初始化项目，由ModelHub XC社区提供模型

Model: promotion/ronpo-qwen3-8b-fair-inpo-avg-s42
Source: Original Platform

.gitattributes

@@ -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

README.md

@@ -0,0 +1,12 @@
+---
+base_model: Qwen/Qwen3-8B
+library_name: transformers
+tags:
+- preference-optimization
+- ronpo
+- qwen3
+---
+
+# Qwen3-8B fair-demo checkpoint: inpo_avg
+
+Validation-selected candidate: `inpo_b`. Exact evaluator, selection, sweep, and training metadata are stored under `experiment/`.

all_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 0.8599068434252956,
+    "total_flos": 0.0,
+    "train_loss": 24.184869588216145,
+    "train_runtime": 7109.0777,
+    "train_samples": 16746,
+    "train_samples_per_second": 2.026,
+    "train_steps_per_second": 0.127
+}

chat_template.jinja

@@ -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 %}

config.json

@@ -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
+}

config.yaml

@@ -0,0 +1,58 @@
+model_name_or_path: /NHNHOME/26msit001_A/BASE/AIPR/sjkim/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/26msit001_A/BASE/AIPR/sjkim/revision_qwen3_8b/full_iter1/flagship_20260712/precomputed/avg: 1.0
+dataset_splits:
+- train
+- test
+preprocessing_num_workers: 4
+bf16: true
+loss_type: inpo
+eta: 0.1
+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: 0.075
+preference_sft_weight: 0.0075
+ht_target_column: ht_target
+ht_target_scale: 1.0
+beta: 10.0
+learning_rate: 5.0e-07
+lr_scheduler_type: cosine
+warmup_ratio: 0.2
+optim: adamw_torch
+weight_decay: 0.0
+max_grad_norm: 1.0
+seed: 42
+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/26msit001_A/BASE/AIPR/sjkim/revision_qwen3_8b/full_iter1/fair_demo_20260715/sweep/candidates/inpo_b
+run_name: qwen3-8b-fair-demo-inpo_b

experiment/evaluator_lock.json

@@ -0,0 +1,85 @@
+{
+  "status": "LOCKED_BEFORE_ANY_NEW_METHOD_RANKING",
+  "locked_at": "2026-07-15T20:20:29+09:00",
+  "objective_signals": [
+    "skywork",
+    "armo_safety",
+    "length_conciseness"
+  ],
+  "objective_semantics": [
+    "helpfulness",
+    "safety",
+    "conciseness"
+  ],
+  "primary": {
+    "name": "open_weight_panel_mean_prompt_worst_vs_base",
+    "definition": "For each prompt and objective, average win=1/tie=0.5/loss=0 over two A/B positions and two judges; take the minimum objective, then mean over prompts.",
+    "judges": [
+      {
+        "model": "openai/gpt-oss-120b",
+        "revision": "b5c939de8f754692c1647ca79fbf85e8c1e70f8a"
+      },
+      {
+        "model": "Qwen/Qwen3-32B",
+        "revision": "9216db5781bf21249d130ec9da846c4624c16137"
+      }
+    ],
+    "decode": {
+      "temperature": 0.0,
+      "top_p": 1.0,
+      "max_new_tokens": 512,
+      "seed": 42
+    },
+    "position_swap": true,
+    "validation_selection_rule": "Highest eligible mean prompt-level worst panel score within each method; an exact numeric tie is broken by candidate_id lexical order."
+  },
+  "secondary": {
+    "normalization": "For each locked signal, divide each evaluation prompt's candidate-minus-base raw delta by the population SD of the diagnostic base and matched-control scores; take the prompt-level minimum across objectives, then the mean. No per-prompt min-max.",
+    "diagnostic_control_scale": {
+      "skywork": 22.807301785782975,
+      "armo_safety": 0.3098083353203591,
+      "length_conciseness": 1.5891969646897472
+    },
+    "reward_signal_provenance": {
+      "skywork": {
+        "model": "Skywork/Skywork-Reward-V2-Llama-3.1-8B",
+        "revision": "cba2f842f3f1af2f1b2f0d35e794d789976390c5",
+        "semantics": "helpfulness"
+      },
+      "armo_safety": {
+        "model": "RLHFlow/ArmoRM-Llama3-8B-v0.1",
+        "revision": "eb2676d20da2f2d41082289d23c59b9f7427f955",
+        "head": "beavertails-is_safe",
+        "transform": "identity"
+      },
+      "length_conciseness": {
+        "deterministic": "-log1p(response_word_count)",
+        "tokenization": "Python str.split whitespace words"
+      }
+    },
+    "report_raw_paired_deltas": true
+  },
+  "bootstrap": {
+    "paired_prompt_resamples": 2000,
+    "seed": 42,
+    "interval": "percentile_95"
+  },
+  "power": {
+    "target_absolute_effect": 0.05,
+    "paired_sd": 0.13831629563357775,
+    "required_prompts_80pct_power": 61,
+    "planned_fresh_test_prompts": 1024,
+    "alpha_two_sided": 0.05,
+    "power": 0.8
+  },
+  "fresh_test_source": {
+    "dataset": "HuggingFaceH4/ultrachat_200k",
+    "revision": "8049631c405ae6576f93f445c6b8166f76f5505a",
+    "split": "test_sft"
+  },
+  "diagnostic_summary_sha256": "e365ed814926ac5c17777e0fc65d0299a45b81bd390542889bbd55b5453d4ffe",
+  "judge_diagnostic_lock_sha256": "fce2d97806053cabdf3d358806f4baa80df7f89d71d8c1c298883327d4e02635",
+  "prereg_sha256": "71cbd1ac17c3e807f2ece1cb0a82374bbe74edb654c2845b97c3222ed680f0d0",
+  "method_ranking_computed": false,
+  "spent_sealed_split_touched": false
+}

