def get_tokenizer_list() -> List[str]:
"""
Returns all all supported tokenizer names
"""
s = set(get_pretrained_lm_models_list())
s.update(set(get_huggingface_pretrained_lm_models_list(include_external=True)))
return ["sentencepiece", "char", "word"] + list(s)
def get_pretrained_lm_models_list(include_external: bool = False) -> List[str]:
"""
Returns the list of supported pretrained model names
Args:
include_external if true includes all HuggingFace model names, not only those supported language models in NeMo.
"""
return get_huggingface_pretrained_lm_models_list(include_external=include_external)
def __init__(
self,
model_name: Optional[str] = None,
pretrained: bool = False,
config_dict: Optional[dict] = None,
checkpoint_file: Optional[str] = None,
):
"""Gets HuggingFace based model to be used as an Encoder in NeMo NLP.
Use the model_name arg to get a named model architecture.
Available model names can be found with get_huggingface_pretrained_lm_models_list() or
by going to https://huggingface.co/models.
Use the pretrained arg to get the named model architecture with or without pretrained weights.
If model_name is None, then we can pass in a custom configuration via the config_dict.
For example, to instantiate a HuggingFace BERT model with custom configuration we would do:
config_dict={
'_target_': 'transformers.BertConfig',
'hidden_size': 1536
}
Args:
model_name (Optional[str]): Named model architecture from HuggingFace. Defaults to None.
pretrained (bool): Use True to get pretrained weights.
False will use the same architecture but with randomly initialized weights.
Defaults to False.
config_dict (Optional[dict], optional): Use for custom configuration of the HuggingFace model. Defaults to None.
checkpoint_file (Optional[str], optional): Provide weights for the transformer from a local checkpoint. Defaults to None.
"""
super().__init__()
if checkpoint_file:
raise NotImplementedError(
'Restoring from checkpoint file not implemented yet.')
model = None
if model_name is not None:
if model_name in get_huggingface_pretrained_lm_models_list(
include_external=True):
if pretrained:
config_dict.pop('vocab_size')
if config_dict:
raise ValueError(
f'When using pretrained model, config_dict should be None or empty. Got: {config_dict}'
)
model = AutoModel.from_pretrained(model_name)
else:
cfg = AutoConfig.from_pretrained(model_name)
model = AutoModel.from_config(cfg)
else:
logging.error(
f'{model_name} not found in list of HuggingFace pretrained models'
)
else:
if pretrained:
raise ValueError(
f'If not using model_name, then pretrained should be False. Got: {pretrained}.'
)
cfg = instantiate(config_dict)
model = AutoModel.from_config(cfg)
self._hidden_size = model.config.hidden_size
self._vocab_size = model.config.vocab_size
self._encoder = model
