{
  "model_name": "unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit",
  "max_seq_length": 4096,
  "lora_config": {
    "default": {
      "base_model_name_or_path": "unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit",
      "r": 32,
      "lora_alpha": 32,
      "lora_dropout": 0,
      "target_modules": [
        "v_proj",
        "down_proj",
        "k_proj",
        "q_proj",
        "up_proj",
        "gate_proj",
        "o_proj"
      ]
    }
  },
  "tokenizer_config": {
    "add_bos_token": false,
    "chat_template": "{% for message in messages %}{% if message['role'] == 'user' %}{{'<|im_start|>user\n' + message['content'] + '<|im_end|>\n'}}{% elif message['role'] == 'assistant' %}{{'<|im_start|>assistant\n' + message['content'] + '<|im_end|>\n' }}{% else %}{{ '<|im_start|>system\n' + message['content'] + '<|im_end|>\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
    "vocab_size": 151643,
    "special_tokens_map": {
      "eos_token": "<|im_end|>",
      "pad_token": "<|PAD_TOKEN|>"
    }
  },
  "trainer_config": {
    "per_device_train_batch_size": 4,
    "gradient_accumulation_steps": 4,
    "warmup_steps": 20,
    "num_train_epochs": 2,
    "learning_rate": 2e-05,
    "weight_decay": 0.1,
    "max_grad_norm": 0.3,
    "lr_scheduler_type": "linear",
    "bf16": true,
    "fp16": false,
    "dataloader_num_workers": 4,
    "dataset_num_proc": 4,
    "remove_unused_columns": true,
    "optim": "adamw_8bit"
  }
}