Python module
registry
Model registry, for tracking various model variants.
PipelineRegistry
class max.pipelines.registry.PipelineRegistry(architectures: list[max.pipelines.registry.SupportedArchitecture])
get_active_huggingface_config()
get_active_huggingface_config(huggingface_repo: HuggingFaceRepo) → AutoConfig
Retrieves or creates a cached HuggingFace AutoConfig for the given model configuration.
This method maintains a cache of HuggingFace configurations to avoid reloading them unnecessarily which incurs a huggingface hub API call. If a config for the given model hasn’t been loaded before, it will create a new one using AutoConfig.from_pretrained() with the model’s settings.
-
Parameters:
huggingface_repo – The HuggingFaceRepo containing the model.
-
Returns:
The HuggingFace configuration object for the model.
-
Return type:
AutoConfig
get_active_tokenizer()
get_active_tokenizer(huggingface_repo: HuggingFaceRepo) → PreTrainedTokenizer | PreTrainedTokenizerFast
Retrieves or creates a cached HuggingFace AutoTokenizer for the given model configuration.
This method maintains a cache of HuggingFace tokenizers to avoid reloading them unnecessarily which incurs a huggingface hub API call. If a tokenizer for the given model hasn’t been loaded before, it will create a new one using AutoTokenizer.from_pretrained() with the model’s settings.
-
Parameters:
huggingface_repo – The HuggingFaceRepo containing the model.
-
Returns:
The HuggingFace tokenizer for the model.
-
Return type:
PreTrainedTokenizer | PreTrainedTokenizerFast
register()
register(architecture: SupportedArchitecture)
Add new architecture to registry.
reset()
reset() → None
retrieve()
retrieve(pipeline_config: PipelineConfig, task: PipelineTask = PipelineTask.TEXT_GENERATION) → tuple[PipelineTokenizer, TokenGenerator | EmbeddingsGenerator]
retrieve_architecture()
retrieve_architecture(huggingface_repo: HuggingFaceRepo) → SupportedArchitecture | None
retrieve_factory()
retrieve_factory(pipeline_config: PipelineConfig, task: PipelineTask = PipelineTask.TEXT_GENERATION) → tuple[PipelineTokenizer, Callable[[], TokenGenerator | EmbeddingsGenerator]]
SupportedArchitecture
class max.pipelines.registry.SupportedArchitecture(name: str, example_repo_ids: list[str], default_encoding: SupportedEncoding, supported_encodings: dict[max.pipelines.config_enums.SupportedEncoding, list[max.nn.kv_cache.cache_params.KVCacheStrategy]], pipeline_model: type[max.pipelines.pipeline.PipelineModel], task: PipelineTask, tokenizer: type[Union[max.pipelines.tokenizer.TextTokenizer, max.pipelines.tokenizer.TextAndVisionTokenizer]], default_weights_format: WeightsFormat, multi_gpu_supported: bool = False, rope_type: RopeType = RopeType.none, weight_adapters: dict[max.graph.weights.format.WeightsFormat, Callable[..., dict[str, max.graph.weights.weights.WeightData]]] | None = None)
Initializes a model architecture supported by MAX pipelines.
New architectures should be registered into the PipelineRegistry.
-
Parameters:
- name – Architecture name.
- example_repo_ids – Hugging Face repo_id which runs this architecture.
- default_encoding – Default encoding for the model.
- supported_encodings – Alternate encodings supported.
- pipeline_model –
PipelineModelclass that defines the model graph and execution. - task – Which pipeline task should the model run with.
- tokenizer – Tokenizer used to preprocess model inputs.
- default_weights_format – The weights format used in pipeline_model.
- weight_converters – A dictionary of weight loaders to use if the input checkpoint has a different format than the default.
get_pipeline_for_task()
max.pipelines.registry.get_pipeline_for_task(task: PipelineTask, pipeline_config: PipelineConfig) → type[TextGenerationPipeline] | type[EmbeddingsPipeline] | type[SpeculativeDecodingTextGenerationPipeline]
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!