For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Python data

PIPELINE_REGISTRY

PIPELINE_REGISTRY

max.pipelines.lib.registry.PIPELINE_REGISTRY = <max.pipelines.lib.registry.PipelineRegistry object>

source

Global registry of supported model architectures and their pipelines.

This singleton is automatically populated with all built-in architectures when you import max.pipelines.

Use PIPELINE_REGISTRY to:

• Register custom architectures: Call register() to add a new model architecture.
• Query supported models: Call retrieve_architecture() to check whether a Hugging Face model repository is supported.
• Access cached configs: Use get_active_huggingface_config() and get_active_tokenizer() for cached access to model configurations and tokenizers.

See PipelineRegistry for the full API.