kernels

Kernel integrations for sparse attention: Triton FA and diffusers/LTX backends.

Functions

get_skip_softmax_context	Return whether skip-softmax softmax patching is active.
register_diffusers_triton_attention	Register modelopt_triton backend in diffusers.
register_ltx_triton_attention	Patch all ltx_core.Attention modules for Triton dispatch.
set_skip_softmax_context	Set whether skip-softmax softmax patching is active (thread-local).

get_skip_softmax_context()

Return whether skip-softmax softmax patching is active.

Return type:

bool

register_diffusers_triton_attention()

Register modelopt_triton backend in diffusers.

Safe to call multiple times; registration happens only once.

Return type:

None

register_ltx_triton_attention(model)

Patch all ltx_core.Attention modules for Triton dispatch.

Parameters:

model (Module)

Return type:

None

set_skip_softmax_context(active)

Set whether skip-softmax softmax patching is active (thread-local).

Parameters:

active (bool)

Return type:

None