add-inference-model

Onboarding a new pre-trained model involves automating class implementation, registry wiring, tests, and weight packaging. This ensures consistent model onboarding across multiple architectures and backends with optional workflow blocks and adapters.

Model registry wiring connects new core pre-trained models to the inference pipeline via script-based registration. It standardizes weight packaging and submission to the registry, ensuring models are accessible for production workflows and HTTP endpoints.

Packaging model weights requires creating a zip file and submitting the model package to the registry. This process is automated alongside class creation, registry wiring, and unit tests to maintain consistent model onboarding.

Yes, the scaffolding applies to multiple architectures and backends. It provides templates and guidance for optional surfaces like foundation blocks for workflows and adapters for inference endpoints, ensuring broad compatibility.

Adding an inference model requires both unit tests and integration tests. The scaffolding also includes documentation templates to ensure the new core pre-trained model meets required testing standards before registry submission.

Yes, you can expose models via workflows using foundation blocks or through HTTP endpoints using adapters. These optional surfaces are supported during the end-to-end onboarding of new core pre-trained models.