045654297, -3.697265625, 73.56010437011719, 133.87159729003906, 89.95223999023438, 147.495849609375, -44.18364715576172, 63.94561767578125, 168.0498046875, 166.8526611328125, 156.68365478515625, 134.83740234375, 67.56571197509766, 98.3812255859375, -196.1786346435547, 382.46307373046875, 263.91156005859375, 180.95559692382812, 215.1564178466797, 319.40545654296875, 111.1204833984375, -4.891693115234375, 155.19869995117188, -18.666297912597656, 246.54562377929688, 234.33547973632812, 70.412109375, -29.233840942382812, 130.45481872558594, 147.5570068359375, 205.86480712890625, 194.1528778076172, 196.52696228027344, 441.9627685546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000404.npy"}
{"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 81.42776489257812, "std": 124.46420288085938, "min": -232.01177978515625, "p10": -41.93114547729492, "median": 62.65028381347656, "p90": 274.80504760742184, "max": 359.0108947753906, "pos_frac": 0.7421875, "sample": [40.7860107421875, -9.77935791015625, 114.52566528320312, 0.0, -31.10931396484375, -37.482421875, 69.41229248046875, 137.593505859375, 145.2374267578125, 55.730804443359375, 5.35693359375, -34.53643798828125, 53.465614318847656, 59.244598388671875, 155.89100646972656, -41.90229034423828, -166.04000854492188, 53.9449462890625, 210.1473846435547, 332.99786376953125, 112.31369018554688, 178.977783203125, 25.423364639282227, -34.980796813964844, 157.3159637451172, -18.992637634277344, -71.60910034179688, 85.40048217773438, 322.818115234375, 137.86827087402344, 152.1634521484375, 15.566551208496094, 79.35836791992188, 18.897247314453125, 346.8735046386719, 176.70578002929688, 86.05403137207031, -18.44744873046875, 0.20763587951660156, 26.20331573486328, -20.097957611083984, 326.5854187011719, 101.07036590576172, -2.0464725494384766, 36.117828369140625, 351.6580810546875, 307.9967346191406, 17.5053768157959, 199.61309814453125, -121.30389404296875, 66.05596923828125, 116.51678466796875, 287.0978088378906, 139.3928680419922, 34.66278076171875, -41.99847412109375, 113.90463256835938, 359.0108947753906, 75.08195495605469, -49.65472412109375, -136.55828857421875, 80.38605499267578, 98.58535766601562, -145.38986206054688, 99.00421142578125, -11.660980224609375, 230.06088256835938, -86.482666015625, 9.0693359375, 191.070556640625, 46.98777770996094, 164.98672485351562, 120.16256713867188, 219.42269897460938, 198.97113037109375, 298.71990966796875, -177.9541473388672, 177.98289489746094, 0.12920379638671875, 97.34683227539062, 174.938720703125, 6.220632553100586, -70.288818359375, -154.64901733398438, 56.71038818359375, 28.966537475585938, 2.9658355712890625, 277.729248046875, 39.816375732421875, 132.68991088867188, 200.62738037109375, 44.18340301513672, -165.50918579101562, 29.6165771484375, 313.8350524902344, 169.37713623046875, -12.532867431640625, -1.0629425048828125, 216.61627197265625, 165.3971710205078, -232.01177978515625, 100.0926513671875, 100.95401000976562, 293.900390625, -7.967613220214844, 128.15585327148438, 54.01848602294922, 148.45541381835938, 193.173828125, 22.389617919921875, 132.98126220703125, 68.09996032714844, 200.9105224609375, -25.83980369567871, 8.853790283203125, -10.426513671875, 19.743789672851562, 16.530412673950195, -37.999603271484375, 18.256717681884766, 122.93002319335938, 273.55181884765625, 218.70538330078125, -36.28778076171875, 43.39872741699219, 343.1215515136719, -31.21087646484375, 177.01412963867188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000405.npy"}
{"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 62.30121994018555, "std": 114.0311050415039, "min": -200.62234497070312, "p10": -69.41697387695312, "median": 48.42415237426758, "p90": 212.6324035644531, "max": 290.60662841796875, "pos_frac": 0.71875, "sample": [194.45956420898438, -26.639785766601562, -73.58258056640625, 149.456298828125, -67.6317138671875, 194.5089111328125, 93.07470703125, 144.2006072998047, 154.53558349609375, 271.6297302246094, 2.5872039794921875, 5.090694427490234, -24.7222900390625, 250.393310546875, 100.7086181640625, -171.751953125, 290.60662841796875, -24.888572692871094, 111.88578796386719, 141.2156982421875, -50.7681884765625, -164.37936401367188, 36.50067138671875, 90.92715454101562, 163.82980346679688, 106.502685546875, 11.666145324707031, 129.899658203125, 9.744171142578125, 160.6968994140625, 167.44171142578125, 259.2874755859375, -185.256591796875, -58.852386474609375, 149.98269653320312, -98.63070678710938, 113.66496276855469, -8.596939086914062, 41.24790954589844, 10.008514404296875, -153.94329833984375, -59.838165283203125, -56.91194152832031, -63.939697265625, 167.59947204589844, 81.59130859375, 26.935302734375, 169.3888397216797, 209.63076782226562, 141.68421936035156, -61.24101257324219, 98.55184936523438, 137.22775268554688, -22.857284545898438, -192.9163818359375, 24.0938720703125, 127.78399658203125, 164.95254516601562, -41.96179962158203, 32.8175048828125, 8.717437744140625, 281.87982177734375, -104.717041015625, 109.9715576171875, 90.59059143066406, 80.45123291015625, -0.643646240234375, -65.41586303710938, 67.35714721679688, 203.8658447265625, 259.2182922363281, 65.94086456298828, 21.8692626953125, 9.46649169921875, 189.64236450195312, 73.98301696777344, 37.518959045410156, 73.01641845703125, 8.223487854003906, 189.05673217773438, 190.69607543945312, 10.862224578857422, -200.62234497070312, 24.88136863708496, 246.73509216308594, 223.5321044921875, 171.39190673828125, -90.11500549316406, 217.6181640625, 9.030670166015625, 151.5271759033203, 21.17721939086914, 145.73390197753906, 197.39076232910156, 74.12887573242188, 273.62652587890625, 173.31640625, -65.72642517089844, 147.27386474609375, 50.45904541015625, 16.01995849609375, 22.95062255859375, 271.48223876953125, 23.94580078125, 6.75347900390625, 55.22381591796875, -24.479217529296875, 238.40997314453125, -5.8639373779296875, 103.032470703125, 106.34649658203125, 66.04153442382812, -27.294677734375, 24.4359130859375, 217.36227416992188, -0.0748138427734375, 15.436847686767578, -59.30462646484375, -86.775146484375, -32.61309814453125, 86.20223999023438, 16.487640380859375, -25.6201171875, -88.0650634765625, 210.60531616210938, -136.91561889648438, 32.825645446777344, 46.389259338378906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000406.npy"}
{"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 80.39791870117188, "std": 133.46949768066406, "min": -318.0177001953125, "p10": -57.40926818847656, "median": 47.857479095458984, "p90": 270.99880371093747, "max": 405.09417724609375, "pos_frac": 0.6953125, "sample": [9.133251190185547, 122.71853637695312, -4.67156982421875, 41.22819519042969, 256.01556396484375, 25.905960083007812, 170.48992919921875, 291.1917724609375, -79.6072998046875, 111.44839477539062, -22.158302307128906, 29.967864990234375, -10.212196350097656, -0.068939208984375, -157.63046264648438, 204.01573181152344, -20.532562255859375, -30.018569946289062, 42.88763427734375, -1.18060302734375, 81.30833435058594, 2.569122314453125, -5.391204833984375, 92.52764892578125, 237.976806640625, -14.056076049804688, -20.0872802734375, 46.11052703857422, -110.94966125488281, -26.276580810546875, 0.0959014892578125, 93.485107421875, -63.256011962890625, 12.873931884765625, 135.84359741210938, 191.73370361328125, 122.44900512695312, 257.022705078125, -40.30596160888672, 40.7003173828125, 121.88665771484375, -53.13299560546875, 174.22659301757812, -13.168617248535156, -61.401092529296875, 259.98175048828125, 210.15875244140625, 148.4657440185547, 217.15292358398438, 35.040924072265625, 18.047317504882812, 199.28179931640625, 4.48046875, 405.09417724609375, 69.50350952148438, 280.28448486328125, -34.802337646484375, 39.92840576171875, 1.419097900390625, 129.48138427734375, 17.980308532714844, 68.12855529785156, 194.9518585205078, -12.211233139038086, 164.0579833984375, 54.637542724609375, 55.34326171875, -150.28985595703125, 106.3004150390625, 43.211273193359375, 49.60443115234375, 13.0526123046875, 304.52606201171875, 146.88467407226562, 274.13299560546875, 6.833553314208984, 18.042449951171875, 79.69296264648438, 0.0, 188.12322998046875, 269.65557861328125, -136.3601837158203, 259.4995422363281, -55.698486328125, 240.32138061523438, -318.0177001953125, 355.2978515625, 21.546707153320312, -44.220703125, 90.4385986328125, 380.494384765625, -2.35028076171875, 15.1783447265625, 127.07989501953125, 164.9180908203125, 7.860626220703125, -36.70892333984375, 10.923065185546875, 59.9638671875, 101.72320556640625, 134.689697265625, -83.32048034667969, 195.46890258789062, -173.47369384765625, 180.6202392578125, -121.21157836914062, 134.07574462890625, 295.60772705078125, 311.5529479980469, -35.679290771484375, 282.86016845703125, 234.525146484375, 345.806884765625, -122.07014465332031, 337.1119079589844, -44.940101623535156, 59.97076416015625, -30.65032386779785, 83.54680633544922, 361.414794921875, 156.6357421875, 252.42791748046875, 21.46697998046875, -109.33613586425781, -15.7384033203125, 140.33319091796875, 199.46524047851562, 0.0], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000407.npy"}
{"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 79.80348205566406, "std": 127.1549072265625, "min": -198.5545654296875, "p10": -52.51606979370117, "median": 60.181983947753906, "p90": 238.47684631347656, "max": 463.2310791015625, "pos_frac": 0.7265625, "sample": [-52.17084503173828, 44.25262451171875, -124.13636779785156, 107.5604248046875, 64.07845306396484, 160.9349822998047, 1.34466552734375, 308.245849609375, 110.01156616210938, 192.10186767578125, -47.27607727050781, 58.76948547363281, 137.40481567382812, 122.74676513671875, -8.7073974609375, 132.79412841796875, 101.5087890625, 5.26361083984375, 18.81976318359375, -68.52388763427734, -74.01406860351562, -21.058307647705078, -28.7266845703125, -49.750526428222656, 463.2310791015625, 39.720176696777344, 386.40435791015625, 110.3670654296875, -49.819915771484375, -9.06402587890625, -10.45562744140625, 333.8590087890625, 214.42465209960938, 27.520172119140625, -147.05343627929688, 137.92710876464844, 315.78887939453125, 157.83123779296875, 201.30393981933594, 40.290008544921875, 42.19708251953125, -134.10589599609375, -174.23863220214844, 3.9976882934570312, 228.62661743164062, 120.26464080810547, -11.943954467773438, -20.278709411621094, -198.5545654296875, 261.5033264160156, -20.983123779296875, 66.103759765625, 39.310455322265625, -29.179306030273438, 148.608642578125, 140.82135009765625, 145.49456787109375, 215.21563720703125, -93.79232788085938, -3.27044677734375, 247.22389221191406, 235.3367462158203, 236.30322265625, 44.7020263671875, 7.159423828125, 191.39834594726562, -53.32159423828125, 95.23652648925781, 140.47705078125, 3.85986328125, 206.88714599609375, -7.3046722412109375, 178.17868041992188, 10.00836181640625, 46.7811279296875, 251.67593383789062, -56.602203369140625, 90.29556274414062, 27.40130615234375, 92.405517578125, 191.4180908203125, 146.7830810546875, -47.451332092285156, 45.563232421875, 60.67730712890625, 43.63877868652344, 259.9048767089844, 86.7672119140625, 156.22369384765625, -191.3765869140625, 97.66452026367188, 65.41777038574219, 262.18255615234375, 59.68666076660156, 390.54833984375, -40.16157531738281, -78.69549560546875, 148.43182373046875, 209.2978515625, 164.10089111328125, -189.9412078857422, 237.88595581054688, 35.245758056640625, 126.59033966064453, 187.9040069580078, -37.085235595703125, 9.951370239257812, 65.52566528320312, 145.81683349609375, -0.86846923828125, 21.006423950195312, 31.85198974609375, 71.04229736328125, 0.0, 9.361701965332031, 239.8555908203125, 165.40675354003906, 6.489019393920898, -40.20921325683594, -36.3779296875, 34.862831115722656, 40.71331787109375, 95.15435791015625, 179.53079223632812, 95.81439208984375, 205.07974243164062, 29.51511001586914, 436.453857421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000408.npy"}
{"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 81.6343994140625, "std": 144.6450958251953, "min": -407.9169921875, "p10": -78.51330261230468, "median": 70.48420333862305, "p90": 260.52679443359375, "max": 710.0324096679688, "pos_frac": 0.7578125, "sample": [179.279541015625, 67.71236419677734, 86.52992248535156, -116.6287841796875, 213.46673583984375, -407.9169921875, 155.07215881347656, 302.07666015625, -45.38592529296875, 167.90211486816406, 89.7408447265625, 153.59060668945312, 106.54945373535156, 34.4671630859375, -18.17523193359375, 283.91949462890625, 96.15202331542969, 63.79559326171875, 185.7610626220703, -96.8692626953125, -57.37773132324219, 69.3812484741211, -3.8893375396728516, 242.0531005859375, 135.36660766601562, -67.75802612304688, 71.587158203125, -176.91696166992188, 102.51408386230469, -291.77996826171875, 356.66497802734375, 148.29037475585938, 12.618988037109375, 42.383270263671875, 8.412841796875, 164.40704345703125, 67.08848571777344, 44.650054931640625, 267.3939208984375, 301.1362609863281, 118.18254089355469, 286.07330322265625, -204.75393676757812, 98.82980346679688, -9.122665405273438, 63.06719970703125, 37.94061279296875, 65.06033325195312, 59.188629150390625, 223.75433349609375, 121.62152099609375, 213.51168823242188, 286.20294189453125, 1.6391754150390625, 242.926513671875, -5.828948974609375, -188.55270385742188, 62.28106689453125, 16.455352783203125, -78.24113464355469, 247.64981079101562, 83.10809326171875, 124.5280990600586, 93.07159423828125, 292.91925048828125, -146.73114013671875, 27.826171875, -43.433746337890625, -7.6223602294921875, -188.02691650390625, 149.235595703125, -51.8721923828125, -31.125701904296875, 224.55999755859375, 6.240562438964844, 245.3065185546875, 710.0324096679688, 365.526123046875, 27.610618591308594, 40.425811767578125, 124.50674438476562, 20.482162475585938, 0.750885009765625, 0.0, 98.0533218383789, 359.9205322265625, 193.19290161132812, 257.583740234375, 189.57275390625, -149.404541015625, -37.815467834472656, 207.52182006835938, 66.0867919921875, 15.484474182128906, 53.085784912109375, -1.5273246765136719, 65.56924438476562, 290.4168701171875, 128.0324249267578, 118.216064453125, 17.68609619140625, 161.3089599609375, 99.36843872070312, 225.36380004882812, 86.43588256835938, -14.49664306640625, 0.8473663330078125, -114.78176879882812, 64.65643310546875, 116.6858901977539, 166.54800415039062, -9.36212158203125, -79.14836120605469, 3.8434982299804688, 116.565185546875, 59.18389892578125, 76.95438385009766, 171.58221435546875, 97.2286605834961, -21.456100463867188, 270.95965576171875, 5.45465087890625, 2.448944091796875, 151.34864807128906, -146.38287353515625, 101.06736755371094, 129.0872802734375, 193.74624633789062], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000409.npy"}
{"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 75.84941864013672, "std": 131.65402221679688, "min": -241.31283569335938, "p10": -85.67915649414061, "median": 67.9599380493164, "p90": 234.34454345703125, "max": 412.1626892089844, "pos_frac": 0.6796875, "sample": [144.9893798828125, 38.22624969482422, 22.048614501953125, -70.58355712890625, 39.036781311035156, 163.8070831298828, -20.187015533447266, 229.53829956054688, -43.8267822265625, 192.69618225097656, 227.49984741210938, 24.02996826171875, 164.78016662597656, -90.98516845703125, 188.60598754882812, -241.31283569335938, -6.281494140625, 162.94395446777344, 307.3982238769531, 57.825103759765625, 71.54644775390625, -107.3206787109375, 185.28945922851562, 9.571144104003906, 210.36643981933594, -61.79034423828125, -4.2437286376953125, 395.0026550292969, 99.29595947265625, 204.1304168701172, 51.5322265625, -23.630271911621094, -1.17510986328125, 128.60299682617188, -4.0152587890625, 210.50350952148438, 19.403308868408203, 70.49868774414062, 96.60272216796875, 271.911865234375, 163.0560302734375, 84.5986328125, 21.606170654296875, 296.97906494140625, -145.176513671875, 188.98028564453125, -129.39340209960938, 47.806793212890625, 20.3338623046875, 11.326492309570312, 347.0555419921875, 412.1626892089844, 166.94744873046875, -130.1290283203125, 31.905929565429688, 69.22882080078125, -65.16927337646484, 214.96942138671875, 68.35630798339844, 67.56356811523438, -183.20376586914062, 82.240234375, 286.88873291015625, 82.59246826171875, -2.8258495330810547, 0.101531982421875, -66.26895141601562, -15.93231201171875, -7.434196472167969, 29.838546752929688, 100.54177856445312, 158.62612915039062, -1.405364990234375, -32.1744384765625, -118.75648498535156, 0.0, 236.8673095703125, 38.95001220703125, 234.64276123046875, 150.95428466796875, 161.75115966796875, -4.4415283203125, 3.294769287109375, -113.44674682617188, -56.22906494140625, 143.85171508789062, 101.32585906982422, 16.01287841796875, -147.6712646484375, 84.635009765625, -199.3865509033203, 5.729179382324219, 23.175933837890625, 175.89956665039062, -18.81695556640625, 201.31625366210938, -152.5453643798828, 196.07025146484375, 121.47476959228516, -1.8217315673828125, 168.1002197265625, 159.01473999023438, 72.23075866699219, -29.77410888671875, 91.08843994140625, 134.81361389160156, 370.60845947265625, 146.35305786132812, 370.1639709472656, 4.135162353515625, 188.89120483398438, -12.288330078125, 185.1457977294922, 100.72745513916016, -45.769012451171875, 173.06475830078125, 170.41864013671875, -157.63864135742188, 234.21673583984375, 89.365234375, -0.050395965576171875, 240.83328247070312, 203.502685546875, -83.4051513671875, 40.798065185546875, -4.02996826171875, 332.77178955078125, -4.323795318603516], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000410.npy"}
{"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 88.84025573730469, "std": 122.2098388671875, "min": -206.00830078125, "p10": -44.030043029785155, "median": 63.17475891113281, "p90": 236.10244140625, "max": 664.3037109375, "pos_frac": 0.7890625, "sample": [-7.799932479858398, 192.40213012695312, 189.089599609375, 112.860107421875, 179.87054443359375, 34.1124267578125, 3.1492462158203125, 63.101593017578125, 52.61175537109375, 173.08493041992188, 270.5093994140625, 100.08285522460938, 25.084529876708984, -5.8399658203125, -65.3165283203125, -35.773040771484375, 297.3231201171875, 19.267623901367188, 63.2479248046875, 12.199127197265625, 55.296234130859375, 234.26593017578125, 108.10745239257812, 156.56149291992188, 35.614654541015625, 141.23825073242188, -11.010833740234375, 664.3037109375, 82.238525390625, -15.905792236328125, 244.45620727539062, 5.623666763305664, -44.18260192871094, 159.1693878173828, 154.52621459960938, 145.21343994140625, 246.14556884765625, 196.9760284423828, 53.77001953125, 13.482406616210938, 201.8282928466797, 75.10708618164062, -79.81321716308594, 159.05874633789062, 27.14373779296875, -79.63491821289062, 232.66680908203125, 80.17945861816406, -46.94024658203125, 152.44515991210938, 243.07998657226562, 88.28244018554688, -37.490753173828125, 5.085662841796875, 142.33395385742188, 208.021728515625, -60.467926025390625, 90.5609130859375, 51.376220703125, -25.852081298828125, -9.508270263671875, 120.7125244140625, 32.9652099609375, -206.00830078125, 137.88375854492188, 10.623199462890625, 168.98391723632812, 262.2693176269531, 156.30880737304688, 167.72622680664062, 8.618095397949219, 173.17169189453125, 90.85174560546875, 240.38763427734375, 18.698020935058594, 201.99847412109375, 41.5426025390625, -25.222564697265625, 319.5718994140625, 64.29544067382812, 42.54725646972656, -167.92095947265625, 176.18960571289062, 6.550374984741211, 8.44482421875, -8.694595336914062, -9.947296142578125, 19.6231689453125, 119.22364044189453, 265.2298278808594, 144.77442932128906, -68.15007019042969, 22.5047607421875, 213.302734375, 158.35812377929688, 73.93805694580078, -15.00750732421875, 204.25326538085938, 58.847198486328125, -91.29728698730469, 206.37835693359375, 17.833816528320312, -110.49887084960938, -37.644805908203125, 32.80968475341797, 49.7867431640625, 42.561798095703125, 57.7174072265625, 130.2158966064453, 171.6010284423828, 48.22834014892578, 50.019775390625, 223.01962280273438, 3.5546035766601562, 54.67997741699219, -137.50872802734375, 4.46661376953125, 276.210693359375, 144.20852661132812, 179.80230712890625, 324.43170166015625, 439.53204345703125, 87.5220947265625, 216.68344116210938, 121.51861572265625, -63.758544921875, -43.96466064453125, 27.37835693359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000411.npy"}
{"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 69.16047668457031, "std": 129.75437927246094, "min": -185.15696716308594, "p10": -92.66393432617187, "median": 64.81553649902344, "p90": 225.37869720458983, "max": 469.24810791015625, "pos_frac": 0.6640625, "sample": [-70.12530517578125, 330.39654541015625, -2.904094696044922, 152.01031494140625, 162.285888671875, 186.16259765625, 141.43515014648438, 139.31768798828125, -9.375152587890625, -110.79425048828125, -114.98452758789062, 70.63296508789062, 1.415191650390625, -35.20653533935547, 65.52566528320312, -184.850341796875, -39.5521240234375, -137.52459716796875, -81.62481689453125, 86.24504089355469, 35.331268310546875, 12.4111328125, 63.4442138671875, 101.32525634765625, 228.4230499267578, -4.3650970458984375, 195.00698852539062, 109.73654174804688, -81.7095947265625, 99.77459716796875, 167.48709106445312, 155.0965576171875, 72.29562377929688, 127.65057373046875, 129.01690673828125, 164.76771545410156, 213.04933166503906, 74.58979797363281, 69.8125, 99.52413940429688, 209.2778778076172, -72.36618041992188, 169.67105102539062, -106.5384521484375, -29.325584411621094, -34.78656005859375, 9.114151000976562, -3.77288818359375, -133.8472900390625, -3.741455078125, 222.98223876953125, 301.6708679199219, 131.81903076171875, -98.646484375, 184.93223571777344, -49.564605712890625, -8.620361328125, 51.90785217285156, 89.6034927368164, -97.71539306640625, -111.59365844726562, 2.954193115234375, 271.53839111328125, 64.033447265625, -13.51593017578125, 353.1883544921875, 104.09786987304688, 469.24810791015625, 291.0655212402344, 130.28372192382812, -90.4990234375, 169.39016723632812, -185.15696716308594, 10.103546142578125, 67.61851501464844, 38.192108154296875, 211.4007110595703, 102.08969116210938, -35.7125244140625, -131.71807861328125, 59.255950927734375, 131.34405517578125, 72.13668823242188, 3.0959320068359375, 144.11346435546875, -71.40835571289062, 303.5738830566406, 66.11682891845703, 436.39227294921875, -66.19013977050781, -8.058425903320312, 207.1046905517578, 224.073974609375, -69.04644775390625, -66.19782257080078, 85.064697265625, -99.93157958984375, 52.372589111328125, 64.10540771484375, 150.7136993408203, 74.03912353515625, -0.7431621551513672, 459.7493896484375, 81.82292175292969, 85.35931396484375, 44.63298797607422, 15.366134643554688, 156.5341796875, -112.10447692871094, -55.7783203125, 232.52044677734375, 45.578094482421875, 0.0, 270.0594482421875, -34.552734375, 121.2864990234375, -20.776123046875, 44.29035949707031, 62.07928466796875, 2.522125244140625, 3.5087051391601562, 177.6260986328125, -63.90550231933594, 266.262939453125, 104.37541198730469, -32.85914611816406, 124.42868041992188, 152.37155151367188], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000412.npy"}
{"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 65.62258911132812, "std": 119.61023712158203, "min": -202.77825927734375, "p10": -87.34760131835938, "median": 50.66144561767578, "p90": 225.32243041992186, "max": 420.81231689453125, "pos_frac": 0.71875, "sample": [61.25676727294922, 37.621849060058594, 32.56953430175781, 41.170143127441406, -8.608428955078125, 101.69160461425781, 23.890167236328125, -42.4454345703125, -178.52603149414062, 9.615020751953125, 244.46182250976562, 135.14141845703125, 239.58773803710938, -45.5447998046875, -36.69813537597656, 162.03305053710938, 99.72052001953125, 203.74566650390625, 145.35000610351562, -164.72142028808594, -12.00433349609375, 175.4580841064453, 139.03976440429688, 15.3143310546875, -24.457366943359375, 4.603851318359375, 42.43865966796875, 246.2318115234375, 108.08309936523438, -55.51458740234375, 94.50759887695312, 32.75958251953125, -82.20536804199219, 80.26473999023438, 54.131134033203125, 202.6314697265625, 361.2900390625, 144.51107788085938, 230.11111450195312, 31.937198638916016, 174.33917236328125, -10.156463623046875, -110.03939819335938, 160.85519409179688, 117.474609375, -40.170440673828125, 65.967529296875, 100.1112060546875, 110.3131103515625, 83.52481079101562, 16.814910888671875, 3.32476806640625, -36.50520324707031, 16.725860595703125, -17.92694091796875, -90.21255493164062, 277.306396484375, -197.9829559326172, 108.83087158203125, 41.344322204589844, 259.2437438964844, 200.2587127685547, -18.512847900390625, 138.71182250976562, 89.80075073242188, 285.7799072265625, 37.57863998413086, 182.0758056640625, 86.66021728515625, -202.77825927734375, 132.74383544921875, 41.97590255737305, -127.35042572021484, 409.8887634277344, 97.9767074584961, -95.18211364746094, 12.083038330078125, 117.23362731933594, -1.52838134765625, 10.386444091796875, 18.637100219726562, 38.29254150390625, -80.689697265625, 223.60739135742188, 179.24951171875, -87.57070922851562, 72.54052734375, 117.55868530273438, -6.964696884155273, -42.210174560546875, 229.32418823242188, -6.634332656860352, -87.25198364257812, 81.8017578125, 202.4879150390625, -84.84688568115234, 8.083290100097656, -65.98150634765625, 83.37237548828125, 124.48321533203125, 154.10226440429688, -45.13603591918945, 46.74822998046875, 174.5108642578125, 156.9449462890625, 112.2686767578125, 12.0977783203125, 25.488311767578125, 111.034912109375, 72.16458129882812, 4.339019775390625, 94.91552734375, 11.189079284667969, 80.56261444091797, 14.637603759765625, -96.07086181640625, 248.29742431640625, 186.07400512695312, -103.61428833007812, 420.81231689453125, 135.52976989746094, 270.8221740722656, 47.19175720214844, -115.48504638671875, 216.74920654296875, 114.87252807617188, -136.77362060546875, -13.29913330078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000413.npy"}
