| name | local-test |
| description | Build and run a model server locally via Docker Compose, then test it with curl. Use when the engineer wants to test a model server locally before committing. |
| argument-hint | [service-name] |
| allowed-tools | ["Bash(docker *)","Bash(curl *)","Read"] |
Local Test — Inference Services
Build, run, and test a model server locally.
Steps
- Ensure a
.env file exists in the repo root with PATH_TO_<MODEL>_MODEL=/local/path/to/model.
- Build the image:
docker compose build <service>.
- Start the service:
docker compose up <service>.
- Once it's running, send a test request. The endpoint is
POST /v1/models/<model-name>:predict where <model-name> is the MODEL_NAME env var set for that service in docker-compose.yml. Check src/models/<service>/model_server/model.py for the expected input schema. Example for revert-risk language agnostic model: curl localhost:8080/v1/models/revertrisk-language-agnostic:predict -X POST -d '{"lang":"en","rev_id":12345}' -H "Content-type: application/json".
- To stop:
Ctrl+C or docker compose down.
Notes
- ARM Macs: If the service doesn't start, add
platform: linux/amd64 to the service in docker-compose.yml.
- The service listens on
localhost:8080 by default. Check docker-compose.yml if the port is mapped differently. If :8080 is already in use, stop the other service or remap the port.
- Model files must be downloaded separately from
https://analytics.wikimedia.org/published/wmf-ml-models/ and placed at the path referenced in .env.
- To run detached:
docker compose up -d <service>. Logs: docker compose logs <service>.
Input
$ARGUMENTS — the docker-compose service name (e.g., revertrisk-language-agnostic, articlequality, reference-need).
