Repository navigation

All pull requests

Add bottom-right causal alignment to FlyDSL backward
#559amd-wsung102 opened Oct 19, 2026Collaborator
1
Reject unsupported group sizes in Hopper BF16/INT4 GEMM
#557Lorenzo-Mazza opened Oct 17, 2026
3
Blackwell FMHA: fix NaNs, hang and bad masking for empty rows
#553bottler opened Oct 15, 2026Contributor
1
Skip bf16_grouped_fallback_test in OSS ROCm CI (#548)
#548bottler opened Oct 11, 2026Contributor
1
Remove dead int32-KV branch from both split-K decode ops
#545aryaman-gupta opened Oct 9, 2026Contributor3 of 4 tasks
2
Route short paged query blocks to the head-packed decode kernel
#538RichardChamberlain1 opened Sep 30, 2026Contributor
1
Support sliding-window attention on paged KV and size split-K from chain length
#537RichardChamberlain1 opened Sep 30, 2026Contributor
1
Memoise paged launch metadata on the attention bias
#535RichardChamberlain1 opened Sep 30, 2026Contributor
3
Suppress type errors for Pyre upgrade - mslk
#532maggiemoss opened Sep 29, 2026Contributor
1
FlyDSL paged decode: maintain MFMA occupancy on short-query workload
#531RichardChamberlain1 opened Sep 29, 2026Contributor
1
Skip dispatch reporting during torch.compile (#530)
#530amolagrawalmeta opened Sep 28, 2026
1
Stop the CUDA conda-build CI job from OOMing the worker
#511sryap opened Sep 4, 2026Contributor
1
Autotune async copy for rowwise FP8 GEMM (#494)
#494warrendeng opened Aug 21, 2026
Add Triton FP8×INT4 rowwise GEMM for ROCm gfx942/gfx950
#479apicciau opened Aug 11, 2026Contributor
Fix flash_attn_bench crash on the flash_attn.cute (FA4) backend: pass window_size=(-1, -1) instead of None
#397Maurits-de-Groot opened Jun 19, 2026
3
[CUDA] [PERFORMANCE] Increase speed of bf16bf16bf16_grouped_wgrad via indicating that ElementC is void / nullptr
#329benediktjohannes opened Apr 19, 2026
2
[WIP] Migrate MOE kernels to libtorch ABI stable
#325janeyx99 opened Apr 14, 2026Contributor

Pull requests