初始化项目，由ModelHub XC社区提供模型
Model: promotion/qwen3-8b-aaai27-flagship-sppo-avg-s44 Source: Original Platform
This commit is contained in:
ModelHub XC
36
.gitattributes
vendored
Normal file
36
.gitattributes
vendored
Normal 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
|
||||
24
README.md
Normal file
24
README.md
Normal file
@@ -0,0 +1,24 @@
|
||||
---
|
||||
library_name: transformers
|
||||
base_model: Qwen/Qwen3-8B
|
||||
tags:
|
||||
- ronpo
|
||||
- mnpo
|
||||
- preference-optimization
|
||||
---
|
||||
|
||||
# qwen3-8b-aaai27-flagship-sppo-avg-s44
|
||||
|
||||
Research checkpoint for the RONPO AAAI revision experiments.
|
||||
|
||||
- Method: sppo_avg
|
||||
- Base model: `Qwen/Qwen3-8B`
|
||||
- Seed: 44
|
||||
- Local source at upload time: `/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/flagship_20260712/full/sppo_avg/seed44/attempt2`
|
||||
- Uploaded at UTC: 2026-07-13T06:41:09Z
|
||||
- Run status metadata: `{"attempt": 2, "effective_batch_size": 16, "method": "sppo_avg", "objective_protocol": "objective_protocol.json", "optimizer_steps": 900, "resource_profile": "resource_profile.json", "seed": 44, "split_manifest": "split_manifest.json", "stability_gate": "stability_gate.json", "status": "completed", "wandb_run_id": "0bbaa979875e", "wandb_url": "https://wandb.ai/promotion-kim/mnpo/runs/0bbaa979875e"}`
|
||||
|
||||
AAAI-27 P1 matched budget: 900 optimizer steps, effective batch 16; passed non-thinking and collapse stability gates; sealed-test metrics are not used for checkpoint selection.
|
||||
|
||||
Intended use: reproducibility and evaluation for the RONPO paper. This
|
||||
checkpoint is not intended as a production assistant.
|
||||
9
all_results.json
Normal file
9
all_results.json
Normal file
@@ -0,0 +1,9 @@
|
||||
{
|
||||
"epoch": 0.8599068434252956,
|
||||
"total_flos": 0.0,
|
||||
"train_loss": 0.688900715245141,
|
||||
"train_runtime": 7064.5608,
|
||||
"train_samples": 16746,
|
||||
"train_samples_per_second": 2.038,
|
||||
"train_steps_per_second": 0.127
|
||||
}
|
||||
89
chat_template.jinja
Normal file
89
chat_template.jinja
Normal file
@@ -0,0 +1,89 @@
|
||||
{%- if tools %}
|
||||
{{- '<|im_start|>system\n' }}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- messages[0].content + '\n\n' }}
|
||||
{%- endif %}
|
||||
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
||||
{%- for tool in tools %}
|
||||
{{- "\n" }}
|
||||
{{- tool | tojson }}
|
||||
{%- endfor %}
|
||||
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
||||
{%- else %}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
||||
{%- for message in messages[::-1] %}
|
||||
{%- set index = (messages|length - 1) - loop.index0 %}
|
||||
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
||||
{%- set ns.multi_step_tool = false %}
|
||||
{%- set ns.last_query_index = index %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- for message in messages %}
|
||||
{%- if message.content is string %}
|
||||
{%- set content = message.content %}
|
||||
{%- else %}
|
||||
{%- set content = '' %}
|
||||
{%- endif %}
|
||||
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
||||
{%- elif message.role == "assistant" %}
|
||||
{%- set reasoning_content = '' %}
|
||||
{%- if message.reasoning_content is string %}
|
||||
{%- set reasoning_content = message.reasoning_content %}
|
||||
