train_weak_to_strong.py [56:61]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        gradient_checkpointing=True,
        model_parallel=True,
        custom_kwargs={
            "trust_remote_code": True,
            "torch_dtype": torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32,
        },
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



train_weak_to_strong.py [91:97]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        gradient_checkpointing=True,
        model_parallel=True,
        # note: you will probably not be able to run this without bf16 support and many gpus
        custom_kwargs={
            "trust_remote_code": True,
            "torch_dtype": torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32,
        },
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -