semantic_router.encoders.clip

CLIPEncoder Objects

class CLIPEncoder(DenseEncoder)

Multi-modal dense encoder for text and images using CLIP-type models via

HuggingFace.

Arguments:

• name (str): The name of the model to use.
• tokenizer_kwargs (Dict): Keyword arguments for the tokenizer.
• processor_kwargs (Dict): Keyword arguments for the processor.
• model_kwargs (Dict): Keyword arguments for the model.
• device (Optional[str]): The device to use for the model.
• str0 (str1): The tokenizer for the model.
• str2 (str1): The processor for the model.
• str4 (str1): The model.
• str6 (str1): The torch library.
• str8 (str1): The PIL library.

__init__

def __init__(**data)

Initialize the CLIPEncoder.

Arguments:

• **data (Dict): Keyword arguments for the encoder.

__call__

def __call__(docs: List[Any],
             batch_size: int = 32,
             normalize_embeddings: bool = True) -> List[List[float]]

Encode a list of documents. Can handle both text and images.

Arguments:

• docs (List[Any]): The documents to encode.
• batch_size (int): The batch size for the encoding.
• normalize_embeddings (bool): Whether to normalize the embeddings.

Returns:

List[List[float]]: A list of embeddings.