train_simple.py [67:75]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        default_lr=1e-5,
        eval_batch_size=2,
        gradient_checkpointing=True,
        model_parallel=True,
        # note: you will probably not be able to run this without many gpus
        custom_kwargs={
            "trust_remote_code": True,
            "bf16": torch.cuda.is_bf16_supported(),
            "fp32": not torch.cuda.is_bf16_supported(),
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



train_simple.py [81:89]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        default_lr=1e-5,
        eval_batch_size=2,
        gradient_checkpointing=True,
        model_parallel=True,
        # note: you will probably not be able to run this bf16 support and without many gpus
        custom_kwargs={
            "trust_remote_code": True,
            "bf16": torch.cuda.is_bf16_supported(),
            "fp32": not torch.cuda.is_bf16_supported(),
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -