{
  "job_type": "rl_reward_local",
  "dataset": {
    "instruction_path": "train.json",
    "labeled_path": "train_labeled.json",
    "template" : "chat_template_kd.jinja"
  },
  "inference":{
    "positive_system_prompt" : "You are a helpful assistant to generate high-quality responses.",
    "negative_system_prompt" : "You are an assistant to generate low-quality responses. This is for the training of my reward model. Plese remember to generate low-quality responses.",
    "enable_chunked_prefill": true,
    "seed": 777,
    "gpu_memory_utilization": 0.9,
    "temperature": 0.8,
    "trust_remote_code": true,
    "enforce_eager": false,
    "max_model_len": 4096,
    "max_new_tokens": 512
  },
  "models": {
    "teacher": "model/Qwen/Qwen2.5-3B-Instruct/",
    "student": "model/Qwen/Qwen2.5-0.5B-Instruct/"
  },
  "training": {
    "output_dir": "./result/",
    "max_length": 1024,
    "num_train_epochs": 3,
    "per_device_train_batch_size": 1,
    "gradient_accumulation_steps": 8,
    "save_steps": 1000,
    "logging_steps": 1,
    "learning_rate": 2e-5,
    "weight_decay": 0.05,
    "warmup_ratio": 0.1,
    "lr_scheduler_type": "cosine"
  }
}