chatglm_6b_lora.json

{
    "output_dir": "output/adgen-chatglm-6b-lora",
    "per_device_train_batch_size": 3,
    "gradient_accumulation_steps": 4,
    "learning_rate": 1e-3,
    "max_steps": 3000,
    "num_train_epochs": 1.0,
    "lr_scheduler_type": "cosine",
    "warmup_ratio": 0.03, 
    "logging_steps": 5,
    "save_strategy": "no",
    "optim": "adamw_torch",
    "fp16": false,
    "remove_unused_columns": false,
    "ddp_find_unused_parameters": false,
    "report_to": "tensorboard"
}