Skip to content

Support MOE Export for Nemotron H #447

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

Open
wants to merge 9 commits into
base: main
Choose a base branch
from
Open

Support MOE Export for Nemotron H #447

Changes from 1 commit
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
15a8351
nano3 import export HF
jenchen13 Oct 20, 2025
614f4df
fix naming
jenchen13 Oct 21, 2025
7d3245d
Merge branch 'main' into jennifchen/nmh-moe-export
jenchen13 Oct 21, 2025
0b25bcb
Merge branch 'main' into jennifchen/nmh-moe-export
jenchen13 Oct 27, 2025
95f6c25
fix moe token forward
jenchen13 Oct 27, 2025
41357c8
allow custom router dtype
jenchen13 Oct 28, 2025
8393149
fix lambda function
jenchen13 Oct 29, 2025
5f7158f
fix import and export moe router dtype
jenchen13 Oct 29, 2025
f9066c1
fix router type
jenchen13 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
24 changes: 24 additions & 0 deletions modelopt/torch/export/plugins/mcore_nemotron.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,8 +17,10 @@
"""Custom mapping from Nemotron Hugging Face models to Megatron Core models."""

from .mcore_custom import (
COL_ETP,
COL_TP,
REPLICATE,
ROW_ETP,
ROW_TP,
CustomModuleMapping,
NameRemapping,
Expand Down Expand Up @@ -63,6 +65,20 @@
"pre_mlp_layernorm": NameRemapping("backbone.layers.{}.norm.", REPLICATE),
"linear_fc1": NameRemapping("backbone.layers.{}.mixer.up_proj.", COL_TP),
"linear_fc2": NameRemapping("backbone.layers.{}.mixer.down_proj.", ROW_TP),
# MoE
"router": NameRemapping("model.layers.{}.mlp.gate.", REPLICATE),
"local_experts.linear_fc1": NameRemapping(
"backbone.layers.{}.mixer.experts.{}.up_proj", COL_ETP
),
"local_experts.linear_fc2": NameRemapping(
"backbone.layers.{}.mixer.experts.{}.down_proj.", ROW_ETP
),
"shared_experts.linear_fc1": NameRemapping(
"backbone.layers.{}.mixer.shared_experts.up_proj.", COL_TP
),
"shared_experts.linear_fc2": NameRemapping(
"backbone.layers.{}.mixer.shared_experts.down_proj.", ROW_TP
),
}


Expand All @@ -87,4 +103,12 @@
"pre_mlp_layernorm": NameRemapping("backbone.layers.{}.norm."),
"linear_fc1": NameRemapping("backbone.layers.{}.mixer.up_proj."),
"linear_fc2": NameRemapping("backbone.layers.{}.mixer.down_proj."),
# MoE
"router": NameRemapping("backbone.layers.{}.mlp.gate."),
"local_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.experts.{}.up_proj."),
"local_experts.linear_fc2": NameRemapping("backbone.layers.{}.mixer.experts.{}.down_proj."),
"shared_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.shared_experts.up_proj."),
"shared_experts.linear_fc2": NameRemapping(
"backbone.layers.{}.mixer.shared_experts.down_proj."
),
}