{"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 86.26232147216797, "std": 124.36508178710938, "min": -259.84454345703125, "p10": -49.20083770751953, "median": 73.20423126220703, "p90": 238.64936523437495, "max": 579.6932373046875, "pos_frac": 0.8046875, "sample": [27.82495880126953, -175.86260986328125, 32.467628479003906, 120.57687377929688, 11.777442932128906, 126.726806640625, 96.18597412109375, 15.80194091796875, -50.209320068359375, -27.32720947265625, -25.71923828125, 49.027435302734375, 337.37384033203125, 6.2239990234375, 33.7314453125, -15.332225799560547, 80.4918212890625, -69.75868225097656, 119.6171875, -12.648067474365234, -123.196533203125, -86.8809814453125, 197.88726806640625, 11.0692138671875, 140.42193603515625, -84.26557159423828, -12.7015380859375, 85.28692626953125, 198.72348022460938, 174.16236877441406, 14.616764068603516, 67.22262573242188, 41.95814514160156, 166.89907836914062, -192.51705932617188, 153.78640747070312, 176.31443786621094, -67.70748901367188, 269.86883544921875, 117.06831359863281, 1.649261474609375, 205.01071166992188, 49.709716796875, 75.674560546875, 69.01686096191406, 371.8642578125, 186.2532958984375, -7.908023834228516, 83.00897216796875, 330.1585693359375, -28.6300048828125, 4.66778564453125, 11.073160171508789, 361.799072265625, 92.81974792480469, 12.93505859375, 261.90777587890625, 224.23281860351562, 177.38677978515625, 88.26971435546875, 39.445404052734375, 172.6468048095703, 16.838027954101562, 202.25436401367188, 40.79449462890625, 32.62504577636719, 212.0978240966797, 73.06364440917969, 80.2933349609375, 44.22425842285156, 188.26675415039062, -25.825050354003906, 73.34481811523438, 234.5133056640625, 139.31503295898438, 193.56243896484375, 162.355712890625, 0.0204315185546875, 184.9239501953125, 72.72734832763672, -20.47064208984375, 84.9169921875, -259.84454345703125, 579.6932373046875, 359.25048828125, 134.8966827392578, 147.1165771484375, 20.298011779785156, 0.7771453857421875, 89.01937866210938, -131.34701538085938, 86.65631103515625, 106.30538940429688, 165.00607299804688, 12.861637115478516, 189.44024658203125, -31.041061401367188, 96.361083984375, -60.40373229980469, 74.52760314941406, 21.569046020507812, 96.77301025390625, 248.3001708984375, 37.222084045410156, 2.7532482147216797, 69.28140258789062, 64.17427062988281, 37.07237243652344, 299.3515625, 224.19290161132812, 95.09552001953125, 72.54846954345703, 296.3070068359375, 83.313720703125, 10.020748138427734, 61.91131591796875, -48.76863098144531, 328.8074035644531, 20.27398681640625, 138.24932861328125, -50.9322509765625, -9.4237060546875, -78.8492431640625, 65.70977783203125, 131.18667602539062, 323.2030029296875, 106.96624755859375, 113.87307739257812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000414.npy"}
{"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 78.24754333496094, "std": 112.82600402832031, "min": -234.7509765625, "p10": -60.180676651000965, "median": 59.30937957763672, "p90": 221.95872802734374, "max": 394.71502685546875, "pos_frac": 0.7578125, "sample": [40.184173583984375, 250.20277404785156, 182.46981811523438, -1.8071174621582031, -87.15303802490234, -19.746200561523438, 106.12271881103516, 14.697372436523438, 5.7457122802734375, 4.010368347167969, -15.919906616210938, -57.46187973022461, 93.28401184082031, 174.18478393554688, 26.407859802246094, -33.182044982910156, 394.71502685546875, 26.092864990234375, 112.70703125, 41.989952087402344, 156.35264587402344, 21.741455078125, 197.96176147460938, 42.47697448730469, 77.417724609375, 67.67276000976562, 29.358428955078125, 142.5054931640625, 316.08624267578125, 117.71957397460938, -16.557220458984375, 217.55780029296875, 246.04562377929688, 44.45111083984375, -172.46856689453125, 62.536041259765625, 67.31597900390625, 353.97930908203125, 53.446136474609375, 36.68109130859375, -16.533241271972656, 139.49240112304688, 15.913345336914062, 40.443939208984375, 24.881004333496094, -86.59809112548828, 210.06292724609375, 159.967529296875, 196.44854736328125, -36.97422790527344, -66.5245361328125, 141.1441650390625, -94.2247314453125, -52.545654296875, 106.05142211914062, 137.20623779296875, 8.22991943359375, 217.77093505859375, 135.894287109375, -19.098419189453125, 50.235107421875, 45.09910583496094, -27.806671142578125, 55.58642578125, 221.13848876953125, -27.58197021484375, 167.09121704101562, 199.82803344726562, -3.167205810546875, -1.0569400787353516, -89.19140625, 133.0718994140625, 230.57723999023438, 161.79278564453125, -18.889434814453125, 10.471282958984375, 102.37637329101562, 183.77947998046875, 14.15927505493164, -79.10186767578125, 118.27679443359375, 30.422607421875, 49.208282470703125, -57.04974365234375, 202.59033203125, -72.00350952148438, -138.412109375, 90.418701171875, 104.92678833007812, 106.830322265625, 62.134979248046875, 40.26043701171875, 156.97067260742188, 243.5689697265625, 140.95428466796875, -18.369842529296875, 117.31805419921875, 73.6585693359375, 251.78610229492188, -234.7509765625, 180.24539184570312, 83.66857147216797, 172.58441162109375, -28.285980224609375, 45.604339599609375, 313.7999572753906, -107.72889709472656, 17.83819580078125, 314.56915283203125, 149.11643981933594, 254.98199462890625, 53.94903564453125, 182.73785400390625, 35.12017059326172, -95.25448608398438, 133.549560546875, 56.48377990722656, 175.8784942626953, 223.87261962890625, 32.0328369140625, -115.51300048828125, 127.33746337890625, 194.4677734375, 41.93470764160156, 138.541259765625, 116.6026611328125, 8.959762573242188, 230.6077880859375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000415.npy"}
{"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 93.40155029296875, "std": 124.9984359741211, "min": -261.9876403808594, "p10": -51.0067268371582, "median": 82.90023803710938, "p90": 252.0712936401367, "max": 370.93017578125, "pos_frac": 0.75, "sample": [272.82098388671875, 256.5091552734375, 316.6174621582031, 44.426025390625, 20.128761291503906, 35.59001159667969, -49.107330322265625, 110.4937973022461, -43.65473937988281, 157.6927032470703, -29.89410400390625, 196.52740478515625, 41.612091064453125, 45.61912536621094, -39.19189453125, 83.86595916748047, 178.468505859375, 256.1039733886719, 334.9832763671875, 65.81265258789062, 164.7154541015625, 100.72486877441406, 7.1388397216796875, 231.28074645996094, -78.58465576171875, 370.93017578125, -26.320663452148438, 129.66961669921875, -43.008888244628906, 18.3740234375, 238.85968017578125, 235.185791015625, 250.34300231933594, 8.568634033203125, 58.42441940307617, 184.52655029296875, -261.9876403808594, 218.45074462890625, 239.8970947265625, -39.23492431640625, 130.332275390625, 27.826419830322266, -61.920928955078125, -260.19921875, -159.81394958496094, 302.2974548339844, -20.352035522460938, 235.1515655517578, 169.5235595703125, 197.609130859375, 25.93170166015625, 290.03814697265625, 26.99578857421875, 23.556602478027344, -27.643386840820312, 243.63775634765625, 227.05067443847656, 134.8392333984375, -78.313232421875, 61.552947998046875, 65.46478271484375, 114.35626220703125, 3.5233154296875, 82.66717529296875, 241.23037719726562, 102.57077026367188, 130.79383850097656, 156.12628173828125, 21.349029541015625, 181.71261596679688, -0.644378662109375, -15.540618896484375, 81.68829345703125, 296.33245849609375, 320.2819519042969, 37.48317337036133, 32.23406982421875, 96.03753662109375, 121.63674926757812, -55.43865203857422, -33.232666015625, 16.7972412109375, 187.4696044921875, -57.7398796081543, -7.712371826171875, 74.979736328125, -7.379730224609375, 83.13330078125, -68.13284301757812, 16.154754638671875, 202.69171142578125, 246.30606079101562, 49.875946044921875, 171.5767822265625, -108.25238037109375, 172.16854858398438, 298.67718505859375, -0.7014923095703125, 150.7780303955078, 224.3131103515625, -99.86163330078125, 117.53507995605469, 244.92864990234375, 40.115325927734375, 126.1241455078125, -62.95147705078125, 183.88641357421875, 22.72503662109375, 270.660888671875, 94.16232299804688, -19.07471466064453, 241.49566650390625, 69.70005798339844, -129.40478515625, -4.38006591796875, 127.37486267089844, 110.05596923828125, 157.4600830078125, 169.2879180908203, -48.85553741455078, 45.20428466796875, -0.31890869140625, 211.1129150390625, 52.891273498535156, 249.18115234375, 62.26225280761719, 83.6895980834961, 263.2783203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000416.npy"}
{"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 75.71672058105469, "std": 134.1065673828125, "min": -242.47705078125, "p10": -86.81962585449219, "median": 71.80548858642578, "p90": 239.29038085937498, "max": 603.1317749023438, "pos_frac": 0.7109375, "sample": [136.80654907226562, 242.80093383789062, 157.3089599609375, 101.03761291503906, -41.24140930175781, 350.7026062011719, 35.65203857421875, -149.177001953125, 284.2928466796875, 217.18017578125, 16.943572998046875, 132.9774169921875, 134.50521850585938, 108.29684448242188, -211.414794921875, -18.870849609375, 17.089889526367188, -139.4229736328125, -129.0936279296875, 75.3375244140625, 302.5634765625, -7.802703857421875, 359.29412841796875, -86.45721435546875, -11.833475112915039, -156.6637420654297, 93.667724609375, 6.830524444580078, -87.66525268554688, 357.6642150878906, 231.666015625, -24.94940185546875, 77.77496337890625, 121.95355987548828, -13.642364501953125, 347.8488464355469, 66.025634765625, -86.34103393554688, 196.69500732421875, 138.599609375, 109.25747680664062, 80.40374755859375, -6.2442626953125, 0.646759033203125, 136.36782836914062, 183.83554077148438, 38.935089111328125, 85.5789794921875, 19.028579711914062, 164.66229248046875, 168.7686767578125, -5.563177108764648, -4.75701904296875, -85.81971740722656, -119.48822021484375, 192.24598693847656, -41.0045166015625, 119.3585205078125, 65.6009521484375, -37.1177978515625, 74.42327880859375, 204.5834197998047, -11.074651718139648, 12.900932312011719, 117.77725219726562, -91.11166381835938, 222.29592895507812, 67.38723754882812, -233.47421264648438, 8.100929260253906, -78.73934936523438, 129.53817749023438, 194.29937744140625, 85.49429321289062, 59.66473388671875, 11.5546875, 21.929244995117188, 172.75677490234375, 0.0, 13.40087890625, 145.97828674316406, 0.0, 80.6817626953125, 95.49519348144531, 133.3860626220703, 58.3956298828125, 109.52372741699219, 47.36119079589844, -4.34234619140625, 15.823455810546875, -79.59707641601562, 233.42684936523438, 140.44064331054688, 159.80581665039062, 603.1317749023438, 123.21624755859375, -117.197509765625, 108.23988342285156, 373.34368896484375, 69.18769836425781, 77.39849853515625, -73.25906372070312, -242.47705078125, -0.28437042236328125, 152.81141662597656, 99.73150634765625, 274.82623291015625, 47.910980224609375, -118.742431640625, 12.469696044921875, 40.057373046875, -17.943939208984375, 19.030311584472656, 261.0496826171875, 98.71920776367188, 185.1197509765625, -11.950592041015625, 120.48777770996094, 262.04205322265625, 9.421100616455078, 288.37969970703125, 114.61859130859375, 171.61891174316406, -91.69599914550781, 189.1495819091797, 62.279022216796875, 237.78585815429688, 33.54217529296875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000417.npy"}
{"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 76.35926055908203, "std": 131.44146728515625, "min": -228.9466552734375, "p10": -82.48203659057617, "median": 64.68030548095703, "p90": 233.85665588378905, "max": 440.74713134765625, "pos_frac": 0.7265625, "sample": [-197.8585968017578, 194.64840698242188, 61.775238037109375, 194.0286865234375, 64.58251953125, -54.74482727050781, 356.3953857421875, 230.590576171875, -191.24151611328125, -216.74673461914062, 87.35746765136719, 113.59211730957031, 65.19448852539062, -36.51768493652344, 53.0321044921875, 70.62879943847656, 227.37994384765625, -136.82713317871094, -68.65786743164062, 29.602874755859375, 389.56134033203125, 113.8492431640625, -16.0640869140625, 23.40373420715332, -45.6444091796875, -16.487701416015625, -67.37362670898438, 261.3687744140625, 41.36431884765625, 388.87872314453125, 196.20663452148438, 94.36065673828125, 311.10650634765625, 178.13369750976562, 79.73765563964844, 389.6506652832031, -6.8986053466796875, 149.9117431640625, -98.00982666015625, 39.991783142089844, -35.034812927246094, 124.45587158203125, -26.670867919921875, 215.45925903320312, 33.65800476074219, 64.7684326171875, -133.29920959472656, 61.938926696777344, 139.85174560546875, 15.413818359375, -6.8990478515625, 440.74713134765625, 91.62835693359375, 191.96241760253906, 86.10298156738281, -7.5638427734375, 30.9622802734375, 75.43536376953125, 220.61715698242188, 41.799407958984375, -72.62451171875, 97.48117065429688, 155.82907104492188, 153.02981567382812, 254.7891845703125, 8.572662353515625, 191.51808166503906, -81.38153839111328, 23.836334228515625, 86.94613647460938, -15.6571044921875, 46.11885070800781, 54.353172302246094, 43.59380340576172, 61.7738037109375, 43.495033264160156, 234.83660888671875, -103.98440551757812, 158.92816162109375, -97.33645629882812, 144.73709106445312, 171.50674438476562, -85.04986572265625, 108.92813110351562, 196.40371704101562, 259.28228759765625, 135.68914794921875, -22.844348907470703, 111.3734130859375, -34.9581298828125, 78.00961303710938, -137.70315551757812, -20.8369140625, 155.9710693359375, 179.0518798828125, 83.22203063964844, 233.43667602539062, 91.82215118408203, -24.160247802734375, 200.91107177734375, -101.65155029296875, 5.973564147949219, 158.37037658691406, 112.65385437011719, -169.89593505859375, 71.5240478515625, 62.42242431640625, 199.42294311523438, -228.9466552734375, 113.99261474609375, 86.73468780517578, 74.98583984375, 50.434112548828125, 17.99904441833496, 64.59217834472656, 91.49779510498047, 356.65228271484375, 63.94757080078125, 17.775917053222656, 293.4947204589844, 46.647125244140625, -21.304901123046875, -40.71160888671875, 47.4171142578125, 27.20396614074707, 175.01629638671875, 328.2468566894531, -78.01425170898438], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000418.npy"}
{"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 81.0560302734375, "std": 134.75833129882812, "min": -425.7598876953125, "p10": -92.70715026855468, "median": 83.76134490966797, "p90": 234.78904724121094, "max": 500.3858642578125, "pos_frac": 0.734375, "sample": [185.03707885742188, -63.670440673828125, -34.862457275390625, 239.63739013671875, 89.43309020996094, 151.80331420898438, 24.850303649902344, 106.84423828125, -16.622024536132812, -126.09408569335938, -17.971744537353516, 57.25604248046875, -117.58041381835938, 191.85888671875, 26.184959411621094, 84.11053466796875, 207.52142333984375, 59.10266876220703, 85.25057983398438, 189.7445068359375, 202.63726806640625, 112.565185546875, 161.34564208984375, 108.12825012207031, 285.75537109375, 79.15631103515625, 90.69219970703125, 15.112060546875, -82.827392578125, -93.85586547851562, 264.46319580078125, 140.57278442382812, -8.198131561279297, -22.933685302734375, 21.20281982421875, -38.5582275390625, 263.0657958984375, -2.5919952392578125, 104.96746826171875, -241.76608276367188, 193.8946533203125, -71.54481506347656, 87.181396484375, 83.136474609375, 305.6348876953125, 32.21745300292969, 266.8699035644531, -152.1214599609375, 64.6168212890625, 130.59811401367188, 127.4979248046875, 319.7108154296875, -14.262214660644531, 174.85006713867188, 149.24090576171875, 138.83554077148438, 500.3858642578125, 235.69940185546875, 107.34881591796875, 124.64419555664062, -100.371826171875, 70.79672241210938, 49.817970275878906, -68.36735534667969, 211.4578857421875, -99.31217956542969, 268.2613830566406, 313.7945251464844, 19.337265014648438, 83.41215515136719, -120.28141784667969, 112.17300415039062, 225.17831420898438, -11.7344970703125, 82.23954772949219, 0.0, 14.139923095703125, -195.08465576171875, 7.010589599609375, 0.0, 18.286041259765625, 144.865478515625, 173.35604858398438, 173.16065979003906, -12.569915771484375, 19.56182861328125, -44.96101379394531, 230.57073974609375, 134.76458740234375, -28.175003051757812, 211.71298217773438, 53.539794921875, 180.80621337890625, 40.986541748046875, 8.775985717773438, 225.8935546875, 186.11227416992188, 123.2623291015625, 88.7420883178711, 334.99859619140625, 234.39889526367188, 38.69970703125, 356.2991943359375, -61.51799011230469, 171.73583984375, -92.21484375, 96.18905639648438, 217.4641876220703, 80.457763671875, 230.8057403564453, -118.9193115234375, -128.42095947265625, 159.4366455078125, 231.19796752929688, -425.7598876953125, 67.52278137207031, -177.54397583007812, 107.29149627685547, 9.027557373046875, 52.4910888671875, 128.23765563964844, 219.50436401367188, 19.38238525390625, 35.771514892578125, 145.04409790039062, -10.723350524902344, 157.25930786132812, 20.69635009765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000419.npy"}
{"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 93.06560516357422, "std": 150.3154296875, "min": -343.28924560546875, "p10": -73.51566467285154, "median": 77.36842346191406, "p90": 274.39507446289065, "max": 498.6522521972656, "pos_frac": 0.734375, "sample": [-85.46174621582031, 11.148101806640625, 117.91818237304688, -243.56573486328125, -68.71151733398438, 133.99839782714844, 77.4996337890625, 249.43353271484375, -55.69708251953125, 156.5836181640625, 122.65172576904297, -5.8245849609375, -116.31948852539062, 24.051544189453125, 41.90717315673828, 22.52252197265625, 192.0946502685547, -27.313674926757812, 136.4896240234375, -67.49497985839844, 283.5259094238281, 68.64552307128906, -25.58075714111328, 66.1055908203125, -85.28138732910156, 498.6522521972656, 113.73873901367188, 4.472541809082031, 89.23744201660156, 105.51300048828125, -93.80242919921875, 269.7048645019531, 227.83860778808594, 193.46109008789062, 58.725341796875, 90.68437194824219, 234.38690185546875, 467.47943115234375, 225.891845703125, 47.42437744140625, 141.57073974609375, 41.78263854980469, -44.59062194824219, 427.523681640625, 99.52874755859375, 161.73574829101562, 269.2586669921875, -10.752761840820312, -2.3119659423828125, 79.85064697265625, -65.68223571777344, 169.24884033203125, 127.37605285644531, 274.09161376953125, 115.63848876953125, -5.2347869873046875, 0.0, 34.67840576171875, -293.6748046875, -57.421531677246094, 199.1339569091797, 194.02890014648438, -2.013885498046875, -17.66863250732422, 21.566818237304688, 77.23721313476562, 28.11822509765625, 240.7701416015625, 355.9283447265625, 185.0469970703125, 87.8990478515625, 50.60838317871094, 13.113037109375, 239.66270446777344, -107.12408447265625, 368.0374450683594, 33.37120819091797, 63.85795593261719, 331.80157470703125, 19.233734130859375, -12.327163696289062, 19.212764739990234, 275.1031494140625, -96.84390258789062, 347.19189453125, 151.01638793945312, -36.478302001953125, 17.258014678955078, 251.03219604492188, 190.5364990234375, -36.64581298828125, 13.551116943359375, 492.6280517578125, 174.79681396484375, 55.42628479003906, 57.027587890625, 31.91632080078125, 135.55007934570312, 250.02001953125, -12.5172119140625, 284.12396240234375, 193.02731323242188, -318.866943359375, 77.15218353271484, 113.14568328857422, 116.08584594726562, 182.99813842773438, -3.39520263671875, -343.28924560546875, -64.14917755126953, 184.79681396484375, 205.9565887451172, 125.36111450195312, 221.34571838378906, 192.79612731933594, 31.68084716796875, 160.373046875, -136.20501708984375, 282.56219482421875, 179.2999725341797, 219.22564697265625, 276.4713134765625, -84.725341796875, 67.34139251708984, 25.12298583984375, 229.80615234375, -104.80516052246094, 29.748565673828125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000420.npy"}
{"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 82.51280212402344, "std": 117.81835174560547, "min": -350.27374267578125, "p10": -56.54836540222166, "median": 82.2603530883789, "p90": 234.7085968017578, "max": 395.6175231933594, "pos_frac": 0.75, "sample": [245.46148681640625, 21.23187255859375, 157.51626586914062, 6.609893798828125, 17.242401123046875, 32.88191223144531, 49.56915283203125, 26.362274169921875, -105.035888671875, -84.0982894897461, 212.67819213867188, -3.547882080078125, -100.35111999511719, 88.244384765625, 25.459396362304688, 56.59888458251953, 59.50384521484375, -34.753326416015625, 129.0233154296875, 108.92820739746094, 26.665573120117188, -100.11921691894531, 92.40130615234375, 114.88941192626953, 93.93221282958984, 395.6175231933594, 199.833740234375, 75.63037109375, 165.125, 115.60576629638672, 0.0, -350.27374267578125, 105.36872863769531, 101.3022232055664, 31.532974243164062, 23.271224975585938, 186.1734619140625, 156.64752197265625, 120.09553527832031, 148.01043701171875, 195.03683471679688, 224.55905151367188, 69.00531005859375, 11.126216888427734, -11.184326171875, 43.60547637939453, 115.24375915527344, -39.01458740234375, 260.469970703125, 140.64321899414062, 177.77960205078125, -155.85305786132812, 102.70733642578125, 131.26364135742188, 25.669158935546875, 354.748046875, -7.480842590332031, 233.94119262695312, -6.86492919921875, 181.16738891601562, 54.217041015625, 9.841522216796875, 270.435302734375, 156.44882202148438, 191.03460693359375, -98.75553131103516, 137.77976989746094, 132.54833984375, 219.03408813476562, 289.57061767578125, 227.1612548828125, 344.299072265625, 102.46401977539062, 148.10354614257812, 15.965776443481445, 157.78616333007812, 84.43904113769531, 62.765380859375, 91.66676330566406, 98.0241470336914, 210.1707000732422, -1.2068862915039062, 102.6917724609375, 259.78094482421875, -5.703857421875, 57.833038330078125, 236.49920654296875, -68.544189453125, 120.88934326171875, -65.4704360961914, 80.0816650390625, 54.125701904296875, 260.3438720703125, 168.85342407226562, -0.11559486389160156, 47.82818603515625, 151.42474365234375, -52.7246208190918, -93.40689086914062, 120.64396667480469, -92.7940673828125, -41.050140380859375, 57.711334228515625, 150.51316833496094, 107.3333740234375, 61.00286865234375, 244.37506103515625, 18.29547119140625, 123.66333770751953, -219.0574951171875, -28.37615966796875, 62.56207275390625, 134.75857543945312, 38.97196960449219, -141.5293731689453, 222.108154296875, 143.91758728027344, 0.0, -24.179931640625, -19.277145385742188, 267.3246154785156, 217.25146484375, 79.45083618164062, 0.0, -18.970504760742188, 54.56219482421875, 255.80239868164062, -29.35772705078125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000421.npy"}
{"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 80.62918853759766, "std": 134.11708068847656, "min": -283.1241760253906, "p10": -77.66038665771484, "median": 53.003822326660156, "p90": 256.5387756347656, "max": 430.5399169921875, "pos_frac": 0.6796875, "sample": [18.1405029296875, -283.1241760253906, 112.53244018554688, -145.09132385253906, 118.93084716796875, -82.49057006835938, -6.81390380859375, -120.6485595703125, 121.58197021484375, -24.27789306640625, 215.22055053710938, 129.1281280517578, 155.22125244140625, 118.50579833984375, 48.47869873046875, -122.64938354492188, -33.8309326171875, -3.16876220703125, -82.42196655273438, 160.3435516357422, 157.93116760253906, 45.143768310546875, 58.750762939453125, 57.52894592285156, -22.47894287109375, 430.5399169921875, -108.60232543945312, 210.1534881591797, 156.36990356445312, 345.9549560546875, 29.062908172607422, -31.006500244140625, -40.98573303222656, 373.37933349609375, -24.8648681640625, 192.8712158203125, 31.113300323486328, -74.70901489257812, 177.2825927734375, 23.23712158203125, -68.33674621582031, 253.29556274414062, -1.425384521484375, 40.644775390625, 37.924530029296875, -192.92401123046875, -12.097076416015625, 250.88137817382812, 173.33694458007812, 211.77572631835938, 169.02435302734375, 117.21240234375, 171.62722778320312, 33.78013229370117, 224.0523681640625, 86.11880493164062, -76.80772399902344, 112.61138153076172, -74.948486328125, 66.9654541015625, 319.5639953613281, 12.96585464477539, -13.4317626953125, -167.42138671875, 264.1527404785156, 38.642791748046875, -3.2611846923828125, 115.05088806152344, -34.30921936035156, 258.12005615234375, 125.57870483398438, -79.64993286132812, 290.0325012207031, -29.680023193359375, 79.82025146484375, 28.70464324951172, 129.03785705566406, 14.559829711914062, 234.35179138183594, 25.35442352294922, 255.861083984375, -36.494529724121094, 182.04624938964844, 43.310935974121094, 203.453369140625, 156.88748168945312, 152.9713134765625, -5.08428955078125, 254.951171875, 38.82919692993164, 0.03304290771484375, -4.6292724609375, 162.45632934570312, -35.75030517578125, 156.1841278076172, 129.4827880859375, 20.34185791015625, 10.048110961914062, 284.25714111328125, -31.928504943847656, 199.6919403076172, 319.5318908691406, 302.557861328125, 141.7305908203125, -55.919036865234375, 148.93218994140625, 143.79443359375, -94.71710205078125, 20.04913330078125, 321.3466796875, 0.0, -143.85281372070312, 217.38442993164062, 200.61541748046875, -27.7906494140625, 67.51644897460938, -135.32374572753906, 228.82992553710938, 46.5941162109375, 11.718170166015625, -30.459442138671875, 119.56763458251953, -9.293846130371094, 411.73944091796875, 95.43435668945312, 163.20362854003906, 39.981937408447266, 269.3126525878906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000422.npy"}
{"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 84.88964080810547, "std": 129.20545959472656, "min": -303.7426452636719, "p10": -71.05910110473631, "median": 82.65880584716797, "p90": 252.07151184082028, "max": 529.1273803710938, "pos_frac": 0.75, "sample": [-11.232864379882812, 529.1273803710938, -107.1024169921875, -15.039825439453125, 261.2821044921875, 36.30596923828125, 125.97502136230469, -12.61627197265625, -14.687454223632812, 192.793701171875, 27.343017578125, 39.594482421875, 53.21099853515625, 101.21287536621094, -96.10247802734375, -22.76080322265625, 212.85198974609375, 300.88189697265625, 129.92767333984375, 25.95062255859375, 33.2943229675293, 8.637451171875, 49.70719909667969, 200.9591064453125, -168.01312255859375, 124.39163208007812, 196.62210083007812, 129.38986206054688, 101.67255401611328, 204.02407836914062, -7.983734130859375, 41.9053955078125, 85.981201171875, 76.99148559570312, 21.315032958984375, 246.8442840576172, 309.8089599609375, 37.712860107421875, -22.290191650390625, 140.09552001953125, 130.12261962890625, 215.10736083984375, -21.908905029296875, 160.05604553222656, -3.466461181640625, -11.720664978027344, 71.10295104980469, 225.61956787109375, 64.94952392578125, -68.786865234375, 275.6900939941406, -139.02096557617188, -303.7426452636719, 156.85845947265625, 185.59033203125, 4.880645751953125, -28.570587158203125, -147.8829803466797, 120.56231689453125, 185.31063842773438, 47.436100006103516, 39.9329833984375, 98.40742492675781, 120.48091125488281, -76.3609848022461, 158.93505859375, -18.30266571044922, 121.8587646484375, 3.638601303100586, 184.42703247070312, 124.98452758789062, 120.29434204101562, 134.029296875, 95.44827270507812, 67.4189453125, 85.87979125976562, -155.3582763671875, 103.53968811035156, 272.11688232421875, -62.287139892578125, 43.07427978515625, 301.4117431640625, 84.93394470214844, 269.3877258300781, 217.13226318359375, 40.317657470703125, 213.67459106445312, 79.25655364990234, 21.07855224609375, 348.6553955078125, 261.295166015625, 96.07839965820312, 67.203857421875, -112.37655639648438, 293.88458251953125, 208.55197143554688, -102.80113983154297, -66.80667114257812, 146.22183227539062, 117.21697998046875, 92.83563232421875, 127.12496948242188, 299.04296875, 176.5805206298828, 137.29266357421875, -214.5906982421875, 25.048309326171875, 216.85140991210938, 163.44970703125, 34.30561065673828, -66.33404541015625, 18.921823501586914, 3.7027587890625, 80.3836669921875, -50.725887298583984, 150.61209106445312, 248.12411499023438, 42.1781005859375, 312.82415771484375, 41.247467041015625, 178.15673828125, 207.85845947265625, 238.46112060546875, -37.723541259765625, 28.59222412109375, -77.2984619140625, -6.65074348449707, -137.040283203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000423.npy"}
