[fix]解决EP不开启mtp时某些size报错

vllm/v1/worker/gpu_model_runner.py

@@ -2151,7 +2151,10 @@ class GPUModelRunnerBase(LoRAModelRunnerMixin):
         if not self.ep_sp:
             num_scheduled_tokens_list[-1] += num_tokens % num_reqs
         else:
-            num_scheduled_tokens_list[-1] += num_tokens % min_tokens_per_req
+            if self.speculative_config is not None:
+                num_scheduled_tokens_list[-1] += num_tokens % min_tokens_per_req
+            else:
+                num_scheduled_tokens_list[-1] += num_tokens % num_reqs
         assert sum(num_scheduled_tokens_list) == num_tokens
         assert len(num_scheduled_tokens_list) == num_reqs
         num_scheduled_tokens = np.array(num_scheduled_tokens_list,