MindIE 启动 DeepSeek-R1-0528-Qwen3-8B 报错_MindIE_昇腾论坛

# MindIE 启动 DeepSeek-R1-0528-Qwen3-8B 报错_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-0279186994636045004-1-1.html

**生成时间**: 2025-08-27 10:33:59

---

## 问题描述

问题使用MindIE启动DeepSeek-R1-0528-Qwen3-8B模型失败 系统欧拉 6.6.0-28.0.0.34.oe2403.aarch64 MindIE版本镜像 mindie:2.0.RC1-300I-Duo-py311-openeuler24.03-lts 说明mindie部署后启动DeepSeek-R1-Distill-Qwen-14B模型 或 Qwen3-14B均可正常启动并访问更换 DeepSeek-R1-0528-Qwen3-8B 模型就失败了 报错信息似乎是这个 ---- /root/mindie/log/debug/mindie-llm_427_20250704144051390.log [2025-07-04 14:40:57.435] [427] [281472955441120] [llm] [ERROR] [model_deploy_config.cpp:102] Failed to get vocab size from tokenizer wrapper with exception: TypeError: RopeScaling.__init__() got an unexpected keyword argument 'attn_factor' At: /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(190): parse_obj /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(143): __init__ /usr/local/Ascend/atb-models/atb_llm/models/qwen2/config_qwen2.py(33): __init__ /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(157): from_dict /usr/local/Ascend/atb-models/atb_llm/models/qwen2/router_qwen2.py(79): get_config /usr/local/Ascend/atb-models/atb_llm/models/base/router....

## 完整内容

问题使用MindIE启动DeepSeek-R1-0528-Qwen3-8B模型失败 系统欧拉 6.6.0-28.0.0.34.oe2403.aarch64 MindIE版本镜像 mindie:2.0.RC1-300I-Duo-py311-openeuler24.03-lts 说明mindie部署后启动DeepSeek-R1-Distill-Qwen-14B模型 或 Qwen3-14B均可正常启动并访问更换 DeepSeek-R1-0528-Qwen3-8B 模型就失败了 报错信息似乎是这个 ---- /root/mindie/log/debug/mindie-llm_427_20250704144051390.log [2025-07-04 14:40:57.435] [427] [281472955441120] [llm] [ERROR] [model_deploy_config.cpp:102] Failed to get vocab size from tokenizer wrapper with exception: TypeError: RopeScaling.__init__() got an unexpected keyword argument 'attn_factor' At: /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(190): parse_obj /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(143): __init__ /usr/local/Ascend/atb-models/atb_llm/models/qwen2/config_qwen2.py(33): __init__ /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(157): from_dict /usr/local/Ascend/atb-models/atb_llm/models/qwen2/router_qwen2.py(79): get_config /usr/local/Ascend/atb-models/atb_llm/models/base/router.py(111): config /usr/local/Ascend/atb-models/atb_llm/runner/tokenizer_wrapper.py(19): __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 [2025-07-04 14:40:57.435] [427] [281472955441120] [llm] [ERROR] [llm_manager_impl.cpp:784] Config manager init exception: TypeError: RopeScaling.__init__() got an unexpected keyword argument 'attn_factor' At: /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(190): parse_obj /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(143): __init__ /usr/local/Ascend/atb-models/atb_llm/models/qwen2/config_qwen2.py(33): __init__ /usr/local/Ascend/atb-models/atb_llm/models/base/config.py(157): from_dict /usr/local/Ascend/atb-models/atb_llm/models/qwen2/router_qwen2.py(79): get_config /usr/local/Ascend/atb-models/atb_llm/models/base/router.py(111): config /usr/local/Ascend/atb-models/atb_llm/runner/tokenizer_wrapper.py(19): __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 请问你的Qwen3-14B的运行都使用修改了哪些设置目前我的Qwen3-14B运行时报错 出错的意思好像是说atb-models/atb_llm/models/下没有qwen3的模型是否是权重路径config.json中的"model_type": "qwen3"改为qwen2或者其它 qwen3发布时间和MindIE2.0发布时间确认下

---

## 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**: 昇腾, AI, MindIE
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。*