Comfy-Org/comfy-kitchen

Fast kernel library for Diffusion inference with multiple compute backends.

Stars: 220
Forks: 91
Open beginner issues: 0
Indexed issues: 35
Avg merge: 1d 7h
Merged PRs (30d): 12

Dominant language: Python
License: Apache-2.0
Last GitHub push: Sep 15, 2026
Latest indexed: Sep 15, 2026
Contributing guide: Contributing guide
Code of conduct: No code of conduct
Beginner labels: No beginner labels indexed

Open All

35 open issues indexed

Bug: Scales do not persist in fp32 when LoRA is applied. Open

Comfy-Org/comfy-kitchen#2 · 4 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
cannot import name '_C' from partially initialized module 'comfy_kitchen.backends.cuda Open

Comfy-Org/comfy-kitchen#5 · 13 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
INT8 support Open

Comfy-Org/comfy-kitchen#9 · 7 comments · 4 reactions · 0 assignees · Indexed Sep 15, 2026
memory bug Open

Comfy-Org/comfy-kitchen#10 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
RTX30xx FP8 feather matmul triton kernel? Open

Comfy-Org/comfy-kitchen#16 · 6 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Hope to support older CUDA version backends Open

Comfy-Org/comfy-kitchen#21 · 1 comment · 2 reactions · 0 assignees · Indexed Sep 15, 2026
Clarification regarding the readme example Open

Comfy-Org/comfy-kitchen#27 · 4 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
The --no-cuda argument can't be passed through pyproject.toml Open

Comfy-Org/comfy-kitchen#31 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
FP8 E4M3 quantized models inference is slower than same model quantized in MXFP8 Open

Comfy-Org/comfy-kitchen#38 · 4 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
100 cpu when import Open

Comfy-Org/comfy-kitchen#44 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Python Python 3.13.11 Open

Comfy-Org/comfy-kitchen#45 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Fatal access violation (0xC0000005) running int8_tensorwise model on ROCm Windows (gfx1101 / RDNA3) Open

Comfy-Org/comfy-kitchen#61 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Memory access fault using Krea 2 INT8 CONVROT with Triton backend on AMD gfx1151 Open

Comfy-Org/comfy-kitchen#64 · 12 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
INT8 matmul autotune uses NVIDIA-tuned configs on AMD RDNA Open

Comfy-Org/comfy-kitchen#68 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
ComfyUI crashes using Flux Klein 9b with Triton backend - dequantize_nvfp4_kernel_tl Open

Comfy-Org/comfy-kitchen#71 · 6 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
eager backend advertises int8_linear on MPS but dispatches to CUDA-only torch._int_mm Open

Comfy-Org/comfy-kitchen#92 · 7 comments · 2 reactions · 0 assignees · Indexed Sep 15, 2026
CUTLASS INT8 dequant selects poor config for tall MiniMax H3 shapes on RTX A6000 Open

Comfy-Org/comfy-kitchen#98 · 2 comments · 1 reaction · 0 assignees · Indexed Sep 15, 2026
retune int8 GEMM tiles for gfx1103 and vectorize rowwise quantize Open

Comfy-Org/comfy-kitchen#102 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
ck-attention STILL crashes on AMD after #104 fix Open

Comfy-Org/comfy-kitchen#110 · 14 comments · 1 reaction · 0 assignees · Indexed Sep 15, 2026
na3d custom-op registration uses PEP-585 annotations, breaking import on torch < 2.7 Open

Comfy-Org/comfy-kitchen#114 · 1 comment · 4 reactions · 0 assignees · Indexed Sep 15, 2026
Feature Request: Streamed QKV Projection and Packed Low-Bit K/V Support for Extreme Long-Sequence Attention Open

Comfy-Org/comfy-kitchen#119 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 15, 2026
[Bug] GPU LOST (driver device lost) with comfy kitchen attention (INT8) on NVIDIA RTX 5060 Ti (sm_120), Windows -- same workflow with native attention is stable Open

Comfy-Org/comfy-kitchen#120 · 2 comments · 1 reaction · 0 assignees · Indexed Sep 15, 2026
Segfault in HIP `dequantize_per_tensor_fp8` on gfx1201 (RDNA4) Open

Comfy-Org/comfy-kitchen#125 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Flash Attention Decode availability detection raises ValueError for CPU devices instead of returning a boolean Open

Comfy-Org/comfy-kitchen#131 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
nanobind 3.0 fails with Clang targeting Windows: dllimport vectorcall aliases are not constant expressions Open

Comfy-Org/comfy-kitchen#132 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
INT8 GEMM kernels index the output in int32: illegal memory access once m*n >= 2**31 (MiniMax H3 at high resolution x length) Open

Comfy-Org/comfy-kitchen#136 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
MiniMax ref2va Noise output with comfy-kitchen attention Open

Comfy-Org/comfy-kitchen#140 · 10 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
CUDA backend cannot build with sm_70 (V100) in arch list: ptxas ldmatrix/cvt.pack errors + missing flash.h submodule Open

Comfy-Org/comfy-kitchen#144 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
Proposal: `xpu` backend for `sol_attn` (Intel Arc / Xe-HPG), VSA subset first Open

Comfy-Org/comfy-kitchen#152 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
H3 (MiniMax) on RTX 5060 Ti (sm_120): text conditioning ignored + numerical blowup in DiT blocks Open

Comfy-Org/comfy-kitchen#163 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
hello, anyone interested in corrections for Apple Silicon? I had codex correct a new issue after updating comfy. Open

Comfy-Org/comfy-kitchen#164 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
[Bug] Triton rms_rope kernel exceeds CUDA gridDim.z limit (65535) when seq_len >= 65536 Open

Comfy-Org/comfy-kitchen#169 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
HIP backend: kernels taking `__bf16` never launch on ROCm 7.1 / clang 21 (host stub and device symbol mangled differently) Open

Comfy-Org/comfy-kitchen#170 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 15, 2026
int8_attention: masked path costs ~1.8x unmasked, and a request for optional return_lse Open

Comfy-Org/comfy-kitchen#177 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026
int8_attention: OverflowError past 299,520 key rows (int32 strides), reachable on long-video workloads Open

Comfy-Org/comfy-kitchen#178 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 15, 2026

A short digest of beginner-friendly GitHub issues.