{%- else %}
|
||||
{%- if '</think>' in content %}
|
||||
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
||||
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- if loop.index0 > ns.last_query_index %}
|
||||
{%- if loop.last or (not loop.last and reasoning_content) %}
|
||||
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- if message.tool_calls %}
|
||||
{%- for tool_call in message.tool_calls %}
|
||||
{%- if (loop.first and content) or (not loop.first) %}
|
||||
{{- '\n' }}
|
||||
{%- endif %}
|
||||
{%- if tool_call.function %}
|
||||
{%- set tool_call = tool_call.function %}
|
||||
{%- endif %}
|
||||
{{- '<tool_call>\n{"name": "' }}
|
||||
{{- tool_call.name }}
|
||||
{{- '", "arguments": ' }}
|
||||
{%- if tool_call.arguments is string %}
|
||||
{{- tool_call.arguments }}
|
||||
{%- else %}
|
||||
{{- tool_call.arguments | tojson }}
|
||||
{%- endif %}
|
||||
{{- '}\n</tool_call>' }}
|
||||
{%- endfor %}
|
||||
{%- endif %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- elif message.role == "tool" %}
|
||||
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
||||
{{- '<|im_start|>user' }}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_response>\n' }}
|
||||
{{- content }}
|
||||
{{- '\n</tool_response>' }}
|
||||
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- if add_generation_prompt %}
|
||||
{{- '<|im_start|>assistant\n' }}
|
||||
{%- if enable_thinking is defined and enable_thinking is false %}
|
||||
{{- '<think>\n\n</think>\n\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
71
config.json
Normal file
71
config.json
Normal file
@@ -0,0 +1,71 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3ForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": null,
|
||||
"dtype": "bfloat16",
|
||||
"eos_token_id": 151645,
|
||||
"head_dim": 128,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 4096,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 12288,
|
||||
"layer_types": [
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention"
|
||||
],
|
||||
"max_position_embeddings": 40960,
|
||||
"max_window_layers": 36,
|
||||
"model_type": "qwen3",
|
||||
"num_attention_heads": 32,
|
||||
"num_hidden_layers": 36,
|
||||
"num_key_value_heads": 8,
|
||||
"pad_token_id": 151643,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_parameters": {
|
||||
"rope_theta": 1000000,
|
||||
"rope_type": "default"
|
||||
},
|
||||
"sliding_window": null,
|
||||
"tie_word_embeddings": false,
|
||||
"transformers_version": "5.13.0",
|
||||
"use_cache": true,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936
|
||||
}
|
||||
58
config.yaml
Normal file
58
config.yaml
Normal file
@@ -0,0 +1,58 @@
|
||||
model_name_or_path: /NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/flagship_20260712/cache/huggingface/hub/models--Qwen--Qwen3-8B/snapshots/b968826d9c46dd6066d109eabc6255188de91218
|
||||
torch_dtype: null
|
||||
attn_implementation: sdpa
|
||||
dataset_mixer:
|
||||
/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/flagship_20260712/precomputed/avg: 1.0
|
||||
dataset_splits:
|
||||
- train
|
||||
- test
|
||||
preprocessing_num_workers: 4
|
||||
bf16: true
|
||||
loss_type: sppo
|
||||
eta: 0.0075
|
||||
ratio: 0.3333
|
||||
max_history_t: 1
|
||||
history_weights:
|
||||
- 1.0
|
||||
dpo_beta: 0.05
|
||||
simpo_beta: 2.0
|
||||
