Ethanfel 82d7f4997a Add configurable attention backend with SageAttention variant support ...

Replace the auto-detect xformers shim with a runtime dispatcher that
always intercepts xformers.ops.memory_efficient_attention. A new
dropdown on STARModelLoader (and --attention CLI arg) lets users
explicitly select: sdpa (default), xformers, sageattn, or specific
SageAttention kernels (fp16 triton/cuda, fp8 cuda). Only backends
that successfully import appear as options.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-02-15 00:12:26 +01:00

22 KiB

Executable File

Raw Blame History

View Raw