Skip to content

Use dual stream for DS MoE whenever cuda graph is used (instead of with token threshold) #9405

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 3 commits into from
Nov 22, 2025
+3 −2
Merged

Use dual stream for DS MoE whenever cuda graph is used (instead of with token threshold) #9405

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f25b0b7
Add env var SGLANG_DUAL_STREAM_TOKEN_THRESHOLD
trevor-m Aug 20, 2025
ee978de
Use dual stream for moe when in cuda graph mode
trevor-m Nov 19, 2025
ff1d5f4
Merge branch 'main' into token-threshold
Fridge003 Nov 22, 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
5 changes: 3 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 @@ -787,12 +787,13 @@ def forward(
gemm_output_zero_allocator: BumpAllocator = None,
) -> torch.Tensor:
if not self._enable_a2a_moe:
DUAL_STREAM_TOKEN_THRESHOLD = 1024
from sglang.srt.model_executor.cuda_graph_runner import get_is_capture_mode

if (
self.alt_stream is not None
and self.num_fused_shared_experts == 0
and hidden_states.shape[0] > 0
and hidden_states.shape[0] <= DUAL_STREAM_TOKEN_THRESHOLD
and get_is_capture_mode()
):
return self.forward_normal_dual_stream(
hidden_states,
Expand Down
Loading