simpo_gamma: 0.6
|
||||
ronpo_alpha: 0.5
|
||||
ronpo_tau: 0.05
|
||||
ronpo_target_column: ronpo_target
|
||||
reference_anchor_weight: 0.02
|
||||
preference_sft_weight: 0.002
|
||||
ht_target_column: ht_target
|
||||
ht_target_scale: 1.0
|
||||
beta: 10.0
|
||||
learning_rate: 2.5e-07
|
||||
lr_scheduler_type: cosine
|
||||
warmup_ratio: 0.1
|
||||
optim: adamw_torch
|
||||
weight_decay: 0.0
|
||||
max_grad_norm: 1.0
|
||||
seed: 44
|
||||
gradient_accumulation_steps: 16
|
||||
gradient_checkpointing: true
|
||||
num_train_epochs: 1
|
||||
max_steps: 900
|
||||
per_device_train_batch_size: 1
|
||||
per_device_eval_batch_size: 1
|
||||
max_length: 2048
|
||||
max_prompt_length: 1800
|
||||
do_eval: false
|
||||
eval_strategy: 'no'
|
||||
logging_steps: 5
|
||||
log_level: info
|
||||
generate_during_eval: false
|
||||
load_best_model_at_end: false
|
||||
save_strategy: steps
|
||||
save_steps: 900
|
||||
save_total_limit: 1
|
||||
save_only_model: true
|
||||
save_safetensors: true
|
||||
push_to_hub: false
|
||||
report_to:
|
||||
- wandb
|
||||
output_dir: /NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/flagship_20260712/full/sppo_avg/seed44/attempt2
|
||||
run_name: aaai27-flagship-full-sppo_avg-s44-a2
|
||||
12
generation_config.json
Normal file
12
generation_config.json
Normal file
@@ -0,0 +1,12 @@
|
||||
{
|
||||
"do_sample": true,
|
||||
"eos_token_id": [
|
||||
151645,
|
||||
151643
|
||||
],
|
||||
"pad_token_id": 151643,
|
||||
"temperature": 0.6,
|
||||
"top_k": 20,
|
||||
"top_p": 0.95,
|
||||
"transformers_version": "5.13.0"
|
||||
}
|
||||
23
job_status.json
Normal file
23
job_status.json
Normal file
@@ -0,0 +1,23 @@
|
||||
{
|
||||
"status": "completed",
|
||||
"stage": "full",
|
||||
"method": "sppo_avg",
|
||||
"seed": 44,
|
||||
"attempt": 2,
|
||||
"gpu": 6,
|
||||
"gpus": [
|
||||
6
|
||||
],
|
||||
"returncode": 0,
|
||||
"finite_metrics": true,
|
||||
"oom": false,
|
||||
"traceback": false,
|
||||
"reason": "passed",
|
||||
"optimizer_steps": 900,
|
||||
"effective_batch_size": 16,
|
||||
"wandb_run_id": "0bbaa979875e",
|
||||
"wandb_url": "https://wandb.ai/promotion-kim/mnpo/runs/0bbaa979875e",
|
||||
"output_dir": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/flagship_20260712/full/sppo_avg/seed44/attempt2",
|
||||
"log": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/flagship_20260712/logs/full_sppo_avg_s44_a2.log",
|
||||
"completed_at": "2026-07-13T06:38:46Z"
|
||||
}
|
||||
3
model.safetensors
Normal file
3
model.safetensors
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:ca34e78a293479e4a1f1f8140eed232f2cc11a5077f6f329f2bc00a0a7cd863c
|
||||
size 16381517208
|
||||
69
objective_protocol.json
Normal file
69
objective_protocol.json
Normal file
@@ -0,0 +1,69 @@
|
||||
{
|
||||
"schema_version": 1,
|
||||
"frozen_at_kst": "2026-07-12T21:40:00+09:00",
|
||||
"base_model": "Qwen/Qwen3-8B",
|
||||
"primary_triple": {
|
||||
"helpfulness": {
|
||||
"model": "RLHFlow/ArmoRM-Llama3-8B-v0.1",
|
||||
"head": "ultrafeedback-helpfulness",
|
||||
"transform": "identity"
|
||||
},
|
||||
"safety": {
|
||||
"model": "RLHFlow/ArmoRM-Llama3-8B-v0.1",
|
||||
"head": "beavertails-is_safe",
|
||||
"transform": "identity"
|
||||
},
|
||||
"conciseness": {
|
||||
"model": "RLHFlow/ArmoRM-Llama3-8B-v0.1",
|
||||
"head": "helpsteer-verbosity",
|
||||
"transform": "negate"
