fresh-stack/freshstack

This repository helps you evaluate your models on the FreshStack benchmark!

50
/ 100
Established

Supports both dense embedding models (via BEIR) and multi-vector retrieval systems (via PyLate/ColBERT), enabling comprehensive evaluation of diverse retrieval architectures. The framework automatically constructs benchmarks from real Stack Overflow queries and GitHub repository code, evaluated using nugget-based metrics (α-NDCG, coverage, recall) that capture partial relevance across multiple valid answers. Datasets and evaluation scripts are provided for five technical topics with standardized BEIR format compatibility.

Available on PyPI.

Maintenance 6 / 25
Adoption 11 / 25
Maturity 24 / 25
Community 9 / 25

How are scores calculated?

Stars

33

Forks

3

Language

Python

License

Apache-2.0

Last pushed

Dec 09, 2025

Monthly downloads

43

Commits (30d)

0

Dependencies

3

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/embeddings/fresh-stack/freshstack"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.