ray.train.huggingface.TransformersCheckpoint.from_model#

classmethod TransformersCheckpoint.from_model(model: Union[transformers.modeling_utils.PreTrainedModel, torch.nn.Module], tokenizer: Optional[transformers.PreTrainedTokenizer] = None, *, path: os.PathLike, preprocessor: Optional[Preprocessor] = None) → TransformersCheckpoint[source]#

Create a Checkpoint that stores a HuggingFace model.

Parameters

  • model – The pretrained transformer or Torch model to store in the checkpoint.

  • tokenizer – The Tokenizer to use in the Transformers pipeline for inference.

  • path – The directory where the checkpoint will be stored.

  • preprocessor – A fitted preprocessor to be applied before inference.

Returns

A TransformersCheckpoint containing the specified model.