300iDUO适配MiniCPM-V-2_6微调后大模型失败_MindIE_昇腾论坛

# 300iDUO适配MiniCPM-V-2_6微调后大模型失败_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-0270187946254662053-1-1.html

**生成时间**: 2025-08-27 10:33:58

---

## 问题描述

环境本身能够适配MiniCPM-V-2_6进行推理 模型在英伟达RTX3090下使用swift微调后再进行适配失败 运行mindieservice_daemon的错误信息如下 [WARNING] APP(6382,mindieservice_daemon):2025-07-15-14:28:44.941.093 [log_inner.cpp:79]6384 build/CMakeFiles/torch_npu.dir/compiler_depend.ts:get_default_custom_lib_path:118: "[PTA]:"config.ini is not exists"" [INFO] APP(6382,mindieservice_daemon):2025-07-15-14:28:44.941.202 [log_inner.cpp:82]6384 build/CMakeFiles/torch_npu.dir/compiler_depend.ts:GetDispatchTimeout:672: "[PTA]:"set dispatchTimeout_ 600 s."" [2025-07-15 14:28:46.347] [6382] [281473075437536] [llm] [ERROR] [model_deploy_config.cpp:102] Failed to get vocab size from tokenizer wrapper with exception: ValueError: Get instance from Minicpmqwen2v2Config failed. Please check the input parameters model_path and kwargs. If the input parameters are valid and the required files exist in model_path, make sure the folder's owner has execute permission. Otherwise, please check the function stack for detailed exception information. At: /usr/local/A...

## 完整内容

环境本身能够适配MiniCPM-V-2_6进行推理 模型在英伟达RTX3090下使用swift微调后再进行适配失败 运行mindieservice_daemon的错误信息如下 [WARNING] APP(6382,mindieservice_daemon):2025-07-15-14:28:44.941.093 [log_inner.cpp:79]6384 build/CMakeFiles/torch_npu.dir/compiler_depend.ts:get_default_custom_lib_path:118: "[PTA]:"config.ini is not exists"" [INFO] APP(6382,mindieservice_daemon):2025-07-15-14:28:44.941.202 [log_inner.cpp:82]6384 build/CMakeFiles/torch_npu.dir/compiler_depend.ts:GetDispatchTimeout:672: "[PTA]:"set dispatchTimeout_ 600 s."" [2025-07-15 14:28:46.347] [6382] [281473075437536] [llm] [ERROR] [model_deploy_config.cpp:102] Failed to get vocab size from tokenizer wrapper with exception: ValueError: Get instance from Minicpmqwen2v2Config failed. Please check the input parameters model_path and kwargs. If the input parameters are valid and the required files exist in model_path, make sure the folder's owner has execute permission. Otherwise, please check the function stack for detailed exception information. At: /usr/local/Ascend/atb-models/atb_llm/models/base/model_utils.py(312): wrapper /usr/local/Ascend/atb-models/atb_llm/models/minicpm_qwen2_v2/router_minicpm_qwen2_v2.py(41): get_config /usr/local/Ascend/atb-models/atb_llm/models/base/router.py(109): config /usr/local/Ascend/atb-models/atb_llm/models/minicpm_qwen2_v2/router_minicpm_qwen2_v2.py(20): __post_init__ <string>(23): __init__ /usr/local/Ascend/atb-models/atb_llm/models/__init__.py(110): get_model /usr/local/Ascend/atb-models/atb_llm/runner/tokenizer_wrapper.py(18): __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 [2025-07-15 14:28:46.347] [6382] [281473075437536] [llm] [ERROR] [llm_manager_impl.cpp:784] Config manager init exception: ValueError: Get instance from Minicpmqwen2v2Config failed. Please check the input parameters model_path and kwargs. If the input parameters are valid and the required files exist in model_path, make sure the folder's owner has execute permission. Otherwise, please check the function stack for detailed exception information. At: /usr/local/Ascend/atb-models/atb_llm/models/base/model_utils.py(312): wrapper /usr/local/Ascend/atb-models/atb_llm/models/minicpm_qwen2_v2/router_minicpm_qwen2_v2.py(41): get_config /usr/local/Ascend/atb-models/atb_llm/models/base/router.py(109): config /usr/local/Ascend/atb-models/atb_llm/models/minicpm_qwen2_v2/router_minicpm_qwen2_v2.py(20): __post_init__ <string>(23): __init__ /usr/local/Ascend/atb-models/atb_llm/models/__init__.py(110): get_model /usr/local/Ascend/atb-models/atb_llm/runner/tokenizer_wrapper.py(18): __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 LLMInferEngine failed to init LLMInferModels [2025-07-15 14:28:46.348+08:00] [6382] [6384] [server] [ERROR] [llm_infer_model_instance.cpp:284] : MIE05E040001[llm_backend] [LLMInferModelInstance::Init] llmManager_ init fail! [2025-07-15 14:28:46.348+08:00] [6382] [6384] [server] [ERROR] [llm_infer_model.cpp:28] : MIE05E040001[llm_backend] ModelInstance init failed 您好根据您的报错信息显示建议排查思路 1. 和模型适配的transformers 版本对不对 2. config.json 文件和模型权重路径和文件权限都修改为640 3. 模型权重文件下是不是缺少config.json 文件 4. 将trust_remote_code 参数修改为 ture 您好请问该问题是否已经解决3个工作日无回复该贴会自动关闭有新的问题可开新帖继续提问

---

## 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, NPU, 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。*