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 class

TaylorSeerState

TaylorSeerState

class max.pipelines.diffusion.TaylorSeerState(factor_0=None, factor_1=None, factor_2=None, last_compute_step=None)

source

Bases: object

Per-request mutable state for TaylorSeer caching.

Allocated fresh for each denoising request via TaylorSeer.create_state().

Parameters:

• factor_0 (Tensor | None)
• factor_1 (Tensor | None)
• factor_2 (Tensor | None)
• last_compute_step (int | None)

factor_0

factor_0: Tensor | None = None

source

Cached 0th-order factor (function value).

factor_1

factor_1: Tensor | None = None

source

Cached 1st-order factor (first derivative approximation).

factor_2

factor_2: Tensor | None = None

source

Cached 2nd-order factor (second derivative approximation).

last_compute_step

last_compute_step: int | None = None

source

Step index of the last full transformer computation.