Release version inconsistency: 0.3.8 in-tree vs 0.3.7 on PyPI (last shipped 2026-03-08)
Summary
There's a significant version skew between the in-tree version and what's published on PyPI. Users installing from PyPI get a 6-month-old release that predates recent security hardening.
Current State
- PyPI last shipped: v0.3.7 (2026-03-08) — 6 months stale
- In-tree versions: 0.3.8 (hnsw, diskann, leann-core, leann packages)
- Backend variants: ivf and flashlib stuck at 0.3.6
- Dockerfiles: Pin 0.3.6
- Self-classified: Beta (Development Status :: 4)
Why This Matters
Users installing via pip install leann get v0.3.7, which predates:
This means new users adopt the package without the latest security posture improvements.
Suggested Approach
When the team is ready to cut a release:
- Align all in-tree package versions (0.3.8 → shared tag)
- Include the September 5 security hardening commits
- Update Dockerfiles to match
- Ship to PyPI with clear release notes (security improvements, storage claims validation, etc.)
This is not blocking for local/development use, but it's worth resolving before recommending LEANN for downstream integration or production workloads.
Note: The MLSys 2026 Best Paper recognition has driven strong adoption (12.9K stars, 1.1K forks). A fresh PyPI release with the latest hardening would reinforce confidence in the project's maturity.
Release version inconsistency: 0.3.8 in-tree vs 0.3.7 on PyPI (last shipped 2026-03-08)
Summary
There's a significant version skew between the in-tree version and what's published on PyPI. Users installing from PyPI get a 6-month-old release that predates recent security hardening.
Current State
Why This Matters
Users installing via
pip install leannget v0.3.7, which predates:This means new users adopt the package without the latest security posture improvements.
Suggested Approach
When the team is ready to cut a release:
This is not blocking for local/development use, but it's worth resolving before recommending LEANN for downstream integration or production workloads.
Note: The MLSys 2026 Best Paper recognition has driven strong adoption (12.9K stars, 1.1K forks). A fresh PyPI release with the latest hardening would reinforce confidence in the project's maturity.