colbert_utils.py

""" colbert_utils.py

Utilities for building (and using) a ColBERT (retrieval) model.

:author: Didier Guillevic
:email: [email protected]
:creation: 2024-12-21
"""

import logging
logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)

from ragatouille import RAGPretrainedModel


def build_colbert_model(
        documents: list[str],
        metadatas: list[dict[str, str]],
        pretrained_model: str='antoinelouis/colbert-xm',
        index_name: str='colbert_index'
    ) -> RAGPretrainedModel:
    """Build a ColBERT model for retrieval.

    Args:
        documents: list of documents to index
        metadatas: list of metadata for each document
        index_name: name of the index built with given documents
        pretrined_model: name of the pretrained model to use

    Returns:
        the ColBERT retrieval model built witt the given documents.
    """
    model = RAGPretrainedModel.from_pretrained(pretrained_model)
    model.index(
        collection=documents, 
        #document_ids=document_ids, # no unique IDs at the moment
        document_metadatas=metadatas,
        index_name=index_name,
        max_document_length=180, 
        split_documents=True,
        use_faiss=False # cannot get it to work...
    )
    return model