experiment/selection_lock.json

@@ -0,0 +1,99 @@
+{
+  "status": "VALIDATION_SELECTION_LOCKED_BEFORE_FRESH_TEST",
+  "locked_at": "2026-07-15T20:37:11+09:00",
+  "selection_metric": "open_weight_panel_mean_prompt_worst_vs_base",
+  "tie_break": "candidate_id lexical order",
+  "selected_by_method": {
+    "dpo": {
+      "candidate_id": "dpo_b",
+      "validation_primary": 0.40625,
+      "validation_primary_ci95": [
+        0.3837890625,
+        0.427734375
+      ],
+      "eligible_candidates": [
+        "dpo_a",
+        "dpo_b"
+      ]
+    },
+    "ht_mnpo_helpfulness": {
+      "candidate_id": "ht_help_b",
+      "validation_primary": 0.4072265625,
+      "validation_primary_ci95": [
+        0.3828125,
+        0.4296875
+      ],
+      "eligible_candidates": [
+        "ht_help_b"
+      ]
+    },
+    "ht_mnpo_safety": {
+      "candidate_id": "ht_safety_a",
+      "validation_primary": 0.4150390625,
+      "validation_primary_ci95": [
+        0.3925537109375,
+        0.4384765625
+      ],
+      "eligible_candidates": [
+        "ht_safety_a",
+        "ht_safety_b"
+      ]
+    },
+    "inpo_avg": {
+      "candidate_id": "inpo_b",
+      "validation_primary": 0.40625,
+      "validation_primary_ci95": [
+        0.3818359375,
+        0.427734375
+      ],
+      "eligible_candidates": [
+        "inpo_b"
+      ]
+    },
+    "ronpo_full_expect": {
+      "candidate_id": "ronpo_full_a",
+      "validation_primary": 0.4033203125,
+      "validation_primary_ci95": [
+        0.37890625,
+        0.42580566406249987
+      ],
+      "eligible_candidates": [
+        "ronpo_full_a",
+        "ronpo_full_b"
+      ]
+    },
+    "ronpo_k_only": {
+      "candidate_id": "ronpo_top_a",
+      "validation_primary": 0.412109375,
+      "validation_primary_ci95": [
+        0.3876953125,
+        0.43557128906249987
+      ],
+      "eligible_candidates": [
+        "ronpo_top_a"
+      ]
+    },
+    "sppo_avg": {
+      "candidate_id": "sppo_b",
+      "validation_primary": 0.4111328125,
+      "validation_primary_ci95": [
+        0.3866943359375,
+        0.4345703125
+      ],
+      "eligible_candidates": [
+        "sppo_b"
+      ]
+    }
+  },
+  "selected_ronpo_overall": "ronpo_top_a",
+  "failed_methods": [
+    "ht_mnpo_conciseness",
+    "ipo",
+    "simpo"
+  ],
+  "panel_summary_sha256": "8850f23e16cadf56a99e3d562d224ce5337ab36fa588d8a843382853e6ad17bd",
+  "evaluator_lock_sha256": "ab7b12c84932ed57acfec7edbcbc1ff1df5e9acd203869b1458343adf6e5b2c8",
+  "grid_sha256": "abfd982ef10c3bc71364752241150286ebf06be5e97f087596adf80e55d459e0",
+  "fresh_test_opened": false,
+  "spent_sealed_split_touched": false
+}