{"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 82.17053985595703, "std": 115.03003692626953, "min": -229.15374755859375, "p10": -41.591000366210935, "median": 68.86027526855469, "p90": 202.34798736572264, "max": 545.5046997070312, "pos_frac": 0.8046875, "sample": [81.42628479003906, 266.9390869140625, 159.64027404785156, -34.313720703125, 51.417240142822266, 0.67510986328125, 53.155670166015625, -20.6466064453125, 25.288787841796875, 29.46221923828125, 15.90753173828125, -70.22030639648438, 370.8096923828125, 200.91212463378906, -172.04562377929688, 161.04319763183594, 7.85858154296875, 142.93743896484375, 169.64804077148438, 206.24256896972656, -16.467147827148438, 51.05517578125, 110.1680908203125, 192.60525512695312, 87.214111328125, 115.88056945800781, 60.984336853027344, -93.92971801757812, 155.505126953125, 136.07106018066406, 121.5255126953125, 268.95086669921875, 152.09429931640625, 19.65593147277832, 155.2763671875, 182.28216552734375, 132.31243896484375, 200.07058715820312, 213.70941162109375, -40.866607666015625, 62.60791015625, 149.1649932861328, 205.69833374023438, -229.15374755859375, 16.283836364746094, 195.02093505859375, 207.44500732421875, 172.0811767578125, 43.418785095214844, 11.00689697265625, 19.07064437866211, 70.2855224609375, 16.80828857421875, 97.25347900390625, 35.534149169921875, 316.63519287109375, -32.02967834472656, 183.15771484375, 150.13430786132812, 64.141357421875, -23.449371337890625, -45.71942138671875, 173.86361694335938, 80.66781616210938, 350.56475830078125, 178.00164794921875, 245.16705322265625, 39.3424072265625, -223.829833984375, 15.8577880859375, -43.28125, 296.0513000488281, 96.6903076171875, 159.54046630859375, 198.683837890625, 85.79931640625, -113.3919906616211, 227.04092407226562, 133.02818298339844, -5.48211669921875, 121.67166137695312, 163.85565185546875, 0.28208160400390625, 18.78319549560547, 66.31837463378906, -18.333585739135742, 35.72412109375, 0.0, 38.11688232421875, 47.551971435546875, -4.744720458984375, 135.89950561523438, 89.96849822998047, 19.562042236328125, 82.3988037109375, 174.132080078125, 103.91021728515625, 144.18240356445312, 170.98736572265625, 70.6146240234375, -22.440792083740234, -14.65997314453125, 67.43502807617188, 53.717498779296875, 176.50921630859375, -72.885009765625, -45.88467025756836, 35.537506103515625, -118.62495422363281, 15.573036193847656, 35.766353607177734, -99.6025390625, 105.875244140625, 149.120849609375, -156.58782958984375, 5.253366470336914, 17.63227653503418, 96.27587890625, 17.829254150390625, 545.5046997070312, 58.47283935546875, 167.7974853515625, 43.07452392578125, 141.82342529296875, 199.2198486328125, 59.05931091308594, 4.802013397216797, 161.40841674804688], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000424.npy"}
{"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 72.02163696289062, "std": 128.01959228515625, "min": -242.30218505859375, "p10": -72.32389526367187, "median": 62.55567169189453, "p90": 235.81650390625, "max": 559.559326171875, "pos_frac": 0.7265625, "sample": [153.09051513671875, 248.17837524414062, 115.00894165039062, 15.619758605957031, -12.013916015625, 180.38983154296875, 64.88088989257812, 9.795894622802734, 23.051681518554688, -179.64651489257812, 241.02676391601562, 85.39556884765625, 194.984130859375, 60.23045349121094, 113.266357421875, 1.3816776275634766, 105.13002014160156, 273.40087890625, 2.3322315216064453, 38.68878173828125, 212.4542236328125, 151.73419189453125, 1.0949859619140625, 11.41763687133789, 119.473876953125, 34.69683074951172, 3.6086692810058594, 87.21771240234375, -32.642303466796875, 6.333160400390625, 107.4324951171875, -10.914398193359375, -164.94760131835938, 39.68708801269531, 237.135498046875, -9.488431930541992, 79.13192749023438, 171.48983764648438, 17.627716064453125, 121.142578125, -13.52960205078125, 195.97344970703125, 171.92007446289062, 122.83648681640625, 87.3055419921875, -4.817363739013672, -142.6066131591797, -70.2744140625, -133.29541015625, 235.251220703125, 88.98556518554688, 120.11900329589844, -124.607666015625, 128.93350219726562, 258.5012512207031, 54.134735107421875, 200.79647827148438, 307.4732971191406, 5.25775146484375, 104.99772644042969, -139.82640075683594, 252.1713409423828, 68.21405029296875, -7.891143798828125, 212.368896484375, -37.412391662597656, 70.55516052246094, -8.24220085144043, 272.0195007324219, 43.123504638671875, 166.472412109375, 33.867340087890625, 159.6513671875, 9.99237060546875, -161.654052734375, -6.0731964111328125, -48.803871154785156, 171.28448486328125, 9.015541076660156, 156.93804931640625, 0.4200401306152344, 69.5562744140625, 38.9644775390625, 298.072265625, 184.27749633789062, -124.02333068847656, 16.50018310546875, -100.09834289550781, -8.279106140136719, -32.28851318359375, -26.79541015625, -39.20094680786133, 83.28531646728516, -242.30218505859375, 88.6412353515625, -72.8314208984375, 138.58294677734375, -61.764442443847656, 154.44879150390625, -58.29296875, 58.66436767578125, 211.26168823242188, 209.6876220703125, 67.22476196289062, 158.19956970214844, 105.16907501220703, 357.8589782714844, 34.71022033691406, -106.0434799194336, 191.662109375, 0.0, -25.80572509765625, 200.30615234375, 72.84210205078125, 559.559326171875, -6.356708526611328, 21.136409759521484, 280.1579895019531, -195.65786743164062, 9.2613525390625, 162.2972412109375, 390.8316650390625, 138.9629364013672, 172.37051391601562, 165.29013061523438, -72.10638427734375, 12.428253173828125, 10.98016357421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000425.npy"}
{"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 83.74078369140625, "std": 126.9645004272461, "min": -382.8851623535156, "p10": -52.23658370971678, "median": 66.06764221191406, "p90": 269.3550048828125, "max": 491.68841552734375, "pos_frac": 0.75, "sample": [193.3563232421875, -17.978439331054688, 2.178323745727539, -44.75250244140625, 83.64871215820312, -48.48350524902344, 131.91860961914062, 2.7492103576660156, 50.42401123046875, -11.242362976074219, -382.8851623535156, -41.01611328125, -74.75833129882812, 171.74549865722656, 178.95425415039062, 107.76954650878906, 192.177978515625, 66.93304443359375, 115.15690612792969, 58.790435791015625, -161.19036865234375, 44.5616455078125, 242.5555419921875, 60.04834747314453, 25.483009338378906, -20.04766845703125, 174.4576873779297, 275.1943359375, 98.23748779296875, 268.0924072265625, 50.63433837890625, 36.489009857177734, 305.4158935546875, 74.88455963134766, 268.959228515625, 12.023920059204102, 67.42753601074219, 141.442138671875, 204.881103515625, 140.4747314453125, 54.465240478515625, 265.61627197265625, 4.5556640625, -18.93968963623047, 275.0775146484375, 279.4723205566406, 10.97613525390625, 206.71914672851562, 66.23220825195312, -92.498291015625, -90.19708251953125, 111.01473999023438, 91.64635467529297, -19.868743896484375, 128.07550048828125, 270.23681640625, -129.50323486328125, 28.065725326538086, 329.71490478515625, 147.54254150390625, 222.81182861328125, 133.6429443359375, -94.04580688476562, -28.804244995117188, -4.650787353515625, 271.004150390625, 233.84225463867188, 80.20172119140625, 64.81220245361328, 26.26470947265625, 65.903076171875, 85.29812622070312, -22.268875122070312, 138.0132293701172, 8.375202178955078, 284.18145751953125, -79.93795013427734, 42.64787292480469, 9.314315795898438, 69.16839599609375, 23.622032165527344, 129.27354431152344, 214.5135498046875, 104.9879379272461, 333.6908874511719, 32.81341552734375, 33.90570068359375, 62.16767883300781, 115.90399932861328, 230.7222900390625, 54.3756103515625, 51.259765625, -23.122203826904297, -127.00851440429688, 269.3445129394531, 269.3794860839844, 70.00006103515625, 72.99444580078125, -5.4064788818359375, -44.49360656738281, 160.4735107421875, 92.73394775390625, 0.0, 26.658424377441406, -19.879913330078125, 491.68841552734375, 161.47373962402344, -22.955413818359375, 74.9169921875, 126.16822814941406, 183.06158447265625, 40.53155517578125, -64.65985107421875, -60.99376678466797, -20.89947509765625, 62.53570556640625, -121.38198852539062, 284.60498046875, 236.75799560546875, 234.54388427734375, 9.393989562988281, -35.02490997314453, -124.18328857421875, 16.91839599609375, 217.14816284179688, 113.44727325439453, 38.6707763671875, 279.25836181640625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000426.npy"}
{"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 108.29835510253906, "std": 147.9292755126953, "min": -376.8973388671875, "p10": -45.16470947265625, "median": 95.62600708007812, "p90": 304.50888671875, "max": 515.1129150390625, "pos_frac": 0.7578125, "sample": [20.823394775390625, 312.648681640625, 66.39599609375, -149.37890625, 41.53424072265625, 94.41380310058594, 101.77865600585938, 272.061279296875, 214.06509399414062, 119.702392578125, 101.06033325195312, 54.66984558105469, 267.58868408203125, -11.77560043334961, 356.3550720214844, 313.81732177734375, 96.29034423828125, -4.596977233886719, 241.951171875, 336.54156494140625, 231.1968994140625, 59.44757080078125, 14.947921752929688, 2.15753173828125, 385.6619873046875, 23.98223876953125, 254.60211181640625, 249.830322265625, 85.29757690429688, 83.30511474609375, 312.91748046875, 259.4122314453125, -41.87542724609375, 52.735198974609375, 36.573272705078125, 65.1607666015625, 102.66809844970703, 11.996732711791992, -21.171783447265625, -215.43150329589844, 254.4635772705078, 28.9913330078125, -183.5227508544922, 95.02639770507812, -126.22747802734375, 124.81065368652344, -18.448139190673828, 377.0372314453125, -31.33355712890625, 198.71856689453125, 149.08987426757812, -19.007659912109375, 68.09465026855469, 223.74755859375, 252.94821166992188, 422.049072265625, 193.15582275390625, -70.56085205078125, -3.854583740234375, -52.13609313964844, 158.16075134277344, 99.18108367919922, 240.706787109375, -10.04150390625, 44.8729248046875, 111.68016052246094, -15.30572509765625, 81.68161010742188, 206.50267028808594, -190.14669799804688, -376.8973388671875, 96.06088256835938, 57.612274169921875, -204.44131469726562, 515.1129150390625, 13.322072982788086, -45.13386535644531, 197.24154663085938, 210.82850646972656, -6.512672424316406, 280.545166015625, -158.4362335205078, 183.17068481445312, -70.167236328125, -45.23667907714844, 237.77406311035156, 214.35049438476562, 165.1141357421875, 260.0857849121094, 128.592041015625, 152.54315185546875, 175.04833984375, -12.44012451171875, 95.19113159179688, 359.76776123046875, 297.3323059082031, 183.23870849609375, 306.914794921875, -7.7165985107421875, 21.976329803466797, 56.422271728515625, 140.01907348632812, 68.576171875, 0.5635833740234375, 158.11767578125, 335.0289306640625, 63.284210205078125, 66.30319213867188, 122.68122863769531, 162.09353637695312, 303.477783203125, -20.621246337890625, 126.11454772949219, 10.36956787109375, 143.634521484375, -65.49029541015625, 227.97665405273438, 41.52214050292969, -5.9854583740234375, -17.08612060546875, 298.3711242675781, 130.76019287109375, -30.69671630859375, 7.478294372558594, 69.80636596679688, 220.53704833984375, 348.23822021484375, 266.15728759765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000427.npy"}
{"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 99.28484344482422, "std": 135.45887756347656, "min": -441.034912109375, "p10": -40.89291076660156, "median": 90.42204284667969, "p90": 269.97097167968747, "max": 513.4866943359375, "pos_frac": 0.7734375, "sample": [79.36231994628906, -75.07681274414062, -54.65813446044922, 354.836181640625, 339.96392822265625, -118.3719482421875, 83.1795654296875, 238.8917236328125, -35.34065246582031, -29.873214721679688, 513.4866943359375, 51.142974853515625, 228.6792449951172, 117.2811279296875, -28.14544677734375, 176.43862915039062, 29.5582275390625, 24.28564453125, 42.631080627441406, -21.1883544921875, 134.13751220703125, 44.725372314453125, -204.105224609375, 362.05718994140625, 319.4561767578125, 277.39166259765625, 227.73666381835938, 84.3718032836914, 139.90896606445312, 110.05426025390625, 151.97610473632812, 286.6976013183594, 11.233724594116211, 53.58624267578125, -8.281982421875, 90.82823181152344, -20.450366973876953, 192.79095458984375, 40.58856201171875, 89.26565551757812, 173.0401611328125, -15.528133392333984, -84.6500244140625, 148.43060302734375, 245.5625, 227.60360717773438, -168.90341186523438, -16.606754302978516, 108.9754638671875, -10.458641052246094, 210.5788116455078, 223.9609832763672, 190.5880126953125, 45.09919738769531, 107.42999267578125, 20.098846435546875, 267.81671142578125, 83.066162109375, -441.034912109375, 341.1058654785156, -11.160728454589844, 338.408935546875, 148.77517700195312, 209.72198486328125, 250.24276733398438, 25.4864501953125, 252.3428955078125, 98.81880187988281, 234.03463745117188, 91.95904541015625, 27.73583984375, 33.09519958496094, 90.01585388183594, 108.59272766113281, 169.1494598388672, 15.49053955078125, 118.35490417480469, 13.072174072265625, -64.74908447265625, 59.39329528808594, 54.73753356933594, 23.950393676757812, 96.80120086669922, 131.84457397460938, -103.78919982910156, -107.05007934570312, -17.91619873046875, 5.0445556640625, 28.34100341796875, 134.44386291503906, 359.95953369140625, 263.4010925292969, 82.54507446289062, 118.73464965820312, 217.634521484375, -141.10675048828125, -78.89422607421875, 149.66159057617188, 204.9300537109375, 0.0, 91.96694946289062, 210.84707641601562, 268.78045654296875, 169.31619262695312, 189.50030517578125, 78.80982208251953, 80.45401000976562, 156.5667724609375, 19.595088958740234, 176.52877807617188, 3.47637939453125, 81.83137512207031, 0.04040718078613281, -4.861412048339844, 226.85354614257812, 290.6883544921875, 148.9298095703125, 98.263671875, 203.9583740234375, -39.69610595703125, -33.1783447265625, 19.99530029296875, 281.74609375, -8.63332748413086, -43.685455322265625, 66.48696899414062, 272.74884033203125, 111.84928894042969], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000428.npy"}
{"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 85.4117660522461, "std": 149.34942626953125, "min": -633.1908569335938, "p10": -70.31536407470702, "median": 80.33732604980469, "p90": 250.4173828125, "max": 405.20855712890625, "pos_frac": 0.703125, "sample": [87.63836669921875, -401.882080078125, -121.93304443359375, 96.04585266113281, 220.77374267578125, 204.85995483398438, 118.85272979736328, 42.1019287109375, 21.0455322265625, 221.69387817382812, -8.254493713378906, -119.85560607910156, 84.13385009765625, 164.79812622070312, 239.3712158203125, 210.489013671875, 65.41081237792969, 355.2205810546875, -96.73199462890625, 76.667724609375, 165.14263916015625, 191.17922973632812, -126.98428344726562, 190.61514282226562, -29.31670379638672, 405.20855712890625, 142.31842041015625, 227.37506103515625, 66.48971557617188, 202.73681640625, -117.38400268554688, 163.13412475585938, 150.06015014648438, -54.519195556640625, 253.5032958984375, 198.68984985351562, -148.06661987304688, 15.859771728515625, 216.68035888671875, 375.7328796386719, 144.3004150390625, 213.4459991455078, 28.26849365234375, -19.929359436035156, 87.29920959472656, -11.697219848632812, 203.78204345703125, 23.07769775390625, 22.512786865234375, -68.22344970703125, 2.170574188232422, 153.76123046875, 183.737548828125, 234.13577270507812, 59.66680908203125, 62.60699462890625, 396.16802978515625, -84.7642822265625, -60.36180114746094, 173.58871459960938, -8.148681640625, 92.80459594726562, 201.30474853515625, 285.6568603515625, -67.073486328125, -633.1908569335938, 324.3819580078125, -5.613555908203125, 225.78245544433594, 204.59552001953125, -109.002197265625, -0.4197502136230469, -74.46914672851562, -15.488296508789062, 276.62579345703125, -55.18904113769531, 254.86618041992188, -6.2364044189453125, 34.17633056640625, 207.97744750976562, 4.835607528686523, 0.0, 266.8173828125, 202.10714721679688, 121.74667358398438, 109.6619873046875, 221.8374481201172, 55.030128479003906, -25.360565185546875, 11.548288345336914, 244.67877197265625, -39.038604736328125, 230.7464141845703, -14.315109252929688, -68.53517150878906, 183.4925537109375, 19.907283782958984, -18.03870391845703, 36.146759033203125, 25.080902099609375, -20.52960205078125, -13.34600830078125, -123.05230712890625, 58.91627502441406, -90.78497314453125, 214.6080322265625, 63.65948486328125, 315.3795166015625, -64.17945861816406, 100.46029663085938, 403.1611633300781, 249.0948486328125, 71.07545471191406, 177.03457641601562, 106.38180541992188, 84.00692749023438, 20.483646392822266, 53.84492492675781, 129.11392211914062, 280.89959716796875, 62.49345397949219, 144.10272216796875, -19.614151000976562, 42.206268310546875, 117.11856079101562, 120.63484191894531, 96.97100830078125, -41.57232666015625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000429.npy"}
{"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 73.51123809814453, "std": 128.8090362548828, "min": -208.71621704101562, "p10": -79.48546752929687, "median": 62.636383056640625, "p90": 231.14080200195312, "max": 444.537109375, "pos_frac": 0.6796875, "sample": [-68.96612548828125, -140.28134155273438, -25.977218627929688, 65.42059326171875, 120.47444915771484, 52.73944091796875, -89.9622802734375, 115.89361572265625, -10.067840576171875, -5.3016510009765625, 63.50726318359375, 19.8594970703125, 87.41941833496094, -208.71621704101562, 61.7655029296875, -124.15414428710938, 427.70184326171875, 124.22328186035156, 17.112667083740234, 201.6039276123047, 40.35491943359375, -21.852554321289062, 387.0895080566406, 135.524169921875, 96.19757080078125, 1.829376220703125, 43.7366943359375, 0.0, 264.08392333984375, -72.818603515625, 86.408935546875, -1.85601806640625, 209.52471923828125, 54.45524597167969, 75.31466674804688, 0.0, 15.662490844726562, -25.181182861328125, 40.7957763671875, 189.73333740234375, -81.915771484375, 274.73675537109375, 48.638397216796875, 74.392578125, 216.9442138671875, -13.983535766601562, -43.83489227294922, 162.84014892578125, -70.50775146484375, 206.8890380859375, 107.31725311279297, 78.01376342773438, 320.3133850097656, 54.49267578125, 148.36721801757812, 178.71273803710938, 73.48985290527344, -88.71014404296875, 229.252197265625, -78.44390869140625, 312.819580078125, 286.31011962890625, 210.29180908203125, 198.16604614257812, -10.206924438476562, 203.89443969726562, 182.45294189453125, 175.71133422851562, 10.802003860473633, 215.5594482421875, 85.31195068359375, 148.708251953125, 236.40838623046875, 330.0333251953125, 11.485771179199219, -4.806522369384766, 16.009246826171875, 99.82180786132812, -26.223085403442383, 43.084930419921875, 191.28924560546875, 54.185699462890625, -11.189826965332031, 168.3728485107422, -141.30877685546875, 44.8380126953125, -49.351287841796875, 278.28497314453125, 0.0, -29.354766845703125, 67.6044692993164, 42.626220703125, 78.34408569335938, -32.423492431640625, 444.537109375, -4.894247055053711, 135.29071044921875, 150.09039306640625, 7.581827163696289, -61.67559814453125, 235.54754638671875, 204.16302490234375, 120.22425842285156, -192.7208709716797, -43.333831787109375, 104.68232727050781, 177.45591735839844, 51.551475524902344, 164.28668212890625, 238.10879516601562, -202.30352783203125, -134.496337890625, 222.05419921875, -16.12646484375, 115.97549438476562, 79.95355224609375, 21.109390258789062, 84.42156982421875, 8.672149658203125, -136.070068359375, -41.0347900390625, -174.85202026367188, 112.0311279296875, 224.72735595703125, 155.47430419921875, -12.430816650390625, -85.40093994140625, 68.98503112792969], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000430.npy"}
{"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 64.9871826171875, "std": 119.96318817138672, "min": -201.3651123046875, "p10": -65.74711380004882, "median": 44.84406280517578, "p90": 241.91072845458984, "max": 411.916259765625, "pos_frac": 0.6953125, "sample": [-64.73344421386719, -118.69122314453125, 149.35946655273438, 90.12594604492188, 22.42120361328125, 157.56280517578125, -30.826202392578125, 157.05511474609375, -34.24884033203125, -2.36956787109375, 33.77685546875, -2.205047607421875, 56.008544921875, 75.90054321289062, -144.702392578125, 240.74208068847656, -147.30633544921875, 249.12911987304688, 155.44772338867188, 16.75647735595703, -31.28955078125, 160.99664306640625, 74.64216613769531, -125.80674743652344, 78.12103271484375, 114.74896240234375, -0.6763916015625, -201.3651123046875, 269.22955322265625, 324.2626953125, 105.51455688476562, 173.96603393554688, -33.61358642578125, -38.98229217529297, -140.93406677246094, 360.53106689453125, -55.76072692871094, -17.74298095703125, -68.11234283447266, 64.17819213867188, 52.424522399902344, -130.76165771484375, 5.132774353027344, 74.20123291015625, 130.30177307128906, 258.7630615234375, 142.33709716796875, -24.42962646484375, -2.985637664794922, -174.17813110351562, -54.12531280517578, 46.2296142578125, 15.7288818359375, 119.33673095703125, -12.536712646484375, 59.17669677734375, 65.12069702148438, 175.57711791992188, 152.4405517578125, 4.268341064453125, 17.53131103515625, 114.1307373046875, 24.066299438476562, 47.90765380859375, 42.29241943359375, 31.899627685546875, 30.51129150390625, 83.40106201171875, -15.125785827636719, -105.44046020507812, 244.6375732421875, 13.033275604248047, 50.958984375, 254.55340576171875, 75.84176635742188, 249.95138549804688, 56.309722900390625, 223.40530395507812, 304.7264404296875, 233.6138153076172, 215.3089599609375, 313.20220947265625, 35.77333068847656, -96.04678344726562, 7.3321380615234375, -19.756134033203125, 108.01717376708984, 105.78976440429688, -48.960113525390625, 26.93438720703125, 317.2677001953125, -76.03932189941406, 165.46575927734375, 22.741226196289062, 9.843116760253906, 318.96771240234375, 25.60211181640625, 206.0157470703125, 43.96913146972656, 26.3948974609375, 158.6406707763672, 119.20450592041016, -10.236846923828125, 45.718994140625, 59.07794189453125, 17.139930725097656, 179.77708435058594, 89.83924865722656, 19.949573516845703, -42.982940673828125, 19.717697143554688, -9.116226196289062, 182.20266723632812, 211.87588500976562, 33.88104248046875, 68.62753295898438, -12.268585205078125, -5.620819091796875, -23.335411071777344, 132.4891815185547, -155.77272033691406, -3.6434326171875, -43.843475341796875, 223.72622680664062, 11.516387939453125, 411.916259765625, 93.98556518554688, 86.73241424560547], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000431.npy"}
{"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 86.66366577148438, "std": 134.4788055419922, "min": -224.084228515625, "p10": -39.49714050292969, "median": 56.870849609375, "p90": 256.13945922851565, "max": 560.968505859375, "pos_frac": 0.71875, "sample": [-8.459747314453125, 57.887908935546875, 45.89186096191406, 376.5245361328125, 256.5238952636719, 215.9930419921875, 149.9869384765625, 90.28990173339844, -224.084228515625, -186.59881591796875, 560.968505859375, 32.7513427734375, 233.0570068359375, 37.4962158203125, 48.811065673828125, 12.058250427246094, 64.91390991210938, 227.28958129882812, 157.48507690429688, 67.80115509033203, 154.12896728515625, -162.3861083984375, -66.53125, -8.466268539428711, -150.55975341796875, -10.590288162231445, -59.8863525390625, 206.9116973876953, -16.505897521972656, 74.42489624023438, 58.1075439453125, 332.2334289550781, 193.97341918945312, 270.6937561035156, -120.2733154296875, -47.62782287597656, 180.0772705078125, -40.2335205078125, 94.67385864257812, 4.16058349609375, 33.05754852294922, 127.18115234375, -27.3067626953125, 5.057371139526367, -9.946296691894531, 58.02122497558594, 130.1461181640625, 141.8536376953125, -171.93609619140625, 9.07916259765625, 175.2899932861328, 63.854042053222656, -28.16461181640625, 408.6346130371094, -25.905120849609375, -27.46484375, -5.346992492675781, 93.54606628417969, 34.88447570800781, 5.872398376464844, 41.344085693359375, 13.647628784179688, 237.4571533203125, -21.438201904296875, 189.93475341796875, 29.167922973632812, 199.79153442382812, 311.1173095703125, 246.8480224609375, 69.85150909423828, -19.51384735107422, 46.47474670410156, 194.15960693359375, -15.392822265625, 11.29791259765625, 338.61822509765625, 75.34278869628906, 199.21142578125, 242.53152465820312, 282.353759765625, 7.61358642578125, 55.853790283203125, 268.1590881347656, -25.835784912109375, 153.67715454101562, -3.14111328125, 136.44454956054688, 192.3406982421875, -19.57147216796875, 244.6361541748047, 43.443115234375, -116.30909729003906, 0.0, 111.3245849609375, 141.00558471679688, -60.2696533203125, -115.48855590820312, 32.39208984375, -18.308929443359375, 46.380889892578125, 79.93930053710938, -22.51993179321289, -7.160585403442383, 224.9483642578125, 147.56910705566406, 20.573326110839844, 255.97470092773438, 113.61808776855469, 183.0531005859375, 156.7430877685547, 3.5684814453125, 238.47396850585938, 63.951812744140625, 399.47332763671875, -8.310861587524414, 52.802154541015625, 260.2834777832031, 189.1405487060547, 11.080831527709961, 45.403564453125, 140.7003936767578, 188.95619201660156, -19.73797607421875, 9.910697937011719, 438.32855224609375, 90.4775390625, 34.4188117980957, -39.181549072265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000432.npy"}
{"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 73.83741760253906, "std": 117.47842407226562, "min": -224.5504150390625, "p10": -65.82347259521482, "median": 63.937503814697266, "p90": 216.05090942382813, "max": 421.94561767578125, "pos_frac": 0.765625, "sample": [7.65110969543457, -196.22406005859375, -74.96746826171875, 242.57138061523438, 70.78038024902344, 51.17108154296875, 216.81373596191406, -10.19024658203125, 2.8969860076904297, 70.43460083007812, 117.05221557617188, 44.8001708984375, 181.2152557373047, 146.72323608398438, 88.01803588867188, 12.94232177734375, -159.6600341796875, 186.39093017578125, 95.382080078125, 89.30104064941406, 68.53875732421875, 123.64028930664062, 132.11416625976562, -25.592849731445312, 6.62109375, -107.27838134765625, 202.0074462890625, 14.138233184814453, 273.3836364746094, 88.20062255859375, 163.54322814941406, 86.2076644897461, 119.50534057617188, 121.52728271484375, -4.768035888671875, 151.4283447265625, 195.96067810058594, 134.27737426757812, -123.84893798828125, 122.3099365234375, -130.66873168945312, 17.680282592773438, 209.33065795898438, 299.7532043457031, 41.4149169921875, 241.82598876953125, 38.5736083984375, -46.9892463684082, 133.34912109375, 27.727863311767578, 158.84112548828125, 173.81784057617188, 179.73770141601562, 39.946319580078125, -23.780914306640625, -149.34091186523438, 161.203369140625, 196.40097045898438, 45.41372299194336, 15.98314094543457, -91.0474853515625, 50.52336120605469, 123.67363739013672, 88.46136474609375, 183.50477600097656, 28.735687255859375, 162.8585968017578, -224.5504150390625, 118.26611328125, 118.82624816894531, -6.397064208984375, 35.547142028808594, 90.87075805664062, -0.8853073120117188, 52.27252197265625, 61.688140869140625, 150.36837768554688, 175.83358764648438, 130.32815551757812, -91.86083984375, -83.44113159179688, -58.87677001953125, 215.43295288085938, 65.16488647460938, 43.3134765625, -195.521240234375, -61.90461730957031, -43.108375549316406, 285.8747863769531, 45.44696044921875, -2.560211181640625, 0.37192535400390625, -57.995849609375, -6.9112548828125, 62.710121154785156, 17.797218322753906, 160.10592651367188, 197.5491485595703, 122.37357330322266, -15.3648681640625, 149.87681579589844, -22.875770568847656, -2.3264236450195312, 421.94561767578125, 6.58282470703125, 12.52008056640625, 274.0050964355469, 226.01548767089844, 324.6851501464844, 18.887958526611328, 227.32583618164062, 38.6522216796875, 112.5333251953125, -32.73402404785156, 0.00042724609375, 34.094085693359375, 215.72398376464844, 28.73351287841797, 96.6265640258789, 334.39697265625, 176.26181030273438, 104.49688720703125, -177.15203857421875, 18.783485412597656, 105.03306579589844, 35.060577392578125, 290.19378662109375, 31.12912368774414], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000433.npy"}
