Back to news
Jan 02, 2026
2 min read

🧯 QonQrete v1.0.1-stable β€” HuggingFace Cache Hotfix for Hardened Docker

Fixes Qontextor complex mode failures in hardened Docker by pre-caching sentence-transformers and forcing HF cache env vars.

🧯 QonQrete v1.0.1-stable Released

January 2, 2026 | Hotfix Release


If you were running Qontextor in local_mode: complex inside the hardened container and hitting PermissionError on model downloads β€” this one’s for you.

What Happened

The hardened runtime mounts a tmpfs over /home/qrane/.cache, which blocked sentence-transformers from downloading or locking model files at runtime.

What We Fixed

  • Pre-cached all-MiniLM-L6-v2 during the Docker build into /opt/hf_cache
  • Force-set HF_HOME, SENTENCE_TRANSFORMERS_HOME, and TRANSFORMERS_CACHE to the new path
  • Added a graceful fallback: if permissions still fail, Qontextor drops to AST-only analysis instead of crashing

Upgrade

Just rebuild:

./qonqrete.sh init

No config changes needed. Drop-in replacement.


#QonQrete #v1.0.1 #Hotfix #LocalFirst #Docker