experiment/sweep_grid.json

@@ -0,0 +1,46 @@
+{
+  "schema_version": 1,
+  "status": "frozen_before_training_and_validation_ranking",
+  "frozen_at_kst": "2026-07-15T09:01:45+09:00",
+  "seed": 42,
+  "common": {
+    "optimizer_steps": 900,
+    "effective_batch_size": 16,
+    "per_device_train_batch_size": 1,
+    "gradient_accumulation_steps": 16,
+    "gradient_checkpointing": true,
+    "bf16": true,
+    "attn_implementation": "sdpa",
+    "cudnn_sdpa": false,
+    "lr_scheduler_type": "cosine",
+    "max_length": 2048,
+    "max_prompt_length": 1800,
+    "save_total_limit": 1,
+    "wandb_entity": "promotion-kim",
+    "wandb_project": "mnpo",
+    "wandb_group": "qwen3-8b-fair-demo-20260715"
+  },
+  "budget_rule": "Exactly two configs per reported method; no extension after validation rankings are visible.",
+  "candidates": [
+    {"id":"ronpo_full_a","method":"ronpo_full_expect","dataset":"ronpo_full_expect_kall","learning_rate":1e-7,"warmup_ratio":0.15,"ronpo_alpha":0.25,"ronpo_tau":0.10,"reference_anchor_weight":0.075,"preference_sft_weight":0.0075,"expected_support_k":0},
+    {"id":"ronpo_full_b","method":"ronpo_full_expect","dataset":"ronpo_full_expect_k6","learning_rate":2.5e-7,"warmup_ratio":0.20,"ronpo_alpha":0.35,"ronpo_tau":0.05,"reference_anchor_weight":0.10,"preference_sft_weight":0.01,"expected_support_k":6},
+    {"id":"ronpo_top_a","method":"ronpo_k_only","dataset":"ronpo_k_only_k1","learning_rate":5e-8,"warmup_ratio":0.20,"ronpo_alpha":0.15,"ronpo_tau":0.10,"reference_anchor_weight":0.10,"preference_sft_weight":0.01,"expected_support_k":1},
+    {"id":"ronpo_top_b","method":"ronpo_k_only","dataset":"ronpo_k_only_k2","learning_rate":1e-7,"warmup_ratio":0.15,"ronpo_alpha":0.25,"ronpo_tau":0.10,"reference_anchor_weight":0.075,"preference_sft_weight":0.0075,"expected_support_k":2},
+    {"id":"dpo_a","method":"dpo","dataset":"avg","learning_rate":2.5e-7,"warmup_ratio":0.10,"dpo_beta":0.05,"reference_anchor_weight":0.02,"preference_sft_weight":0.002},
+    {"id":"dpo_b","method":"dpo","dataset":"avg","learning_rate":5e-7,"warmup_ratio":0.15,"dpo_beta":0.10,"reference_anchor_weight":0.05,"preference_sft_weight":0.005},
+    {"id":"ipo_a","method":"ipo","dataset":"avg","learning_rate":2.5e-7,"warmup_ratio":0.15,"dpo_beta":0.20,"reference_anchor_weight":0.05,"preference_sft_weight":0.005},
+    {"id":"ipo_b","method":"ipo","dataset":"avg","learning_rate":5e-7,"warmup_ratio":0.20,"dpo_beta":0.50,"reference_anchor_weight":0.075,"preference_sft_weight":0.0075},
+    {"id":"simpo_a","method":"simpo","dataset":"avg","learning_rate":2.5e-7,"warmup_ratio":0.10,"simpo_beta":1.0,"simpo_gamma":0.3,"reference_anchor_weight":0.02,"preference_sft_weight":0.002},
+    {"id":"simpo_b","method":"simpo","dataset":"avg","learning_rate":5e-7,"warmup_ratio":0.15,"simpo_beta":2.0,"simpo_gamma":0.6,"reference_anchor_weight":0.05,"preference_sft_weight":0.005},
+    {"id":"sppo_a","method":"sppo_avg","dataset":"avg","learning_rate":2.5e-7,"warmup_ratio":0.15,"eta":0.005,"reference_anchor_weight":0.05,"preference_sft_weight":0.005},
+    {"id":"sppo_b","method":"sppo_avg","dataset":"avg","learning_rate":5e-7,"warmup_ratio":0.20,"eta":0.010,"reference_anchor_weight":0.075,"preference_sft_weight":0.0075},
+    {"id":"inpo_a","method":"inpo_avg","dataset":"avg","learning_rate":2.5e-7,"warmup_ratio":0.15,"eta":0.050,"ratio":0.3333,"reference_anchor_weight":0.05,"preference_sft_weight":0.005},
+    {"id":"inpo_b","method":"inpo_avg","dataset":"avg","learning_rate":5e-7,"warmup_ratio":0.20,"eta":0.100,"ratio":0.3333,"reference_anchor_weight":0.075,"preference_sft_weight":0.0075},
+    {"id":"ht_help_a","method":"ht_mnpo_helpfulness","dataset":"ht_mnpo_helpfulness","learning_rate":2.5e-7,"warmup_ratio":0.10,"eta":0.005,"reference_anchor_weight":0.02,"preference_sft_weight":0.002},
+    {"id":"ht_help_b","method":"ht_mnpo_helpfulness","dataset":"ht_mnpo_helpfulness","learning_rate":5e-7,"warmup_ratio":0.15,"eta":0.010,"reference_anchor_weight":0.05,"preference_sft_weight":0.005},
+    {"id":"ht_safety_a","method":"ht_mnpo_safety","dataset":"ht_mnpo_safety","learning_rate":2.5e-7,"warmup_ratio":0.10,"eta":0.005,"reference_anchor_weight":0.02,"preference_sft_weight":0.002},
+    {"id":"ht_safety_b","method":"ht_mnpo_safety","dataset":"ht_mnpo_safety","learning_rate":5e-7,"warmup_ratio":0.15,"eta":0.010,"reference_anchor_weight":0.05,"preference_sft_weight":0.005},
+    {"id":"ht_concise_a","method":"ht_mnpo_conciseness","dataset":"ht_mnpo_conciseness","learning_rate":2.5e-7,"warmup_ratio":0.10,"eta":0.005,"reference_anchor_weight":0.02,"preference_sft_weight":0.002},
+    {"id":"ht_concise_b","method":"ht_mnpo_conciseness","dataset":"ht_mnpo_conciseness","learning_rate":5e-7,"warmup_ratio":0.15,"eta":0.010,"reference_anchor_weight":0.05,"preference_sft_weight":0.005}
+  ]
+}

