commit 93f5abc78e61c9e3ac917f44b1da39f0d92600d7
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jun 15 00:16:02 2026 +0200

dev-python/comfy-kitchen-bin: add 0.2.10

Fast diffusion-inference kernel library. Provides comfy_kitchen.apply_rope,
which ComfyUI's flux/lumina/z-image families call unconditionally through
comfy.quant_ops.ck during sampling -- so it is required for those models
despite the import being try/except-wrapped -- plus the FP8/FP4/NVFP4
quantization kernels.

Upstream ships no sdist. USE=cuda installs the cp312-abi3 x86_64 manylinux
wheel (CUDA + Triton kernels); without cuda the py3-none-any wheel (eager +
Triton backends, no CUDA) for the CPU/ROCm path.