K100-AI推理Qwen3-30B-A3B报错RuntimeError: Worker failed with error 'only BW and set LMSLIM_USE_LIGHTOP=1 support Marlin W16A16 MoE', please check the stack trace above for the root cause 镜像`harbor.sourcefind.cn:5443/dcu/admin/base/vllm:0.9.2-ubuntu22.04-dtk25.04.2-tx-1226-das1.7-py3.10-20251226`是不是不支持K100-AI卡啊 docker-compose.yaml ```yaml services: vllm: image: harbor.sourcefind.cn:5443/dcu/admin/base/vllm:0.9.2-ubuntu22.04-dtk25.04.2-tx-1226-das1.7-py3.10-20251226 container_name: vllm-test environment: - HIP_VISIBLE_DEVICES=0,1 - HSA_FORCE_FINE_GRAIN_PCIE=1 - NCCL_MIN_NCHANNELS=16 - NCCL_MAX_NCHANNELS=16 - NCCL_P2P_DISABLE=1 - VLLM_USE_FUSED_RMS_ROPE=1 - VLLM_USE_MARLIN_W16A16_MOE=1 - VLLM_USE_FLASH_ATTN_PA=0 #- VLLM_LOGGING_LEVEL=DEBUG #- CUDA_LAUNCH_BLOCKING=1 #- NCCL_DEBUG=TRACE #- VLLM_TRACE_FUNCTION=1 command: > python3 -m vllm.entrypoints.openai.api_server --model /workspace/models/Qwen3-30B-A3B --tensor-parallel-size 2 --gpu-memory-utilization 0.9 --served-model-name Qwen3-30B-A3B --dtype auto --max-model-len 32768 --trust-remote-code --enforce-eager --host 0.0.0.0 --port 8000 network_mode: host ipc: host shm_size: 128G devices: - "/dev/kfd:/dev/kfd" - "/dev/mkfd:/dev/mkfd" - "/dev/dri:/dev/dri" volumes: - /opt/hyhal:/opt/hyhal - /home/whkej/llm/models:/workspace/models - /sys/kernel/debug:/sys/kernel/debug cap_add: - SYS_PTRACE security_opt: - seccomp=unconfined group_add: - video restart: unless-stopped ``` 错误日志 ```text vllm-test | (VllmWorker rank=0 pid=696) ERROR 12-31 08:40:47 [multiproc_executor.py:522] AssertionError: only BW and set LMSLIM_USE_LIGHTOP=1 support Marlin W16A16 MoE vllm-test | ERROR 12-31 08:40:47 [core.py:603] EngineCore failed to start. vllm-test | ERROR 12-31 08:40:47 [core.py:603] Traceback (most recent call last): vllm-test | ERROR 12-31 08:40:47 [core.py:603] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 594, in run_engine_core vllm-test | ERROR 12-31 08:40:47 [core.py:603] engine_core = EngineCoreProc(*args, **kwargs) vllm-test | ERROR 12-31 08:40:47 [core.py:603] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 421, in __init__ vllm-test | ERROR 12-31 08:40:47 [core.py:603] super().__init__(vllm_config, executor_class, log_stats, vllm-test | ERROR 12-31 08:40:47 [core.py:603] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 84, in __init__ vllm-test | ERROR 12-31 08:40:47 [core.py:603] self._initialize_kv_caches(vllm_config) vllm-test | ERROR 12-31 08:40:47 [core.py:603] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 144, in _initialize_kv_caches vllm-test | ERROR 12-31 08:40:47 [core.py:603] available_gpu_memory = self.model_executor.determine_available_memory() vllm-test | ERROR 12-31 08:40:47 [core.py:603] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/abstract.py", line 76, in determine_available_memory vllm-test | ERROR 12-31 08:40:47 [core.py:603] output = self.collective_rpc("determine_available_memory") vllm-test | ERROR 12-31 08:40:47 [core.py:603] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 215, in collective_rpc vllm-test | ERROR 12-31 08:40:47 [core.py:603] result = get_response(w, dequeue_timeout) vllm-test | ERROR 12-31 08:40:47 [core.py:603] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 202, in get_response vllm-test | ERROR 12-31 08:40:47 [core.py:603] raise RuntimeError( vllm-test | ERROR 12-31 08:40:47 [core.py:603] RuntimeError: Worker failed with error 'only BW and set LMSLIM_USE_LIGHTOP=1 support Marlin W16A16 MoE', please check the stack trace above for the root cause vllm-test | ERROR 12-31 08:40:53 [multiproc_executor.py:135] Worker proc VllmWorker-0 died unexpectedly, shutting down executor. vllm-test | Process EngineCore_0: vllm-test | Traceback (most recent call last): vllm-test | File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _bootstrap vllm-test | self.run() vllm-test | File "/usr/lib/python3.10/multiprocessing/process.py", line 108, in run vllm-test | self._target(*self._args, **self._kwargs) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 607, in run_engine_core vllm-test | raise e vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 594, in run_engine_core vllm-test | engine_core = EngineCoreProc(*args, **kwargs) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 421, in __init__ vllm-test | super().