Pixano-Inference

Overview

Pixano-Inference serves multimodal inference models behind a REST API powered by Ray Serve. Models are declared in Python config files, deployed as Ray actors, and invoked through either the HTTP API or the Python client.

Key capabilities:

• Python-config deployment -- Declare models in models.py with typed validation
• GPU-aware actors -- Assign CPU/GPU resources per model deployment
• Autoscaling and batching -- Tune replicas and batch size per model
• Multi-model serving -- Run several deployments in one server
• Custom models -- Register your own InferenceModel subclasses

How it works

1. Write a Python config with pixano_inference.configs.ModelConfig
2. Start the server with pixano-inference --config models.py
3. Send requests via the Python client or HTTP API

Next steps

• Getting Started -- Install, configure, and run your first inference
• Server Deployment -- Advanced configuration, autoscaling, and custom models
• HTTP API Reference -- Full endpoint and schema documentation

Contributing

We welcome contributions from the community! Please open issues and pull requests for any bugs or feature requests you may have.

License

Pixano-Inference is released under the terms of the CeCILL-C license.