micm_nlp.models.xpe.config

Configuration dataclass for XPE.

Classes

CrossPromptEncoderConfig

This is the configuration class to store the configuration of a [CrossPromptEncoder].

Module Contents

class micm_nlp.models.xpe.config.CrossPromptEncoderConfig

Bases: peft.PromptEncoderConfig

This is the configuration class to store the configuration of a [CrossPromptEncoder].

Parameters:

  • encoder_embedding_init_type (str) – The type of initialization to use for the embedding.

  • encoder_init_state_dict_path (str) – The path to pretraine encoder state for initialization shared embeddings and encoder heads.

  • encoder_embedding_freeze (bool) – The indicator of frozen or trainable embedding.

  • modules_to_save (Optional[List[str]]) – List of modules apart from CrossPromptEncoder layers to be set as trainable and saved in the final checkpoint.

  • encoder_embedding_normalize (str) – The type of normalization to use for the embedding (None,unit, clip).

  • encoder_embedding_normalize_max_norm (float) – The maximum norm for the embedding.

  • encoder_input_size (int) – The input size for the encoder.

  • encoder_num_heads (int) – The number of attention heads in the encoder.

classmethod from_peft_type(**kwargs)

Loads the configuration from a set of kwargs. Present for API parity with upstream PeftConfig subclasses; defers to the dataclass __init__.

encoder_embedding_freeze: bool = True
encoder_embedding_init_type: str = 'hf_default'
encoder_embedding_normalize: str | None = None
encoder_embedding_normalize_max_norm: float | None = None
encoder_freeze: bool = True
encoder_init_state_dict_path: str = None
encoder_input_size: int | None = None
encoder_num_heads: int = 8
encoder_ratio: float = 0.7
modules_to_save: list[str] | None = None