Auto Classes

AutoDeltaConfig

class AutoDeltaConfig(*args, **kwargs)

This is a generic configuration class that will be instantiated as one of the configuration classes of the library when created with the from_finetuned() or from_dict() class method. This class cannot be instantiated directly using __init__() (throws an error).

classmethod from_dict(config_dict: Dict[str, Any], **kwargs)

Instantiate a DeltaConfig according to the dict. Automatically load the config specified by delta_type.

Parameters

• config_dict (dict) – The dict of configs of delta model.

• kwargs – Other keyword argument pass to initialize the config.

Examples:

config = AutoDeltaConfig.from_dict({"delta_type":"lora"}) # This will load the dault lora config.
config = AutoDeltaConfig.from_dict({"delta_type":"lora", "lora_r":5}) # Will load the default lora config, with lora_r = 5

classmethod from_finetuned(finetuned_delta_path, **kwargs)

Instantiate one of the configuration classes of the library from a finetuned delta model configuration. The configuration class to instantiate is selected based on the delta_type property of the config object that is loaded.

Parameters

• finetuned_delta_path (str or os.PathLike, optional) –

  Can be either:

  • A string, the model id of a finetuned delta model configuration hosted inside a model repo on huggingface.co. Valid model ids can be located at the root-level, like Davin/lora, or namespaced under a user or organization name, like DeltaHub/lora_t5-base_mrpc.

  • A path to a directory containing a configuration file saved using the save_finetuned() method, e.g., ./my_model_directory/.

  • A path or url to a saved configuration JSON file, e.g.,``./my_model_directory/configuration.json``.

• cache_dir (str or os.PathLike, optional) – Path to a directory in which a downloaded pretrained model configuration should be cached if the standard cache should not be used.

Examples:

from transformers import AutoConfig
delta_config = AutoDeltaConfig.from_finetuned("thunlp/FactQA_T5-large_Adapter")

AutoDeltaModel

class AutoDeltaModel(*args, **kwargs)

classmethod from_config(config, backbone_model, **kwargs) → DeltaBase

Automatically instantiates a delta model based on the config. The delta model correspond to the delta config will be loaded and initialized using the arguments in config.

Note

Only using from_config() method will not load the finetuned weight file (e.g., pytorch_model.bin). Please use from_finetuned directly.

Parameters

• config (BaseDeltaConfig) –

• backbone_model (nn.Module) –

Examples:

config = AutoDeltaConfig.from_finetuned("DeltaHub/lora_t5-base_mrpc")
delta_model = AutoDeltaModel.from_config(config, backbone_model)

classmethod from_finetuned(finetuned_delta_path, backbone_model, *model_args, **kwargs) → DeltaBase

Automatically instantiated a delta model and load the finetuned checkpoints based on the finetuned_delta_path, which can either be a string pointing to a local path or a url pointint to the delta hub. It will check the hash after loading the delta model to see whether the correct backbone and delta checkpoint are used.

Parameters

• finetuned_delta_path (str or os.PathLike, optional) –

  Can be either:

  • A string, the model name of a finetuned delta model configuration hosted inside a model repo on Delta Center, like thunlp/FactQA_T5-large_Adapter.

  • A path to a directory containing a configuration file saved using the save_finetuned() method, e.g., ./my_model_directory/.

  • A path or url to a saved configuration JSON file, e.g., ./my_model_directory/configuration.json.The last two option are not tested but inherited from huggingface.

• backbone_model (nn.Module) – The backbone model to be modified.

• model_args – Other argument for initialize the model. See :DeltaBase.from_finetuned for details.

• kwargs – Other kwargs that will be passed into DeltaBase.from_finetuned. See DeltaBase.from_finetuned for details.

Example:

delta_model = AutoDeltaModel.from_finetuned("thunlp/FactQA_T5-large_Adapter", backbone_model=5)