{"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 75.56307220458984, "std": 123.56360626220703, "min": -328.060791015625, "p10": -68.64090881347656, "median": 68.877685546875, "p90": 227.3228546142578, "max": 532.2653198242188, "pos_frac": 0.75, "sample": [124.54342651367188, 66.40782165527344, -5.226531982421875, 30.691879272460938, 126.70208740234375, 104.65226745605469, 33.01976776123047, 165.90811157226562, 81.44354248046875, 201.597900390625, -73.52165985107422, 164.6144256591797, 22.076263427734375, 0.31500244140625, -179.09368896484375, 19.106201171875, 34.34613037109375, 152.05059814453125, -104.29336547851562, 6.929229736328125, 209.02859497070312, 421.3956298828125, 9.94610595703125, 344.33489990234375, 201.64923095703125, -179.1397705078125, 0.0, 24.64141845703125, -52.95263671875, 127.14495849609375, 311.0165710449219, 183.217041015625, -8.81304931640625, 70.24557495117188, 146.56301879882812, 0.30377197265625, 41.9678955078125, 48.53033447265625, 7.91864013671875, 230.11428833007812, 9.57720947265625, 287.20367431640625, 185.06192016601562, -98.8798828125, 87.16380310058594, -67.83688354492188, 70.29425048828125, 256.85986328125, 38.412078857421875, 107.25091552734375, 89.81076049804688, 256.9432373046875, 114.01238250732422, 155.4881591796875, 115.49635314941406, 103.58406066894531, 135.26133728027344, -5.356109619140625, 87.67617797851562, 136.54336547851562, 122.02481079101562, 67.50979614257812, 236.878173828125, -26.09280014038086, -34.0191650390625, 4.471435546875, -101.44609069824219, 177.42919921875, 66.6976318359375, 2.7634143829345703, 226.12652587890625, -32.41600036621094, 20.30957794189453, 200.00747680664062, -83.89312744140625, 130.49691772460938, 94.50723266601562, 234.90866088867188, 251.12783813476562, 98.8719482421875, 137.8865966796875, -148.27304077148438, 114.58401489257812, 206.55807495117188, 76.87255859375, 156.98126220703125, 189.4395751953125, -328.060791015625, 64.28659057617188, -70.5169677734375, 142.66030883789062, -51.00755310058594, -139.56259155273438, 0.0, 123.3829345703125, 55.83769226074219, -31.801025390625, 532.2653198242188, 226.10195922851562, 14.314441680908203, 99.87826538085938, -19.321365356445312, 20.035966873168945, 97.1231689453125, 49.92037582397461, 35.515071868896484, -3.2176437377929688, -56.750694274902344, 240.8521728515625, 120.56840515136719, 37.48553466796875, 11.602712631225586, 0.0, 225.67941284179688, 9.323883056640625, -27.3731689453125, 15.633392333984375, 129.41720581054688, 58.680023193359375, -8.763916015625, -156.9957733154297, 91.37730407714844, 95.89715576171875, -21.625442504882812, -94.03594970703125, 256.3976135253906, 152.40921020507812, 210.1990966796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000434.npy"}
{"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 86.36825561523438, "std": 126.9636001586914, "min": -277.7374572753906, "p10": -44.60570335388183, "median": 68.87860107421875, "p90": 223.65296936035156, "max": 706.5430908203125, "pos_frac": 0.765625, "sample": [214.12106323242188, 61.04949951171875, -8.054466247558594, 141.58596801757812, 115.14166259765625, 284.8826904296875, 238.28570556640625, -78.772216796875, 66.96487426757812, -5.45855712890625, -0.10800552368164062, 31.782005310058594, -277.7374572753906, 130.10574340820312, 34.49774932861328, 22.10662841796875, 73.0767822265625, 90.83203125, 365.6557922363281, 113.12045288085938, 69.01498413085938, 18.069868087768555, 211.7884521484375, 0.0, 16.3388614654541, -40.26495361328125, 67.1156005859375, 76.93831634521484, 140.91680908203125, -3.8526382446289062, 64.06745910644531, 192.3776092529297, 160.96298217773438, 152.99240112304688, -29.355613708496094, 278.2833251953125, 171.23861694335938, 223.11404418945312, 129.42562866210938, -9.11468505859375, 251.94732666015625, 196.55096435546875, 224.91046142578125, 47.64263916015625, 188.54042053222656, 42.45574951171875, 102.93362426757812, -57.25897216796875, 142.81045532226562, 30.150360107421875, 145.38677978515625, 569.013916015625, 103.22219848632812, 226.1292724609375, 7.598682403564453, 109.53639221191406, 19.982269287109375, 14.287368774414062, 191.23031616210938, -70.63346862792969, -74.73567199707031, 137.20294189453125, 134.22921752929688, -76.46791076660156, 109.56998443603516, 2.01397705078125, 176.826904296875, 155.54049682617188, -18.99083137512207, 105.70439147949219, 68.74221801757812, -0.98248291015625, 27.658447265625, 286.0113525390625, 217.71890258789062, 65.64736938476562, 169.5019989013672, 178.13174438476562, -25.073348999023438, 97.43753051757812, -139.80596923828125, 181.6326904296875, 368.49517822265625, 188.36167907714844, -91.21795654296875, 63.538330078125, 151.09112548828125, -2.8309574127197266, -8.141426086425781, 107.4617919921875, -43.14358901977539, 119.55026245117188, 134.4422607421875, 706.5430908203125, 103.47238159179688, 3.1053009033203125, 135.42718505859375, -68.3468017578125, -151.10125732421875, 228.42315673828125, -17.707748413085938, 177.22991943359375, 4.167671203613281, -29.09858512878418, 146.2030029296875, -111.65777587890625, 24.494842529296875, 70.83267211914062, 143.62413024902344, 17.01264190673828, -34.35157775878906, 144.3209228515625, 4.84600830078125, 54.93394470214844, 11.489013671875, 270.6004638671875, 16.669464111328125, 12.30072021484375, 56.875152587890625, 28.22283935546875, 49.31402587890625, 88.9561767578125, 71.28972625732422, -59.37699890136719, -48.017303466796875, 149.00369262695312, 35.11993408203125, 65.62385559082031], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000435.npy"}
{"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 67.73057556152344, "std": 138.30702209472656, "min": -294.9237060546875, "p10": -74.98263092041016, "median": 47.996646881103516, "p90": 251.47090301513668, "max": 715.90283203125, "pos_frac": 0.7265625, "sample": [35.987060546875, -294.9237060546875, 245.32064819335938, 46.5538330078125, -13.699108123779297, -12.943801879882812, 128.35189819335938, 25.64067840576172, -67.9814453125, 715.90283203125, 11.37274169921875, 157.38583374023438, 48.024192810058594, 65.1622314453125, -60.71337890625, 376.8918762207031, 72.9736328125, -58.09291076660156, -193.2935791015625, -238.31146240234375, -18.966445922851562, 54.87596893310547, 107.98930358886719, 202.1783447265625, 301.0746154785156, -37.41456604003906, 165.36981201171875, 13.412353515625, 180.00997924804688, 35.8873291015625, 122.3047103881836, 33.25379943847656, -33.52374267578125, 31.56707763671875, 159.10093688964844, 238.63644409179688, 47.96910095214844, 108.1698989868164, 262.4432373046875, 38.562103271484375, 177.31761169433594, -75.47544860839844, 75.08214569091797, 140.57772827148438, 96.77496337890625, 46.423095703125, 14.954116821289062, -154.17568969726562, 20.56396484375, 282.71173095703125, 179.85696411132812, -62.24090576171875, -48.223785400390625, -60.583740234375, 132.18362426757812, 83.88887786865234, 199.58004760742188, -126.91156005859375, -100.65179443359375, 278.7161560058594, 210.76504516601562, 81.7406005859375, 59.722747802734375, -29.66783905029297, 180.50631713867188, 247.5959930419922, -7.784637451171875, -92.55667114257812, 36.438507080078125, 286.8780517578125, 11.164459228515625, 279.045166015625, 148.69039916992188, 43.92523193359375, 61.720855712890625, 97.0042724609375, 17.512187957763672, 126.33248901367188, 47.060821533203125, 7.230560302734375, 216.5899658203125, 41.19085693359375, 64.93862915039062, -7.023078918457031, 136.6749267578125, 120.7801513671875, 93.82725524902344, 22.338218688964844, 104.05137634277344, -101.4207763671875, 394.6940002441406, 60.7259521484375, -10.935911178588867, -213.87283325195312, -9.765411376953125, 129.37255859375, 276.62451171875, -33.81182861328125, 11.164020538330078, -138.16717529296875, 12.8316650390625, -195.01202392578125, 127.89079284667969, 56.925079345703125, 63.22770690917969, 63.25567626953125, -74.77142333984375, 75.68744659423828, 352.69384765625, -2.8993759155273438, -73.55020141601562, 192.50421142578125, 24.742889404296875, 260.5123596191406, -56.01045227050781, 168.8258819580078, -18.30816650390625, 280.1846008300781, 27.71149444580078, 119.45062255859375, 12.00933837890625, 59.707763671875, 0.267730712890625, 6.536346435546875, -107.0283203125, 123.97531127929688, 5.2230072021484375, 88.7537841796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000436.npy"}
{"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 96.0469970703125, "std": 137.38531494140625, "min": -222.22967529296875, "p10": -80.1524185180664, "median": 83.95561981201172, "p90": 259.5937744140625, "max": 441.417236328125, "pos_frac": 0.7734375, "sample": [339.40087890625, 125.8109130859375, 130.25689697265625, 119.64837646484375, -137.3663330078125, 441.417236328125, -24.99322509765625, 81.57276916503906, 120.69012451171875, 108.19873046875, 324.7261962890625, 81.78900146484375, 26.182830810546875, 114.63522338867188, 16.559967041015625, 204.20187377929688, -101.30706787109375, 38.74433898925781, 76.37689208984375, 119.60604858398438, 85.43925476074219, 410.93988037109375, 404.1044616699219, 172.104248046875, 59.057830810546875, 219.39288330078125, -11.27042007446289, 82.47198486328125, 149.99398803710938, 52.07072448730469, -44.98236083984375, 106.1092529296875, 30.14910888671875, 48.25700378417969, 132.98764038085938, -10.394855499267578, 22.82408905029297, 235.196533203125, -43.76454162597656, 49.5640869140625, 247.49526977539062, 142.92791748046875, 136.13653564453125, 325.1595153808594, -118.9327392578125, 24.6815185546875, 77.53350830078125, -7.517333984375, 129.11904907226562, 225.81573486328125, 135.9457244873047, 235.47311401367188, -51.282066345214844, 259.9068603515625, 142.04745483398438, 47.4149169921875, -108.40353393554688, -3.6801071166992188, 387.9401550292969, 178.98074340820312, 81.98382568359375, 36.32183837890625, 129.01480102539062, 186.65655517578125, 235.03802490234375, -164.95565795898438, 1.6203536987304688, 195.51931762695312, -84.2974853515625, 26.181541442871094, 19.726242065429688, -0.807281494140625, -222.22967529296875, 107.7618408203125, -45.629791259765625, 348.6519775390625, 7.81922721862793, 138.8961181640625, -122.594482421875, 0.0, 187.3602752685547, -78.37596130371094, 17.5113525390625, 61.0401725769043, 225.88644409179688, 126.60403442382812, -112.00567626953125, 197.0121307373047, 7.573486328125, 158.11416625976562, 263.3270263671875, 131.92776489257812, 2.5946102142333984, 80.39950561523438, -47.0345458984375, -16.64581298828125, 117.52874755859375, 129.30029296875, 158.3853759765625, 28.81060028076172, 259.4595947265625, 219.82003784179688, 171.4091796875, 337.5611572265625, 102.2740249633789, 20.380538940429688, 205.53268432617188, -173.8148193359375, 408.141357421875, -182.01242065429688, 57.902984619140625, -169.0345458984375, 183.39427185058594, -150.37884521484375, 28.071014404296875, 258.31298828125, 49.983551025390625, 211.19117736816406, 3.901885986328125, 354.9005432128906, 22.362472534179688, -36.173583984375, 213.77908325195312, 160.79315185546875, 0.0, 41.724273681640625, 167.2811279296875, 250.09710693359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000437.npy"}
{"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 44.992794036865234, "std": 131.1546173095703, "min": -411.6578369140625, "p10": -101.12745056152343, "median": 33.29372978210449, "p90": 194.86862792968748, "max": 451.2742919921875, "pos_frac": 0.640625, "sample": [160.9783935546875, -257.8518371582031, 155.2459716796875, -60.4322509765625, -49.31804275512695, 164.52883911132812, 265.7193298339844, 87.32745361328125, 0.0, 26.754451751708984, 367.2724304199219, -130.86349487304688, 32.249961853027344, 143.9914093017578, 33.48126220703125, 47.759765625, 43.02581787109375, -17.191009521484375, 49.49066925048828, -44.66344451904297, -108.52227783203125, -108.18536376953125, -76.84034729003906, -17.53595733642578, 10.361053466796875, -24.235610961914062, 84.721923828125, 59.60559844970703, 2.12005615234375, 124.63629150390625, 33.106197357177734, 330.8878173828125, 76.0518798828125, 65.69436645507812, 128.40863037109375, 394.3317565917969, -55.94602966308594, -120.82919311523438, 40.29254150390625, -97.80374145507812, -171.32400512695312, -122.40504455566406, 13.9134521484375, -84.4158706665039, 129.67657470703125, 41.6158447265625, 2.612640380859375, 35.0767822265625, -112.84927368164062, -11.726470947265625, -15.641357421875, -202.52798461914062, 14.500946044921875, -16.475830078125, -42.8248291015625, 145.85662841796875, 22.47576904296875, 141.19720458984375, 144.4991455078125, 118.20309448242188, -44.469398498535156, 1.4977188110351562, -2.132232666015625, -96.92874145507812, 71.70573425292969, -21.242351531982422, 100.10519409179688, 72.94213104248047, 82.58677673339844, -45.68800354003906, 222.6290283203125, -81.04290008544922, 159.33999633789062, -71.47586059570312, 187.72015380859375, 229.37667846679688, 2.20697021484375, -11.958076477050781, 216.70156860351562, 131.6969757080078, -182.07131958007812, 191.45303344726562, -51.491424560546875, -42.187774658203125, 200.55450439453125, -240.1708984375, 185.51602172851562, 244.42578125, -411.6578369140625, 181.2352294921875, 139.46730041503906, 37.67613983154297, 82.35208892822266, 91.18557739257812, 181.74725341796875, 163.6482696533203, 22.731582641601562, 292.77191162109375, 273.10089111328125, 13.877410888671875, 121.02446746826172, -62.72174072265625, 167.162353515625, -24.745895385742188, 63.19037628173828, -65.463134765625, 13.933013916015625, -3.2216033935546875, -32.945556640625, 74.0125732421875, 235.77566528320312, 11.089067459106445, -57.87908172607422, 178.9119415283203, 0.0, 12.552398681640625, 13.701875686645508, 52.915565490722656, -98.10263061523438, 74.21864318847656, 192.43182373046875, -144.11119079589844, 0.18161964416503906, 92.65484619140625, 451.2742919921875, 51.87681579589844, 103.41096496582031, 72.95257568359375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000438.npy"}
{"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 82.15447998046875, "std": 122.40888977050781, "min": -217.74465942382812, "p10": -53.59618682861328, "median": 72.66865539550781, "p90": 243.21727447509764, "max": 395.677490234375, "pos_frac": 0.734375, "sample": [-138.690673828125, -123.804443359375, -98.41848754882812, 387.5042724609375, 5.6220703125, 102.48123168945312, 173.85206604003906, 115.44339752197266, 214.62521362304688, 197.31149291992188, -12.007818222045898, 29.55718994140625, -13.891075134277344, 70.3306884765625, 209.1051025390625, -34.61993408203125, 10.226150512695312, 166.38172912597656, 5.9388885498046875, 29.763427734375, 203.92596435546875, 208.49412536621094, 29.45318603515625, 125.53836059570312, 18.522552490234375, 136.9766845703125, 105.20306396484375, 212.95440673828125, 30.111602783203125, 24.35338020324707, -15.636054992675781, 137.33079528808594, -23.384563446044922, 227.24954223632812, 15.776214599609375, 138.26187133789062, 329.12945556640625, 68.68710327148438, 354.7371520996094, 74.68701171875, -33.589263916015625, 42.83479690551758, 263.7474365234375, 249.53140258789062, -125.634765625, 242.3253631591797, 175.54971313476562, 141.86407470703125, -5.99566650390625, 61.933380126953125, -91.0589599609375, 52.467872619628906, -189.0668487548828, 213.5625, 174.9888916015625, 265.47015380859375, 76.46195220947266, -53.02703094482422, 35.44961166381836, 0.0, 17.042922973632812, 0.0, 27.31951904296875, -6.950958251953125, 179.639892578125, 285.08184814453125, 75.2969970703125, 92.56524658203125, -17.23698616027832, -18.04986572265625, 52.9783935546875, 247.47113037109375, -50.71142578125, -41.783660888671875, -54.924217224121094, 120.19422149658203, 32.34514617919922, -7.03045654296875, -157.39089965820312, -8.7506103515625, 17.606239318847656, 126.1744613647461, 40.21260070800781, 35.66557312011719, 179.25277709960938, 88.9617919921875, 66.67236328125, 12.635467529296875, -187.01084899902344, 101.84815979003906, 126.80497741699219, -146.23699951171875, 0.0, 204.1995849609375, 276.07244873046875, 110.86032104492188, 264.1809387207031, 195.28443908691406, 50.084197998046875, 24.70294189453125, 15.02862548828125, 136.39352416992188, -114.16424560546875, -0.643585205078125, 79.90057373046875, 97.18301391601562, 97.7825927734375, 216.2965087890625, 134.9286651611328, 100.27178955078125, 84.1204833984375, 133.997314453125, 176.376708984375, 268.3131103515625, 196.62876892089844, 0.0, -217.74465942382812, 217.54345703125, 128.376220703125, 395.677490234375, 245.29840087890625, 152.23301696777344, 70.65029907226562, -120.30677795410156, 234.05030822753906, 0.0, 172.23214721679688, 63.374549865722656], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000439.npy"}
{"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 60.95854187011719, "std": 116.2919692993164, "min": -346.92327880859375, "p10": -61.935684204101555, "median": 50.67587471008301, "p90": 214.6867767333984, "max": 384.34759521484375, "pos_frac": 0.6953125, "sample": [-141.51605224609375, 0.0, 92.1734619140625, -143.64944458007812, -2.9617557525634766, 30.46685791015625, -26.5157470703125, 59.50291442871094, -124.71669006347656, 55.823638916015625, 117.085693359375, -9.79852294921875, -56.457763671875, -7.27374267578125, 177.47021484375, 101.19561767578125, 3.5623855590820312, 116.20436096191406, 119.4835433959961, 156.06549072265625, 95.51470947265625, 179.11334228515625, -18.535690307617188, 10.835906982421875, -11.67010498046875, -52.314910888671875, -346.92327880859375, 101.428955078125, -55.324371337890625, 23.191802978515625, 264.0115966796875, 292.87835693359375, 14.039573669433594, -9.1395263671875, -242.63446044921875, -16.310516357421875, 203.95077514648438, -153.57513427734375, 0.0, 174.375244140625, -205.61895751953125, 5.22979736328125, -9.59783935546875, 46.20640182495117, 220.97415161132812, 8.379661560058594, -59.268157958984375, 8.135993957519531, 34.31438064575195, 184.55020141601562, 109.93641662597656, -17.908782958984375, 146.8450164794922, -53.53066635131836, 74.06304931640625, 155.66708374023438, -2.544118881225586, -68.49030303955078, -68.159912109375, 130.2444610595703, -102.18832397460938, 17.767333984375, 145.96636962890625, -36.630462646484375, 237.961669921875, 23.916900634765625, 384.34759521484375, 86.47642517089844, 24.831298828125, 83.0359115600586, 15.44963264465332, -18.3779296875, 6.320640563964844, 182.40322875976562, 228.09622192382812, 63.30457305908203, -11.258773803710938, -26.194080352783203, 147.74435424804688, 99.04013061523438, 12.32830810546875, 81.38162231445312, 210.69561767578125, 157.55587768554688, 245.12911987304688, -69.509033203125, 64.6448974609375, 121.5220947265625, 74.7550048828125, -1.536294937133789, 211.9921875, -9.213298797607422, 192.2969207763672, 0.702789306640625, 230.43463134765625, 20.032470703125, 56.11326599121094, 25.351158142089844, 262.77630615234375, 286.3741455078125, 123.6029052734375, 55.45063781738281, 29.272972106933594, 98.74044799804688, 100.5068359375, 38.5938720703125, 27.005950927734375, 280.94244384765625, 90.89019775390625, 149.6868896484375, 163.95416259765625, 139.40310668945312, -14.761962890625, 234.65985107421875, 105.56658935546875, 138.8509063720703, 179.196533203125, 9.92181396484375, 154.57530212402344, 99.58245849609375, -84.66336822509766, 248.42364501953125, 44.85993957519531, 55.145347595214844, -6.610107421875, 24.029312133789062, -172.37698364257812, 153.91848754882812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000440.npy"}
{"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 87.71729278564453, "std": 118.64030456542969, "min": -169.64434814453125, "p10": -69.39617691040037, "median": 80.83242797851562, "p90": 241.85914306640623, "max": 407.0641174316406, "pos_frac": 0.7578125, "sample": [-27.485397338867188, 30.42388916015625, -81.09368896484375, 169.55587768554688, 4.634754180908203, -85.4952392578125, 277.3907165527344, 124.810302734375, 119.96800231933594, -2.52203369140625, -142.14694213867188, 407.0641174316406, 142.11712646484375, 196.3138427734375, 95.01129150390625, 20.431304931640625, 89.47085571289062, 180.1910400390625, 163.24032592773438, 129.37295532226562, 64.205322265625, 192.80252075195312, 202.4141387939453, 67.04144287109375, 173.40631103515625, -54.74578857421875, 199.27978515625, 195.83860778808594, 104.3819580078125, -13.27880859375, 198.292236328125, 208.6439208984375, 64.30601501464844, -105.33467864990234, 169.89727783203125, 205.7225341796875, 51.4022216796875, 262.33929443359375, 229.93759155273438, -52.489891052246094, 124.11019897460938, 183.83306884765625, -30.966567993164062, -116.18560791015625, 179.0626220703125, -1.4526939392089844, 94.84664916992188, 173.5320281982422, 65.35702514648438, 151.4044189453125, 51.968589782714844, 2.38983154296875, 89.535400390625, -3.8395919799804688, 45.92466735839844, 56.101409912109375, -14.0987548828125, 3.617095947265625, 58.98069763183594, -96.79241943359375, -58.078094482421875, 22.675491333007812, 10.803596496582031, 240.9071044921875, 15.52410888671875, 0.0, -12.187698364257812, -25.660781860351562, 77.58069610595703, 219.95602416992188, 248.178955078125, -169.64434814453125, 11.886604309082031, 105.54222869873047, 0.0, 108.93138122558594, 74.34481811523438, 68.4691162109375, 139.5785369873047, 310.6103820800781, 56.51921081542969, 188.954345703125, 142.56158447265625, -147.90835571289062, 121.12667846679688, -64.55010986328125, 101.74566650390625, 147.93109130859375, -83.56208801269531, 79.0150146484375, 25.91424560546875, 82.64984130859375, 103.60293579101562, 105.79562377929688, 36.52871322631836, 51.83473205566406, 190.0044403076172, -80.70366668701172, 327.3304443359375, 40.6707763671875, 142.31100463867188, -100.84246826171875, 278.2430419921875, 119.65274047851562, -123.218505859375, -38.309844970703125, 305.0956726074219, -101.49411010742188, 123.14244079589844, -61.89947509765625, 63.40156555175781, 101.25515747070312, 266.15673828125, 346.8272705078125, 214.8284912109375, 136.51168823242188, 191.11825561523438, 17.3389892578125, -52.95892333984375, 13.177276611328125, 317.62939453125, 41.49653625488281, 139.98922729492188, 167.94012451171875, 353.6500244140625, 75.14689636230469, 244.08056640625, 40.02166748046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000441.npy"}
{"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 66.92411804199219, "std": 132.7486114501953, "min": -271.1363525390625, "p10": -108.33909606933592, "median": 54.2833251953125, "p90": 244.88180847167968, "max": 398.4538879394531, "pos_frac": 0.6953125, "sample": [264.9132080078125, 51.477569580078125, 14.495101928710938, -178.071533203125, 232.44140625, 129.67315673828125, 4.36851692199707, -91.24844360351562, 174.50933837890625, 16.728973388671875, 141.38377380371094, 89.50260925292969, -103.54217529296875, -19.427597045898438, 247.25833129882812, -82.35708618164062, 149.22967529296875, 53.122344970703125, 101.736083984375, -22.709144592285156, 28.67877197265625, 34.71770477294922, 31.444427490234375, -4.232166290283203, 262.1121826171875, -5.200469970703125, 122.22694396972656, -143.12564086914062, 266.9012451171875, 189.7991943359375, -75.6865234375, -70.375, 12.81585693359375, 234.11270141601562, 115.30484008789062, 138.54965209960938, -23.989028930664062, -75.6121826171875, -152.25433349609375, 58.962196350097656, 137.5874481201172, -39.937103271484375, 52.07221984863281, 77.55923461914062, 57.1370849609375, -223.5343017578125, 55.405303955078125, 115.55828857421875, 40.088844299316406, -89.85284423828125, 103.17781066894531, 114.26022338867188, 105.13690948486328, 157.00408935546875, 152.3699493408203, 54.78436279296875, 90.82501220703125, 0.0, 23.500778198242188, -12.864486694335938, -126.173583984375, -17.613174438476562, -177.89541625976562, 137.78863525390625, -271.1363525390625, 207.62026977539062, 197.76223754882812, 17.32489013671875, 100.7239990234375, 154.99447631835938, -15.59443473815918, -145.04705810546875, 268.83905029296875, 73.15032958984375, 226.19284057617188, 146.60421752929688, 272.58837890625, 53.7235107421875, 27.72564697265625, -3.353546142578125, 68.65818786621094, 48.955841064453125, -7.369007110595703, 398.4538879394531, 208.22665405273438, 329.808837890625, -115.17471313476562, -55.828216552734375, -71.28858947753906, 28.41777801513672, 116.53814697265625, 13.616012573242188, -214.70877075195312, 295.84271240234375, -44.35408020019531, 152.51339721679688, 355.7530212402344, 174.09466552734375, -43.693756103515625, 251.26483154296875, 236.27908325195312, 182.08294677734375, 231.01971435546875, 246.34613037109375, 12.712112426757812, 368.13641357421875, -16.55908203125, -116.06423950195312, -186.1739501953125, 214.79373168945312, 161.61993408203125, 78.75822448730469, -19.109161376953125, 194.18756103515625, 126.21636962890625, 42.87425231933594, 6.28955078125, 53.311279296875, 32.109619140625, -116.264892578125, 139.882568359375, 87.46855163574219, 244.25424194335938, 53.78228759765625, -105.4095458984375, 22.08319091796875, 129.83885192871094, 154.95570373535156], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000442.npy"}
