使用minide 最新镜像启动Qwen3-Coder-30B-A3B-Instruct模型报错，weight model.layers.0.self_attn.q_proj.bias does not exist

驱动版本 npu-smi 25.0.rc1.1               Version: 25.0.rc1.1

镜像版本 ：mindie:2.1.RC1-800I-A2-py311-openeuler24.03-lts

npu 为910B2

[2025-08-07 15:43:01,284] [391] [281463458623840] [llmmodels] [INFO] [flash_causal_qwen2_moe.py-96] : >>>> qwen_MoeDecoderModel is called. 

[2025-08-07 15:43:01.007+0800] [391] [281473500311904] [batchscheduler] [ERROR] [model.py:61] : [Model] >>> Exception:weight model.layers.0.self_attn.q_proj.bias does not exist 

Traceback (most recent call last): 

  File "/usr/local/lib/python3.11/site-packages/model_wrapper/model.py", line 59, in initialize 

    return self.python_model.initialize(config) 

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

  File "/usr/local/lib/python3.11/site-packages/model_wrapper/standard_model.py", line 133, in initialize 

    self.generator = Generator( 

                     ^^^^^^^^^^ 

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/generator.py", line 228, in __init__ 

    self.generator_backend = get_generator_backend(model_config) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/__init__.py", line 32, in get_generator_backend 

    return generator_cls(model_config) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_torch.py", line 136, in __init__ 

    super().__init__(model_config) 

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_backend.py", line 141, in __init__ 

    self.model_wrapper = get_model_wrapper(model_config, backend_type) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/__init__.py", line 15, in get_model_wrapper 

    return wrapper_cls(**model_config) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/atb/atb_model_wrapper.py", line 84, in __init__ 

    self.model_runner.load_weights() 

  File "/usr/local/Ascend/atb-models/atb_llm/runner/model_runner.py", line 214, in load_weights 

    self.model = self.model_cls(self.config, 

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

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/flash_causal_qwen2_moe.py", line 39, in __init__ 

    self.model = FlashQwenModel(config, weights) 

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

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/modeling_qwen2_moe.py", line 282, in __init__ 

    [ 

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/modeling_qwen2_moe.py", line 283, in <listcomp> 

    FlashQwenLayer( 

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/modeling_qwen2_moe.py", line 219, in __init__ 

    self.self_attn = FlashQwenAttention( 

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

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2/modeling_qwen2.py", line 163, in __init__ 

    self.c_attn = load_column_multi( 

                  ^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/Ascend/atb-models/atb_llm/utils/layers/__init__.py", line 50, in load_column_multi 

    b = [weights.get_sharded(f"{p}.bias", dim=0, gqa_size=head_size) for p in prefixes] 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/layers/__init__.py", line 50, in <listcomp> 

    b = [weights.get_sharded(f"{p}.bias", dim=0, gqa_size=head_size) for p in prefixes] 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/weights.py", line 349, in get_sharded 

    slice_ = self._get_slice(tensor_name) 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/weights.py", line 857, in _get_slice 

    filename, tensor_name = self.get_filename(tensor_name) 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/weights.py", line 164, in get_filename 

    raise AssertionError(f"weight {tensor_name} does not exist") 

AssertionError: weight model.layers.0.self_attn.q_proj.bias does not exist 

[2025-08-07 15:43:01.007+0800] [391] [281473500311904] [batchscheduler] [ERROR] [model.py:64] : [MIE04E13030A] [Model]  >>> return initialize error result: {'status': 'error', 'npuBlockNum': '0', 'cpuBlockNum': '0', 'memPoolId': '-1'} 

[2025-08-07 15:43:01.521+08:00] [123] [125] [server] [WARN] [llm_daemon.cpp:74] : [MIE04W01011A] [daemon] Received exit signal[17] 

[2025-08-07 15:43:01.521+08:00] [123] [125] [server] [ERROR] [llm_daemon.cpp:90] : [MIE04E010109] [daemon] Process 391 was terminated by signal 9 (Killed) 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

/usr/lib64/python3.11/multiprocessing/resource_tracker.py:254: UserWarning: resource_tracker: There appear to be 30 leaked semaphore objects to clean up at shutdown 

  warnings.warn('resource_tracker: There appear to be %d ' 

[2025-08-07 15:43:04,552] [393] [281462769840480] [llmmodels] [INFO] [dist.py-81] : initialize_distributed has been Set 

[2025-08-07 15:43:04,724] [393] [281462769840480] [llmmodels] [INFO] [flash_causal_qwen2_moe.py-96] : >>>> qwen_MoeDecoderModel is called. 

