Skip to content

[perf] Enable concurrent execution of "shared_experts" and "selected_experts" in qwen3-next #27578

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 3 commits into from
Oct 29, 2025
Merged

[perf] Enable concurrent execution of "shared_experts" and "selected_experts" in qwen3-next #27578

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c786eb8
init
ZJY0516 Oct 27, 2025
d56e4be
Merge branch 'main' into dual_stram
ZJY0516 Oct 29, 2025
32e914b
Merge branch 'main' into dual_stram
ZJY0516 Oct 29, 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
17 changes: 12 additions & 5 deletions vllm/model_executor/models/qwen3_next.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,6 +159,7 @@ def __init__(self, vllm_config: VllmConfig, prefix: str = ""):

self.experts = SharedFusedMoE(
shared_experts=self.shared_expert,
gate=self.gate,
num_experts=self.n_routed_experts,
top_k=config.num_experts_per_tok,
hidden_size=config.hidden_size,
Expand All @@ -181,11 +182,17 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
if self.is_sequence_parallel:
hidden_states = sequence_parallel_chunk(hidden_states)

# router_logits: (num_tokens, n_experts)
router_logits, _ = self.gate(hidden_states)
final_hidden_states = self.experts(
hidden_states=hidden_states, router_logits=router_logits
)
if self.experts.is_internal_router:
# In this case, the gate/router runs inside the FusedMoE class
final_hidden_states = self.experts(
hidden_states=hidden_states, router_logits=hidden_states
)
else:
# router_logits: (num_tokens, n_experts)
router_logits, _ = self.gate(hidden_states)
final_hidden_states = self.experts(
hidden_states=hidden_states, router_logits=router_logits
)

if self.shared_expert is not None:
final_hidden_states = final_hidden_states[0] + final_hidden_states[1]
Expand Down