Skip to content

[Bug] fix not desired disable fused share experts caused by rocm logic #14432

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
BBuf merged 1 commit into from
Dec 5, 2025
Merged

[Bug] fix not desired disable fused share experts caused by rocm logic #14432

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: 4 additions & 2 deletions python/sglang/srt/models/deepseek_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3270,8 +3270,10 @@ def determine_num_fused_shared_experts(
"Only Deepseek V3/R1 on NV-platform with capability >= 80 "
"or AMD-platform with capability >= gfx942(MI30x) can use shared experts fusion optimization."
)
elif get_moe_expert_parallel_world_size() > 1 and (
not _is_hip or torch.cuda.get_device_capability("cuda") < (9, 4)
elif (
get_moe_expert_parallel_world_size() > 1
and _is_hip
and torch.cuda.get_device_capability("cuda") < (9, 4)
):
disable_reason = "Only Deepseek V3/R1 on AMD-platform with capability >= gfx942(MI30x) can use shared experts fusion optimization under expert parallelism."
elif disable_reason is None and get_moe_a2a_backend().is_deepep():
Expand Down
Loading