Skip to content

[NemotronH] Use ReplicatedLinear for fc1_latent_proj #31807

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
tlrmchlsmth merged 1 commit into from
Jan 6, 2026
+1 −5
Merged

[NemotronH] Use ReplicatedLinear for fc1_latent_proj #31807

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: 1 addition & 5 deletions vllm/model_executor/models/nemotron_h.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,16 +210,12 @@ def __init__(
)

if self.use_latent_moe:
# TODO: check if using ReplicatedLinear is better than
# ColumnParallelLinear + all_gather
self.fc1_latent_proj = ColumnParallelLinear(
self.fc1_latent_proj = ReplicatedLinear(
input_size=config.hidden_size,
output_size=self.moe_hidden_size,
bias=config.mlp_bias,
quant_config=quant_config,
disable_tp=self.is_sequence_parallel,
# We need to gather the output to prepare input for moe
gather_output=True,
prefix=f"{prefix}.fc1_latent_proj",
)
self.fc2_latent_proj = ReplicatedLinear(
Expand Down