|
||||
}
|
||||
},
|
||||
"gate_thresholds": {
|
||||
"raw_score_std_strictly_greater_than": 0.01,
|
||||
"aggregate_prompt_pair_spearman_median_strictly_less_than": 0.0,
|
||||
"aggregate_pairwise_top1_mismatch_at_least": 0.5
|
||||
},
|
||||
"gate_pool": "source-train Qwen3-8B response pool split into deterministic train/validation; sealed source-test excluded",
|
||||
"training_seeds": [42, 43, 44],
|
||||
"optimizer_steps": 900,
|
||||
"effective_batch_size": 16,
|
||||
"decode": {
|
||||
"enable_thinking": false,
|
||||
"temperature": 0.7,
|
||||
"top_p": 0.9,
|
||||
"max_new_tokens": 2048,
|
||||
"decode_seed": 42
|
||||
},
|
||||
"methods": [
|
||||
"ronpo_full_expect",
|
||||
"ronpo_k_only",
|
||||
"dpo",
|
||||
"ipo",
|
||||
"simpo",
|
||||
"kto",
|
||||
"sppo_avg",
|
||||
"inpo_avg",
|
||||
"ht_mnpo_helpfulness",
|
||||
"ht_mnpo_safety",
|
||||
"ht_mnpo_conciseness"
|
||||
],
|
||||
"selection_policy": "No sealed-test selection. One preregistered configuration per method; failures use only the declared stability retry recipe, at most three attempts.",
|
||||
"stability_gate": {
|
||||
"validation_prompts": 128,
|
||||
"min_candidate_base_mean_word_ratio": 0.33,
|
||||
"max_candidate_base_mean_word_ratio": 2.0,
|
||||
"max_consecutive_identical_word_run": 20,
|
||||
"think_tag_count": 0,
|
||||
"empty_response_count": 0,
|
||||
"retry_recipe": "attempt 2 halves learning rate, enables gradient checkpointing, and adds reference=0.02/SFT=0.002 anchoring to unanchored unified losses; attempt 3 quarters learning rate and uses reference=0.05/SFT=0.005; RONPO uses reference=0.05/SFT=0.005 from attempt 1"
|
||||
},
|
||||
"primary_metric": "mean prompt-level minimum of per-prompt min-max-normalized helpfulness, safety, conciseness",
|
||||
"bootstrap": {
|
||||
"unit": "prompt",
|
||||
"paired": true,
|
||||
"resamples": 2000,
|
||||
"interval": "percentile_95",
|
||||
"seed": 42
|
||||
}
|
||||
}
|
||||
34
pair_manifest.json
Normal file
34
pair_manifest.json
Normal file
@@ -0,0 +1,34 @@
|
||||
{
|
||||
"schema_version": 1,
|
||||
"objectives": [
|
||||
"helpfulness",
|
||||
"safety",
|
||||
"conciseness"
|
||||
],
|
||||
"normalization": "per_prompt_minmax",
|
||||
"adversary": {
|
||||
"steps": 25,
|
||||
"alpha": 1.0,
|
||||
"kappa": 0.05,
|
||||
"preference_scale": 8.0
|
||||
},
|
||||
"ronpo_estimator": "Rao-Blackwellized full sigma expectation with three common response pairs per prompt",
|
||||
"counts": {
|
||||
"train": {
|
||||
"avg": 16746,
|
||||
"ronpo_full_expect": 50340,
|
||||
"ronpo_k_only": 50340,
|
||||
"ht_mnpo_helpfulness": 16743,
|
||||
"ht_mnpo_safety": 16764,
|
||||
"ht_mnpo_conciseness": 16755
|
||||
},
|
||||
"validation": {
|
||||
"avg": 1859,
|
||||
"ronpo_full_expect": 5592,
|
||||
"ronpo_k_only": 5592,
|
||||
"ht_mnpo_helpfulness": 1864,
|
||||
"ht_mnpo_safety": 1862,
|
||||
"ht_mnpo_conciseness": 1858
|
||||
}
|
||||
}
|
||||
}
|
||||
13
provenance.json
Normal file
13
provenance.json
Normal file
@@ -0,0 +1,13 @@
|
||||
{
|
||||
"method": "sppo_avg",
|
||||
"seed": 44,
|
||||
"attempt": 2,
|
||||
"optimizer_steps": 900,
|
||||
"effective_batch_size": 16,
|
||||
"wandb_run_id": "0bbaa979875e",
|
||||
"wandb_url": "https://wandb.ai/promotion-kim/mnpo/runs/0bbaa979875e",
|
||||
"split_manifest": "split_manifest.json",
|
||||
"objective_protocol": "objective_protocol.json",
