[bugfix] fix deepseek bug when tp_size == 1 (#1755)

### What this PR does / why we need it?
fix deepseek bug when tp_size == 1

### Does this PR introduce _any_ user-facing change?


### How was this patch tested?


Signed-off-by: zzzzwwjj <1183291235@qq.com>

Author: zzzzwwjj

vllm_ascend/ops/fused_moe.py

@@ -1242,13 +1242,15 @@ def forward(self,
             if isinstance(e_hidden_states, tuple):
                 e_hidden_states, shared_hidden_states = e_hidden_states
 
-        if tp_size > 1 and fused_moe_state != FusedMoEState.AllGather:
-            dist.all_gather(list(chunk_hidden_states), e_hidden_states,
-                            self.tp_group)
-            final_hidden_states = torch.cat(chunk_hidden_states, dim=0)
+        if fused_moe_state != FusedMoEState.AllGather:
+            if tp_size > 1:
+                dist.all_gather(list(chunk_hidden_states), e_hidden_states,
+                                self.tp_group)
+                final_hidden_states = torch.cat(chunk_hidden_states, dim=0)
+            else:
+                final_hidden_states = e_hidden_states
             if num_tokens < forward_context.padded_num_tokens:
                 final_hidden_states = final_hidden_states[:num_tokens]
-            dispose_tensor(e_hidden_states)
         elif self.dp_size > 1 and fused_moe_state == FusedMoEState.AllGather:
             final_hidden_states = dist._functional_collectives.reduce_scatter_tensor(
                 e_hidden_states,