Implementation:AnswerDotAI RAGatouille RAGPretrainedModel From Index

Knowledge Sources	RAGatouille RAGatouille Docs
Domains	NLP, Information_Retrieval, Index_Management
Last Updated	2026-02-12 12:00 GMT

Overview

Concrete tool for loading a ColBERT model and its associated PLAID index from an existing index directory provided by the RAGatouille library.

Description

The RAGPretrainedModel.from_index() class method restores a complete retrieval system from a previously built index. It creates a RAGPretrainedModel instance and initializes the underlying ColBERT object with load_from_index=True, which triggers loading the configuration from the index directory, restoring the PLAID model index, and deserializing the document collection and metadata mappings from disk.

Usage

Use this method when you have a previously built index and want to resume search or update operations without re-indexing. This is the standard entry point when deploying a search service or continuing work on an existing document collection.

Code Reference

Source Location

• Repository: RAGatouille
• File: ragatouille/RAGPretrainedModel.py
• Lines: L76-96

Signature

@classmethod
def from_index(
    cls,
    index_path: Union[str, Path],
    n_gpu: int = -1,
    verbose: int = 1,
) -> "RAGPretrainedModel":
    """Load an Index and the associated ColBERT encoder from an existing document index.

    Parameters:
        index_path (Union[str, Path]): Path to the index directory.
        n_gpu (int): Number of GPUs to use. -1 means auto-detect.
        verbose (int): Verbosity level. 1 filters most internal logs.

    Returns:
        RAGPretrainedModel: Initialized instance with model and index loaded.
    """

Import

from ragatouille import RAGPretrainedModel

I/O Contract

Inputs

Name	Type	Required	Description
index_path	Union[str, Path]	Yes	Path to an existing index directory built by RAGatouille
n_gpu	int	No	Number of GPUs to use. -1 (default) auto-detects available GPUs
verbose	int	No	Verbosity level. 1 (default) filters most internal ColBERT logs

Outputs

Name	Type	Description
return	RAGPretrainedModel	Initialized instance with self.model set to a ColBERT object. The model has its index loaded, collection restored, and pid_docid_map deserialized.

Usage Examples

Load from a Previously Built Index

from ragatouille import RAGPretrainedModel

# Load model and index from disk
RAG = RAGPretrainedModel.from_index(".ragatouille/colbert/indexes/my_index")

# Immediately search without re-indexing
results = RAG.search("What is ColBERT?", k=5)

Load and Update an Existing Index

from ragatouille import RAGPretrainedModel

# Load existing index
RAG = RAGPretrainedModel.from_index(".ragatouille/colbert/indexes/my_index")

# Add new documents to the existing index
RAG.add_to_index(
    new_collection=["New document text here."],
    new_document_ids=["doc_new_1"],
)

Related Pages

Implements Principle

• Principle:AnswerDotAI_RAGatouille_Index_Loading

Requires Environment

• Environment:AnswerDotAI_RAGatouille_Python_ColBERT_Dependencies