diff --git a/vllm_ascend/ascend_forward_context.py b/vllm_ascend/ascend_forward_context.py
index 582e4957e6b..d013be34db2 100644
--- a/vllm_ascend/ascend_forward_context.py
+++ b/vllm_ascend/ascend_forward_context.py
@@ -69,9 +69,7 @@ def set_ascend_forward_context(
 
         from vllm_ascend.ops.fused_moe.moe_comm_method import get_moe_comm_method
 
-        max_num_tokens = int(num_tokens_across_dp.max().item()) if num_tokens_across_dp is not None else num_tokens
-        moe_comm_type = select_moe_comm_method(max_num_tokens, vllm_config, is_draft_model)
-
+        moe_comm_type = select_moe_comm_method(num_tokens, vllm_config, is_draft_model)
         forward_context.moe_comm_type = moe_comm_type
         forward_context.moe_comm_method = get_moe_comm_method(moe_comm_type)
 
diff --git a/vllm_ascend/worker/model_runner_v1.py b/vllm_ascend/worker/model_runner_v1.py
index 677e9925a82..89af64bc731 100644
--- a/vllm_ascend/worker/model_runner_v1.py
+++ b/vllm_ascend/worker/model_runner_v1.py
@@ -1235,7 +1235,6 @@ def execute_model(
                     num_scheduled_tokens_np=num_scheduled_tokens_np,
                     max_num_scheduled_tokens=max_num_scheduled_tokens,
                     use_cascade_attn=cascade_attn_prefix_lens is not None,
-                    force_eager=self.model_config.enforce_eager,
                     num_encoder_reqs=len(scheduler_output.scheduled_encoder_inputs),
                 )
 
@@ -1854,7 +1853,6 @@ def _sync_batch_across_dp(
         self,
         num_tokens_padded: int | None = None,
         cudagraph_mode: int = 0,
-        allow_dp_padding: bool = False,
     ) -> tuple[bool, torch.Tensor | None, int]:
         """
         Coordinates amongst all DP ranks to determine if and how the full batch
@@ -1898,16 +1896,11 @@ def _sync_batch_across_dp(
 
         num_tokens_across_dp = tensor[0, :]
         max_num_tokens = int(num_tokens_across_dp.max().item())
-
-        if allow_dp_padding:
-            num_tokens_after_padding = torch.tensor(
-                [max_num_tokens] * len(num_tokens_across_dp),
-                device="cpu",
-                dtype=torch.int32,
-            )
-        else:
-            num_tokens_after_padding = num_tokens_across_dp.cpu()
-
+        num_tokens_after_padding = torch.tensor(
+            [max_num_tokens] * len(num_tokens_across_dp),
+            device="cpu",
+            dtype=torch.int32,
+        )
         # Synchronize cudagraph_mode across ranks (take min)
         synced_cudagraph_mode = _post_process_cudagraph_mode(tensor)
         return False, num_tokens_after_padding, synced_cudagraph_mode
@@ -1976,7 +1969,6 @@ def dispatch_cudagraph(num_tokens, disable_full=False, valid_modes=None):
             _, num_tokens_across_dp, synced_cudagraph_mode = self._sync_batch_across_dp(
                 num_tokens_padded=num_tokens_padded,
                 cudagraph_mode=cudagraph_mode.value,
-                allow_dp_padding=cudagraph_mode != CUDAGraphMode.NONE,
             )
 
             # Extract DP padding if there is any