[2025-08-07 15:43:04.007+0800] [393] [281472812314976] [batchscheduler] [ERROR] [model.py:61] : [Model] >>> Exception:weight model.layers.0.self_attn.q_proj.bias does not exist 

Traceback (most recent call last): 

  File "/usr/local/lib/python3.11/site-packages/model_wrapper/model.py", line 59, in initialize 

    return self.python_model.initialize(config) 

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

  File "/usr/local/lib/python3.11/site-packages/model_wrapper/standard_model.py", line 133, in initialize 

    self.generator = Generator( 

                     ^^^^^^^^^^ 

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/generator.py", line 228, in __init__ 

    self.generator_backend = get_generator_backend(model_config) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/__init__.py", line 32, in get_generator_backend 

    return generator_cls(model_config) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_torch.py", line 136, in __init__ 

    super().__init__(model_config) 

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_backend.py", line 141, in __init__ 

    self.model_wrapper = get_model_wrapper(model_config, backend_type) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/__init__.py", line 15, in get_model_wrapper 

    return wrapper_cls(**model_config) 

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

  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/atb/atb_model_wrapper.py", line 84, in __init__ 

    self.model_runner.load_weights() 

  File "/usr/local/Ascend/atb-models/atb_llm/runner/model_runner.py", line 214, in load_weights 

    self.model = self.model_cls(self.config, 

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

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/flash_causal_qwen2_moe.py", line 39, in __init__ 

    self.model = FlashQwenModel(config, weights) 

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

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/modeling_qwen2_moe.py", line 282, in __init__ 

    [ 

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/modeling_qwen2_moe.py", line 283, in <listcomp> 

    FlashQwenLayer( 

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2_moe/modeling_qwen2_moe.py", line 219, in __init__ 

    self.self_attn = FlashQwenAttention( 

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

  File "/usr/local/Ascend/atb-models/atb_llm/models/qwen2/modeling_qwen2.py", line 163, in __init__ 

    self.c_attn = load_column_multi( 

                  ^^^^^^^^^^^^^^^^^^ 

  File "/usr/local/Ascend/atb-models/atb_llm/utils/layers/__init__.py", line 50, in load_column_multi 

    b = [weights.get_sharded(f"{p}.bias", dim=0, gqa_size=head_size) for p in prefixes] 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/layers/__init__.py", line 50, in <listcomp> 

    b = [weights.get_sharded(f"{p}.bias", dim=0, gqa_size=head_size) for p in prefixes] 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/weights.py", line 349, in get_sharded 

    slice_ = self._get_slice(tensor_name) 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/weights.py", line 857, in _get_slice 

    filename, tensor_name = self.get_filename(tensor_name) 

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

  File "/usr/local/Ascend/atb-models/atb_llm/utils/weights.py", line 164, in get_filename 

    raise AssertionError(f"weight {tensor_name} does not exist") 

AssertionError: weight model.layers.0.self_attn.q_proj.bias does not exist 

[2025-08-07 15:43:04.007+0800] [393] [281472812314976] [batchscheduler] [ERROR] [model.py:64] : [MIE04E13030A] [Model]  >>> return initialize error result: {'status': 'error', 'npuBlockNum': '0', 'cpuBlockNum': '0', 'memPoolId': '-1'} 

[2025-08-07 15:43:04.942+08:00] [123] [123] [server] [WARN] [llm_daemon.cpp:74] : [MIE04W01011A] [daemon] Received exit signal[17] 

[2025-08-07 15:43:04.942+08:00] [123] [123] [server] [ERROR] [llm_daemon.cpp:90] : [MIE04E010109] [daemon] Process 393 was terminated by signal 9 (Killed) 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

[ERROR] TBE Subprocess[task_distribute] raise error[], main process disappeared! 

/usr/lib64/python3.11/multiprocessing/resource_tracker.py:254: UserWarning: resource_tracker: There appear to be 30 leaked semaphore objects to clean up at shutdown 

  warnings.warn('resource_tracker: There appear to be %d ' 

Daemon is killing... 

[2025-08-07 15:43:06.521+08:00] [123] [zysk] [system] [server] [stop endpoint] [success] 

[2025-08-07 15:43:06.521+08:00] [123] [zysk] [system] [server] [stop mindie server] [success]