This release adds 2 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+14 more
Summary
AI summaryGraceful GPU→CPU fallback prevents crashes when CUDA initialization fails.
Full changelog
What's New
- CUDA DLL auto-discovery — Automatically finds CUDA 12 DLLs from pip-installed NVIDIA packages (
nvidia-cublas-cu12, etc.). No manual PATH configuration needed. - Graceful GPU→CPU fallback — If CUDA init fails (missing drivers, wrong version), falls back to CPU with a
[WARN]log instead of crashing. - Explicit CPU provider — When
gpu: false, forcesCPUExecutionProvideronly. No noisy CUDA probe errors in logs. - BASE_DIR resolution fix — Correctly prefers directories with
config.yamlover those with onlyconfig.example.yaml(fixes editable installs).
GPU Setup (NVIDIA)
pip install knowledge-rag[gpu]
pip install nvidia-cublas-cu12 nvidia-cudnn-cu12 nvidia-cuda-runtime-cu12
# config.yaml
models:
embedding:
gpu: true
Install / Upgrade
pip install --upgrade knowledge-rag
Full Changelog: https://github.com/lyonzin/knowledge-rag/compare/v3.5.1...v3.5.2
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About lyonzin/knowledge-rag
Local RAG system for Claude Code with hybrid search (BM25 + semantic), cross-encoder reranking, markdown-aware chunking, query expansion, and 12 MCP tools. Runs entirely offline with zero external servers.
Related context
Beta — feedback welcome: [email protected]