Python class

DenoisingCacheConfig

`DenoisingCacheConfig`

class max.pipelines.lib.DenoisingCacheConfig(*, config_file=None, section_name=None, first_block_caching=False, taylorseer=False, taylorseer_cache_interval=None, taylorseer_warmup_steps=None, taylorseer_max_order=None, teacache=False, teacache_rel_l1_thresh=None, teacache_coefficients=None)

source

Bases: ConfigFileModel

Pipeline-level cache configuration for diffusion model denoising.

Controls First-Block Cache (step cache) and TaylorSeer optimizations that skip redundant transformer passes during the denoising loop.

Parameters:

config_file (str | None)
section_name (str | None)
first_block_caching (bool)
taylorseer (bool)
taylorseer_cache_interval (int | None)
taylorseer_warmup_steps (int | None)
taylorseer_max_order (int | None)
teacache (bool)
teacache_rel_l1_thresh (float | None)
teacache_coefficients (list[float] | None)

`first_block_caching`

first_block_caching: bool

source

`model_config`

model_config: ClassVar[ConfigDict] = {'extra': 'forbid', 'frozen': False, 'strict': False}

source

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

`taylorseer`

taylorseer: bool

source

`taylorseer_cache_interval`

taylorseer_cache_interval: int | None

source

`taylorseer_max_order`

taylorseer_max_order: int | None

source

`taylorseer_warmup_steps`

taylorseer_warmup_steps: int | None

source

`teacache`

teacache: bool

source

`teacache_coefficients`

teacache_coefficients: list[float] | None

source

`teacache_rel_l1_thresh`

teacache_rel_l1_thresh: float | None

source

DenoisingCacheConfig​

first_block_caching​

model_config​

taylorseer​

taylorseer_cache_interval​

taylorseer_max_order​

taylorseer_warmup_steps​

teacache​

teacache_coefficients​

teacache_rel_l1_thresh​