Skip to content

Commit 02be52e

Browse files
committed
lint
Signed-off-by: Jennifer Chen <[email protected]>
1 parent 271bb98 commit 02be52e

File tree

1 file changed

+14
-8
lines changed

1 file changed

+14
-8
lines changed

modelopt/torch/export/plugins/mcore_nemotron.py

Lines changed: 14 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -19,8 +19,8 @@
1919
from .mcore_custom import (
2020
COL_ETP,
2121
COL_TP,
22-
ROW_ETP,
2322
REPLICATE,
23+
ROW_ETP,
2424
ROW_TP,
2525
CustomModuleMapping,
2626
NameRemapping,
@@ -41,7 +41,7 @@
4141
"linear_fc2": NameRemapping("model.layers.{}.mlp.down_proj."),
4242
"final_layernorm": NameRemapping("model.norm."),
4343
"output_layer": NameRemapping("lm_head."),
44-
}
44+
}
4545

4646

4747
nemotron_h_causal_lm_import: dict[str, CustomModuleMapping] = {
@@ -67,13 +67,18 @@
6767
"linear_fc2": NameRemapping("backbone.layers.{}.mixer.down_proj.", ROW_TP),
6868
# MoE
6969
"router": NameRemapping("model.layers.{}.mlp.gate.", REPLICATE),
70-
"local_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.experts.{}.up_proj", COL_ETP),
71-
"local_experts.linear_fc2": NameRemapping("backbone.layers.{}.mixer.experts.{}.down_proj.", ROW_ETP),
72-
"shared_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.shared_experts.up_proj.", COL_TP),
70+
"local_experts.linear_fc1": NameRemapping(
71+
"backbone.layers.{}.mixer.experts.{}.up_proj", COL_ETP
72+
),
73+
"local_experts.linear_fc2": NameRemapping(
74+
"backbone.layers.{}.mixer.experts.{}.down_proj.", ROW_ETP
75+
),
76+
"shared_experts.linear_fc1": NameRemapping(
77+
"backbone.layers.{}.mixer.shared_experts.up_proj.", COL_TP
78+
),
7379
"shared_experts.linear_fc2": NameRemapping(
7480
"backbone.layers.{}.mixer.shared_experts.down_proj.", ROW_TP
7581
),
76-
7782
}
7883

7984

@@ -103,6 +108,7 @@
103108
"local_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.experts.{}.up_proj."),
104109
"local_experts.linear_fc2": NameRemapping("backbone.layers.{}.mixer.experts.{}.down_proj."),
105110
"shared_experts.linear_fc1": NameRemapping("backbone.layers.{}.mixer.shared_experts.up_proj."),
106-
"shared_experts.linear_fc2": NameRemapping("backbone.layers.{}.mixer.shared_experts.down_proj."),
107-
111+
"shared_experts.linear_fc2": NameRemapping(
112+
"backbone.layers.{}.mixer.shared_experts.down_proj."
113+
),
108114
}

0 commit comments

Comments
 (0)