[bugfix] fix mcore_bridge deepseek-v3 (#6508)

Jintao-Huang · web-flow · commit f6883d1cce9d · 2025-11-09T20:55:28.000+08:00
diff --git a/swift/megatron/model/gpt_bridge.py b/swift/megatron/model/gpt_bridge.py
@@ -75,6 +75,7 @@ def _get_tp_split_dim(mg_key: Optional[str]) -> Optional[int]:
             'linear_qkv',
             # mla
             'linear_q_proj',
+            'linear_q_up_proj',
             'linear_kv_up_proj'
         }
         # RowLinear

Original file line number	Diff line number	Diff line change
`@@ -75,6 +75,7 @@ def _get_tp_split_dim(mg_key: Optional[str]) -> Optional[int]:`
`75`	`75`	`'linear_qkv',`
`76`	`76`	`# mla`
`77`	`77`	`'linear_q_proj',`
	`78`	`+ 'linear_q_up_proj',`
`78`	`79`	`'linear_kv_up_proj'`
`79`	`80`	`}`
`80`	`81`	`# RowLinear`