{
  "schema_version": 1,
  "task": "countdown_rlvr_initial_phase",
  "model": {
    "kind": "hub",
    "repo_id": "Qwen/Qwen3-4B",
    "revision": "1cfa9a7208912126459214e8b04321603b3df60c"
  },
  "training_data": {
    "path": "data/countdown/train.jsonl",
    "samples": 1024,
    "hash_basis": "file_bytes",
    "sha256": "aa7abb6242d8ada72e55a6d8d0917e3618473ddf2f0f880b288814394b231131"
  },
  "system_prompt": "제시된 숫자는 각 항목을 정확히 한 번씩 사용하여 목표값을 만드세요. 허용 연산은 +, -, *, /이며 괄호를 사용할 수 있습니다. 연산 기호는 두 수 사이에만 사용하세요. 숫자를 이어 붙여 새 수를 만들지 마세요. 풀이를 간결하게 검토한 뒤 <answer>와 </answer> 사이에는 등호와 계산 결과 없이 수식만 작성하세요.",
  "phase": {
    "name": "초기 구간",
    "start_step": 1,
    "end_step": 120,
    "scheduler_end_step": 480,
    "max_completion_length": 3072,
    "vllm_gpu_memory_utilization": 0.25,
    "resume_from_checkpoint": null
  },
  "training_phases": [
    {
      "name": "초기 구간",
      "start_step": 1,
      "end_step": 120,
      "scheduler_end_step": 480,
      "max_completion_length": 3072,
      "vllm_gpu_memory_utilization": 0.25,
      "resume_from_checkpoint": null
    },
    {
      "name": "재개 구간",
      "start_step": 121,
      "end_step": 480,
      "scheduler_end_step": 480,
      "max_completion_length": 3072,
      "vllm_gpu_memory_utilization": 0.25,
      "resume_from_checkpoint": "checkpoint-120"
    }
  ],
  "runtime": {
    "python": "3.12.3",
    "packages": {
      "torch": "2.11.0+cu130",
      "transformers": "5.14.1",
      "trl": "1.9.2",
      "peft": "0.20.0",
      "vllm": "0.24.0"
    },
    "allocator_environment": {
      "PYTORCH_ALLOC_CONF": "expandable_segments:True",
      "PYTORCH_CUDA_ALLOC_CONF": "expandable_segments:True"
    },
    "training_script_sha256": "4886fc449a84929704516cc67eae9196d531afdc8cc4d35fac92799038955fb4",
    "dependency_lock": {
      "path": "requirements-lock.txt",
      "sha256": "e8d7d01893ceed1278caee3b21bed24ed3350ce4e649f103c91570e4c61b4562"
    }
  },
  "lora": {
    "r": 16,
    "lora_alpha": 32,
    "lora_dropout": 0.05,
    "bias": "none",
    "target_modules": [
      "q_proj",
      "k_proj",
      "v_proj",
      "o_proj",
      "gate_proj",
      "up_proj",
      "down_proj"
    ],
    "task_type": "CAUSAL_LM"
  },
  "rewards": {
    "functions": [
      "format_rewards",
      "expression_rewards",
      "correctness_rewards"
    ],
    "weights": [
      0.3,
      0.2,
      0.5
    ],
    "implementation_sha256": "d779da290b9c7376298160a4e9c722e17af6e60d9e45a7bfdb4952ccf19b07f3"
  },
  "grpo": {
    "seed": 42,
    "per_device_train_batch_size": 2,
    "gradient_accumulation_steps": 4,
    "steps_per_generation": 4,
    "generation_batch_size": 8,
    "num_generations": 8,
    "max_completion_length": 3072,
    "temperature": 0.6,
    "top_p": 0.95,
    "top_k": 20,
    "mask_truncated_completions": true,
    "use_vllm": true,
    "vllm_mode": "colocate",
    "vllm_gpu_memory_utilization": 0.25,
    "learning_rate": 5e-06,
    "lr_scheduler_type": "SchedulerType.COSINE",
    "warmup_steps": 0.05,
    "optim": "OptimizerNames.PAGED_ADAMW_8BIT",
    "beta": 0.001,
    "epsilon": 0.2,
    "loss_type": "grpo",
    "bf16": true,
    "gradient_checkpointing": true,
    "chat_template_kwargs": {
      "enable_thinking": true
    },
    "model_init_kwargs": {
      "dtype": "bfloat16",
      "revision": "1cfa9a7208912126459214e8b04321603b3df60c"
    },
    "max_steps": 480
  },
  "checkpoint": {
    "label": "checkpoint-120",
    "global_step": 120,
    "files": {
      "README.md": {
        "bytes": 5193,
        "sha256": "37e16a61fcdb14836debbbe0d1d63c181cd3e98ca2049d22b513c82bf2b55fd1"
      },
      "adapter_config.json": {
        "bytes": 1146,
        "sha256": "04f4e9988d744c123dd84078aa823e2afc9b78c9cdedae44edeea72fe876089e"
      },
      "adapter_model.safetensors": {
        "bytes": 132187888,
        "sha256": "e422bfa8c0c6088002925a85f57d554771757b890036639ad8d9345c9287562c"
      },
      "chat_template.jinja": {
        "bytes": 4168,
        "sha256": "a55ee1b1660128b7098723e0abcd92caa0788061051c62d51cbe87d9cf1974d8"
      },
      "optimizer.pt": {
        "bytes": 67694085,
        "sha256": "880ef606fe577d606b01fea31837f72ca083ed1710427cd64c0f2cbd8c2db2f1"
      },
      "rng_state.pth": {
        "bytes": 14645,
        "sha256": "c329ead9ac4b1f5a08c6d75afef1853aca4b1d7a9242bb4a9224b2682fd2e1ef"
      },
      "scheduler.pt": {
        "bytes": 1465,
        "sha256": "b21770590c70fe1297817882fd4f61724612625618a8c998fe2fcbd51ceeb8aa"
      },
      "tokenizer.json": {
        "bytes": 11422650,
        "sha256": "be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506"
      },
      "tokenizer_config.json": {
        "bytes": 694,
        "sha256": "04b1682c59acbd057f4c9072297faa73d56fc9de053094c659cdb4c464f58f86"
      },
      "trainer_state.json": {
        "bytes": 40791,
        "sha256": "a41022ab39cc63b52b47dfe5e00be141b14fb788058f0994ec3bea739ec9f4c8"
      },
      "training_args.bin": {
        "bytes": 7761,
        "sha256": "7722ce538c162271ac3dcf8ef53ba88fe9b2eb0d02a4347cd5ac692162992ad9"
      }
    }
  }
}