Atlas 800TA2单机8卡

[2025-08-07 10:11:24.889] [920] [281473177280128] [llm] [INFO] [llm_manager_impl.cpp:76] LLMRuntime init success!

[2025-08-07 10:11:30.225] [920] [281473177280128] [llm] [ERROR] [model_deploy_config.cpp:101] Failed to get vocab size from tokenizer wrapper with exception: ValueError: safe_get_tokenizer_from_pretrained failed. Please check the input parameters model_path, kwargs and the version of transformers. If the input parameters are valid, the required files exist in model_path and the version of transformers is correct, make sure the folder's owner has execute permission. Otherwise, please check the function stack for detailed exception information and the logs of the llmmodels.

At:

  /usr/local/Ascend/atb-models/atb_llm/models/base/model_utils.py(261): wrapper

  /usr/local/Ascend/atb-models/atb_llm/models/qwen2/router_qwen2.py(84): get_tokenizer

  /usr/local/Ascend/atb-models/atb_llm/models/base/router.py(207): tokenizer

  /usr/local/Ascend/atb-models/atb_llm/runner/tokenizer_wrapper.py(32): __init__

  /usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/atb/atb_tokenizer_wrapper.py(8): __init__

  /usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/__init__.py(28): get_tokenizer_wrapper

[2025-08-07 10:11:30.226] [920] [281473177280128] [llm] [ERROR] [llm_manager_impl.cpp:846] Config manager init exception: ValueError: safe_get_tokenizer_from_pretrained failed. Please check the input parameters model_path, kwargs and the version of transformers. If the input parameters are valid, the required files exist in model_path and the version of transformers is correct, make sure the folder's owner has execute permission. Otherwise, please check the function stack for detailed exception information and the logs of the llmmodels.

At:

  /usr/local/Ascend/atb-models/atb_llm/models/base/model_utils.py(261): wrapper

  /usr/local/Ascend/atb-models/atb_llm/models/qwen2/router_qwen2.py(84): get_tokenizer

  /usr/local/Ascend/atb-models/atb_llm/models/base/router.py(207): tokenizer

  /usr/local/Ascend/atb-models/atb_llm/runner/tokenizer_wrapper.py(32): __init__

  /usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/atb/atb_tokenizer_wrapper.py(8): __init__

  /usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/__init__.py(28): get_tokenizer_wrapper

llmmodel.log

[2025-08-07 10:11:30,225] [920] [281473177280128] [llmmodels] [ERROR] [logging.py-54] : data did not match any variant of untagged enum ModelWrapper at line 757479 column 3

mindieserver.log

[2025-08-07 10:11:30,225] [920] [281473177280128] [llmmodels] [ERROR] [logging.py-54] : data did not match any variant of untagged enum ModelWrapper at line 757479 column 3

[root@localhost debug]# cat mindie-server_920_202508071011226.log

[2025-08-07 10:11:30.226+08:00] [920] [922] [server] [ERROR] [llm_infer_model_instance.cpp:344] : MIE05E040001[llm_backend] [LLMInferModelInstance::Init] llmManager_ init fail!

[2025-08-07 10:11:30.226+08:00] [920] [922] [server] [ERROR] [llm_infer_model.cpp:28] : MIE05E040001[llm_backend] ModelInstance init failed

[2025-08-07 10:11:30.226+08:00] [920] [922] [server] [ERROR] [llm_infer_engine.cpp:41] : MIE05E040001[llm_backend] LLMInferEngine initialize LLMInferModels failed. please check the error stacks for more details

[2025-08-07 10:11:30.226+08:00] [920] [922] [server] [ERROR] [infer_backend_manager.cpp:216] : [MIE04E06011C] [infer_backend_manager] Failed to init engine. backendName is mindieservice_llm_engine

[2025-08-07 10:11:30.226+08:00] [920] [922] [server] [ERROR] [endpoint.cpp:45] : [MIE04E02011C] [endpoint] Failed to init engine! Please check in the env MINDIE_LOG_PATH (default ~/mindie/log) or console output.

使用mindie2.0可以启动qwen3这个模型，升级2.1后无法启动，报错日志如上。

环境