diff --git a/python/sglang/srt/layers/communicator.py b/python/sglang/srt/layers/communicator.py
index 30f068595188..909085402255 100644
--- a/python/sglang/srt/layers/communicator.py
+++ b/python/sglang/srt/layers/communicator.py
@@ -1301,8 +1301,12 @@ def _scatter_hidden_states_moe(
 
         # DP scatter (if DP attention is enabled)
         if context.attn_dp_size > 1:
+            if get_tensor_model_parallel_world_size() == get_attention_dp_size():
+                group = get_tp_group()
+            else:
+                group = get_attention_tp_group()
             hidden_states_output, global_hidden_states = (
-                get_local_dp_buffer(),
+                get_local_dp_buffer(group),
                 hidden_states,
             )
             dp_scatter(hidden_states_output, global_hidden_states, forward_batch)
diff --git a/python/sglang/srt/models/deepseek_v4.py b/python/sglang/srt/models/deepseek_v4.py
index f4128884ad58..0a4740fa4a98 100644
--- a/python/sglang/srt/models/deepseek_v4.py
+++ b/python/sglang/srt/models/deepseek_v4.py
@@ -17,7 +17,11 @@
     fused_rope_inplace,
 )
 from sglang.srt.configs.deepseek_v4 import DeepSeekV4Config
-from sglang.srt.distributed import get_pp_group, get_tensor_model_parallel_world_size
+from sglang.srt.distributed import (
+    get_pp_group,
+    get_tensor_model_parallel_world_size,
+    get_tp_group,
+)
 from sglang.srt.environ import envs
 from sglang.srt.eplb.expert_location import ModelConfigForExpertLocation
 from sglang.srt.layers.attention.dsv4.compressor import Compressor
@@ -32,11 +36,13 @@
 from sglang.srt.layers.dp_attention import (
     _DpGatheredBufferWrapper,
     attn_tp_all_gather,
-    dp_gather_partial,
+    attn_tp_all_reduce,
+    dp_gather_replicate,
     dp_scatter,
     get_attention_cp_rank,
     get_attention_cp_size,
     get_attention_dp_size,
+    get_attention_tp_group,
     get_attention_tp_rank,
     get_attention_tp_size,
     get_global_dp_buffer,
@@ -303,7 +309,8 @@ def __init__(
             self.hidden_size,
             bias=False,
             quant_config=quant_config,
-            reduce_results=attn_tp_size > 1,
+            reduce_results=attn_tp_size == get_tensor_model_parallel_world_size()
+            and attn_tp_size > 1,
             prefix=add_prefix("wo_b", prefix),
             tp_rank=attn_tp_rank,
             tp_size=attn_tp_size,
@@ -516,9 +523,6 @@ def forward(
         forward_batch: ForwardBatch,
     ) -> torch.Tensor:
         if not get_attn_tp_context().input_scattered and x.shape[0] == 0:
-            assert (
-                not self.wo_b.reduce_results
-            ), "short-circuiting allreduce will lead to hangs"
             return x
 
         attn_backend = forward_batch.attn_backend
@@ -601,6 +605,8 @@ def forward(
             o = torch.einsum("tgd,grd->tgr", o, wo_a)
 
         o, _ = self.wo_b(o.flatten(1))
+        if self.tp_size > 1 and self.tp_size < get_tensor_model_parallel_world_size():
+            o = attn_tp_all_reduce(o)
 
         return o
 
@@ -830,8 +836,14 @@ def forward(
             input_ids = input_ids[cp_rank::cp_size].contiguous()
             input_ids_global = input_ids
         elif _use_tp_moe_gather:
-            hidden_states, local_hidden_states = get_global_dp_buffer(), hidden_states
-            dp_gather_partial(hidden_states, local_hidden_states, forward_batch)
+            hidden_states, local_hidden_states = (
+                get_global_dp_buffer(get_tp_group()),
+                hidden_states,
+            )
+            # hidden_states here follow TP_ATTN_FULL semantics: they are replicated
+            # within an attention-TP group. Use replicate gather to avoid summing the
+            # same activations across attention-TP ranks before entering MLP/MoE.
+            dp_gather_replicate(hidden_states, local_hidden_states, forward_batch)
         _a2a_scatter_chunks: Optional[List[torch.Tensor]] = None
         if _use_tp_attn_a2a_scatter:
             s, r = get_attention_tp_size(), get_attention_tp_rank()
@@ -846,7 +858,10 @@ def forward(
             input_ids_global=input_ids_global,
         )
         if _use_tp_moe_gather:
-            hidden_states, global_hidden_states = get_local_dp_buffer(), hidden_states
+            hidden_states, global_hidden_states = (
+                get_local_dp_buffer(get_attention_tp_group()),
+                hidden_states,
+            )
             dp_scatter(hidden_states, global_hidden_states, forward_batch)
         if _use_tp_attn_a2a_scatter:
             assert _a2a_scatter_chunks is not None
@@ -950,7 +965,9 @@ def forward(
                 dtype=input_ids.dtype,
                 device=input_ids.device,
             )
-            dp_gather_partial(input_ids_global, input_ids[:, None], forward_batch)
+            # Token ids are replicated within an attention-TP group. Use replicate
+            # gather here to avoid summing duplicated ids when attention_tp_size > 1.
+            dp_gather_replicate(input_ids_global, input_ids[:, None], forward_batch)
             input_ids_global = input_ids_global.squeeze(-1)
         else:
             input_ids_global = input_ids