__init__(vllm_config, executor_class, log_stats, vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 84, in __init__ vllm-test | self._initialize_kv_caches(vllm_config) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 144, in _initialize_kv_caches vllm-test | available_gpu_memory = self.model_executor.determine_available_memory() vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/abstract.py", line 76, in determine_available_memory vllm-test | output = self.collective_rpc("determine_available_memory") vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 215, in collective_rpc vllm-test | result = get_response(w, dequeue_timeout) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 202, in get_response vllm-test | raise RuntimeError( vllm-test | RuntimeError: Worker failed with error 'only BW and set LMSLIM_USE_LIGHTOP=1 support Marlin W16A16 MoE', please check the stack trace above for the root cause vllm-test | Traceback (most recent call last): vllm-test | File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main vllm-test | return _run_code(code, main_globals, None, vllm-test | File "/usr/lib/python3.10/runpy.py", line 86, in _run_code vllm-test | exec(code, run_globals) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/entrypoints/openai/api_server.py", line 1495, in vllm-test | uvloop.run(run_server(args)) vllm-test | File "/usr/local/lib/python3.10/dist-packages/uvloop/__init__.py", line 69, in run vllm-test | return loop.run_until_complete(wrapper()) vllm-test | File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete vllm-test | File "/usr/local/lib/python3.10/dist-packages/uvloop/__init__.py", line 48, in wrapper vllm-test | return await main vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/entrypoints/openai/api_server.py", line 1431, in run_server vllm-test | await run_server_worker(listen_address, sock, args, **uvicorn_kwargs) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/entrypoints/openai/api_server.py", line 1451, in run_server_worker vllm-test | async with build_async_engine_client(args, client_config) as engine_client: vllm-test | File "/usr/lib/python3.10/contextlib.py", line 199, in __aenter__ vllm-test | return await anext(self.gen) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/entrypoints/openai/api_server.py", line 158, in build_async_engine_client vllm-test | async with build_async_engine_client_from_engine_args( vllm-test | File "/usr/lib/python3.10/contextlib.py", line 199, in __aenter__ vllm-test | return await anext(self.gen) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/entrypoints/openai/api_server.py", line 194, in build_async_engine_client_from_engine_args vllm-test | async_llm = AsyncLLM.from_vllm_config( vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/async_llm.py", line 162, in from_vllm_config vllm-test | return cls( vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/async_llm.py", line 124, in __init__ vllm-test | self.engine_core = EngineCoreClient.make_async_mp_client( vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core_client.py", line 96, in make_async_mp_client vllm-test | return AsyncMPClient(*client_args) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core_client.py", line 666, in __init__ vllm-test | super().__init__( vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core_client.py", line 403, in __init__ vllm-test | with launch_core_engines(vllm_config, executor_class, vllm-test | File "/usr/lib/python3.10/contextlib.py", line 142, in __exit__ vllm-test | next(self.gen) vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/utils.py", line 444, in launch_core_engines vllm-test | wait_for_engine_startup( vllm-test | File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/utils.py", line 494, in wait_for_engine_startup vllm-test | raise RuntimeError("Engine core initialization failed. " vllm-test | RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {} ```