{"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 77.63595581054688, "std": 123.90055084228516, "min": -272.7330627441406, "p10": -60.0868049621582, "median": 76.11209869384766, "p90": 233.2900146484375, "max": 382.47467041015625, "pos_frac": 0.734375, "sample": [191.06674194335938, 10.25909423828125, 188.84022521972656, 129.48890686035156, -9.124267578125, 179.26007080078125, 252.667724609375, 8.857574462890625, 289.18145751953125, -129.28512573242188, -4.926483154296875, 53.6025390625, -5.187904357910156, 228.45590209960938, -33.722137451171875, 86.56405639648438, -48.867950439453125, 2.5510330200195312, 133.3508758544922, 35.78730773925781, 162.4671630859375, 181.17440795898438, -22.273117065429688, 121.3721923828125, 44.3072509765625, 91.24894714355469, 152.5509033203125, 198.5196990966797, 154.6928253173828, 8.032218933105469, -209.22802734375, 0.70147705078125, -213.7679443359375, 21.649436950683594, -33.49365234375, 48.454620361328125, 164.92901611328125, -14.115829467773438, 88.63335418701172, -2.522968292236328, 187.92874145507812, -127.7608642578125, 22.300064086914062, -0.177001953125, 115.38023376464844, 150.68045043945312, 281.0199279785156, -39.80909729003906, 120.44522094726562, -40.6175537109375, 103.82891845703125, -190.04257202148438, -3.5252227783203125, 118.3680419921875, 131.00311279296875, 249.4656982421875, 63.17500305175781, 350.1905212402344, 46.84442138671875, 233.4525146484375, 74.09283447265625, 232.07962036132812, 62.684722900390625, 101.40470886230469, 163.84353637695312, 151.05194091796875, 30.747451782226562, -150.60888671875, 168.24000549316406, 178.91986083984375, -11.646896362304688, 40.77299499511719, 146.25762939453125, 86.23761749267578, 335.692626953125, 75.120849609375, -33.975921630859375, 160.21090698242188, -119.1190185546875, -18.835601806640625, 140.52745056152344, 165.4114990234375, 147.08544921875, 10.374385833740234, 153.56655883789062, -112.24095916748047, 211.5625457763672, -58.778770446777344, 105.10220336914062, 233.44027709960938, 223.978759765625, 233.22561645507812, -81.79295349121094, 267.1612548828125, -10.751739501953125, 116.15489196777344, 12.61932373046875, 382.47467041015625, 152.84629821777344, 147.86288452148438, 107.51026916503906, 27.544586181640625, -182.26992797851562, 77.10334777832031, 129.5494384765625, -49.98297119140625, 87.53936767578125, 119.34359741210938, 54.297943115234375, 21.086868286132812, 180.74960327148438, 68.61701965332031, -63.138885498046875, 116.24256134033203, 38.74937438964844, -38.668678283691406, 29.416656494140625, 267.81072998046875, 57.402706146240234, -272.7330627441406, 215.29498291015625, 268.9909973144531, 321.0130615234375, -43.895042419433594, 54.03338623046875, -130.25973510742188, 19.287033081054688, 69.46392822265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000443.npy"}
{"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 85.54299926757812, "std": 138.1156463623047, "min": -300.41864013671875, "p10": -83.71796569824218, "median": 72.59943771362305, "p90": 254.78377380371091, "max": 533.528564453125, "pos_frac": 0.7421875, "sample": [-20.28820037841797, 174.214111328125, 180.807373046875, 1.151611328125, 242.55401611328125, 114.91999816894531, -148.24685668945312, 42.254608154296875, 59.55689239501953, 84.99176025390625, -23.310943603515625, 92.01513671875, -135.53857421875, 186.088623046875, 102.92813110351562, 93.64007568359375, 186.8995361328125, 171.98045349121094, -60.494781494140625, 45.14305877685547, 127.28395080566406, 47.917510986328125, 172.03118896484375, -94.6705322265625, 11.312652587890625, -96.582275390625, 227.81396484375, 31.92055892944336, 19.132644653320312, 12.547515869140625, 101.42056274414062, 141.67538452148438, 233.21981811523438, -52.4088134765625, 421.07403564453125, -155.13339233398438, -54.24993896484375, 124.8515625, 229.9306640625, 261.7497863769531, 70.3919677734375, 184.1365966796875, -21.56500244140625, -41.445255279541016, -47.078887939453125, 216.78302001953125, 14.4979248046875, 71.54976654052734, -34.65631103515625, -109.17318725585938, 247.59027099609375, 64.3094482421875, 56.92735290527344, 294.1699523925781, 29.704792022705078, 251.79833984375, 277.1473388671875, 334.427734375, 12.91693115234375, 93.72674560546875, 10.694849014282227, 388.36248779296875, 131.61024475097656, 13.30157470703125, 103.79816436767578, -81.853759765625, 124.432373046875, 209.29214477539062, 104.81672668457031, 374.0826416015625, 206.41143798828125, 119.9546127319336, -97.86773681640625, 207.98931884765625, 73.64910888671875, 59.19512939453125, 180.59915161132812, 181.60430908203125, 224.77407836914062, 110.3052978515625, 533.528564453125, -93.99028778076172, 8.378501892089844, 379.90057373046875, 26.469900131225586, 279.947998046875, 80.99742889404297, -94.55874633789062, 223.1221160888672, -163.65805053710938, -88.06777954101562, 219.28390502929688, 88.24413299560547, 62.149505615234375, 38.6585693359375, 36.49530029296875, -36.71356201171875, 125.34054565429688, 119.63134765625, -48.02301025390625, -35.3310546875, 41.943870544433594, 275.072998046875, 32.95415496826172, -43.385284423828125, 175.0166015625, 18.44769859313965, 177.3163604736328, 41.159942626953125, 323.7634582519531, -77.23588562011719, 123.2203369140625, -26.220184326171875, 173.64736938476562, 85.96881103515625, 25.01482582092285, 208.73614501953125, 275.1675720214844, -66.16058349609375, 110.27010345458984, -9.748350143432617, 35.64024353027344, -33.31890869140625, -31.399816513061523, -214.23910522460938, 63.2520751953125, 189.81822204589844, -300.41864013671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000444.npy"}
{"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 73.35039520263672, "std": 137.0552520751953, "min": -322.341552734375, "p10": -77.18220977783203, "median": 66.3267993927002, "p90": 233.5085220336914, "max": 632.1759033203125, "pos_frac": 0.734375, "sample": [58.612159729003906, -240.28297424316406, 204.15757751464844, 47.2711181640625, 68.66183471679688, 338.5577392578125, -35.227783203125, 85.13599395751953, 222.93922424316406, -182.92001342773438, 57.85059356689453, 235.58766174316406, 58.83855438232422, -69.50982666015625, 131.91189575195312, 39.786712646484375, 24.35162353515625, 90.69099426269531, 133.343994140625, 29.44225311279297, 632.1759033203125, 42.84686279296875, 49.18499755859375, -125.14496612548828, -1.55499267578125, 80.78529357910156, 112.4388427734375, 3.449085235595703, 228.95042419433594, 83.34689331054688, 205.443359375, 130.510009765625, 8.018798828125, 26.882415771484375, -207.94729614257812, -21.226852416992188, 17.853023529052734, 104.1445541381836, 272.5932312011719, 96.82614135742188, 10.3311767578125, 207.13975524902344, 283.9598388671875, 4.25244140625, 192.98101806640625, 76.11781311035156, -11.091644287109375, 218.8106689453125, 5.570762634277344, -80.39974975585938, -98.04052734375, 26.38421630859375, 107.6304931640625, 17.863357543945312, 191.57730102539062, 13.822647094726562, 85.3099365234375, -30.11212158203125, -19.494384765625, 356.34552001953125, 71.18417358398438, 219.98977661132812, 309.46337890625, 154.58676147460938, 158.87496948242188, 186.30471801757812, 85.76115417480469, -322.341552734375, 232.61746215820312, 291.94720458984375, 102.551025390625, 115.75537109375, -13.9969482421875, 171.09994506835938, 88.41563415527344, 72.14070129394531, 21.518798828125, -154.96389770507812, 114.41802978515625, 174.51536560058594, 81.84439086914062, -21.201614379882812, -75.85335540771484, -32.3450927734375, 63.991764068603516, -45.68632507324219, 138.7415313720703, -78.27810668945312, 123.60398864746094, -14.541976928710938, -33.755462646484375, 132.37274169921875, 143.61944580078125, 116.25750732421875, 88.02011108398438, 142.86968994140625, -4.210487365722656, 51.973724365234375, 196.77139282226562, 0.5540771484375, 81.9005355834961, 13.121688842773438, 281.0499267578125, 172.64297485351562, -40.98236083984375, 143.90975952148438, 9.783126831054688, 70.4413070678711, 362.3695068359375, -11.62298583984375, -74.24057006835938, 130.3938751220703, 0.0, -8.804512023925781, 46.71223449707031, -109.76800537109375, 45.681480407714844, 57.022705078125, 62.069427490234375, 299.8722229003906, -102.29522705078125, -216.7344970703125, -172.99298095703125, 0.07495307922363281, 96.3699951171875, 279.64520263671875, -76.71253967285156, 397.61968994140625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000445.npy"}
{"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 80.4897232055664, "std": 130.60394287109375, "min": -274.79803466796875, "p10": -41.56148223876953, "median": 72.40709686279297, "p90": 238.25934753417965, "max": 612.8271484375, "pos_frac": 0.7421875, "sample": [265.4187927246094, 215.59197998046875, 139.53504943847656, -12.20684814453125, 41.864105224609375, 53.161773681640625, 165.6602783203125, -23.340293884277344, 72.78263854980469, -167.37054443359375, -27.499298095703125, 9.728607177734375, 25.643295288085938, 15.094024658203125, 102.1771240234375, 47.75065612792969, 171.10211181640625, 13.606803894042969, -75.55217742919922, 14.056785583496094, -33.937225341796875, 72.03155517578125, 124.95062255859375, 129.65704345703125, 30.36681365966797, -106.53074645996094, 229.7510528564453, -9.994789123535156, -13.895095825195312, 11.34222412109375, -2.101287841796875, 21.98724365234375, -192.03201293945312, 51.84947967529297, -197.48321533203125, 61.17371368408203, 282.4580078125, 52.96836853027344, 128.17257690429688, 111.3718490600586, 132.114990234375, 93.79281616210938, 141.61509704589844, 61.279510498046875, 48.58111572265625, 186.60791015625, 75.06060791015625, 283.4533386230469, 279.875, -261.8097229003906, 89.53689575195312, 120.24365234375, 235.9820556640625, -8.391395568847656, 167.66387939453125, 203.5286865234375, 145.6686248779297, 204.70889282226562, -274.79803466796875, 10.697494506835938, 162.63955688476562, -44.080322265625, 95.09674072265625, 0.0, 268.31658935546875, 181.20199584960938, 55.480194091796875, 135.7017364501953, 612.8271484375, 183.77114868164062, -87.8175048828125, 277.49493408203125, 163.34225463867188, 94.37261962890625, 105.3984375, 211.53343200683594, 243.57302856445312, 147.1022491455078, -2.56707763671875, 0.0, 128.62454223632812, 13.482406616210938, 262.8397521972656, 10.971832275390625, 90.11156463623047, -25.36077880859375, 49.43359375, 219.71875, 112.55938720703125, 177.7357940673828, -71.72567749023438, 467.9343566894531, 104.991943359375, 172.05043029785156, 162.468505859375, 129.00985717773438, 149.99090576171875, -16.90386962890625, 47.299644470214844, 220.0277099609375, 66.4433822631836, -10.5870361328125, 253.2421875, 255.8631591796875, -215.91006469726562, -166.87741088867188, 1.4566421508789062, 181.13058471679688, -0.904876708984375, -10.68896484375, -2.084442138671875, 0.0, 21.342620849609375, 133.04913330078125, 167.1103973388672, -94.82516479492188, 128.30992126464844, -39.74456787109375, 316.1482238769531, 21.191360473632812, 22.249267578125, 79.93191528320312, 12.666397094726562, -40.48197937011719, 150.35153198242188, 11.905464172363281, 77.33769989013672, 41.688690185546875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000446.npy"}
{"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 94.25474548339844, "std": 123.9744873046875, "min": -217.45257568359375, "p10": -46.17676696777343, "median": 85.45893478393555, "p90": 254.1689819335937, "max": 501.56927490234375, "pos_frac": 0.78125, "sample": [-19.482681274414062, 4.416128158569336, 248.1510009765625, 186.3388671875, -48.61358642578125, 31.252933502197266, 352.2177429199219, 120.3559341430664, -37.220703125, 289.41693115234375, 167.67498779296875, 121.50205993652344, 62.462646484375, 382.503173828125, 172.9223175048828, 161.48690795898438, 0.0, 0.273834228515625, 165.2183837890625, 29.389163970947266, 42.00433349609375, 91.097412109375, 234.8203125, 100.88153076171875, 116.79629516601562, 31.487823486328125, 44.4263916015625, 84.95113372802734, -41.8985595703125, 321.2368469238281, 82.84717559814453, 79.28253173828125, 187.80926513671875, 168.3633575439453, 113.595703125, 212.54959106445312, 163.96629333496094, 181.24156188964844, 46.535499572753906, 300.7252197265625, -9.29339599609375, 28.66705322265625, -99.19955444335938, 191.25811767578125, 68.4420166015625, -10.86151123046875, 167.06582641601562, 218.005615234375, 122.28459930419922, 0.0, 75.92999267578125, 47.46137237548828, 87.52182006835938, 86.12882232666016, -59.81781005859375, -121.2003173828125, -178.6800537109375, 160.08404541015625, 83.7195053100586, 276.7651672363281, 64.13809204101562, 54.400840759277344, -81.754150390625, 69.34695434570312, -134.38125610351562, 214.08236694335938, 290.7439270019531, 31.334609985351562, -12.69049072265625, 77.24813842773438, 204.32244873046875, 48.3922119140625, -173.73606872558594, 63.38726806640625, 121.85940551757812, 4.138214111328125, -68.48551940917969, 102.33012390136719, 40.7872314453125, 156.66635131835938, 245.57522583007812, -15.21185302734375, -34.81840515136719, 90.62225341796875, -217.45257568359375, 0.0, 38.004493713378906, 199.49513244628906, -54.29191589355469, 161.10931396484375, 85.96673583984375, 81.39068603515625, 57.0819091796875, 61.677001953125, 238.31227111816406, 93.39419555664062, 268.357666015625, 398.0976257324219, 0.0, 137.8684539794922, 121.89704895019531, 146.8958740234375, 39.06817626953125, 89.02865600585938, 42.340911865234375, 156.212890625, 41.041595458984375, -208.84532165527344, 222.45230102539062, 94.6529541015625, 40.56536865234375, -16.859268188476562, 195.69027709960938, 311.92205810546875, 501.56927490234375, 134.065673828125, 19.44580841064453, 37.1927490234375, -45.132415771484375, -7.190305709838867, 129.183837890625, 268.2109375, 161.54356384277344, 108.39643859863281, 128.4610595703125, 321.2489929199219, 121.9254150390625, -84.95294189453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000447.npy"}
{"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 85.2490234375, "std": 122.41460418701172, "min": -179.8521728515625, "p10": -55.79908447265625, "median": 70.8840560913086, "p90": 250.7902069091797, "max": 462.12158203125, "pos_frac": 0.75, "sample": [78.73867797851562, 90.54408264160156, 58.62347412109375, 32.56280517578125, 138.76336669921875, 200.764892578125, 176.75, -123.93206787109375, 71.87928771972656, 137.976318359375, -54.090576171875, 8.91644287109375, 69.4727783203125, 62.4189453125, 23.98455810546875, 78.27850341796875, 83.4156494140625, 193.09521484375, 42.60662841796875, -2.986299514770508, 82.08212280273438, 320.2655334472656, 280.0203857421875, -30.165332794189453, -36.559852600097656, 48.05497741699219, 61.113487243652344, -9.815120697021484, 26.17108154296875, -54.355682373046875, -45.828330993652344, 392.3677062988281, 222.35317993164062, 27.80767822265625, 174.26473999023438, -100.99493408203125, 139.80438232421875, 403.34014892578125, 86.6407470703125, 21.780059814453125, 95.05313110351562, 115.74589538574219, 170.23150634765625, 100.43431854248047, -45.8143310546875, 175.33880615234375, 203.73699951171875, -15.371498107910156, 49.57121276855469, 354.81451416015625, 22.794837951660156, 102.72299194335938, 4.72857666015625, 135.5196990966797, 165.01046752929688, 19.592864990234375, 43.36500549316406, 191.46603393554688, 65.45025634765625, 242.1077880859375, 237.8271942138672, 128.14389038085938, 72.16168212890625, 21.42304229736328, 208.4298095703125, 296.66668701171875, 75.49331665039062, 77.4085693359375, 257.5008850097656, 327.87890625, 278.55023193359375, 204.99826049804688, 27.623519897460938, 160.50177001953125, 21.931617736816406, 57.87225341796875, 250.25274658203125, 63.274078369140625, 125.74459838867188, 114.84823608398438, -118.32174682617188, 257.9112243652344, -110.09748077392578, 254.900390625, 69.88882446289062, -41.337982177734375, -134.18539428710938, -179.8521728515625, 233.9998779296875, -5.5302581787109375, -59.167022705078125, -4.587890625, 116.73187255859375, 156.35964965820312, 198.5995330810547, -70.319580078125, -42.58111572265625, -116.04810333251953, 55.4947509765625, 200.83447265625, 252.04428100585938, 22.096637725830078, 175.4024200439453, -30.698020935058594, 240.66046142578125, -63.27301025390625, -133.6697540283203, -96.99967956542969, 193.82403564453125, -63.60841369628906, 133.60928344726562, 58.905029296875, 73.72013854980469, 56.13800048828125, 16.454559326171875, 28.28363037109375, 39.64825439453125, -41.759857177734375, 93.07273864746094, 0.16026687622070312, 89.67723083496094, -31.94183349609375, -0.008199691772460938, 122.44346618652344, 462.12158203125, -33.92181396484375, 106.6260986328125, -0.9835968017578125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000448.npy"}
{"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 83.43479919433594, "std": 129.0272674560547, "min": -450.89923095703125, "p10": -82.4155174255371, "median": 101.67987060546875, "p90": 228.93702697753906, "max": 390.11578369140625, "pos_frac": 0.78125, "sample": [168.37265014648438, 4.7684173583984375, -20.177047729492188, 117.65901184082031, 87.58306884765625, -22.8798828125, 16.790435791015625, 224.8116455078125, -10.669471740722656, -17.152908325195312, 25.169052124023438, 50.14208984375, 148.3369903564453, 50.574615478515625, 46.74638366699219, 308.5951843261719, 227.78561401367188, 199.59552001953125, 19.541015625, -4.522472381591797, -84.246826171875, 160.659912109375, 210.97735595703125, 207.57130432128906, 140.7430419921875, 77.4732666015625, -52.559906005859375, 103.88796997070312, 10.852069854736328, 163.19775390625, 307.56134033203125, 312.5227966308594, 233.05133056640625, 136.40249633789062, -16.69287109375, 183.3756103515625, 205.95330810546875, 116.06976318359375, 171.49209594726562, 118.04901123046875, 220.69305419921875, 76.36224365234375, 197.8258056640625, 210.54803466796875, 184.5206298828125, 263.77020263671875, 42.139251708984375, 135.9058837890625, 294.70648193359375, 24.597412109375, -134.75350952148438, 40.123046875, -36.87164306640625, -34.92475891113281, 150.3151397705078, 155.76516723632812, 56.98109436035156, 27.503738403320312, 39.68426513671875, -153.86151123046875, 6.334175109863281, 143.4072723388672, 129.89181518554688, 105.64437866210938, 166.54115295410156, 126.98921966552734, 77.58224487304688, 122.63536071777344, 103.20697021484375, 147.2930908203125, -123.89263916015625, 152.6136016845703, 111.28033447265625, 122.46923065185547, -192.687255859375, 13.43484115600586, 23.855979919433594, 142.54476928710938, 82.24632263183594, 238.02908325195312, 309.3330383300781, 112.55962371826172, 257.5259704589844, -28.16192626953125, 225.29537963867188, -3.221771240234375, 236.1297607421875, 26.497100830078125, 167.76681518554688, -52.34454345703125, 22.89935302734375, 192.14288330078125, -154.58013916015625, 231.6236572265625, 142.5289306640625, 155.94866943359375, -135.71932983398438, -35.780479431152344, 91.29379272460938, 100.15277099609375, -276.6343994140625, 214.18992614746094, 1.3271942138671875, -450.89923095703125, -154.58944702148438, 166.36346435546875, -145.5928955078125, 71.87378692626953, 175.74969482421875, 69.4697265625, 19.08422088623047, 69.24798583984375, -83.54029083251953, 5.105873107910156, 178.20938110351562, 48.07716369628906, 191.38284301757812, 59.01719665527344, 17.662139892578125, 390.11578369140625, 113.12322998046875, -90.62272644042969, 267.8623352050781, 169.88641357421875, 107.18829345703125, -69.31272888183594, -81.9334716796875, 48.09327697753906], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000449.npy"}
{"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 74.2271957397461, "std": 116.64873504638672, "min": -246.7523193359375, "p10": -63.3543701171875, "median": 58.202972412109375, "p90": 239.40752563476562, "max": 430.66778564453125, "pos_frac": 0.7109375, "sample": [0.0, -3.876556396484375, 197.05035400390625, -81.05606079101562, 6.50347900390625, 15.88409423828125, 42.547489166259766, -49.4560546875, 211.3314208984375, 53.0230712890625, 47.984222412109375, 368.1371154785156, 16.31787109375, 58.005615234375, 254.29025268554688, 39.596893310546875, -26.82825469970703, 132.92446899414062, 117.16983032226562, -24.146286010742188, 60.46312713623047, 106.89495849609375, 159.69558715820312, 214.9285125732422, 167.26004028320312, 156.51199340820312, -18.117294311523438, 57.04071044921875, -86.11711120605469, 82.96792602539062, -119.10537719726562, -6.2794189453125, 38.15618896484375, -13.554428100585938, 81.09701538085938, -40.9879150390625, 212.14630126953125, -35.442169189453125, 54.767120361328125, 19.89642333984375, 124.99153137207031, 166.1790771484375, 73.31788635253906, 132.12486267089844, -16.044219970703125, 53.25050354003906, 0.41707611083984375, 71.42745971679688, 228.47360229492188, 144.11538696289062, 37.830101013183594, -4.5094451904296875, 22.445079803466797, 37.1903076171875, 291.8594970703125, 71.01821899414062, 127.74435424804688, 197.82015991210938, 244.81961059570312, -91.838134765625, 103.76229858398438, -167.74581909179688, 241.1375732421875, -67.01242065429688, 54.3150634765625, 292.2183837890625, 235.59622192382812, 129.24839782714844, 147.99468994140625, 190.9138641357422, -24.182754516601562, 58.77306365966797, -28.61322021484375, 250.1983642578125, 78.12199401855469, 56.341087341308594, 155.3544921875, 252.43359375, -27.926620483398438, 257.7899169921875, -69.16461181640625, 156.63851928710938, -77.76171112060547, 29.11419677734375, 257.157958984375, 61.862548828125, 8.10004997253418, 69.84225463867188, -57.40612030029297, -0.824462890625, 61.475189208984375, 24.810569763183594, 430.66778564453125, 273.21685791015625, -34.962677001953125, 280.1428527832031, -31.77752685546875, -108.3262939453125, -31.42919921875, -238.9302978515625, 56.2906494140625, 111.33746337890625, 181.79583740234375, 76.08251953125, 238.66607666015625, 71.99172973632812, -246.7523193359375, 145.2715301513672, 146.283203125, 115.79608154296875, 96.48294067382812, 44.539451599121094, 58.40032958984375, -17.23223876953125, 190.09371948242188, -5.618999481201172, -65.42266845703125, 166.039306640625, 212.56576538085938, 159.66058349609375, 0.0, -62.46795654296875, 46.730377197265625, 126.95948791503906, 7.32574462890625, 134.1453399658203, -67.00668334960938, 7.7015838623046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000450.npy"}
{"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 68.1118392944336, "std": 127.73899841308594, "min": -256.3240966796875, "p10": -87.54966735839844, "median": 58.68955039978027, "p90": 236.0334777832031, "max": 440.87945556640625, "pos_frac": 0.7109375, "sample": [108.4493408203125, 152.7372283935547, -115.07443237304688, -27.129493713378906, 286.9991149902344, 142.36236572265625, 175.1616973876953, 10.0677490234375, -147.23892211914062, -35.7235107421875, 39.71832275390625, 90.40239715576172, 149.72177124023438, 17.94677734375, 233.42303466796875, 79.9097900390625, 68.64566040039062, 16.025299072265625, 279.0583190917969, -89.0648193359375, 328.55230712890625, -30.833221435546875, 0.820892333984375, 22.41778564453125, 178.70748901367188, 18.0999755859375, 440.87945556640625, -15.960189819335938, 70.22543334960938, 132.0052490234375, -167.66893005371094, 65.53106689453125, -256.3240966796875, 185.70809936523438, 95.29617309570312, 0.192291259765625, 139.47218322753906, 75.89895629882812, 187.39984130859375, 184.44503784179688, 106.08395385742188, 264.2055358886719, 232.27114868164062, 92.27740478515625, 97.55089569091797, 171.13949584960938, -110.40607452392578, 14.713336944580078, 316.4786071777344, -27.18408203125, -0.55316162109375, 268.26702880859375, -122.89646911621094, 375.2404479980469, 45.331024169921875, -22.129150390625, -10.746650695800781, 188.338134765625, -181.40325927734375, 112.79409790039062, 64.67829895019531, 157.6383819580078, 32.948272705078125, 165.73019409179688, -4.702369689941406, 9.707977294921875, -30.181869506835938, -44.65546417236328, 276.9465637207031, 134.47467041015625, -33.874664306640625, -0.4478759765625, -35.796142578125, 114.45205688476562, 14.210113525390625, -211.2548828125, 14.969573974609375, -88.08917236328125, 48.82171630859375, 102.28472900390625, 182.999755859375, -16.98369598388672, 26.1943359375, 209.0888214111328, 227.80001831054688, 34.768531799316406, 59.03076171875, 195.33245849609375, 9.614742279052734, 3.4863510131835938, 156.33123779296875, 50.32456970214844, -240.38970947265625, 177.5967559814453, -29.514175415039062, -127.7227783203125, -55.66993713378906, 0.5285797119140625, -38.09510803222656, 183.88168334960938, 67.408203125, 173.13815307617188, 242.12451171875, 325.45880126953125, -85.66061401367188, 31.746627807617188, -87.31845092773438, 113.01429748535156, 101.07861328125, 73.95051574707031, 171.7027587890625, -9.902740478515625, 66.2247314453125, 119.97163391113281, 6.030998229980469, 58.34833908081055, 130.7706298828125, 20.800750732421875, -29.293922424316406, 302.58935546875, 102.84616088867188, -19.754241943359375, -150.1641082763672, 36.19961166381836, 16.716476440429688, 101.95260620117188, -1.04052734375, 242.2767791748047], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000451.npy"}
{"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 90.01078796386719, "std": 144.17959594726562, "min": -274.6812438964844, "p10": -61.02685089111328, "median": 73.75714874267578, "p90": 260.15851135253905, "max": 807.888916015625, "pos_frac": 0.7578125, "sample": [21.6497802734375, 165.02267456054688, 246.5455322265625, 807.888916015625, -183.15577697753906, 119.4336929321289, 29.3719482421875, 278.13446044921875, -168.89183044433594, -57.759437561035156, 280.7374267578125, 56.43975830078125, 57.40643310546875, 55.8648681640625, -65.38090515136719, 29.5726318359375, 194.88943481445312, 21.168426513671875, -22.57232666015625, 160.14620971679688, 7.0982818603515625, -87.14955139160156, -83.63186645507812, 128.43167114257812, 74.72119140625, 72.90618896484375, 56.2518310546875, 49.65080261230469, 106.63186645507812, 15.032669067382812, 28.584136962890625, 192.36959838867188, -19.523235321044922, 261.187255859375, 161.33944702148438, -93.08843994140625, 428.2233581542969, 184.187744140625, -34.71562194824219, 76.02006530761719, 133.1746826171875, -19.42999267578125, 21.070648193359375, 26.57696533203125, 188.42919921875, -4.878589630126953, 218.99085998535156, 155.74957275390625, 13.188278198242188, 54.52618408203125, 205.426025390625, 125.819580078125, 147.01910400390625, 3.5534210205078125, 33.627685546875, -159.74453735351562, 171.77943420410156, -18.444091796875, 186.7760009765625, 338.28759765625, 105.6790542602539, 68.53378295898438, 19.5291748046875, 104.546875, -100.24856567382812, 338.89862060546875, 74.60810852050781, 214.81005859375, 218.27874755859375, 6.01983642578125, 113.37429809570312, 239.6864013671875, 146.68930053710938, 21.543212890625, 163.99557495117188, 319.811767578125, 263.72314453125, -139.23477172851562, 66.618896484375, 197.65867614746094, -274.6812438964844, 238.2159423828125, 262.887451171875, 11.06781005859375, 259.7176208496094, 107.78589630126953, -35.158058166503906, 176.56727600097656, 123.5885009765625, 4.3970947265625, 96.96795654296875, 431.605712890625, 42.477294921875, 11.814285278320312, -50.75071716308594, -59.16082763671875, 38.81310272216797, 168.9596710205078, 78.9832763671875, -56.9786376953125, 283.4551086425781, 69.638916015625, 76.01687622070312, 168.28802490234375, 64.10005950927734, 13.764467239379883, -30.15386962890625, 96.46774291992188, 182.35269165039062, 412.24896240234375, 239.323974609375, 185.44363403320312, -55.85163879394531, 76.60565185546875, 203.36688232421875, 0.0, 169.858642578125, -8.73086166381836, -97.6630859375, -114.17245483398438, 114.75869750976562, 80.00723266601562, -175.67556762695312, -5.02447509765625, 20.054962158203125, 202.42324829101562, -32.677886962890625, -11.024921417236328], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000452.npy"}