experiment/training_status.json

@@ -0,0 +1,20 @@
+{
+  "status": "completed",
+  "candidate_id": "inpo_b",
+  "method": "inpo_avg",
+  "seed": 42,
+  "gpu": 0,
+  "pid": 1079779,
+  "config": "/NHNHOME/26msit001_A/BASE/AIPR/sjkim/revision_qwen3_8b/full_iter1/fair_demo_20260715/sweep/candidates/inpo_b/config.yaml",
+  "dataset": "/NHNHOME/26msit001_A/BASE/AIPR/sjkim/revision_qwen3_8b/full_iter1/flagship_20260712/precomputed/avg",
+  "optimizer_steps": 900,
+  "effective_batch_size": 16,
+  "wandb_run_id": "098086245ba4",
+  "wandb_url": "https://wandb.ai/promotion-kim/mnpo/runs/098086245ba4",
+  "started_at": "2026-07-15T15:36:21+09:00",
+  "returncode": 0,
+  "model_complete": true,
+  "finite_results": true,
+  "measured_step": null,
+  "completed_at": "2026-07-15T17:35:44+09:00"
+}

generation_config.json

@@ -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"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13ecf1fe52b368464f8bba1caa7719d0c7a3e6b555c26ef3872918666fcd836c
+size 16381517208

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650

tokenizer_config.json

@@ -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": true,
+  "model_max_length": 2048,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 0.8599068434252956,
+    "total_flos": 0.0,
+    "train_loss": 24.184869588216145,
+    "train_runtime": 7109.0777,
+    "train_samples": 16746,
+    "train_samples_per_second": 2.026,
+    "train_steps_per_second": 0.127
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5117975d32aa6d5ee887a27ed5d586ce2b1ab4f86ce47449497c97462ed54443
+size 6993

training_status.json

@@ -0,0 +1,20 @@
+{
+  "status": "completed",
+  "candidate_id": "inpo_b",
+  "method": "inpo_avg",
+  "seed": 42,
+  "gpu": 0,
+  "pid": 1079779,
+  "config": "/NHNHOME/26msit001_A/BASE/AIPR/sjkim/revision_qwen3_8b/full_iter1/fair_demo_20260715/sweep/candidates/inpo_b/config.yaml",
+  "dataset": "/NHNHOME/26msit001_A/BASE/AIPR/sjkim/revision_qwen3_8b/full_iter1/flagship_20260712/precomputed/avg",
+  "optimizer_steps": 900,
+  "effective_batch_size": 16,
+  "wandb_run_id": "098086245ba4",
+  "wandb_url": "https://wandb.ai/promotion-kim/mnpo/runs/098086245ba4",
+  "started_at": "2026-07-15T15:36:21+09:00",
+  "returncode": 0,
+  "model_complete": true,
+  "finite_results": true,
+  "measured_step": null,
+  "completed_at": "2026-07-15T17:35:44+09:00"
+}