Skip to content

Use deepgemm instead of triton for fused_qkv_a_proj_with_mqa #6890

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jun 5, 2025
Merged

Use deepgemm instead of triton for fused_qkv_a_proj_with_mqa #6890

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
3618f95
Update fp8_utils.py
fzyzcjy Jun 5, 2025
268b267
Update fp8_utils.py
fzyzcjy Jun 5, 2025
74508e9
Update fp8_utils.py
fzyzcjy Jun 5, 2025
2735c50
Merge branch 'main' into fzyzcjy-patch-2
fzyzcjy Jun 5, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions python/sglang/srt/layers/quantization/fp8_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -227,8 +227,8 @@ def deepgemm_w8a8_block_fp8_linear_with_fallback(
output_dtype = input.dtype
dtype_supported = output_dtype == torch.bfloat16

# TODO: add more robust shape check here
shape_supported = weight.shape[0] % 128 == 0 and weight.shape[1] % 128 == 0
# TODO: https://github.com/sgl-project/sglang/pull/6890#issuecomment-2943395737
shape_supported = weight.shape[0] % 64 == 0 and weight.shape[1] % 128 == 0

if not (shape_supported and dtype_supported):
# fall back to triton
Expand Down
Loading