测试命令：python benchmark_throughput_dcu.py  --num-prompts 16 --input-len 512 --output-len 1024 --model /mnt/nvme/DeepSeek-R1-Distill-Qwen-1.5B --tensor-parallel-size 2 --dtype float16 --enforce-eager

报错：

INFO 02-19 11:32:52 selector.py:223] Cannot use _Backend.FLASH_ATTN backend on CPU. 

INFO 02-19 11:32:52 selector.py:154] Using MindIE backend. 

Traceback (most recent call last): 

  File "/mnt/nvme/zmc/benchmark_throughput_dcu.py", line 645, in <module> 

    main(args) 

  File "/mnt/nvme/zmc/benchmark_throughput_dcu.py", line 409, in main 

    elapsed_time = run_vllm(*run_args, args.use_new_beam_search_impl) 

                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 

  File "/mnt/nvme/zmc/benchmark_throughput_dcu.py", line 99, in run_vllm 

    llm = LLM( 

          ^^^^ 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/entrypoints/llm.py", line 214, in __init__ 

    self.llm_engine = LLMEngine.from_engine_args( 

                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/engine/llm_engine.py", line 576, in from_engine_args 

    engine = cls( 

             ^^^^ 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/engine/llm_engine.py", line 282, in __init__ 

    self.model_executor = executor_class( 

                          ^^^^^^^^^^^^^^^ 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/executor/executor_base.py", line 47, in __init__ 

    self._init_executor() 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/executor/npu_executor.py", line 75, in _init_executor 

    self.driver_worker.init_device() 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/worker/npu_worker.py", line 123, in init_device 

    init_worker_distributed_environment( 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/worker/npu_worker.py", line 317, in init_worker_distributed_environment 

    init_distributed_environment(parallel_config.world_size, rank, distributed_init_method, local_rank, "hccl") 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/distributed/parallel_state.py", line 964, in init_distributed_environment 

    _WORLD = init_world_group(ranks, local_rank, backend) 

             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/distributed/parallel_state.py", line 843, in init_world_group 

    return GroupCoordinator( 

           ^^^^^^^^^^^^^^^^^ 

  File "/usr/local/lib/python3.11/site-packages/vllm-0.6.3.dev0+g7193774b.d20250218.npu-py3.11.egg/vllm/distributed/parallel_state.py", line 186, in __init__ 

    cpu_group = torch.distributed.new_group(ranks, backend="gloo") 

                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/lib64/python3.11/site-packages/torch/distributed/c10d_logger.py", line 74, in wrapper 

    func_return = func(*args, **kwargs) 

                  ^^^^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/lib64/python3.11/site-packages/torch/distributed/distributed_c10d.py", line 3881, in new_group 

    return _new_group_with_tag(ranks, timeout, backend, pg_options, None, use_local_synchronization=use_local_synchronization) 

           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/lib64/python3.11/site-packages/torch/distributed/distributed_c10d.py", line 3944, in _new_group_with_tag 

    pg, pg_store = _new_process_group_helper( 

                   ^^^^^^^^^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/lib64/python3.11/site-packages/torch/distributed/distributed_c10d.py", line 1264, in _new_process_group_helper 

    backend_class = ProcessGroupGloo(backend_prefix_store, group_rank, group_size, timeout=timeout) 

                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 

RuntimeError: Socket Timeout 

[ERROR] 2025-02-19-12:33:17 (PID:36839, Device:0, RankID:-1) ERR99999 UNKNOWN application exception