train_args.json

{
    "output_dir": "output",
    "model_name_or_path": "models/Qwen2.5-1.5B-Instruct",
    "deepspeed": "./train_args/ds_z3_config.json",
    "train_file": "train_GPT/conversations_20000_MASK.jsonl",
    "template_name": "qwen",
    "train_mode": "full",
    "num_train_epochs": 3,
    "per_device_train_batch_size": 4,
    "gradient_accumulation_steps": 16,
    "learning_rate": 1e-05,
    "max_seq_length": 4096,
    "logging_steps": 200,
    "save_steps": 200,
    "save_total_limit": 1,
    "lr_scheduler_type": "cosine",
    "warmup_steps": 50,
    "gradient_checkpointing": false,
    "disable_tqdm": false,
    "optim": "adamw_hf",
    "seed": 42,
    "fp16": true,
    "report_to": "tensorboard",
    "dataloader_num_workers": 0,
    "save_strategy": "steps",
    "weight_decay": 0,
    "max_grad_norm": 1.0,
    "remove_unused_columns": false
}