|
||||
"resource_profile": "resource_profile.json",
|
||||
"stability_gate": "stability_gate.json"
|
||||
}
|
||||
21
resource_profile.json
Normal file
21
resource_profile.json
Normal file
@@ -0,0 +1,21 @@
|
||||
{
|
||||
"schema_version": 1,
|
||||
"measured_at_kst": "2026-07-12T22:38:00+09:00",
|
||||
"hardware": "NVIDIA B200 183359 MiB reported total; 178.35 GiB PyTorch-visible capacity",
|
||||
"smoke_evidence": {
|
||||
"attempted_profile": {"per_device_batch_size": 2, "gradient_accumulation_steps": 8, "gradient_checkpointing": false, "effective_batch_size": 16},
|
||||
"outcome": "OOM before completing the 20-step smoke across every launched unified baseline; representative process used 176.87 GiB and failed a 1.62 GiB allocation. KTO also OOMed at the optimizer step.",
|
||||
"paper_metric": false
|
||||
},
|
||||
"selected_full_training_profile": {
|
||||
"per_device_batch_size": 2,
|
||||
"gradient_accumulation_steps": 8,
|
||||
"gradient_checkpointing": false,
|
||||
"attention_implementation": "PyTorch SDPA with cuDNN SDPA disabled",
|
||||
"effective_batch_size": 16,
|
||||
"applies_symmetrically_to_all_unified_pairwise_methods": true,
|
||||
"kto_implementation_constraint": "TRL KTO requires actual batch size > 1. KTO uses batch 2 / accumulation 8 and identical AdamW with foreach=False. PyTorch SDPA is used with cuDNN SDPA disabled. The fixed base reference model is placed on a second B200 and its exact logits are returned to the policy device; the policy/optimizer remain on the first B200. This avoids a measured ~6-hour reference-precompute path while leaving the KTO loss, actual/effective batch, optimizer, and 900-step budget unchanged."
|
||||
},
|
||||
"retry_semantics": "Resource selection is separated from method-stability attempts. Full attempt 1 uses the smoke-validated SDPA batch-2 profile and retains the preregistered method LR/anchor; attempts 2 and 3 fall back to batch 1 / accumulation 16 / gradient checkpointing and apply the declared LR/anchor stabilization recipe.",
|
||||
"provenance_note": "No reward or sealed-evaluation result was available when this hardware profile was selected."
|
||||
}
|
||||
14
run_status.json
Normal file
14
run_status.json
Normal file
@@ -0,0 +1,14 @@
|
||||
{
|
||||
"method": "sppo_avg",
|
||||
"seed": 44,
|
||||
"attempt": 2,
|
||||
"optimizer_steps": 900,
|
||||
"effective_batch_size": 16,
|
||||
"wandb_run_id": "0bbaa979875e",
|
||||
"wandb_url": "https://wandb.ai/promotion-kim/mnpo/runs/0bbaa979875e",
|
||||
"split_manifest": "split_manifest.json",
|
||||
"objective_protocol": "objective_protocol.json",
|
||||
"resource_profile": "resource_profile.json",
|
||||
"stability_gate": "stability_gate.json",
|
||||
"status": "completed"
|
||||
}
|
||||
23
split_manifest.json
Normal file
23
split_manifest.json
Normal file
@@ -0,0 +1,23 @@
|
||||
{
|
||||
"schema_version": 1,
|
||||
"split_rule": "sort source-train prompts by sha256('flagship-v1||'+prompt_id); first round(10%) validation",
|
||||
"validation_fraction": 0.1,
|
||||
"source_dir": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/revision_qwen3_8b/full_iter1/precomputed/avg_oracle",
|
||||
"base_model": "Qwen/Qwen3-8B",
|
||||
"counts": {
|
||||
"train": 16780,
|
||||
"validation": 1864,
|
||||
"sealed_test": 604
|
||||
},
|
||||
"prompt_id_sha256": {
|
||||
