Skip to content

Commit 614f4df

Browse files
committed
fix naming
Signed-off-by: Jennifer Chen <[email protected]>
1 parent 15a8351 commit 614f4df

File tree

1 file changed

+7
-3
lines changed

1 file changed

+7
-3
lines changed

modelopt/torch/export/plugins/mcore_nemotron.py

Lines changed: 7 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -66,9 +66,11 @@
6666
"linear_fc1": NameRemapping("backbone.layers.{}.mixer.up_proj.", COL_TP),
6767
"linear_fc2": NameRemapping("backbone.layers.{}.mixer.down_proj.", ROW_TP),
6868
# MoE
69-
"router": NameRemapping("model.layers.{}.mlp.gate.", REPLICATE),
69+
"router": NameRemapping(
70+
"backbone.layers.{}.mixer.gate.", {"mapping": {"expert_bias": "e_score_correction_bias"}}
71+
),
7072
"local_experts.linear_fc1": NameRemapping(
71-
"backbone.layers.{}.mixer.experts.{}.up_proj", COL_ETP
73+
"backbone.layers.{}.mixer.experts.{}.up_proj.", COL_ETP
7274
),
7375
"local_experts.linear_fc2": NameRemapping(
7476
"backbone.layers.{}.mixer.experts.{}.down_proj.", ROW_ETP
@@ -104,7 +106,9 @@
104106
"linear_fc1": NameRemapping("backbone.layers.{}.mixer.up_proj."),
105107
"linear_fc2": NameRemapping("backbone.layers.{}.mixer.down_proj."),
106108
# MoE
107-
"router": NameRemapping("backbone.layers.{}.mlp.gate."),
109+
"router": NameRemapping(
110+
"backbone.layers.{}.mixer.gate.", {"mapping": {"expert_bias": "e_score_correction_bias"}}
111+
),
108112
"local_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.experts.{}.up_proj."),
109113
"local_experts.linear_fc2": NameRemapping("backbone.layers.{}.mixer.experts.{}.down_proj."),
110114
"shared_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.shared_experts.up_proj."),

0 commit comments

Comments
 (0)