From ef8a29c429f62e0b0d013cc5091264ac75f70f56 Mon Sep 17 00:00:00 2001
From: Gang Chen <13298548+MoonBall@users.noreply.github.com>
Date: Sat, 5 Jul 2025 14:21:56 +0800
Subject: [PATCH] Embedding parallel by attn_tp (#7623)

---
 python/sglang/srt/models/deepseek_v2.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py
index 2ef529fca..1ebd403b6 100644
--- a/python/sglang/srt/models/deepseek_v2.py
+++ b/python/sglang/srt/models/deepseek_v2.py
@@ -1930,7 +1930,7 @@ class DeepseekV2Model(nn.Module):
         self.embed_tokens = VocabParallelEmbedding(
             config.vocab_size,
             config.hidden_size,
-            enable_tp=not global_server_args_dict["enable_dp_attention"],
+            use_attn_tp_group=True,
         )
         self.alt_stream = torch.cuda.Stream() if _is_cuda else None
         self.layers = nn.ModuleList(