{"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 81.39653015136719, "std": 129.8886260986328, "min": -233.2679443359375, "p10": -75.34417114257812, "median": 78.08588790893555, "p90": 245.55182037353512, "max": 516.5848999023438, "pos_frac": 0.71875, "sample": [77.66051483154297, 133.7652587890625, -115.32489013671875, 3.10540771484375, 392.4197998046875, 20.628387451171875, 55.366363525390625, -10.991416931152344, 162.10842895507812, 162.880126953125, 107.40451049804688, -73.7518310546875, 101.61660766601562, 119.24505615234375, -22.25323486328125, -10.948501586914062, 98.30813598632812, -126.66534423828125, 27.03948974609375, 277.7117919921875, -81.51919555664062, 253.38609313964844, 104.36517333984375, 80.08808898925781, 123.259033203125, 73.1619873046875, -81.1650390625, 359.8028564453125, 67.15237426757812, 274.9337158203125, 180.66989135742188, 204.64804077148438, -7.4706573486328125, 185.93934631347656, 146.13832092285156, 213.5005645751953, 30.40283203125, -14.51953125, 259.0833740234375, 161.164794921875, 72.26448059082031, 347.417236328125, 171.25469970703125, -24.960723876953125, 10.580177307128906, -170.025634765625, 44.934814453125, 60.41058349609375, 41.0657958984375, -56.771087646484375, 37.430908203125, 157.78411865234375, -37.120521545410156, 1.3294754028320312, -67.016357421875, 27.5107421875, -68.93771362304688, -40.862640380859375, 267.28192138671875, 66.00547790527344, 118.8974609375, 17.54900360107422, 0.0, 89.11810302734375, 220.2326202392578, 20.47376251220703, 125.02363586425781, 160.65826416015625, 38.592437744140625, 24.80559539794922, 36.5623779296875, 123.12786865234375, 109.80593872070312, -171.71231079101562, 43.19140625, 144.75933837890625, 168.44573974609375, 171.61849975585938, -0.2201557159423828, 127.15582275390625, -25.127037048339844, 335.662353515625, 239.61224365234375, -43.5970458984375, 190.64239501953125, -87.38507080078125, 218.30136108398438, -79.05963134765625, 113.10528564453125, 345.62652587890625, -73.11178588867188, 212.27651977539062, -33.55120849609375, -81.3486328125, 193.37152099609375, 16.042388916015625, -140.7889404296875, 54.24017333984375, -5.712615966796875, 516.5848999023438, 167.9820098876953, 24.889739990234375, -27.05646514892578, 114.16305541992188, 78.51126098632812, 116.60125732421875, -105.34263610839844, -3.7643890380859375, 305.0428466796875, -223.91278076171875, 99.65153503417969, 152.1606903076172, 38.52166748046875, 189.77865600585938, 169.73178100585938, 78.89834594726562, 197.10499572753906, 286.2480163574219, -27.69378662109375, 82.21647644042969, 153.1986083984375, 13.336692810058594, 83.24774169921875, -60.14825439453125, 185.03506469726562, -233.2679443359375, 135.63584899902344, 242.19427490234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000453.npy"}
{"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 93.43684387207031, "std": 138.25462341308594, "min": -343.939208984375, "p10": -55.89330139160156, "median": 81.09077835083008, "p90": 265.19740600585936, "max": 546.2496337890625, "pos_frac": 0.7890625, "sample": [28.118408203125, 184.39193725585938, -157.46905517578125, -12.143539428710938, -67.754638671875, 65.80126953125, 203.2601318359375, 155.62030029296875, 66.94308471679688, 220.23507690429688, 143.4019775390625, -0.8510360717773438, 1.5577125549316406, 147.61422729492188, 180.28070068359375, 158.43246459960938, -200.58291625976562, 25.643945693969727, 1.1558380126953125, 8.795372009277344, 78.339111328125, 12.716865539550781, 241.63616943359375, 275.11505126953125, 318.511474609375, 252.8934326171875, -8.70880126953125, 307.8184814453125, -311.61761474609375, 94.72821044921875, 81.72879791259766, 155.54412841796875, 67.54907989501953, 127.58456420898438, 165.68228149414062, 31.602828979492188, 171.0169677734375, -343.939208984375, 58.01611328125, 4.303985595703125, 141.74603271484375, 111.23233032226562, 214.608642578125, 163.05239868164062, 137.52633666992188, 139.60491943359375, 224.63943481445312, 22.108154296875, 220.70738220214844, 54.13165283203125, 116.56167602539062, 443.80810546875, 264.4510498046875, -91.05863189697266, 96.62612915039062, -28.10345458984375, 77.45254516601562, 180.7876739501953, 323.43011474609375, 308.34454345703125, 153.7479705810547, 149.3258056640625, 291.642333984375, 39.836181640625, 546.2496337890625, 266.93890380859375, 137.38043212890625, 80.4527587890625, 1.041229248046875, -12.254989624023438, 195.82742309570312, 138.40798950195312, -6.99530029296875, 29.980072021484375, 7.43792724609375, 86.15615844726562, -66.36549377441406, 82.41917419433594, -47.3984375, -58.02545166015625, -5.253997802734375, 3.0568008422851562, 402.1896057128906, -64.8372802734375, 23.597259521484375, 241.90155029296875, 179.55201721191406, 95.00787353515625, -18.2774658203125, 196.00918579101562, 28.437454223632812, 40.77787780761719, -4.939910888671875, 246.24713134765625, 175.7623291015625, -14.00616455078125, 113.1989517211914, 37.015472412109375, 360.7630310058594, 21.730077743530273, 6.289833068847656, 19.37213134765625, 383.05694580078125, 20.737062454223633, 64.6153564453125, 51.97198486328125, 336.49237060546875, 95.02809143066406, -123.3150634765625, 193.30120849609375, -43.43212890625, 155.49777221679688, 126.3167724609375, -119.71500396728516, 18.583648681640625, 137.47100830078125, 54.32415771484375, 73.40049743652344, -102.17254638671875, 60.106239318847656, 42.12565231323242, 131.41139221191406, -54.979522705078125, 215.70123291015625, 90.86019897460938, -37.06646728515625, 182.3611297607422, -146.79490661621094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000454.npy"}
{"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 67.53520202636719, "std": 132.49436950683594, "min": -272.66607666015625, "p10": -50.674245452880854, "median": 39.241655349731445, "p90": 212.3256118774414, "max": 536.8297119140625, "pos_frac": 0.6953125, "sample": [259.4932861328125, -53.13629913330078, -8.343124389648438, 9.18353271484375, 174.21786499023438, -26.4598388671875, 23.669784545898438, 364.3003234863281, 54.60441589355469, -42.545867919921875, 10.0537109375, 118.27398681640625, 16.564544677734375, 30.76073455810547, 19.644695281982422, 185.11550903320312, 177.986083984375, 40.5131950378418, 299.72467041015625, 116.469970703125, 20.367156982421875, 66.48342895507812, 172.05038452148438, 185.05999755859375, 493.5907287597656, 226.31703186035156, 227.79000854492188, 89.167724609375, 26.264476776123047, -3.22314453125, -121.51736450195312, -272.66607666015625, -4.745601654052734, 48.54327392578125, 205.29833984375, -91.00527954101562, -45.71453857421875, -24.65087890625, -3.62451171875, 54.307647705078125, 152.24285888671875, 7.1597900390625, -6.024951934814453, 252.3507080078125, -47.9908447265625, 187.3897705078125, 188.2414093017578, 20.423385620117188, 196.24644470214844, 24.4132080078125, 62.85612487792969, 6.96942138671875, 43.0960693359375, -0.07079315185546875, -3.28265380859375, -4.62908935546875, 103.78915405273438, -133.69415283203125, 20.294708251953125, 37.970115661621094, 33.58613586425781, -0.1838226318359375, 58.7716064453125, -64.04340362548828, -106.181640625, -19.65448570251465, 71.82476806640625, -49.61907958984375, 215.5428924560547, 59.414520263671875, -116.67330932617188, 51.43595504760742, 149.8548583984375, 196.5072021484375, 173.6981201171875, -117.54095458984375, 6.9296875, 162.18011474609375, -49.13818359375, 267.36724853515625, -0.46783447265625, 22.93378448486328, 29.380889892578125, -197.74151611328125, 210.94677734375, 5.7319793701171875, -228.33013916015625, 21.420440673828125, -27.924530029296875, 196.85586547851562, -22.250137329101562, 72.21533203125, -35.748291015625, 22.361373901367188, 59.296356201171875, 154.77706909179688, 199.9779052734375, 528.4949951171875, 25.317474365234375, -129.42718505859375, 190.5821533203125, 13.836074829101562, 86.96551513671875, 155.87744140625, 2.51171875, -18.470844268798828, 97.87849426269531, -48.84552001953125, 122.3790283203125, 172.35418701171875, -0.132110595703125, 50.159912109375, 79.31500244140625, 57.37286376953125, 221.44561767578125, -7.151025772094727, -253.29840087890625, 134.26800537109375, 30.472259521484375, 196.63150024414062, 144.63021850585938, 55.482391357421875, 536.8297119140625, 113.91720581054688, 151.2136688232422, 139.1451416015625, 270.55621337890625, -9.25372314453125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000455.npy"}
{"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 85.70262145996094, "std": 159.77330017089844, "min": -249.03375244140625, "p10": -88.6112091064453, "median": 75.65300750732422, "p90": 278.2244384765625, "max": 919.0421752929688, "pos_frac": 0.6875, "sample": [-20.611602783203125, -34.04624938964844, -249.03375244140625, 202.72337341308594, 127.6522216796875, -59.075286865234375, 100.75634765625, -101.39697265625, 229.82659912109375, 72.85842895507812, 183.76885986328125, -244.69808959960938, 66.04539489746094, -17.88262939453125, 9.31744384765625, 158.95037841796875, 119.42605590820312, -108.51910400390625, -91.95050048828125, 31.7908935546875, 302.851318359375, 34.001129150390625, -108.64862060546875, 226.17172241210938, -5.702295303344727, -18.294471740722656, 140.4725341796875, -51.17762756347656, 115.04452514648438, 7.13665771484375, 157.911865234375, -66.17697143554688, 156.19985961914062, -126.93501281738281, 107.81761169433594, 246.853759765625, 104.25582885742188, 297.00140380859375, 94.25798034667969, 70.77877807617188, 402.89141845703125, 12.99456787109375, 272.51934814453125, 11.3150634765625, 173.13650512695312, 201.591796875, 367.17572021484375, 27.561065673828125, 96.18411254882812, 150.76669311523438, 49.174835205078125, 82.42901611328125, 31.29107666015625, 121.97635650634766, -168.8660888671875, 64.15451049804688, -57.75531005859375, -81.40553283691406, -73.0634765625, 283.5880432128906, -25.10736083984375, -15.317752838134766, 14.935791015625, -104.365234375, 41.017486572265625, 347.779296875, 169.55337524414062, -214.57119750976562, 41.489349365234375, 5.37982177734375, 52.96698760986328, 126.02896118164062, -28.43896484375, 150.16366577148438, 80.46310424804688, 253.47747802734375, 281.70794677734375, 112.8214111328125, -2.101531982421875, 97.25994873046875, 182.19830322265625, 494.826904296875, -50.94549560546875, 232.58843994140625, 11.716522216796875, 98.0045166015625, 12.15521240234375, 380.8125915527344, 167.20587158203125, 226.74710083007812, 105.3824462890625, 251.70626831054688, -28.01641845703125, 116.52777099609375, 0.0, -4.1369171142578125, 106.7685775756836, 276.73150634765625, 123.10257720947266, 78.44758605957031, 180.22613525390625, 41.239166259765625, 78.8997802734375, 13.439102172851562, 201.15042114257812, -120.17025756835938, 266.2802429199219, 140.53355407714844, 12.59722900390625, -47.388671875, -87.18008422851562, -18.434654235839844, -54.674285888671875, -81.03103637695312, 140.70941162109375, -6.783588409423828, 919.0421752929688, 140.68710327148438, 339.9267578125, 172.8515625, 221.8275146484375, -20.912261962890625, -18.45184326171875, -209.17990112304688, 292.89312744140625, 16.48260498046875, -121.99876403808594, 383.0076599121094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000456.npy"}
{"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 90.16606140136719, "std": 143.66653442382812, "min": -465.5118103027344, "p10": -85.83499603271484, "median": 99.7484359741211, "p90": 250.1815185546875, "max": 494.0479431152344, "pos_frac": 0.765625, "sample": [221.1029815673828, 124.0738525390625, -107.47787475585938, -17.469749450683594, 254.15316772460938, 220.5439453125, 182.77474975585938, 132.24737548828125, 154.80545043945312, -88.78225708007812, 187.86996459960938, 10.244316101074219, 247.3699951171875, 77.68373107910156, 46.82366943359375, 210.33343505859375, 203.6156005859375, 105.72064208984375, -93.88484191894531, 209.94186401367188, -21.578048706054688, 185.65005493164062, 54.29241180419922, 290.3495178222656, 245.5941162109375, 109.0615234375, 237.20632934570312, -6.3142242431640625, 231.23329162597656, 20.697799682617188, 152.15374755859375, 7.8443603515625, 15.371612548828125, 2.9344482421875, 114.05361938476562, 188.94195556640625, 31.974437713623047, 258.52825927734375, -109.98932647705078, 134.28485107421875, 100.39456176757812, 253.63612365722656, -30.36786651611328, 48.92938232421875, 19.015478134155273, -281.2661437988281, 214.82550048828125, 37.17999267578125, 214.1136016845703, 35.43247985839844, -20.59649658203125, 94.40777587890625, 34.98835754394531, -328.858154296875, 149.0611572265625, 10.074377059936523, -71.234130859375, 262.2104797363281, 101.59393310546875, 29.04486083984375, -2.9324951171875, -93.779541015625, 179.64886474609375, -78.37155151367188, 99.10231018066406, 494.0479431152344, -23.740036010742188, -3.8931655883789062, 231.572998046875, -16.969390869140625, -84.57188415527344, 57.62847900390625, 246.86660766601562, 150.9649658203125, 323.40521240234375, 29.26654052734375, -465.5118103027344, 140.34262084960938, 46.290000915527344, 138.3881072998047, 162.377685546875, 170.97381591796875, 73.98556518554688, 161.2281494140625, 57.05242919921875, -145.3551025390625, 166.41232299804688, 146.5185546875, 47.13201904296875, 329.6910705566406, 127.30166625976562, -166.28244018554688, 39.1427001953125, -93.87173461914062, -31.55352783203125, 328.0390625, -95.28643798828125, 278.84197998046875, -32.78704833984375, 227.99891662597656, 186.783935546875, 81.18707275390625, 148.0370635986328, 152.0498046875, -314.8603515625, 196.17877197265625, 128.0013427734375, 151.15017700195312, 108.19287872314453, 248.7009735107422, 286.1527404785156, 74.1343994140625, 28.7005615234375, 206.41294860839844, 301.03875732421875, 161.48179626464844, 83.16680908203125, -5.075439453125, -13.74798583984375, 348.9808044433594, 229.11715698242188, 6.679025650024414, 0.0, 208.3121337890625, 4.027000427246094, 22.042465209960938, 75.94783020019531, 22.582420349121094], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000457.npy"}
{"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 78.72277069091797, "std": 132.52603149414062, "min": -298.3525390625, "p10": -68.98743438720703, "median": 51.79117202758789, "p90": 244.19324951171873, "max": 462.3548278808594, "pos_frac": 0.7421875, "sample": [21.567367553710938, -84.04779052734375, 69.63760375976562, -213.1277618408203, 217.3479461669922, 28.173263549804688, -48.83721923828125, 184.72250366210938, 205.6968994140625, -141.35806274414062, 47.43841552734375, -51.09700012207031, -298.3525390625, 2.3336009979248047, 2.4341392517089844, 153.30422973632812, 137.12774658203125, -27.43927001953125, -39.67999267578125, 340.07379150390625, 245.04833984375, 25.843219757080078, 92.52191162109375, -28.160980224609375, 86.72354125976562, 3.554441452026367, 200.97897338867188, 374.27862548828125, 17.934112548828125, 165.03271484375, 160.833251953125, -31.790817260742188, 219.62367248535156, 176.04769897460938, 247.57867431640625, 51.4906005859375, 79.64227294921875, 121.65542602539062, 213.40090942382812, -46.40065002441406, 278.4576110839844, -8.770292282104492, -68.27243041992188, -17.3968505859375, 28.1741943359375, -9.5582275390625, -20.94671630859375, 51.80530548095703, 5.074760437011719, 23.626873016357422, 37.90730667114258, -99.98150634765625, -174.17176818847656, 212.34356689453125, 213.4906768798828, -77.862548828125, 12.65228271484375, -16.6904296875, 26.91620635986328, 232.91815185546875, -33.9786376953125, 0.7509040832519531, 153.6787109375, 47.673553466796875, 82.93473815917969, -62.09014892578125, 71.07498168945312, -127.11056518554688, -12.86065673828125, 243.0340576171875, 240.4560546875, 101.09519958496094, 63.2572021484375, 155.542236328125, 66.57647705078125, 178.21507263183594, 38.551971435546875, -77.51020812988281, -61.70640563964844, 163.96798706054688, 271.9129333496094, -94.63653564453125, 91.31900024414062, 110.94572448730469, 280.7114562988281, 30.872573852539062, 398.19647216796875, 51.77703857421875, 86.28482055664062, 188.84912109375, 6.401123046875, 249.79263305664062, -10.541915893554688, 128.55758666992188, 227.40072631835938, 93.96076202392578, 31.58026123046875, 4.42474365234375, 8.116174697875977, 149.21038818359375, 48.33404541015625, 15.33160400390625, 24.60198974609375, 278.33355712890625, 87.34152221679688, -70.65577697753906, 68.4609375, 154.607421875, 340.8409118652344, 226.9999237060547, 5.86688232421875, 462.3548278808594, 104.53822326660156, 201.40496826171875, 377.45611572265625, 213.80584716796875, 116.18658447265625, 134.70608520507812, -175.0078125, -138.92022705078125, 209.6345977783203, 243.8267822265625, -17.27496337890625, 5.074188232421875, -18.101539611816406, 83.91094970703125, 42.02275848388672, 2.6769256591796875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000458.npy"}
{"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 69.82811737060547, "std": 126.82351684570312, "min": -225.61090087890625, "p10": -71.68844909667969, "median": 58.13515090942383, "p90": 243.35519104003907, "max": 479.3461608886719, "pos_frac": 0.6640625, "sample": [143.840576171875, -34.90777587890625, 71.61505889892578, 116.17938995361328, -94.75799560546875, 69.17013549804688, -152.13214111328125, 90.89764404296875, 132.48291015625, 257.74566650390625, 243.13333129882812, -58.81072998046875, 31.27585792541504, -60.01312255859375, 58.068153381347656, -38.485496520996094, 228.8646697998047, 243.87286376953125, -43.112571716308594, 3.430419921875, 60.27618408203125, 118.2578125, -16.377822875976562, 119.68634033203125, 12.21697998046875, -90.22050476074219, 102.89236450195312, 13.028617858886719, 160.57785034179688, 47.26435852050781, -25.59309196472168, 0.0, 47.991607666015625, -62.75152587890625, 58.2021484375, 20.224365234375, 247.86590576171875, 54.47703552246094, -28.137054443359375, 91.61456298828125, 248.3673858642578, -60.691192626953125, 26.38818359375, 109.991455078125, 65.8736572265625, 86.22811889648438, -134.36566162109375, 238.4193572998047, -106.29766082763672, 201.361328125, -26.224258422851562, 39.41941452026367, -0.2552680969238281, 311.37969970703125, 74.00091552734375, 57.470481872558594, 241.53250122070312, -116.38482666015625, 240.7276153564453, 300.653076171875, 136.24319458007812, -78.41934204101562, 273.801513671875, 204.321044921875, 288.0351867675781, 259.9939880371094, 218.82284545898438, -14.450836181640625, -24.799285888671875, -49.460906982421875, -81.43548583984375, 65.67724609375, 175.2206268310547, 199.05368041992188, -61.99354553222656, -225.61090087890625, 142.50119018554688, 161.6349639892578, 23.3721923828125, -66.170166015625, -141.7215576171875, -21.70379638671875, 135.15771484375, -163.3368377685547, 231.48846435546875, 91.22747802734375, 8.5909423828125, 110.41622924804688, 0.0, 38.16126251220703, 83.43864440917969, -26.38555908203125, 79.41192626953125, 163.70396423339844, -43.2918701171875, 137.50100708007812, -42.412200927734375, 130.35821533203125, 366.447998046875, -43.13658142089844, 64.28390502929688, 129.43161010742188, 71.877685546875, -4.97235107421875, -46.2791748046875, -43.47038269042969, 174.766845703125, 12.996368408203125, 44.02009582519531, 37.07118225097656, -210.0155029296875, -72.03231811523438, 479.3461608886719, -71.54107666015625, 234.0072784423828, 191.6881103515625, -11.621261596679688, 33.987220764160156, 212.5872802734375, 102.85675048828125, 24.785316467285156, 317.92724609375, 249.47021484375, -11.946731567382812, 88.2120361328125, 11.02301025390625, 184.83438110351562, 167.0056610107422], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000459.npy"}
{"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 70.35595703125, "std": 121.18965148925781, "min": -234.26502990722656, "p10": -72.46305694580079, "median": 50.52464294433594, "p90": 222.60853271484376, "max": 350.75970458984375, "pos_frac": 0.7265625, "sample": [5.288604736328125, 68.26226806640625, -44.222137451171875, 98.07196044921875, 208.1162109375, 218.30543518066406, 12.876861572265625, 10.761775970458984, -114.36236572265625, 247.92010498046875, 13.054718017578125, 140.27456665039062, 10.50714111328125, 19.486923217773438, -11.219085693359375, -65.9483642578125, 103.6451416015625, 26.85174560546875, 229.2086944580078, 273.8894348144531, -152.7100830078125, 35.00054931640625, 92.89013671875, 16.68426513671875, 179.36537170410156, 48.594871520996094, 108.97109985351562, -27.296661376953125, 283.9781494140625, 22.080352783203125, 50.420166015625, 99.7955322265625, -2.6234130859375, 350.75970458984375, 195.5816650390625, -217.82044982910156, 158.6296844482422, -89.4302978515625, 317.7939453125, -12.03668212890625, -234.26502990722656, 176.837890625, -73.15704345703125, 139.93850708007812, 84.7249755859375, 157.69619750976562, -38.37042236328125, 321.327880859375, 183.1292724609375, -17.083145141601562, 222.6005859375, -47.2042121887207, 14.090635299682617, -72.16563415527344, -81.17826843261719, -2.1109619140625, -15.811203002929688, 2.244384765625, -168.5944061279297, 25.322540283203125, 195.19305419921875, 5.815093994140625, 237.45306396484375, 31.66851806640625, -5.2197723388671875, -14.018310546875, -20.514781951904297, 107.66836547851562, 56.5733642578125, 165.3545379638672, 223.4580078125, -13.062744140625, 12.07293701171875, 58.56463623046875, 186.92916870117188, 17.80448341369629, 30.486724853515625, 187.37734985351562, 164.71826171875, -107.82209777832031, 130.27059936523438, 66.740966796875, 2.370819091796875, 306.3567199707031, 222.6270751953125, 247.83058166503906, 221.89871215820312, 130.85223388671875, 50.629119873046875, 199.8271026611328, 209.66322326660156, 177.98440551757812, 22.4796142578125, 144.41574096679688, 26.715560913085938, 106.78079223632812, 210.99452209472656, -207.2314453125, 172.61642456054688, -170.10720825195312, 155.62171936035156, -127.0419921875, 129.24295043945312, -30.44952392578125, 89.25472259521484, -2.7330169677734375, -5.101799011230469, -27.3262939453125, 147.744384765625, 47.670196533203125, 80.80828857421875, 118.43710327148438, 63.036376953125, 118.31442260742188, -150.0216064453125, 16.801502227783203, 32.84938049316406, 209.82772827148438, -45.33056640625, 320.081298828125, 5.715545654296875, 58.17400360107422, 200.49588012695312, -10.893918991088867, 40.51066589355469, 142.35443115234375, 123.22332763671875, 22.71240234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000460.npy"}
{"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 75.32850646972656, "std": 139.47630310058594, "min": -306.84014892578125, "p10": -82.70699920654296, "median": 62.65332794189453, "p90": 233.51010284423825, "max": 572.4810791015625, "pos_frac": 0.703125, "sample": [86.31117248535156, -100.71994018554688, 227.10147094726562, 37.29794692993164, -41.60017395019531, 230.53533935546875, 175.93504333496094, 62.66180419921875, -27.171789169311523, 314.9861755371094, 204.0513916015625, 0.0, 35.10108947753906, 410.97509765625, 83.47296142578125, 206.8756561279297, 196.84469604492188, 60.641937255859375, 90.34817504882812, 66.69831848144531, 73.0390396118164, 174.146484375, 122.68354797363281, -54.63128662109375, 139.57748413085938, -49.20849609375, 80.76220703125, 572.4810791015625, -83.73765563964844, 241.18190002441406, -45.24137878417969, 221.71206665039062, 130.83184814453125, 167.74986267089844, 10.964508056640625, 154.0432891845703, 213.41421508789062, 227.35092163085938, -306.84014892578125, -71.49234008789062, 343.13067626953125, -163.0712890625, -35.19152069091797, 71.62570190429688, 144.41357421875, 61.85661315917969, 21.927169799804688, 171.371337890625, 193.14459228515625, 192.09393310546875, 179.0087890625, -24.91781234741211, 37.32560729980469, -21.71337890625, 11.940776824951172, 8.34326171875, 161.069580078125, -228.505126953125, -104.21881103515625, 48.341400146484375, 230.6284942626953, 28.790084838867188, 3.1188507080078125, 148.6824951171875, 265.7603454589844, 93.33499145507812, -181.88330078125, 117.38607788085938, 205.0938720703125, 23.87030029296875, 3.7392807006835938, 240.23385620117188, -64.83060455322266, 243.616455078125, 32.67730712890625, 195.514892578125, 389.12890625, -61.779571533203125, -9.2393798828125, -147.2529296875, 224.42727661132812, 42.476837158203125, 39.45953369140625, 116.4139404296875, 156.0975341796875, 40.22096252441406, -16.098739624023438, -38.187530517578125, -30.109586715698242, -94.15847778320312, -65.29269409179688, 70.97637176513672, 5.8642578125, 185.03866577148438, 32.50177001953125, 267.06329345703125, -29.857070922851562, 173.3914794921875, 117.6077880859375, -90.49958801269531, 124.89024353027344, -10.235694885253906, 266.52880859375, 8.108184814453125, -147.97909545898438, -37.913604736328125, 122.65977478027344, 206.3288116455078, 115.70084381103516, -72.22413635253906, -82.26528930664062, 62.64485168457031, -6.474798202514648, 163.47137451171875, 12.2344970703125, 143.62037658691406, 50.295616149902344, 162.24720764160156, -179.41156005859375, -11.936843872070312, 40.64361572265625, 76.24917602539062, -259.91314697265625, 338.92315673828125, -36.571144104003906, 7.438446044921875, 167.0313720703125, 246.92225646972656], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000461.npy"}
{"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 90.26371765136719, "std": 143.2453155517578, "min": -225.95936584472656, "p10": -90.87536697387695, "median": 81.11036682128906, "p90": 268.2608642578125, "max": 625.4949951171875, "pos_frac": 0.7578125, "sample": [266.443359375, 226.66159057617188, 231.44570922851562, 37.2132568359375, 234.86663818359375, 68.13262939453125, 287.18658447265625, 120.756103515625, 124.36820983886719, 84.48614501953125, -196.9378204345703, 194.75372314453125, 409.3231201171875, 160.48788452148438, 246.04324340820312, -74.78889465332031, 193.50015258789062, 213.23529052734375, 97.38671875, -38.930450439453125, 69.64569091796875, 97.97998046875, 106.31219482421875, 270.1339111328125, -151.5416259765625, -91.67697143554688, 72.4083251953125, -10.853313446044922, 212.4774169921875, -90.53182220458984, -170.32785034179688, 235.73971557617188, 133.41769409179688, 23.59075927734375, 31.27557373046875, 165.78485107421875, 267.4581298828125, -25.645721435546875, -98.62625122070312, 31.12591552734375, 208.844970703125, 90.87713623046875, 222.30886840820312, 4.470619201660156, 288.971923828125, 34.436527252197266, 183.21261596679688, 276.9481201171875, 235.09130859375, 36.35285949707031, 255.8445281982422, 32.62994384765625, 92.15693664550781, -62.86406707763672, 22.711318969726562, -18.37987518310547, 224.2401123046875, -28.28863525390625, 7.346549987792969, 14.8564453125, 187.72674560546875, 279.4886474609375, -40.67109680175781, 372.439208984375, 0.351287841796875, 45.21331024169922, -225.95936584472656, 75.5855712890625, 312.482421875, -82.77214050292969, 123.925048828125, 276.5440673828125, 152.3470458984375, 117.21026611328125, 192.82318115234375, 56.51329040527344, 16.859375, 46.822235107421875, -75.93431091308594, -30.98419189453125, 35.76527404785156, -73.2012939453125, -62.20684814453125, 29.702499389648438, -105.76947021484375, 44.89768981933594, 163.9866943359375, 247.56787109375, -80.87947082519531, -106.53160095214844, -13.702018737792969, 282.21734619140625, 122.12872314453125, 56.716552734375, 115.06768798828125, 625.4949951171875, -157.66946411132812, 56.751708984375, 77.73458862304688, 375.4908447265625, 61.613616943359375, 104.86764526367188, 331.3565673828125, 182.16140747070312, 10.14501953125, 61.997283935546875, 117.16291046142578, 33.81463623046875, 112.172607421875, 14.39349365234375, 169.01513671875, 13.1392822265625, 104.56689453125, 223.86911010742188, -188.76870727539062, 189.04917907714844, -122.5836181640625, 179.908447265625, 139.3522491455078, -29.17010498046875, 236.82534790039062, 84.66429138183594, -118.25262451171875, -12.289695739746094, 116.98049926757812, 184.6351318359375, 48.529144287109375, -214.5208740234375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000462.npy"}
