环境：四卡K100+VLLM0.9.2+das.opt1.rc1.dtk25041 尝试使用VLLM部署Qwen3-235B-A22B-2507-AWQ模型时报错，已测试Qwen3-32b-AWQ和Qwen3-30B-A3B-Coder均可正常部署运行 --- 模型config如下： { "name_or_path": "tclf90/Qwen3-235B-A22B-Instruct-2507-AWQ", "architectures": [ "Qwen3MoeForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 151643, "decoder_sparse_step": 1, "eos_token_id": 151645, "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 12288, "max_position_embeddings": 262144, "max_window_layers": 94, "mlp_only_layers": [], "model_type": "qwen3_moe", "moe_intermediate_size": 1536, "norm_topk_prob": true, "num_attention_heads": 64, "num_experts": 128, "num_experts_per_tok": 8, "num_hidden_layers": 94, "num_key_value_heads": 4, "output_router_logits": false, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 5000000, "router_aux_loss_coef": 0.001, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "float16", "transformers_version": "4.51.0", "use_cache": true, "use_sliding_window": false, "vocab_size": 151936, "quantization_config": { "quant_method": "awq", "bits": 4, "group_size": 128, "version": "gemm", "zero_point": true, "modules_to_not_convert": [".mlp.gate"] } } --- 模型运行命令如下： VLLM_USE_MODELSCOPE=true vllm serve './Qwen3-235b-2507-awq/' --tensor-parallel 4 --host 0.0.0.0 --port 30000 --enforce-eager --quantization awq --enable-expert-parallel 报错信息如下： ERROR 10-16 09:04:56 [core.py:589] EngineCore failed to start. ERROR 10-16 09:04:56 [core.py:589] Traceback (most recent call last): ERROR 10-16 09:04:56 [core.py:589] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 580, in run_engine_core ERROR 10-16 09:04:56 [core.py:589] engine_core = EngineCoreProc(*args, **kwargs) ERROR 10-16 09:04:56 [core.py:589] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 407, in __init__ ERROR 10-16 09:04:56 [core.py:589] super().__init__(vllm_config, executor_class, log_stats, ERROR 10-16 09:04:56 [core.py:589] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 84, in __init__ ERROR 10-16 09:04:56 [core.py:589] self._initialize_kv_caches(vllm_config) ERROR 10-16 09:04:56 [core.py:589] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/engine/core.py", line 144, in _initialize_kv_caches ERROR 10-16 09:04:56 [core.py:589] available_gpu_memory = self.model_executor.determine_available_memory() ERROR 10-16 09:04:56 [core.py:589] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/abstract.py", line 76, in determine_available_memory ERROR 10-16 09:04:56 [core.py:589] output = self.collective_rpc("determine_available_memory") ERROR 10-16 09:04:56 [core.py:589] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 215, in collective_rpc ERROR 10-16 09:04:56 [core.py:589] result = get_response(w, dequeue_timeout) ERROR 10-16 09:04:56 [core.py:589] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 202, in get_response ERROR 10-16 09:04:56 [core.py:589] raise RuntimeError( ERROR 10-16 09:04:56 [core.py:589] RuntimeError: Worker failed with error 'Not adapted to the current size!!!', please check the stack trace above for the root cause ---