mindie2.1启动qwen3:32b报错llminferengine failed to init llminfermodels_MindIE_昇腾论坛

# mindie2.1启动qwen3:32b报错llminferengine failed to init llminfermodels_MindIE_昇腾论坛

## 概述

本文档基于昇腾社区论坛帖子生成的技术教程。

**原始链接**: https://www.hiascend.com/forum/thread-0259189916428936146-1-1.html

**生成时间**: 2025-08-27 10:33:56

---

## 问题描述

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/toke...

## 完整内容

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 CANN 8.1.T18 PTAdapter 7.0.RC1.B100 transformers 4.51.0

---

## 技术要点总结

基于以上内容，主要技术要点包括：

1. **问题类型**: 错误处理
2. **涉及技术**: NPU, Atlas, MindIE, 昇腾, CANN, AI
3. **解决方案**: 请参考完整内容中的解决方案

## 相关资源

- 昇腾社区: https://www.hiascend.com/
- 昇腾论坛: https://www.hiascend.com/forum/

---

*本文档由AI自动生成，仅供参考。如有疑问，请参考原始帖子。*