{"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 71.0579833984375, "std": 134.66748046875, "min": -416.87078857421875, "p10": -66.75812225341795, "median": 54.50811767578125, "p90": 231.69867095947265, "max": 568.0389404296875, "pos_frac": 0.7421875, "sample": [73.52880859375, -86.16522216796875, 120.46516418457031, -102.50456237792969, 61.37799072265625, 1.7307720184326172, 206.987548828125, -38.4783935546875, -237.42849731445312, -270.21002197265625, 124.35595703125, 6.175323486328125, 44.85906982421875, -57.76591491699219, 64.53530883789062, 126.16433715820312, -19.13201141357422, 230.78810119628906, -17.855438232421875, 5.4230499267578125, 9.999282836914062, -249.79315185546875, -61.84344482421875, 50.02031326293945, -4.807699203491211, 24.595054626464844, -76.56268310546875, 187.6596221923828, 290.4071044921875, 134.55902099609375, 17.374374389648438, 195.39822387695312, 34.76594543457031, 66.83477783203125, 41.916534423828125, 266.71051025390625, 244.50360107421875, 285.274169921875, 509.97247314453125, 90.58009338378906, 77.904541015625, 86.764892578125, 57.726104736328125, 244.798583984375, 25.10417938232422, -8.32708740234375, 327.9188232421875, 15.072273254394531, 60.817138671875, 45.40727996826172, 35.80701446533203, 49.561614990234375, 247.75296020507812, 233.82333374023438, 2.7313385009765625, 107.58633422851562, 189.13473510742188, 147.62277221679688, 167.01190185546875, 121.89456176757812, 47.0689697265625, 176.49310302734375, 17.996124267578125, -116.20120239257812, -75.74076843261719, 38.36808776855469, 121.52996826171875, 171.17823791503906, 57.953399658203125, 77.4483642578125, 227.97482299804688, -78.33892822265625, 84.81675720214844, 149.02569580078125, 63.797515869140625, 3.8033447265625, -38.710479736328125, 73.81221008300781, -76.5051498413086, 73.63658142089844, 223.12063598632812, 127.26025390625, 218.96536254882812, 368.4485168457031, 568.0389404296875, 17.546724319458008, 13.48775863647461, 235.8301239013672, 341.347900390625, -22.87814712524414, -4.552581787109375, 92.168701171875, -43.02122497558594, 0.0, -76.591064453125, 49.345924377441406, 19.980205535888672, 187.697998046875, 70.23947143554688, -95.64390563964844, -15.4188232421875, 0.2501068115234375, -47.691650390625, 6.85273551940918, 38.38365173339844, 51.290130615234375, -7.404296875, 149.78759765625, 163.06182861328125, -21.957263946533203, -62.908416748046875, 33.521087646484375, 19.433837890625, 123.14640808105469, -32.54693603515625, 99.10662841796875, 229.80886840820312, 88.59770202636719, 193.49656677246094, 135.28857421875, 182.98452758789062, 49.778717041015625, 60.38090515136719, 162.3192138671875, -28.91119384765625, 136.67266845703125, -416.87078857421875, -15.757965087890625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000463.npy"}
{"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 70.50304412841797, "std": 129.88079833984375, "min": -286.4708251953125, "p10": -68.01690063476562, "median": 62.338539123535156, "p90": 231.58275146484374, "max": 441.7698059082031, "pos_frac": 0.6640625, "sample": [0.0, 136.61032104492188, 2.2160720825195312, 251.0834503173828, -157.74908447265625, 205.82261657714844, 86.12246704101562, -117.148681640625, 207.09487915039062, -6.94587516784668, -36.98455810546875, 7.733489990234375, 185.48580932617188, 102.98971557617188, -66.97927856445312, 6.071781158447266, 184.58929443359375, 66.97128295898438, 199.225341796875, -50.93994140625, 336.4525146484375, 157.38385009765625, 26.69061851501465, -104.59481811523438, 84.02896118164062, 158.00216674804688, 2.6308670043945312, -32.5545654296875, 258.5933532714844, -61.91766357421875, -84.67230224609375, 33.00157165527344, -45.13531494140625, 172.46141052246094, 233.10400390625, 250.4304656982422, 71.80776977539062, 118.80107879638672, -19.6669921875, -73.30655670166016, 26.147857666015625, 33.24932861328125, 162.05093383789062, 441.7698059082031, 149.07623291015625, -5.9509124755859375, -45.693359375, -43.39453125, 4.779388427734375, 107.9900894165039, 1.73126220703125, 67.940185546875, -35.42278289794922, 206.0340576171875, -28.148967742919922, 95.475830078125, 138.21209716796875, 97.8357925415039, -246.3264617919922, 244.47329711914062, -66.5950927734375, 230.9307861328125, -21.940719604492188, 201.67300415039062, 351.8860168457031, -3.83837890625, 115.371337890625, -13.017852783203125, 89.27996826171875, 200.22982788085938, -17.4833984375, -234.33657836914062, 32.71983337402344, -12.367706298828125, 97.85671997070312, -1.4522247314453125, -25.87384033203125, 33.700225830078125, 333.408203125, 99.41083526611328, 167.12661743164062, 241.4324188232422, 156.2738037109375, 228.46572875976562, -112.413818359375, 93.33447265625, -1.2317848205566406, -70.43801879882812, 25.9498291015625, 10.649383544921875, 79.34750366210938, -198.38369750976562, 117.31561279296875, -30.214691162109375, -9.386749267578125, 82.23143005371094, 8.382537841796875, -113.50885009765625, 35.69384765625, -40.784305572509766, 47.64353942871094, 392.9871520996094, 66.25035095214844, 208.26425170898438, 279.536865234375, -6.174346923828125, 229.9149169921875, 58.426727294921875, 207.9426727294922, -60.06195068359375, 143.7242431640625, -18.403995513916016, 33.56853485107422, 152.04837036132812, 202.64288330078125, 181.4237060546875, 234.06243896484375, 192.30450439453125, 160.07986450195312, -286.4708251953125, 191.55947875976562, 6.63153076171875, -8.880126953125, 160.2539825439453, 37.11370849609375, -104.42591094970703, 89.3465576171875, 115.03974914550781], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000464.npy"}
{"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 93.48235321044922, "std": 136.4739990234375, "min": -211.00460815429688, "p10": -80.48519897460938, "median": 96.15850448608398, "p90": 274.3390655517578, "max": 467.65264892578125, "pos_frac": 0.75, "sample": [-39.14640808105469, -23.33526611328125, 193.8467254638672, 26.12469482421875, 288.7227783203125, -110.7724380493164, 0.51934814453125, 282.23089599609375, 122.43203735351562, 34.571533203125, 16.4039306640625, 68.36846923828125, -49.050514221191406, -8.90167236328125, -124.012939453125, -43.37744140625, 107.04147338867188, 283.3431396484375, 235.42355346679688, 406.88970947265625, 137.56979370117188, -166.851806640625, 48.60063171386719, 219.02395629882812, 44.95404052734375, 111.59178924560547, 37.5096435546875, 259.0032653808594, 212.33013916015625, 42.93980407714844, 54.95550537109375, 179.0312957763672, 35.4779052734375, 92.91134643554688, 85.5836181640625, -149.79135131835938, -82.121337890625, 234.82839965820312, 70.84016418457031, 316.1765441894531, 84.35662841796875, 125.84039306640625, 5.6956787109375, -82.63232421875, 84.1368408203125, 125.12954711914062, -24.956192016601562, -91.28712463378906, -32.2652587890625, 19.316513061523438, 157.490966796875, 195.25204467773438, 193.808349609375, 161.9320526123047, 129.89524841308594, -157.23806762695312, 277.2203063964844, 101.177490234375, -30.42840576171875, -119.82962036132812, 172.930908203125, 206.53106689453125, 116.81024169921875, 204.39187622070312, 387.22686767578125, 141.55618286132812, 232.5208740234375, 249.99221801757812, 102.09133911132812, 110.81824493408203, 124.56633758544922, 241.5721893310547, 87.21507263183594, 25.829498291015625, 99.4056625366211, -185.58184814453125, -16.5877685546875, 273.104248046875, 7.113250732421875, 115.70848083496094, -64.02053833007812, 10.053863525390625, 11.30487060546875, 104.67904663085938, 14.950164794921875, 255.66107177734375, 192.36788940429688, -38.19734191894531, 105.06217956542969, 44.084014892578125, 467.65264892578125, 297.4124755859375, 295.87548828125, 119.71302795410156, -46.370269775390625, 142.77261352539062, 383.5068054199219, 291.1070556640625, 231.606201171875, 3.9361801147460938, 63.85430908203125, 216.74673461914062, 221.731689453125, 119.34393310546875, 86.33448791503906, 171.10186767578125, -40.132781982421875, 172.13870239257812, -196.47015380859375, 34.5555419921875, -79.78399658203125, -31.894927978515625, 268.7431335449219, 230.8184814453125, 147.34915161132812, 90.44061279296875, -147.2337646484375, 256.9154357910156, 54.084991455078125, -21.141334533691406, -25.361900329589844, 102.69137573242188, -20.70440673828125, 7.7543487548828125, 288.49249267578125, -211.00460815429688, -8.0313720703125, 119.53176879882812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000465.npy"}
{"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 98.67318725585938, "std": 126.22212219238281, "min": -318.8514709472656, "p10": -21.778872680664055, "median": 77.76468658447266, "p90": 261.8885009765625, "max": 582.3428344726562, "pos_frac": 0.765625, "sample": [165.20738220214844, 142.0023193359375, 87.40081787109375, 360.63800048828125, 42.85310363769531, 328.966552734375, 73.1278076171875, 0.0, 33.504119873046875, 127.73014831542969, 116.99787902832031, -27.181655883789062, 72.0697021484375, 70.37728881835938, 139.720458984375, 259.492919921875, -50.25544357299805, -5.9861297607421875, 198.93313598632812, 16.84295654296875, 169.80429077148438, 111.08470153808594, 10.369476318359375, 65.71981811523438, -9.56231689453125, 69.45489501953125, 86.49370574951172, 17.907257080078125, 200.1495819091797, 41.90263366699219, -13.709266662597656, 162.76675415039062, -12.215690612792969, 157.69622802734375, 243.367431640625, -19.463394165039062, 168.59567260742188, 30.384979248046875, 309.2203674316406, 72.5780029296875, 364.207763671875, -100.30462646484375, 212.7333984375, 96.58193969726562, 125.29473876953125, 231.3140869140625, 63.02879333496094, 196.38021850585938, -29.051223754882812, 101.14202880859375, -12.6376953125, 134.43902587890625, -0.1748046875, -3.2354660034179688, 199.43341064453125, 195.05181884765625, 60.82392883300781, -97.1478271484375, 67.03741455078125, 195.37796020507812, 118.18841552734375, 226.5178985595703, 99.42242431640625, 219.86279296875, 135.03610229492188, 52.96109390258789, 155.466796875, 125.82530212402344, 395.7425537109375, 13.795372009277344, 0.0, 185.40098571777344, 207.645751953125, 283.436279296875, -37.668556213378906, 125.88934326171875, 139.9972381591797, -18.757492065429688, 126.52239990234375, 263.3695068359375, 40.159263610839844, 13.33135986328125, 582.3428344726562, -2.9354248046875, 85.228515625, -97.43058776855469, 38.92877197265625, -1.8636474609375, 92.3250732421875, 238.73086547851562, 148.4794464111328, -90.17628479003906, 49.80912780761719, -74.21339416503906, 340.32269287109375, 42.58539581298828, 54.617401123046875, 419.4659423828125, -137.87277221679688, -318.8514709472656, 53.842689514160156, 5.938629150390625, -19.024276733398438, 299.7790832519531, 30.45679473876953, 0.0, 131.927001953125, 10.94140625, -5.490373611450195, 190.1384735107422, 222.16900634765625, 261.2537841796875, -18.29119873046875, 40.820884704589844, 168.4393310546875, 61.88714599609375, 93.46951293945312, 82.40156555175781, 2.56689453125, 108.877685546875, 7.33544921875, 84.02271270751953, 284.139892578125, 69.32624816894531, -34.076629638671875, -71.40058135986328, 47.15362548828125, 264.64422607421875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000466.npy"}
{"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 69.97258758544922, "std": 143.3873748779297, "min": -280.17034912109375, "p10": -100.60778198242187, "median": 62.742671966552734, "p90": 232.2723159790039, "max": 639.8799438476562, "pos_frac": 0.6796875, "sample": [252.90338134765625, 101.31034851074219, 206.17442321777344, 80.291259765625, 43.49201202392578, -25.20459747314453, 57.77099609375, 16.747573852539062, 204.5985107421875, 18.91729736328125, -134.1019287109375, -25.60943603515625, 95.11761474609375, -21.472503662109375, 21.841217041015625, 95.44329833984375, 95.29571533203125, 222.38394165039062, -100.36236572265625, 156.09500122070312, -9.194290161132812, 213.11819458007812, -102.28962707519531, 124.80570983886719, 249.27508544921875, 35.99067687988281, 82.16596984863281, 59.699398040771484, 127.15278625488281, -45.85003662109375, 56.01165771484375, 65.1682357788086, 231.531494140625, 639.8799438476562, 255.1499481201172, 182.4147491455078, 10.297454833984375, 241.17315673828125, 76.59269714355469, -266.25018310546875, 196.65240478515625, 114.01516723632812, -5.54362678527832, 144.53341674804688, 176.82278442382812, 193.6624755859375, -11.067342758178711, -56.614898681640625, 204.79747009277344, 85.87092590332031, 39.319915771484375, -168.7888946533203, -90.94789123535156, 135.79171752929688, -140.2354736328125, 166.3330078125, -11.730134963989258, -185.8702392578125, 205.18020629882812, 349.3779296875, -86.73504638671875, -45.16249084472656, -0.7239227294921875, -9.572906494140625, 157.57687377929688, 33.95343017578125, 33.651268005371094, -32.719886779785156, 97.354248046875, 2.1836795806884766, 59.19720458984375, -171.0122833251953, 74.22848510742188, 0.0, 27.610076904296875, -1.597076416015625, 36.595672607421875, 247.9311981201172, 16.9365234375, -13.312713623046875, 0.0, 85.00564575195312, 176.3258056640625, 18.42608642578125, 150.82806396484375, 46.89186096191406, 213.18048095703125, -237.1256561279297, 206.04220581054688, 108.13961791992188, -161.01797485351562, 287.86492919921875, 101.40631103515625, 92.91093444824219, 173.41714477539062, 69.64071655273438, -49.84642791748047, 102.18017578125, -4.25006103515625, -56.5853271484375, -101.180419921875, 138.6792449951172, 234.0009002685547, -5.24261474609375, -31.38580322265625, 60.317108154296875, -273.31744384765625, -280.17034912109375, -12.5511474609375, 264.9749755859375, 92.89697265625, 321.0560607910156, -16.468505859375, 74.06097412109375, 25.715524673461914, -85.44598388671875, -162.10272216796875, 135.08706665039062, 415.10113525390625, 134.09498596191406, 200.99375915527344, 80.6731185913086, 215.36558532714844, 207.9957733154297, 144.7294921875, 461.1463928222656, 4.2950439453125, 33.32073974609375], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000467.npy"}
{"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 69.12068176269531, "std": 135.21994018554688, "min": -561.593505859375, "p10": -61.46986770629883, "median": 48.484230041503906, "p90": 246.82723083496094, "max": 410.85308837890625, "pos_frac": 0.7109375, "sample": [105.97610473632812, 226.75140380859375, 249.89328002929688, -25.56713104248047, 224.94422912597656, 25.572784423828125, -27.296859741210938, 292.4208984375, -2.885343551635742, 122.154296875, 54.0853271484375, -196.27911376953125, 26.498687744140625, 178.78334045410156, 23.66058349609375, -35.50157928466797, 39.2257080078125, 281.8004150390625, 26.2906494140625, -50.1446533203125, 109.69761657714844, 132.9178466796875, -9.379470825195312, 49.67352294921875, 13.03363037109375, 349.12481689453125, 85.73370361328125, 267.152587890625, 47.1102294921875, -9.93032455444336, 8.876708984375, -60.501434326171875, 247.3634033203125, 4.909929275512695, 186.25204467773438, -34.053466796875, -11.158859252929688, 246.59744262695312, -561.593505859375, -46.203887939453125, 99.88331604003906, 78.658203125, -24.331817626953125, 132.33413696289062, -28.085235595703125, -63.72954559326172, 133.02398681640625, 33.2501220703125, 241.91798400878906, 18.24390411376953, 90.90768432617188, 0.6971759796142578, -163.00128173828125, 27.15240478515625, 47.29493713378906, 23.521453857421875, 39.240875244140625, 151.5674591064453, 6.898780822753906, 110.92024230957031, 410.85308837890625, -53.66925048828125, -56.38330841064453, 0.77362060546875, 384.2407531738281, -26.072433471679688, -15.786136627197266, 166.4296875, -57.291259765625, 104.7230224609375, 163.60888671875, 67.76361083984375, -42.850830078125, 143.4969482421875, 239.08030700683594, 59.8499755859375, 85.67520141601562, -55.22589111328125, -135.72561645507812, 219.76629638671875, 3.2136001586914062, 259.1136474609375, 127.79694366455078, 220.05865478515625, 60.02191162109375, 170.1614990234375, 89.93243408203125, -40.08373260498047, -70.33930969238281, 107.30889892578125, 79.97247314453125, 28.850879669189453, 1.1697998046875, 359.18096923828125, 9.122802734375, 171.35867309570312, 121.90179443359375, 40.35392761230469, 7.82537841796875, 31.78314208984375, 154.7908477783203, 243.08853149414062, 131.17080688476562, 23.643798828125, -299.3287353515625, -75.77914428710938, -72.9215087890625, 96.58990478515625, 99.60260009765625, 337.48419189453125, -16.05255126953125, -74.98971557617188, 144.22775268554688, 109.12857055664062, -89.10867309570312, 109.7071533203125, 187.90142822265625, -32.242149353027344, -84.83927154541016, 42.516265869140625, 157.33314514160156, 188.4658203125, 156.33489990234375, 114.3602294921875, -81.07913208007812, 263.14739990234375, -31.1331787109375, 251.09719848632812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000468.npy"}
{"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 91.9258041381836, "std": 133.58274841308594, "min": -319.3172607421875, "p10": -70.28456726074218, "median": 84.91553497314453, "p90": 256.93507080078126, "max": 487.0152893066406, "pos_frac": 0.7734375, "sample": [94.45686340332031, 300.8613586425781, 0.0, 24.804920196533203, 55.26251220703125, 132.39158630371094, -40.33759307861328, 76.42523193359375, -110.844970703125, 8.303878784179688, 65.03150939941406, 181.42572021484375, 203.9205322265625, -76.79502868652344, 102.40087890625, 43.8262939453125, 275.4784240722656, 203.070556640625, 34.992523193359375, 102.65966796875, -68.07659912109375, 93.30877685546875, 49.246192932128906, -11.203365325927734, -75.43649291992188, 150.80364990234375, 33.502655029296875, 143.416748046875, 209.96408081054688, 34.40240478515625, 161.75967407226562, 126.005126953125, 325.50396728515625, 247.86614990234375, 53.312198638916016, 110.68681335449219, 248.377685546875, 34.09075927734375, 260.95989990234375, 195.1953125, 26.279823303222656, -36.76385498046875, 56.25518798828125, 8.58331298828125, 206.65765380859375, -34.5352783203125, 50.42411804199219, 96.765869140625, 204.37823486328125, 68.92462158203125, -0.5187606811523438, 214.6949462890625, 233.63279724121094, 225.33132934570312, -161.22796630859375, 1.5971221923828125, 116.171142578125, 220.18112182617188, 177.00491333007812, -97.56723022460938, 18.561782836914062, -208.52734375, 281.0181884765625, -89.61749267578125, -203.57614135742188, 255.21014404296875, 111.47785949707031, 75.0333251953125, -112.179443359375, 112.56649780273438, -36.30261993408203, 163.74058532714844, 367.7896728515625, 249.30990600585938, -212.20419311523438, 1.2726383209228516, 100.53575134277344, 220.97898864746094, 14.052032470703125, 351.48193359375, 196.89199829101562, 81.64961242675781, 71.88693237304688, 122.20433044433594, 246.25955200195312, -31.048934936523438, -319.3172607421875, -9.516468048095703, 143.2960968017578, -51.529605865478516, 76.58175659179688, 205.83749389648438, -26.253631591796875, 191.6620635986328, -63.618408203125, -118.28981018066406, 7.8505096435546875, 25.763931274414062, 14.986186981201172, 141.7457275390625, 52.81056213378906, -1.5738449096679688, 93.37408447265625, 487.0152893066406, 0.125640869140625, 94.08456420898438, 214.29302978515625, 126.74071502685547, 185.0325927734375, 304.70220947265625, 309.6343078613281, -2.050018310546875, 286.4850158691406, 88.18145751953125, 81.4668960571289, 72.200439453125, 12.677703857421875, 317.81024169921875, 77.9007568359375, 155.218994140625, 183.63247680664062, 198.9149169921875, 68.72428131103516, -76.7642822265625, 88.54747009277344, 88.6505126953125, 345.4842834472656, -31.771392822265625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000469.npy"}
{"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 84.67138671875, "std": 128.03367614746094, "min": -268.0505676269531, "p10": -46.58047790527343, "median": 71.820068359375, "p90": 239.97210235595702, "max": 487.5111389160156, "pos_frac": 0.7734375, "sample": [-65.5908203125, 45.94061279296875, 187.8892822265625, 206.27755737304688, -90.23483276367188, 20.45587158203125, 13.879219055175781, 82.43675231933594, 113.90283203125, 89.77365112304688, -49.27642822265625, 7.772918701171875, 33.717323303222656, 177.407958984375, 100.9384765625, 126.35543823242188, 312.0375061035156, 23.532089233398438, -48.269744873046875, 37.898590087890625, 35.5465202331543, -111.34222412109375, 22.742340087890625, 55.75274658203125, 361.9642333984375, 101.47048950195312, 89.08977508544922, 108.39617919921875, 137.36434936523438, 100.929443359375, 231.611328125, 128.6898956298828, -28.997406005859375, 17.30010986328125, 140.94210815429688, -243.16220092773438, 185.0328369140625, 139.38522338867188, 29.908447265625, -8.001358032226562, 104.58901977539062, 320.4854736328125, 243.56629943847656, -27.344482421875, 11.655609130859375, 269.37896728515625, 131.04412841796875, 39.12281036376953, 14.715316772460938, 47.9512939453125, 128.89471435546875, 151.45492553710938, 32.236419677734375, 198.4199676513672, -1.4520263671875, 148.03753662109375, 24.084732055664062, -38.576873779296875, 214.77227783203125, -16.15899658203125, 157.0614471435547, 11.203781127929688, 161.39389038085938, -32.69151306152344, 248.23098754882812, 11.522621154785156, -40.156890869140625, -108.9381103515625, 19.61163330078125, 114.81753540039062, -15.278091430664062, 143.30450439453125, 186.13165283203125, 34.053932189941406, -23.7174072265625, 37.789573669433594, 10.84197998046875, 31.2891845703125, 230.3529815673828, 10.9830322265625, 77.83877563476562, 65.80136108398438, 168.1661376953125, 226.385009765625, -41.82879638671875, -194.9068603515625, -26.718063354492188, 177.36749267578125, 229.17523193359375, 364.5707092285156, -211.39845275878906, 487.5111389160156, 159.48976135253906, 175.9140167236328, -60.82098388671875, 15.874267578125, 124.35502624511719, 234.93609619140625, 79.86993408203125, -140.38754272460938, 263.2099609375, 390.24176025390625, 44.05265808105469, 310.25396728515625, 119.82131958007812, 181.7845458984375, 9.01295280456543, 31.607101440429688, 39.6004638671875, -268.0505676269531, 42.498504638671875, -24.279052734375, 151.56787109375, -33.221527099609375, 179.0062255859375, 283.9100341796875, 38.265899658203125, 138.89755249023438, 93.7913818359375, 61.917449951171875, 183.557861328125, -96.86312866210938, 94.4522705078125, -16.6217041015625, 198.4857177734375, 238.43173217773438, 311.11737060546875, -45.85650634765625], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000470.npy"}
{"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 80.33290100097656, "std": 146.1367950439453, "min": -247.0464630126953, "p10": -67.24664306640625, "median": 61.70782470703125, "p90": 235.04477844238278, "max": 973.675537109375, "pos_frac": 0.7578125, "sample": [-67.55607604980469, 143.7068328857422, 32.293212890625, 126.44461059570312, 138.89065551757812, 278.997802734375, 90.42401123046875, 82.47920227050781, 2.879568099975586, 39.615447998046875, -247.0464630126953, -28.79400634765625, 156.74609375, 158.42495727539062, -35.517982482910156, 74.30908203125, 61.457489013671875, 133.71038818359375, -26.054786682128906, 61.958160400390625, 18.95306396484375, 31.6007080078125, 20.211395263671875, -13.26715087890625, -67.11402893066406, 11.154998779296875, 510.81756591796875, -5.74005126953125, -2.9639892578125, 110.40171813964844, 144.2239990234375, 148.28952026367188, 114.15634155273438, -5.94305419921875, -58.04010009765625, 66.8829345703125, 9.0089111328125, 101.0936050415039, 49.05126953125, 11.36541748046875, 57.76123046875, 19.21856689453125, 153.83456420898438, 195.67657470703125, 5.821084976196289, 131.3014678955078, 19.153335571289062, 307.89251708984375, 146.76373291015625, 10.765998840332031, 141.9759521484375, 26.9788818359375, 250.04876708984375, 87.70867919921875, -105.94451904296875, 205.78619384765625, 169.7337646484375, -1.6651382446289062, 5.786033630371094, 138.63763427734375, 241.56756591796875, 973.675537109375, 5.508689880371094, -64.29522705078125, -148.76556396484375, -2.820810317993164, 21.087631225585938, 11.537353515625, -164.95501708984375, -14.17486572265625, 113.9747314453125, 59.86585998535156, 37.62921142578125, 95.98307800292969, -135.91925048828125, 71.30744934082031, 191.20761108398438, 97.36146545410156, 205.12567138671875, 16.675048828125, 121.4534912109375, 2.722564697265625, -160.03997802734375, 14.210552215576172, 103.9493408203125, 94.45341491699219, 75.40278625488281, 118.62350463867188, 97.68170166015625, -21.7166748046875, 225.10792541503906, 179.5528564453125, 76.63937377929688, 79.34103393554688, 29.19171142578125, 191.83706665039062, 408.030029296875, 0.6054153442382812, -98.3355712890625, -83.9912338256836, 193.29287719726562, -4.0014495849609375, 202.8545684814453, 33.64569091796875, 79.47689819335938, 145.17401123046875, 104.189208984375, 232.24929809570312, 54.670379638671875, 52.264434814453125, 56.872283935546875, 268.478271484375, 226.1893310546875, -172.37796020507812, 144.34588623046875, -72.5370864868164, 303.0205078125, 0.0, 112.68283081054688, 11.361595153808594, 338.45904541015625, 365.2109069824219, 302.91680908203125, -41.975067138671875, 250.1341552734375, -52.914703369140625, -142.58013916015625, -113.53338623046875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000471.npy"}
