-
Notifications
You must be signed in to change notification settings - Fork 89
Pull requests: Comfy-Org/comfy-kitchen
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
sol_attn: optional blk_cnt out-parameter for the routed-block count
#168
opened Sep 8, 2026 by
fblissjr
Loading…
MiniMax H3 VAE kernels: fused encoder pad/norm, fp16-accumulate conv3d and GEMM, int8 residual epilogue
#167
opened Sep 8, 2026 by
kijai
Contributor
Loading…
Remove the torch._dynamo import, dead since #3 removed its only user
#160
opened Sep 6, 2026 by
PozzettiAndrea
Loading…
Add a float-GEMM fallback for int8_linear on devices without aten::_int_mm
#145
opened Aug 30, 2026 by
vqt123
Loading…
sm86 INT8 GEMM tuning: measured tile heuristic, per-shape cfg cache, and path logging
#142
opened Aug 29, 2026 by
Many0therFunctions
Loading…
Enable gfx1170/gfx1171 HIP backend support
#141
opened Aug 28, 2026 by
tvukovic-amd
Contributor
Loading…
[ROCm] Add native gfx12 HIP attention and Z-Image INT8 fusions
#139
opened Aug 27, 2026 by
tvukovic-amd
Contributor
Loading…
fix: use typing.List in na3d signature for PyTorch custom_op compatib…
#135
opened Aug 24, 2026 by
itzero-ru
Loading…
fix(eager): fallback int8 matmul on ROCm/HIP to prevent access violation on RDNA2
#133
opened Aug 23, 2026 by
niklastheiler
Loading…
Tune gfx1151 INT8 deep-K tile selection
#130
opened Aug 22, 2026 by
Yasei-no-otoko
Loading…
5 tasks done
HIP: vectorize rowwise INT8 quantization with 16-byte loads
#127
opened Aug 21, 2026 by
tangzzycc
Loading…
Round fused SwiGLU to the storage dtype before ConvRot.
#124
opened Aug 20, 2026 by
niehen6174
Loading…
Previous Next
ProTip!
Type g p on any issue or pull request to go back to the pull request listing page.