Skip to main content

Python module

max.pipelines.modeling.base

Base classes for MAX pipeline model definitions.

Componentsโ€‹

ComponentModelBase interface for component models with weight-backed execution.
TensorStructBase for structured tensor containers used as pipeline inputs/outputs.

First-block cacheโ€‹

FirstBlockCacheStandalone FirstBlockCache module.
FirstBlockCacheStatePer-request mutable state for FirstBlockCache.

Denoising cacheโ€‹

DenoisingCacheConfigPipeline-level cache configuration for diffusion model denoising.
DenoisingCacheStatePer-request mutable cache state for a single denoising stream.
TaylorSeerStandalone TaylorSeer caching module.
TaylorSeerStatePer-request mutable state for TaylorSeer caching.
fbcache_conditional_executionHandle FBCache F.cond branching pattern shared across DiT models.
run_denoising_stepExecute one denoising step with caching logic.