Skip to content

deployment.md

Latest commit

 

History

History
43 lines (30 loc) · 1.21 KB

deployment.md

File metadata and controls

43 lines (30 loc) · 1.21 KB
Raw

Deploying Triage

Modal (production)

Authenticate once with modal setup, then run the pipeline in order:

modal run modal_app/train.py
modal run modal_app/export.py --run-id <run-id-printed-by-training>
modal deploy modal_app/service.py

Training and export share the triage-model-artifacts Modal Volume. Export writes production.json, so the deployed service always loads the promoted INT8 ONNX artifact rather than an arbitrary checkpoint.

The Modal deployment exposes:

GET  /              one-page interactive demo
GET  /health        model readiness
POST /predict       priority prediction
GET  /metrics       recorded evaluation and optimization metrics
GET  /docs          OpenAPI documentation

Local API

The service expects a promoted artifact at $TRIAGE_ARTIFACTS_PATH (default /models).

export TRIAGE_ARTIFACTS_PATH="$PWD/artifacts"
uvicorn app.main:app --reload

Docker

Build the API image and mount the directory containing production.json and runs/:

docker build -t triage-api .
docker run --rm -p 8000:8000 -v "$PWD/artifacts:/models:ro" triage-api

Open http://localhost:8000 to use the demo and http://localhost:8000/docs for the API contract.