ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-29 19:28:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

remove restriction of group mode hd192 no lse

Browse Source
This commit is contained in:
Yaoming Mu
2025-06-02 07:36:31 +00:00
parent 160788cdf4
commit 1c4e6d3b70
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/ck_tile/01_fmha/codegen/ops/fmha_fwd.py
View File

@@ -492,7 +492,7 @@ def get_fwd_blobs(kernel_filter : Optional[str], receipt, mask_impl) -> Tuple[Fm
continue
if hdim == 192 and tile.F_bn1 == 128:
# NOTE: this is used to speedup deepseek prefill case, we don't gen training
if pipeline.F_bias != 'no' or pipeline.F_lse == 't' or pipeline.F_dropout == 't':
if pipeline.F_bias != 'no' or pipeline.F_dropout == 't':
continue
k = FmhaFwdKernel(F_idx=0,
F_hdim=hdim,