Python package
The pipelines package provides end-to-end implementations for text generation, embeddings, audio generation, and speech processing that automatically convert Hugging Face models into performance-optimized MAX graphs. Each pipeline can be served via OpenAI-compatible endpoints for production deployment.
Modules
architectures: Architecture classes for all supported model types.config: Pipeline configuration for text generation, embeddings, audio generation, and more.core: Core pipeline request and response types.hf_utils: Utilities for interacting with Hugging Face repositories and weight files.interfaces: Abstract base classes and protocols for pipeline components.log_probabilities: Log probability computation graphs.lora_config: LoRA adapter configuration and management.model_config: Model configuration dataclasses.pipeline: Pipeline implementations for text generation.registry: Model registry and factory functions.sampling: Token sampling and speculative decoding utilities.tokenizer: Tokenization utilities for pipeline inputs.