Use attn tp group in embedding for more models (#17570)

2026-01-24 13:37:44 +08:00
parent 176da1bbdd
commit fb683be6eb
19 changed files with 19 additions and 19 deletions
--- a/python/sglang/srt/models/llada2.py
+++ b/python/sglang/srt/models/llada2.py
@@ -674,7 +674,7 @@ class LLaDA2MoeModel(nn.Module):
                self.embed_dim,
                quant_config=quant_config,
                prefix=add_prefix("word_embeddings", prefix),
-                enable_tp=not is_dp_attention_enabled(),
+                use_attn_tp_group=is_dp_attention_enabled(),
            )
        else:
            self.word_embeddings = PPMissingLayer()