Spaces:
Running
Running
Commit History
-
Expose the API alongside the Streamlit demo in the HF Space via nginx
Daniel WiesmannAdds nginx as a reverse proxy in front of both the Streamlit demo and the FastAPI app: / goes to Streamlit, /api/ goes to the API, both managed by supervisord in the single HF Space container (nginx.conf) and mirrored for local docker-compose (nginx.local.conf, using container DNS names instead of localhost). Fixes on top of the original approach: - Dockerfile switched to USER root to install the nginx config/perms but never switched back, so the whole container would've run as root at runtime despite HF Spaces requiring UID 1000 - restored USER user before CMD. - GAZET_PUBLIC_API_URL was set but never read anywhere; wired it into the Streamlit sidebar as a link to the API docs so the exposed API is actually discoverable. - The /api/ location was missing proxy_buffering off and generous timeouts (present on the streamlit location but not copied over): without them, GET /search/stream's incremental NDJSON would get buffered by nginx until the response completes, and the…7cc374f
-
hf: update Dockerfile & supervisord to point to q8_0.gguf model
71d368c
-
fix: pull model from models/
d128a38
-
fix: register llama shared libs with ldconfig and LD_LIBRARY_PATH
e753721 -
fix: keep llama-server binary and backend .so files together
d688b0d -
fix: correct llama-server binary path in Space Dockerfile
e4226db -
enh: Add HF spaces support with all-in-one Dockerfile
cb3558e -
enh: Add docker compose file to host llama server, fastapi app & streamlit together
789bf58