llama-index-packs-fuzzy-citation 0.4.0

llama-index packs fuzzy_citation integration

Fuzzy Citation Query Engine Pack

Creates and runs a CustomQueryEngine -- FuzzCitationQueryEngine -- which post-processes response objects to identify source sentences using fuzzy matching.

The identified sentences are available in the response.metadata dictionary, containing a mapping of (response_sentence, source_chunk) -> {"start_char_idx": idx, "end_char_idx": idx, "node" node}. The start/end idxs represent the character indexes in the node text that the source chunk comes from.

The fuzzy matching uses fuzz.ratio() to compare sentences. The default threshold score is 50.

CLI Usage

You can download llamapacks directly using llamaindex-cli, which comes installed with the llama-index python package:

llamaindex-cli download-llamapack FuzzyCitationEnginePack --download-dir ./fuzzy_citation_pack

You can then inspect the files at ./fuzzy_citation_pack and use them as a template for your own project!

Code Usage

You can download the pack to a the ./fuzzy_citation_pack directory:

from llama_index.core import Document, VectorStoreIndex
from llama_index.core.llama_pack import download_llama_pack

# download and install dependencies
FuzzyCitationEnginePack = download_llama_pack(
    "FuzzyCitationEnginePack", "./fuzzy_citation_pack"
)

index = VectorStoreIndex.from_documents([Document.example()])
query_engine = index.as_query_engine()

fuzzy_engine = FuzzyCitationEnginePack(query_engine, threshold=50)

The run() function is a light wrapper around query_engine.query(). The response will have metadata attached to it indicating the fuzzy citations.

response = fuzzy_engine.run("What can you tell me about LLMs?")

# print source sentences
print(response.metadata.keys())

# print full source sentence info
print(response.metadata)

See the notebook on llama-hub for a full example.