[DP encoder] Fix pos_emb layer TP issue when DP encoder enabled for Qwen3 VL (#20788)

python/sglang/srt/models/qwen3_vl.py

@@ -330,6 +330,7 @@ class Qwen3VLMoeVisionModel(nn.Module, RotaryPosMixin):
                 self.hidden_size,
                 quant_config=quant_config,
                 use_attn_tp_group=is_dp_attention_enabled(),
+                enable_tp=False if use_data_parallel else True,
                 prefix=add_prefix("pos_embed", prefix),
             )
         else: