Skip to content

[Fix] A followup fix for TRTLLM BF16 MoE #15303

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
Feb 27, 2026
+4 −3
Merged

[Fix] A followup fix for TRTLLM BF16 MoE #15303

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
112730c
Fix
samuellees Feb 25, 2026
e04068d
Fix
samuellees Feb 25, 2026
92cf69a
Fix
samuellees Feb 25, 2026
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
7 changes: 4 additions & 3 deletions python/sglang/srt/server_args.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2334,8 +2334,9 @@ def _handle_speculative_decoding(self):
):
self.speculative_draft_model_revision = "main"

# Avoid using flashinfer_trtllm for speculative MoE runner backend by default
# TODO: Remove this block after verifying no accuracy regression with flashinfer_trtllm speculative backend
# FlashInfer trtllm moe bf16 only support RenormalizeNaive routing method and Deepseek routing method
# It is hard to tell the routing method in draft model, and the moe layer in draft model is not the bottleneck among
# end to end, so we just avoid using trtllm_moe for speculative decoding.
from sglang.srt.layers.moe.utils import MoeRunnerBackend

if self.speculative_moe_runner_backend is None:
Expand All @@ -2347,7 +2348,7 @@ def _handle_speculative_decoding(self):
else:
assert not MoeRunnerBackend(
self.speculative_moe_runner_backend
).is_flashinfer_trtllm(), "Currently speculative MoE runner backend cannot be flashinfer_trtllm for risk in some draft models."
).is_flashinfer_trtllm(), "Currently speculative MoE runner backend doesn't support flashinfer_trtllm, please use triton or auto backend for speculative moe runner instead."

if self.speculative_algorithm == "NEXTN":
self.speculative_algorithm = "EAGLE"
Expand Down
Loading