safelens
Pluggable safety analysis infrastructure for LLMs, with unified interfaces for probes, runtime monitors, attribution methods, model loading, pipeline execution, and FlagSafe-compatible reporting. Supports configurable model sources including dummy, HuggingFace, and ModelScope backends.
View on Github
Activity
- Latest release
- 18h ago
- Total releases
- 1
- Cadence
- —
- Last 12 months
- 1
Reach
- Stars
- 0
Details
- License
- MIT
- First release
- Sep 14, 2026