Skip to content

Fix flashinfer cutlass MoE output shape for non-FP4-packed inputs #14028

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
Fridge003 merged 1 commit into from
Nov 27, 2025
Merged

Fix flashinfer cutlass MoE output shape for non-FP4-packed inputs #14028

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 5 additions & 1 deletion python/sglang/srt/layers/quantization/modelopt_quant.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1654,11 +1654,15 @@ def apply(

output_dtype = torch.bfloat16

# If x_sf is not None, x is FP4 packed (half size), so we need * 2
# If x_sf is None, x is not packed, so output_col = x.shape[1]
output_col = x.shape[1] * 2 if x_sf is not None else x.shape[1]

with use_symmetric_memory(
get_tp_group(), disabled=not is_allocation_symmetric()
):
symm_output = torch.empty(
x.shape[0], x.shape[1] * 2, dtype=output_dtype, device=x.device
x.shape[0], output_col, dtype=output_dtype, device=x.device
)

output = flashinfer_cutlass_fused_moe(
Expand Down
Loading