"train": "fcf6f2ea5a0b7bcd41179e6f5288f4d850b2d04bdb729da91907bd18bf733a69",
|
||||
"validation": "a87ba6f519192164daa487b109a6547dd8cbde46a27a373d6da201ca2d061b8f",
|
||||
"sealed_test": "b398ced6af6cf73d731e1a5f32a0631d8060323d51eda1e375695566bf5f1867"
|
||||
},
|
||||
"overlaps": {
|
||||
"train_validation": 0,
|
||||
"train_sealed_test": 0,
|
||||
"validation_sealed_test": 0
|
||||
},
|
||||
"sealed_test_policy": "prompts only until S4a; no objective scores or model generations used in S0-S3"
|
||||
}
|
||||
38
stability_gate.json
Normal file
38
stability_gate.json
Normal file
@@ -0,0 +1,38 @@
|
||||
{
|
||||
"status": "passed",
|
||||
"passed": true,
|
||||
"base": {
|
||||
"records": 128,
|
||||
"empty_count": 0,
|
||||
"think_leak_count": 0,
|
||||
"think_leak_indices": [],
|
||||
"mean_words": 388.1484375,
|
||||
"max_words": 1379,
|
||||
"max_repeat_run": 4
|
||||
},
|
||||
"candidate": {
|
||||
"records": 128,
|
||||
"empty_count": 0,
|
||||
"think_leak_count": 0,
|
||||
"think_leak_indices": [],
|
||||
"mean_words": 393.4375,
|
||||
"max_words": 1254,
|
||||
"max_repeat_run": 4
|
||||
},
|
||||
"candidate_base_mean_word_ratio": 1.0136263913209749,
|
||||
"thresholds": {
|
||||
"min_length_ratio": 0.33,
|
||||
"max_length_ratio": 2.0,
|
||||
"max_repeat_run": 20,
|
||||
"expected_records": 128,
|
||||
"think_leakage": 0,
|
||||
"empty_count": 0
|
||||
},
|
||||
"checks": {
|
||||
"complete_records": true,
|
||||
"zero_empty": true,
|
||||
"zero_think_leakage": true,
|
||||
"length_near_base": true,
|
||||
"repeat_run_below_threshold": true
|
||||
}
|
||||
}
|
||||
3
tokenizer.json
Normal file
3
tokenizer.json
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
|
||||
size 11422650
|
||||
30
tokenizer_config.json
Normal file
30
tokenizer_config.json
Normal file
@@ -0,0 +1,30 @@
|
||||
{
|
||||
"add_prefix_space": false,
|
||||
"backend": "tokenizers",
|
||||
"bos_token": null,
|
||||
"clean_up_tokenization_spaces": false,
|
||||
"eos_token": "<|im_end|>",
|
||||
"errors": "replace",
|
||||
"extra_special_tokens": [
|
||||
"<|im_start|>",
|
||||
"<|im_end|>",
|
||||
"<|object_ref_start|>",
|
||||
"<|object_ref_end|>",
|
||||
"<|box_start|>",
|
||||
"<|box_end|>",
|
||||
"<|quad_start|>",
|
||||
"<|quad_end|>",
|
||||
"<|vision_start|>",
|
||||
"<|vision_end|>",
|
||||
"<|vision_pad|>",
|
||||
"<|image_pad|>",
|
||||
"<|video_pad|>"
|
||||
],
|
||||
"is_local": true,
|
||||
"local_files_only": false,
|
||||
"model_max_length": 2048,
|
||||
"pad_token": "<|endoftext|>",
|
||||
"split_special_tokens": false,
|
||||
"tokenizer_class": "Qwen2Tokenizer",
|
||||
"unk_token": null
|
||||
}
|
||||
9
train_results.json
Normal file
9
train_results.json
Normal file
@@ -0,0 +1,9 @@
|
||||
{
|
||||
"epoch": 0.8599068434252956,
|
||||
"total_flos": 0.0,
|
||||
"train_loss": 0.688900715245141,
|
||||
"train_runtime": 7064.5608,
|
||||
"train_samples": 16746,
|
||||
"train_samples_per_second": 2.038,
|
||||
"train_steps_per_second": 0.127
|
||||
}
|
||||
3643
trainer_state.json
Normal file
3643
trainer_state.json
Normal file
File diff suppressed because it is too large
Load Diff
3
training_args.bin
Normal file
3
training_args.bin
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:09132e3eb4c33f3135f06a68a2d822cc9c604599f0e7703c7037fe12e529beaf
|
||||
size 7057
|
||||
Reference in New Issue
Block a user