Skip to content

[Bugfix][MoE] Only unpad routed output before shared expert add #40865

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
tomeras91 merged 4 commits into from
Apr 25, 2026
Merged

[Bugfix][MoE] Only unpad routed output before shared expert add #40865

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
992cd02
[Bugfix][MoE] Only unpad routed output before shared expert add
netanel-haber Apr 25, 2026
9bd6e7b
Merge branch 'main' into bugfix/only-truncate-padded-fused-output-for…
jeejeelee Apr 25, 2026
ab253dd
still truncate if self.routed_output_transform is not None, even if t…
netanel-haber Apr 25, 2026
71acc3e
fix comment
netanel-haber Apr 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
6 changes: 5 additions & 1 deletion vllm/model_executor/layers/fused_moe/runner/moe_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -552,6 +552,8 @@ def forward(

# Record before `_maybe_pad_hidden_states` pads activations to match
# `moe_config.hidden_dim`, e.g. after `align_trtllm_fp4_moe_hidden_dim_for_fi`
# so routed output can be trimmed before
# shared+routed add / latent up proj if needed.
routed_hidden_dim = hidden_states.shape[-1]
hidden_states, og_hidden_dim = self._maybe_pad_hidden_states(
shared_experts_input,
Expand All @@ -577,7 +579,9 @@ def forward(

# Extract outputs from result
shared_output, fused_output = _unpack(result)
if hidden_dim_was_padded:
if (
shared_output is not None or self.routed_output_transform is not None
) and hidden_dim_was_padded:
fused_output = fused_output[..., :routed_hidden_dim]

# If combine kernel already reduced fused, reduce shared to match.
Expand Down
Loading