{"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 92.0108413696289, "std": 130.20956420898438, "min": -185.7627410888672, "p10": -40.04054718017578, "median": 62.95481872558594, "p90": 272.2825927734375, "max": 483.8052978515625, "pos_frac": 0.734375, "sample": [461.33856201171875, 54.4671630859375, -54.47682571411133, 38.13630676269531, 16.96455192565918, 14.615081787109375, 139.71075439453125, 2.745849609375, -7.7926025390625, -7.877655029296875, 9.142730712890625, 30.868698120117188, 277.17901611328125, 130.97360229492188, -78.95054626464844, 174.18890380859375, -135.07870483398438, 61.08290100097656, -43.42474365234375, 115.23306274414062, -47.297088623046875, 77.05242919921875, 123.95335388183594, 62.19342041015625, -35.213645935058594, 224.67523193359375, 123.10836029052734, -72.6937255859375, -17.983360290527344, 196.73858642578125, 158.6273193359375, -5.245674133300781, 59.108428955078125, 384.785400390625, -111.81645202636719, -38.61309814453125, 271.51629638671875, 167.4493408203125, -9.789947509765625, 4.321723937988281, 112.1500244140625, -1.48736572265625, 126.98898315429688, 12.23455810546875, 30.942283630371094, 46.321044921875, 54.37481689453125, 44.61322021484375, 105.3729248046875, 85.3199462890625, 190.47842407226562, 271.115478515625, 321.8717041015625, -106.34469604492188, 49.40809631347656, 288.77935791015625, 406.351318359375, 89.47219848632812, 23.7149658203125, 154.88612365722656, 101.45416259765625, 252.85675048828125, -31.2418212890625, -17.95771598815918, 112.12322998046875, 208.90084838867188, 147.51223754882812, 94.59246826171875, -8.685783386230469, 7.667051315307617, 32.10301971435547, 29.00107192993164, 63.69476318359375, 323.1374816894531, 49.2227783203125, 148.3817138671875, 141.4517364501953, -20.470502853393555, 224.6431121826172, 60.224029541015625, -121.29653930664062, 274.07061767578125, 321.9139709472656, 249.1260986328125, 37.570831298828125, 202.23553466796875, 229.81472778320312, 316.81658935546875, 40.26225280761719, -167.7496337890625, 86.31024169921875, -0.07220458984375, 142.02064514160156, 23.242431640625, 154.35858154296875, 147.16726684570312, 399.4201965332031, 23.825653076171875, -33.38848876953125, 205.46205139160156, 156.27627563476562, -36.72905731201172, 187.56094360351562, 163.92739868164062, -16.158004760742188, 168.630859375, -2.000732421875, 36.6651611328125, -35.339439392089844, 145.69979858398438, 270.5104675292969, 122.88494873046875, -0.31271934509277344, 79.54348754882812, 122.07723999023438, -185.7627410888672, -43.37126159667969, 483.8052978515625, -22.756683349609375, 93.5819091796875, 279.988037109375, 135.212158203125, -161.84237670898438, 85.86981201171875, 62.214874267578125, 234.93701171875, -33.620452880859375, 15.687583923339844], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000472.npy"}
{"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 92.44808959960938, "std": 122.33883666992188, "min": -310.4181213378906, "p10": -47.83176879882812, "median": 85.99369049072266, "p90": 238.8577178955078, "max": 460.2998046875, "pos_frac": 0.7734375, "sample": [17.77147674560547, 236.79641723632812, 54.22450256347656, -67.24186706542969, 62.286865234375, -23.36212158203125, -10.14752197265625, 38.5440673828125, 148.43603515625, -13.65008544921875, 182.52432250976562, 19.006202697753906, -113.12542724609375, 11.730779647827148, -30.85614776611328, 77.88717651367188, -47.58807373046875, -123.07068634033203, 219.21563720703125, 197.42501831054688, 26.8116455078125, 144.74652099609375, 242.43417358398438, 260.634033203125, 60.66839599609375, 43.6925048828125, 4.584747314453125, 96.33364868164062, 131.36865234375, -6.4730682373046875, 204.7771453857422, 174.730712890625, 155.93783569335938, 125.96649169921875, -40.438453674316406, 69.51580810546875, 50.26475524902344, -30.114151000976562, 237.324951171875, 172.09446716308594, 91.44422912597656, 133.43963623046875, 98.57888793945312, 21.482431411743164, 45.066253662109375, -63.24290466308594, -119.72079467773438, 460.2998046875, 108.97323608398438, 122.07647705078125, 27.257286071777344, 145.34072875976562, -5.92921257019043, 295.7767028808594, -48.400390625, 402.8917541503906, 24.897918701171875, 18.566970825195312, 298.5736083984375, -169.30819702148438, 47.913330078125, 191.4334716796875, 200.0277099609375, -1.43377685546875, -12.268341064453125, 105.03277587890625, 213.052001953125, 224.03048706054688, -15.30389404296875, 165.846435546875, 134.7930908203125, 208.90541076660156, 31.561767578125, 133.154541015625, 0.5235652923583984, 212.34681701660156, 235.4125213623047, 42.300079345703125, 151.57403564453125, 276.9869689941406, 35.311195373535156, -37.41374206542969, 293.4967041015625, 204.68515014648438, 134.10928344726562, 259.518310546875, -310.4181213378906, 28.166963577270508, 201.30978393554688, -67.14715576171875, 120.17340087890625, 219.4783935546875, 24.7960205078125, 99.02110290527344, -2.4933319091796875, -34.917945861816406, 64.12906646728516, 99.195068359375, -82.7674331665039, -82.32887268066406, 249.2041015625, 67.97032928466797, 162.07965087890625, 0.0, 180.127197265625, 71.1300048828125, -89.72821044921875, 71.95639038085938, 65.58889770507812, 21.90064811706543, 185.27548217773438, 86.96046447753906, 58.928993225097656, 236.01219177246094, 121.78082275390625, 30.311553955078125, 183.36209106445312, -207.43588256835938, 85.02691650390625, 184.6287384033203, 296.47784423828125, 308.4556884765625, 129.39114379882812, 116.333740234375, 261.74969482421875, 186.02926635742188, 153.4310760498047, 54.88165283203125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000473.npy"}
{"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 74.75314331054688, "std": 120.0405502319336, "min": -266.4739074707031, "p10": -85.96323928833007, "median": 69.87968444824219, "p90": 221.10250549316405, "max": 476.4792785644531, "pos_frac": 0.75, "sample": [26.4150390625, -17.00689697265625, 476.4792785644531, -2.938507080078125, -16.330615997314453, -8.98223876953125, -14.371490478515625, 322.0975341796875, 82.72389221191406, 143.1302947998047, 157.903076171875, 127.70529174804688, -155.626220703125, -103.28042602539062, 22.28765869140625, -177.27906799316406, 25.775650024414062, 40.29924774169922, 68.23907470703125, 252.48324584960938, 125.70254516601562, 175.67564392089844, 72.083251953125, 61.355621337890625, -82.71571350097656, -146.6446533203125, -2.987548828125, 12.70172119140625, -60.434967041015625, 115.23297119140625, 204.72247314453125, 82.44375610351562, 19.570228576660156, 68.52252197265625, 64.47195434570312, 98.6949462890625, 286.54522705078125, 29.973127365112305, -29.58527374267578, 196.76141357421875, 44.23968505859375, 281.8556213378906, 200.46963500976562, 332.6080322265625, 81.04183959960938, 44.52062225341797, 0.0, 28.803977966308594, 89.29815673828125, 35.949424743652344, 29.4873046875, 88.70132446289062, -92.41184997558594, 205.6600799560547, 130.56784057617188, 102.3534927368164, -13.636543273925781, -85.7520980834961, 265.112548828125, 114.61981201171875, 125.81797790527344, -19.595458984375, 42.81245422363281, 249.15875244140625, 197.1441192626953, -8.318496704101562, 266.18792724609375, 148.46755981445312, -98.23907470703125, 94.52377319335938, 281.7471923828125, 35.728271484375, -114.13375091552734, 174.6984100341797, 127.74783325195312, -101.2686767578125, 222.3358154296875, 135.3177490234375, 172.97842407226562, -18.15374755859375, -67.80551147460938, -33.07698059082031, 14.477767944335938, 187.95909118652344, 19.323240280151367, -223.7242431640625, -86.45590209960938, 139.17991638183594, -102.73622131347656, 138.68728637695312, 59.32688903808594, -156.85256958007812, 81.03726196289062, -18.55340576171875, 88.32142639160156, 40.50056076049805, 128.49993896484375, -6.68756103515625, 80.19915771484375, 95.3275146484375, 287.31939697265625, 178.40853881835938, 240.52093505859375, 71.94400024414062, 198.71893310546875, 210.94491577148438, 123.58305358886719, 106.1558837890625, 220.57394409179688, 78.93449401855469, 15.8057861328125, 150.75143432617188, 26.2091064453125, 193.05145263671875, 49.961181640625, 102.06002807617188, 73.94548034667969, 203.52899169921875, 68.06396484375, 58.93693542480469, 14.828125, 202.8975830078125, 35.66845703125, -266.4739074707031, 26.128814697265625, 71.23684692382812, 51.58197021484375, 55.908660888671875], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000474.npy"}
{"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 71.22021484375, "std": 128.572998046875, "min": -233.909912109375, "p10": -64.14126892089843, "median": 44.62589645385742, "p90": 238.3037902832031, "max": 536.2258911132812, "pos_frac": 0.7265625, "sample": [228.892578125, 370.65032958984375, -39.33808898925781, -15.398200988769531, 209.953125, -61.269744873046875, -233.909912109375, 265.18023681640625, -135.19810485839844, -119.00604248046875, -92.27925109863281, -24.177093505859375, -13.65728759765625, -17.01482582092285, 22.371917724609375, 70.75340270996094, 257.98077392578125, 190.7971954345703, 21.090240478515625, 6.2228240966796875, 196.04965209960938, -37.55863952636719, -18.707870483398438, 240.55563354492188, 180.02108764648438, 126.88801574707031, 226.95745849609375, -20.241668701171875, 266.1522216796875, -0.9944381713867188, 42.76116943359375, 40.95855712890625, -59.971527099609375, 206.14703369140625, 32.138458251953125, 90.09754943847656, -79.1030044555664, 145.5015869140625, -57.95787048339844, 210.68572998046875, -44.5111083984375, 80.39132690429688, 180.2598876953125, 28.3134765625, 536.2258911132812, -20.62628173828125, 29.362533569335938, -44.3817138671875, 109.40777587890625, -55.231689453125, 38.06611633300781, 99.23624420166016, 288.85992431640625, 138.53509521484375, -85.48995971679688, 92.129150390625, 193.49417114257812, -15.129077911376953, 41.807098388671875, 39.8583984375, 141.10171508789062, -37.00829315185547, 43.908546447753906, -135.73422241210938, 79.10137939453125, 52.97169494628906, 100.64753723144531, 158.69815063476562, -45.587677001953125, 68.21527099609375, 16.7235107421875, 237.33871459960938, 21.12281036376953, 74.71783447265625, 67.4615478515625, 13.7283935546875, -6.182136535644531, 338.5395202636719, 14.3209228515625, 13.583412170410156, 13.991714477539062, 255.1678466796875, 85.71185302734375, -33.768646240234375, 195.93931579589844, -53.42509460449219, 75.90728759765625, 4.642333984375, 62.956939697265625, 108.79849243164062, 21.99352264404297, 84.47977447509766, 65.33767700195312, 126.73162841796875, 25.43536376953125, 273.01983642578125, 166.96298217773438, 232.06552124023438, 331.7190856933594, 83.83189392089844, -144.58213806152344, -70.84149169921875, 212.24899291992188, 31.31829833984375, 30.242584228515625, 115.39349365234375, -171.05377197265625, 64.04501342773438, 48.59912109375, 18.187511444091797, 206.56134033203125, 16.388565063476562, 29.218982696533203, -125.42306518554688, 170.27349853515625, -156.98806762695312, 108.56246948242188, 63.48834228515625, -87.568115234375, 14.6524658203125, 481.58160400390625, 147.300048828125, 37.69014358520508, 300.0611572265625, 7.7136077880859375, 46.9658203125, 45.34324645996094, 78.03958129882812], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000475.npy"}
{"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 81.7248306274414, "std": 143.4490966796875, "min": -217.116455078125, "p10": -80.8950828552246, "median": 61.16484069824219, "p90": 233.47684936523427, "max": 630.7223510742188, "pos_frac": 0.7265625, "sample": [119.1998291015625, -155.668212890625, -70.61293029785156, -83.94140625, 10.62005615234375, 175.32296752929688, 15.620811462402344, 43.125572204589844, 47.19781494140625, -44.64128112792969, 176.71646118164062, 39.486083984375, 343.17657470703125, 269.5062255859375, 159.36996459960938, 455.0921630859375, 109.95870971679688, 204.30838012695312, 137.39456176757812, -73.91414642333984, -11.684944152832031, 21.91021728515625, 62.85369873046875, 77.93185424804688, 161.4578399658203, 42.833213806152344, 630.7223510742188, -79.58951568603516, -194.6102294921875, 264.6173095703125, 70.01239013671875, 22.8177490234375, -53.891754150390625, 127.7218017578125, 184.59365844726562, 43.553009033203125, 214.3570556640625, -2.9641647338867188, 59.4227294921875, -120.21026611328125, -34.62420654296875, -129.3856201171875, 143.38624572753906, 43.66142272949219, 263.758056640625, 186.2222900390625, 152.74647521972656, 209.22647094726562, 210.02191162109375, -148.66522216796875, 59.475982666015625, 192.2333984375, -65.210205078125, -15.156524658203125, -14.181209564208984, 290.067626953125, 169.45016479492188, -39.44232177734375, 171.42868041992188, 30.695892333984375, 169.7249755859375, 127.01087951660156, -60.485748291015625, 341.675048828125, 136.32623291015625, 178.70147705078125, 124.40673828125, 30.488037109375, 85.07249450683594, 97.546875, 57.21044921875, -110.48683166503906, 281.3718566894531, 27.5291748046875, 106.11233520507812, -24.01766586303711, 183.24896240234375, -217.116455078125, 110.97213745117188, 179.5418701171875, 36.73881530761719, 168.82711791992188, 149.061279296875, 39.91192626953125, 47.79438400268555, 36.37078094482422, 9.91033935546875, 190.40701293945312, -145.68576049804688, 156.69439697265625, -9.667816162109375, 21.2430419921875, 159.9493408203125, 9.142425537109375, 83.18366241455078, -169.64755249023438, -43.8642578125, 117.51326751708984, 17.86102294921875, -71.9294662475586, -69.8463134765625, 50.076873779296875, -5.84088134765625, 215.77774047851562, 57.33612060546875, 119.92585754394531, -145.47830200195312, 51.193016052246094, -34.12645721435547, 606.437744140625, 101.931884765625, 260.26129150390625, 174.07504272460938, 13.50555419921875, 70.18186950683594, -42.411224365234375, 422.5810546875, -105.52505493164062, 163.17872619628906, 65.01547241210938, 221.997802734375, 112.82267761230469, -159.138427734375, 286.58074951171875, 17.57171630859375, 186.86065673828125, -1.2810516357421875, 123.5868911743164], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000476.npy"}
{"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 79.20506286621094, "std": 150.56935119628906, "min": -320.6339111328125, "p10": -100.17669296264648, "median": 71.16647338867188, "p90": 274.3233245849609, "max": 479.29693603515625, "pos_frac": 0.7109375, "sample": [-164.01708984375, -14.3031005859375, 153.323486328125, -61.06036376953125, 239.37673950195312, 56.680389404296875, -146.4302978515625, 31.010574340820312, -189.1700439453125, 11.164566040039062, 3.6703643798828125, 274.7849426269531, -160.19711303710938, 219.4519500732422, -91.84951782226562, 3.84033203125, 247.71450805664062, 154.6114501953125, -307.28863525390625, 78.16207885742188, 39.118377685546875, -56.966064453125, -2.5393295288085938, 258.6062316894531, 378.9751892089844, 156.01025390625, 247.173095703125, -7.249240875244141, 70.46919250488281, 367.7933044433594, 178.14663696289062, 208.487060546875, -62.984649658203125, 433.3326721191406, -14.199493408203125, 86.79998779296875, 247.23919677734375, 85.23177337646484, -80.13461303710938, 0.0, 95.05511474609375, 97.4134750366211, -106.1031494140625, -51.57415771484375, 167.99978637695312, -103.02693176269531, 143.37716674804688, 338.328125, 218.30670166015625, -134.87567138671875, -91.7208251953125, 7.1253509521484375, -320.6339111328125, 199.32049560546875, 71.3460693359375, 14.112548828125, 165.86416625976562, 133.62860107421875, 164.9329833984375, 238.77462768554688, 75.70074462890625, 178.80377197265625, 392.88580322265625, -103.23382568359375, 105.71734619140625, 46.39476776123047, 315.74346923828125, -11.610748291015625, 69.37821960449219, -46.88067626953125, 36.71556091308594, 274.12548828125, 82.88768005371094, 81.81631469726562, 479.29693603515625, 253.50860595703125, -64.21143341064453, 33.89384460449219, 2.6240158081054688, 63.13407897949219, -73.56318664550781, 154.6243438720703, 43.15936279296875, 81.37760925292969, -57.056396484375, -98.95516204833984, 246.70384216308594, 92.0731430053711, 110.49738311767578, 71.41513061523438, -9.329042434692383, 35.382843017578125, 54.19642639160156, 318.7137451171875, 21.487152099609375, 185.44155883789062, 17.91180419921875, 43.527374267578125, 54.867034912109375, 153.49620056152344, -259.7611083984375, 301.35992431640625, -79.18684387207031, -159.8427734375, 166.57763671875, 30.448516845703125, -82.40298461914062, -23.2777099609375, 293.6719970703125, 125.24423217773438, 70.98687744140625, 368.45892333984375, 245.1782684326172, 320.66107177734375, -9.059707641601562, 194.9947967529297, 30.74090576171875, -107.19168090820312, 91.04379272460938, 77.12075805664062, 215.562744140625, 146.2764892578125, 145.0851593017578, 49.309078216552734, -68.6371841430664, 127.25215911865234, 86.860107421875, 7.677276611328125], "npy": "/workspace/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.45-20260427-221551/margin_logs/step_0000477.npy"}

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0958d55d9aa63bf5cf60af91758dfe1b62e0da6f406ef3c08479561160618afb
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bd0b1ba21a2866f6f520d0a636efe0ff98d6169719559bc91cac84cb769740d0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7dbf6efa6449a8baba7298eb22e521fa7a5e98e145685737914cd4b745dd62ed
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f82b324f0b4ec8d25e070bee819a5144ba5532878057c00572724f4440dfc915
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:53d4de4e24f1e6faa76bd5ea95c188b371d2bc0ec7325ca0d5568f1155dc1933
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2cf5d273719b0bc03beb0157466a3e1667a1bc23a493f209bfa01fe145baaef3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5211992d6d4a2172c5fda3bd08194cd2a2d7da114bef1bdfe7ca382978b74970
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f0682e49dfe952cd88c781f84eb5a2cb2025a9472a843bd3c8e588695fecf137
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6c3594dc71b07de3ef3659b0661b39ea38ae96eaa78407aafe3d0adf6746baf6
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a8617a8c80bc4efa57925e427d2276666501e108f3ce2fc9fbbd6528821b722b
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ca9413ea0ada6b8f05bedad6ab36e8143fa8c5981b71c463bb5518e0ff73d6e1
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4d9cc89fe3d782a24178125bb976a0d0648b8e993bebfc98f2c2fef96d2913a4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d271ca7fc00692c24b64befd3be6357ce7046a5a4a011af3abe04fa2ece1c47a
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:83732b4171b69237f555a7973aec2438b0c2928fc7945cffc370e7a74d280ac8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:21a10918b1f4a8aa285393e791b1e9f53547ace9771168cfb3290af794570ac1
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:25665c2033be6989abc2cea6642e4ee9fe8c4a9729d835b008130acbceed217b
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:fb3afb415d49de79fe5ea637cc6fb12c4e41886cf4c27ad816b36b5002e9dd78
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:012a9b719eeb33d1901c1be3583ed4a4360c277fcfe055567669f9686dc9d5ff
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b9ef56433bb868d52eed8ec28c0bc03b385f2fb308f14dfa0d74d4752d2dd289
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:dfa44dd08431925bdcebb29cca7e5187d21f081dbefc95a16d5a2de82f5319de
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:09786aacbcd92306a37a26dd89f375954e978c81735def395ce059684d1f7dd5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0f840808661de7074b71ead9b737d39cbe92a7380e76204dfb07a52b572791f8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:45ca87338e16fffd728eb7ca9b4e2bb91f5e686997dbc83ed977170864dba3a4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d1480610e3417a231857a0961189aea36308d2ba17d918ce39c7976e7f5d825c
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:86c194a9d76c46bbafa8b43fadcb094146b544bc928c349f0804bc7092a7236e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:29ac2d61f9ac4a18e1cda7fe5425aa44313ad3528a35e94df639757d71882f97
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cebb8ea8bedcd10270df73b8c3ca7cc7777fb3b7aa880202c773f5e319cb1fa0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3e50aa39d285b83d7394c6b7357d0754aefb70f2a6916384e7dda52b21a9b916
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e8ba7d22e75fd2e309cdc92dfc2eb4fbe3c2095b9f3e143cb64a3e6ea9c868c6
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:545a9197117de5a2536f812fdcf3f2372cbdcc6dd0c61f02a1cc1af6964b6cb7
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:fb7d3c00ca9ae4761d27ebca69f51d94adde10331d50873859c3a17bcd6544c2
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:74aa1e993891809a4091c347df193726565ba5ffa16f6a1186ce2aa0ebb9d44e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:779f98c854d74c2c6e4ec5be8274cdd7792f083022e2566f1d90a40fa59d86c8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2b9cee52fe50d2c5e48a30456c23cba042fb513bcc713e0c794b16d9f19cdae5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0b850f26a7cea659a20986c06d00c5bd3b3f0e41f870ae6466fe7eeb56f464f1
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f49d243a785dc23efd0bbade83e9638b033b592e2d98ab13b8f9bb799cfde2e9
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7d55257b6403f05ed45f080f35a49b6bba1289de6b2dd1ddef49990fcc2a81e5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cfa43f04f54550b90a011d7e9df7a90dcf342271997113024bf9f64a1f7ba997
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8bf6b4240077922758b2f85fe2d33a5cbeb622a4fd4153812a18a6ac8e7f73d9
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8521fc199dfa05671b12f00e26811239012d1e038dea9f9083463e3347393e9e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bc23abccdce3008132e1167da79703295b6d8eecd24792b3ac6eb63ae233a136
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e1c7c628d6d219f7f1d00dc37383d6499c3026399fee44672ec3dd1c5303ad26
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:54487681458d6c09666ed803716e250fd5558aa1ff8cf2e32c56dcaddc69de1f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:dc85af3877feb5482b044ef747be5218bd0221fce3c1f33ef8de983998746c64
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1c9a75cc914397d6884dc08ba33b0ee52ab1aaf892394f9c9da91047908b8339
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:62059d2b6c04744799c49c1744c073dc89f91b796bbdbbf9e30729f5d6e889f8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2a21e7d9cf248c87b7a472eaf1de3420d388c2fe056699d4d8ffa42a7e21ccd5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:057d15dd867a5279a470eaf7f91e0427c31a73ae8069b99ad537355f7898ff5e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:39548e314c3e90bbe916782f9e3a0db381a8d95acbd9ae250ae939d422b0b681
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ba3e236bb66652c804ca1e84ea7e9b12c1c1ac1c5d982203dc55125c0cc7adc6
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ce74af644fd8750b7d1437cb94c4ec6162e7cc780bc6050d4a7fcdffde560ee5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:204ae350ac9de5f0bb773669275d9c9916366d019b3fa09d498130b6d13ac5b9
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5e710368f2b51673baaeedc7d15833e9c08e62f5ad8e4e75ea17d69c696c0258
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:051747b3e8c14ba936969d9b4d6787245fc9434f1d21b8fc0eb713494fdaacd0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7e982f9a7393c10a6ced6108c18fead4915edf05707738c78e46cafd631ed34f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:01a5e157ecd274419abc7ef2a1517244d2dde35d254857cc09516e50fae3be2d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c234788c327dda9dbf53bfe0300ef276ff9b7e0bf60bfda696c4de29921bae24
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7cfd3874fd15e8738f0c83845196d076b83b5465e45b50d9e01e528d4b424406
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:490f62e1616a7fddc6a4f92bd44760a46aea11c9f57236417d3b854d7e33b408
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cf199ea20ec3cc408772109389fecc5ae62044f6d73aad0bb5e70fcacbfc5783
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:78f35b974339c9a5c4eba731d157dcc695c355c9ead95e4a65435da55cef624d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ca54bd7714847dc9064e80a1335ed5cf97bb7017f36522af203624c249c0784d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6a33bae420b9f30cc632aeccbdd8f35786679ed06a36c0503674099add7226b3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:69adb8409e4dbe51de539666dd38e8bae2f4640ac4958a00619387bfe4299012
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2d9ad86495446cdd0b8a06ebe263623ca48cded896e683dbb4701e171395c9bc
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:70f35ce9977ca5c4a3b1b2a6f678a8adee1a90160ab2634e4bc7f88c86faa7ec
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a5a3f6d625b52aa289e76a4f8fe14b830285b8fee862f179c9316aa7e30c56ab
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e9e0d4768638751348e7fa0fa5f5850d6f7940d07ce6bd6d0a6e5a673571a6dd
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c58fe24479f915eee2e309d92236964dd6c44d5e88914d3158b9fb45d240c297
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:96921b0c6c493b06e113857f155bf2b1e1cb3709eb9c4623ff548f0ad6213e20
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:799a7cc2d5c2962a8fe62ae80e0c5252ea3b7280bf95055d25f66ca86a352344
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8819cfe24246f7e19b37eeca7fc9fbb4ea18fd251f1f44db333eb71fccfb2674
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:09088eb33633abe689ccb550f95d0944658ed0d88903f921b1251997daae55fc
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4f155e86bb185fe8d2889b0b7806a06e8fe68dc30afceaca1be2fd3976b7859f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:76c7f66882789b6cf85546e11830e8d45b5a9b29b7fd863a6adbac9d3a22b0a9
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6259cd8595a5a4541fa80eba7d1dbe0baef4c58bb6b8affadba48798c9d0fb4c
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:090efb6df0b5dc91f8c67aac06a0058f4a3f82be7a80142474912c1a99c224f4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5377bc93cea8bbbba105e49d2cfbaec5aa020dd5cf713ead166fbf32d6aabb61
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2da9ad358c437ec8f3a796c5c5ecad3858bab7995ac8c51f290d57a0ba86dd29
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b20a3eb78fa63fc693e20be53ce9714c7d70e68aa5a4f11cc43f82c22fda09f3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e6833a9f151df682b6d170f60cbbaabfe972f1f48aabd19a998b8b05e11d5121
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5d1393dea2ccc9b02f38eb2ff97734f6464ce24cba000932fdc3867071ef6258
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1741b6c0f8214fb32434318080ef8cd1da1eb26c622b8544096064d7febc4c7e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b51dab35d5f0a43d4324cda509b382d2dc9d4f91b4a470f9d63a54134bef40c6
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b27679cc20d4f1da034ccfd63a96ac76427e180baac2f9cc631bc58c9e08d51a
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b045317a01290f978828a54455b18191ee037ae588138bb98edf3894a02351ab
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0b16c82cab1afbfd485b2108cd2edfc4c75e2f44c002d1ed6091dc2b877ce5e3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e9191e7e404d526bdd057735f31e6806d13781c9c26cf1a12c39bde6741ba035
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:54ad01998142c8c1e21d8f012b8b420fb344b36b4cce277b0ac87392659cb940
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0612366c5e0e5f1320d0393cf37c5e8d7d6e16f7c32a60d3cf5cc51d9693837d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d3370064d40b6c1551aca065e2eff838b314919fe26b964fd2863f2205e5cc25
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1bc9202aed0a08ec92180804318b63f1e043519a46f4c86224890b633370ddc5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ffcc16527f0762a175253fe9affb1c73e72a9a373d11e9f7be6e714adfa4ad56
size 640

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