Adds a new adapter module of the specified type to the model. Args: adapter_name (str): The name of the adapter module t

Adds a biaffine dependency parsing head on top of the model. The parsing head uses the architecture described in "Is Supervised Synta

Adds a new AdapterFusion. Args: adapter_names (List[str]): The names of the adapters to fuse. config (Option

See BertModel.add_fusion_layer

Adds an image classification head on top of the model. Args: head_name (str): The name of the head. num_labe

Add appropriate language model head based on model type

This function calculates the column-wise norm of the pre-trained weights and the LoRA matrices eg: for LoRA, it would be ||w_0 + BA||c

Create Prefix Tuning module based on shim layer infications.

Converts the model output, which consists of a Tuple of Tensors to a Tuple of lists, while preserving the original dimensions. The converted outpu

Given a model run a forward pass with some dummy data. Args: model: The model for which the forward pass is run. model_name: The n

Deletes the embedding with the given name Args: name: The name of the embedding that should be deleted

Deletes the prediction head with the specified name from the model. Args: head_name (str): The name of the prediction to

This method downloads a file and caches it. For more information on why this is needed, refer to the explanation in this Pull Request: https

Unfreezes a given list of adapters, the adapter fusion layer, or both Args: adapter_names: names of adapters to unfreeze

Enables the gradients for the input embeddings specifically for BEiT's tuple output format.

The callable returned by this method is used to extract the module weights to be saved or loaded based on their names. Args:

Sets seeds manually. Args: seed (int, optional): The seed value to use for random number generation.

Freezes all weights of the model.

Returns the id2label dict for the given hea Args: head_name: (str, optional) the name of the head which labels should be

Returns a list of strings representing the testable model names.

Returns a list of strings representing new adapter configurations.

Activates gradient checkpointing for the current model. Note that in other frameworks this feature can be referred to as "activation

Initialize the weights for the compacter module or the shared parameters

Loads a given adapter configuration specifier into a full AdapterConfig instance. Args: config (Union[dict, str]): The c

Loads a pre-trained model from a specified path based on the specified model_name. Args: model_name (str): The name of the model to be loa

Loads a pretrained model with adapters from the given path or url. Parameters: model_name_or_path (`str` or `os.PathLike`, *optional

Concatenates the given states along the batch size dimension. Pads the states before concatenation if necessary. This is e.g. used by the Batc

Parses a potential head configuration from a setup of adapters. Args: adapter_composition: The adapter setup to be parsed. r

Retrieves the hidden_states, query (for Fusion), and residual connection according to the set configuration. Args: adapt

Saves a prediction head module into the given directory. Args: save_directory (str): The directory to save the weights i

Saves an adapter setup to a directory so that it can be shared or reloaded using `load_adapter_setup()`. Args: save_directory (st

Calls the forward function of the model with the given arguments and keyword arguments.