transformers

transformers

TransformersModel

TransformersModel(*args, config_model: Type[PretrainedConfig] = None, automodel: Type[AutoModel] = AutoModel, **kwargs)

Bases: HuggingFaceModel

NNsight wrapper for HuggingFace Transformers models.

Adds AutoConfig / AutoModel support on top of :class:HuggingFaceModel. Handles config loading, meta-tensor initialization via from_config, and full weight loading via from_pretrained.

PARAMETER	DESCRIPTION
*args	Forwarded to :class:HuggingFaceModel. The first positional argument is typically a repo ID string or a pre-loaded torch.nn.Module. DEFAULT: ()
config_model	An explicit HuggingFace config instance to use instead of loading one from the repo. Defaults to None (auto-loaded). TYPE: Optional[Type[PretrainedConfig]] DEFAULT: None
automodel	The AutoModel class to use for loading (e.g. AutoModelForCausalLM). Defaults to AutoModel. TYPE: Type[AutoModel] DEFAULT: AutoModel
**kwargs	Forwarded to from_pretrained / from_config. DEFAULT: {}

ATTRIBUTE	DESCRIPTION
config	The model's HuggingFace configuration. TYPE: PretrainedConfig
automodel	The AutoModel class used for loading. TYPE: Type[AutoModel]

config instance-attribute

config: PretrainedConfig = config_model

automodel instance-attribute

automodel = automodel if not isinstance(automodel, str) else getattr(modeling_auto, automodel)