TuiML Documentation

Serve a trained model over a REST API.

Loads a model and runs it behind HTTP endpoints in the foreground, so other services can request predictions over the network. The server is backed by serving and publishes interactive OpenAPI docs at /docs. Press Ctrl-C to shut it down, or stop a detached server with tuiml stop-server.

Serve a model on the default port:

bash

$ tuiml serve --model-path model.pkl

Serve on a custom port with a specific model ID:

bash

$ tuiml serve --model-path classifier.pkl -m my_classifier -p 9000

Serve with multiple workers, reachable from other machines:

bash

$ tuiml serve --model-path model.pkl -w 4 -H 0.0.0.0

Serve a model already trained in this session, by ID:

bash

$ tuiml serve --model-id abc123

Endpoints

GET /health - Health check
GET /stats - Server statistics
GET /models - List loaded models
POST /models - Load additional models
GET /models/{id} - Get model info
POST /models/{id}/predict - Make predictions
POST /models/{id}/predict_proba - Get probabilities
POST /predict - Predict with default model

API Documentation

After starting the server, visit:

http://localhost:8000/docs - Swagger UI
http://localhost:8000/redoc - ReDoc

Authentication

Every endpoint except / and /health needs a bearer token, printed on startup. Loading a model unpickles a file and predicting runs it, so the server does not expose either without one.

$ curl -H "Authorization: Bearer $TOKEN" localhost:8000/health

serve.py

Functions

tuiml serve

Endpoints

API Documentation

Authentication

Options