raphaelsty/cherche

Composes modular retriever and ranker components—including sparse methods (BM25, TF-IDF), dense encoders (SentenceTransformers), and cross-encoders—into end-to-end pipelines for semantic search and question answering. Designed for batch inference on CPU/GPU with support for offline indexing, enabling efficient retrieval-then-rerank workflows over document collections. Integrates seamlessly with Hugging Face pre-trained models and provides built-in question-answering modules that work within the same pipeline abstraction.

333 stars and 251 monthly downloads. No commits in the last 6 months. Available on PyPI.