Implementation:AnswerDotAI RAGatouille RAGPretrainedModel Search Encoded Docs

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

Overview

Concrete tool for searching through documents previously encoded in memory using exact MaxSim scoring provided by the RAGatouille library.

Description

The RAGPretrainedModel.search_encoded_docs() method searches through documents that were previously encoded via encode(). It delegates to ColBERT.search_encoded_docs() which encodes the query into token embeddings, then calls ColBERT._index_free_search() to compute exact MaxSim scores against the in-memory document tensors. Results include content, score, rank, result_index, and optionally document_metadata.

The delegation chain:

RAGPretrainedModel.search_encoded_docs() → delegates to model
ColBERT.search_encoded_docs() → encodes queries, calls _index_free_search, attaches metadata
ColBERT._index_free_search() → computes MaxSim via _colbert_score, sorts results

Usage

Use after calling encode() to search through in-memory documents. Supports single and batch queries.

Code Reference

Source Location

Repository: RAGatouille
File: ragatouille/RAGPretrainedModel.py
Lines: L386-406

Signature

def search_encoded_docs(
    self,
    query: Union[str, list[str]],
    k: int = 10,
    bsize: int = 32,
) -> list[dict[str, Any]]:
    """Search through documents encoded in-memory.

    Parameters:
        query: Query string or list of queries.
        k: Number of results per query (default 10).
        bsize: Batch size for query encoding (default 32).

    Returns:
        list[dict]: Results with keys: content, score, rank, result_index,
        and optionally document_metadata.
    """

Import

from ragatouille import RAGPretrainedModel

I/O Contract

Inputs

Name	Type	Required	Description
query	Union[str, list[str]]	Yes	Single query string or list of queries
k	int	No	Number of top results per query (default 10)
bsize	int	No	Batch size for query encoding (default 32)

Outputs

Name	Type	Description
return	list[dict[str, Any]]	Results with keys: content (str), score (float), rank (int), result_index (int), and optionally document_metadata (dict)

Usage Examples

Search Encoded Documents

from ragatouille import RAGPretrainedModel

RAG = RAGPretrainedModel.from_pretrained("colbert-ir/colbertv2.0")

# Encode documents
RAG.encode([
    "ColBERT uses late interaction.",
    "BERT is a language model.",
    "Transformers use attention.",
])

# Search
results = RAG.search_encoded_docs("What is ColBERT?", k=2)
for r in results:
    print(f"[{r['rank']}] (score: {r['score']:.4f}) {r['content']}")

Batch Search with Metadata

RAG.encode(
    documents=["Doc A", "Doc B", "Doc C"],
    document_metadatas=[{"id": "a"}, {"id": "b"}, {"id": "c"}],
)

results = RAG.search_encoded_docs(
    ["query one", "query two"],
    k=2,
)
# Returns list of lists for batch queries

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment