{
    "model_name_or_path": "meta-llama/Llama-3.1-8B-Instruct",
    "recompute_baseline": false,
    "cache_dir": "/home/panda/pda-llm/cache/sft-tools",
    "max_length": 4096,
    "trust_remote_code": true,
    "train_datasets": [
        [
            "tools",
            {
                "proportion": 1.0
            }
        ]
    ],
    "eval_datasets": null,
    "safety_ratio_tol": 50.0,
    "important_sft": true,
    "resilient_coeff": 1.0,
    "epochs": 3,
    "per_device_train_batch_size": 1,
    "per_device_eval_batch_size": 1,
    "gradient_accumulation_steps": 48,
    "gradient_checkpointing": true,
    "lr": 0.0001,
    "lr_scheduler_type": "cosine",
    "lr_warmup_ratio": 0.1,
    "weight_decay": 0.0,
    "seed": 42,
    "fp16": false,
    "bf16": true,
    "tf32": false,
    "lora_r": 4,
    "lora_alpha": 32,
    "lora_dropout": 0.05,
    "eval_strategy": "epoch",
    "eval_interval": 1000000,
    "need_eval": true,
    "eval_split_ratio": null,
    "output_dir": "/home/panda/pda-llm/output/sft-tools/run-true-1-50-4-4096",
    "log_type": "wandb",
    "log_dir": "/home/panda/pda-llm/output/sft-tools/run-true-1-50-4-4096",
    "log_project": "TOOLS-SFT",
    "log_run_name": "tools-sft-2025-05-11-14-32-03",
    "save_16bit": false,
    "save_interval": 1000000,
    "local_rank": 0,
    "zero_stage": 0,
    "offload": "none",
    "deepspeed": false,
    "deepspeed_config": null,
    "deepscale": false,
    "deepscale_config": null,
    "global_rank": 0,
    "device": {
        "type": "torch.device",
        "repr": "device(type='cuda', index=0)"
    },
    "num_update_steps_per_epoch": 112,
    "total_training_steps": 336
}