-
Notifications
You must be signed in to change notification settings - Fork 36
Pull requests: Comfy-Org/comfy-kitchen
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Triton Backend - Ensure different dim1/dim2 matched against freqs_cis independently
#77
opened Jul 14, 2026 by
sleppyrobot
Loading…
Add HIP backend for AMD RDNA2-RDNA4 to the existing wheels
#74
opened Jul 13, 2026 by
0xDELUXA
Loading…
RDNA-aware autotune configs for the INT8 matmul kernels
#69
opened Jul 10, 2026 by
liminfei-amd
Loading…
int4_tensorwise: W4A4 layout (+ConvRot) with fused CUDA kernels
#63
opened Jul 6, 2026 by
HK416-TYPED
Contributor
Loading…
eager: dequantize INT8 weights instead of torch._int_mm on AMD RDNA / CPU
#56
opened Jun 26, 2026 by
0xDELUXA
Loading…
Change rounding method in quantization process (triton backend fix for amd gpus)
#54
opened Jun 26, 2026 by
patientx
Loading…
Fix crash when running quantized models through torch.compile
#52
opened Jun 25, 2026 by
jtreminio
Loading…
Try to optimize sdvquant implementation a bit.
#50
opened Jun 23, 2026 by
comfyanonymous
Member
Loading…
[docs] Fix QuantizedTensor.from_float examples to use layout name strings
#43
opened May 24, 2026 by
BhagyeshKothalkar
Loading…
Fix __version__ to read from package metadata instead of hardcoded string
#30
opened Feb 22, 2026 by
davidmoserai
Loading…
Adding c10d ops and additional aten ops for DTensor and FSDP2
#29
opened Feb 6, 2026 by
komikndr
Loading…
ProTip!
Updated in the last three days: updated:>2026-07-13.