取出deepep的部分调试信息

vllm/model_executor/layers/fused_moe/modular_kernel.py

@@ -922,7 +922,7 @@ class DeepGemmDisabledFusedMoEModularKernel(torch.nn.Module):
             num_ht_ll_tokens = envs.VLLM_MOE_HT_THRESHOLD
             num_tokens = hidden_states.size(0)
             logger.info("num_tokens=%d", num_tokens)
-            if num_tokens > num_ht_ll_tokens and False:
+            if num_tokens > num_ht_ll_tokens:
                 
                 prepare_finalize = self.prepare_finalize.ht_prepare_finalize
                 fused_experts = self.fused_experts_ht

vllm/v1/worker/gpu_model_runner.py

@@ -1316,7 +1316,7 @@ class GPUModelRunnerBase(LoRAModelRunnerMixin):
          spec_decode_metadata,
          num_scheduled_tokens_np) = (self._prepare_inputs(scheduler_output))
         num_scheduled_tokens = scheduler_output.total_num_scheduled_tokens
-        logger.info("***********self.cudagraph_batch_sizes_max",self.cudagraph_batch_sizes[-1])
+
         if (self.use_cuda_graph
                 and num_scheduled_tokens <= self.cudagraph_batch_sizes[-1]):
             # Use piecewise CUDA graphs.