experiments/configs.json

{
    "output_dir": "./results",
    "evaluation_strategy": "no",
    "do_eval": false,
    "eval_steps": 0,
    "log_level": "info",
    "logging_first_step": true,
    "logging_steps": 5,
    "logging_dir": "./logs",
    "save_steps": 150,
    "save_total_limit": 1,
    "num_train_epochs": 10,
    "per_device_train_batch_size": 6,
    "optim": "adamw_torch",
    "gradient_accumulation_steps": 4,
    "dataloader_drop_last": true,
    "warmup_steps": 1000,
    "weight_decay": 0.1,
    "learning_rate": 0.0001,
    "deepspeed": "deepspeed.json",
    "local_rank": 0,
    "total_gpus": 8,
    "v_cpus": 64,
    "total_memory_in_gb": 257604.9765625,
    "dataset_limit": 0
}