{
    "model_name_or_path": "cerebras/btlm-3b-8k-base",
    "max_length": 8092,
    "trust_remote_code": true,
    "train_datasets": [
        [
            "bt",
            {
                "proportion": 1.0
            }
        ]
    ],
    "eval_datasets": null,
    "epochs": 16,
    "per_device_train_batch_size": 8,
    "per_device_eval_batch_size": 2,
    "gradient_accumulation_steps": 1,
    "gradient_checkpointing": true,
    "learning_rate": 4.7e-06,
    "lr_scheduler_type": "cosine",
    "num_warmup_steps": 20,
    "weight_decay": 0.0,
    "seed": 42,
    "fp16": false,
    "bf16": true,
    "tf32": true,
    "eval_strategy": "epoch",
    "eval_interval": 1000000,
    "need_eval": false,
    "eval_split_ratio": null,
    "output_dir": "/home/paperspace/safe-rlhf/output/sft",
    "log_type": "wandb",
    "log_dir": "/home/paperspace/safe-rlhf/output/sft",
    "log_project": "BT-Training",
    "log_run_name": "sft-2023-07-25-19-40-13",
    "save_16bit": false,
    "save_interval": 1000000,
    "local_rank": 0,
    "zero_stage": 2,
    "deepspeed": false,
    "deepspeed_config": null,
    "deepscale": false,
    "deepscale_config": null,
    "deepspeed_mpi": false,
    "global_rank": 0,
    "device": {
        "type": "torch.device",
        "repr": "device(type='cuda', index=0)"
    },
    "num_update_steps_per_epoch": 55,
    "total_training_steps": 880
}