系统报错nca load model from file failed, ret 254.【免费下载链接】geGEGraph Engine是面向昇腾的图编译器和执行器提供了计算图优化、多流并行、内存复用和模型下沉等技术手段加速模型执行效率减少模型内存占用。 GE 提供对 PyTorch、TensorFlow 前端的友好接入能力并同时支持 onnx、pb 等主流模型格式的解析与编译。项目地址: https://gitcode.com/cann/ge问题现象量化过程中如果出现如下日志表示可能没有获取到分组信息。[INFO][ncx_interface.cpp:35] ncx lib has been init [INFO][ncx_interface.cpp:35] ncx lib has been init [ERROR][ncx_interface.cpp:208] nca load model from file failed, ret 254. please check detail info from device log.根据上述ERROR信息查看“/home/mdc/var/log/debug/TOOLS_*.log”进一步定位如果出现如下信息可以确定该错误原因是NCS没有获取到分组信息而创建分组失败。[INFO][ncx_interface.cpp:35] ncx lib has been init [INFO][ncx_interface.cpp:35] ncx lib has been init [ERROR][ncx_interface.cpp:208] nca load model from file failed, ret 254. please check detail info from device log. [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[sample_process.cpp:378]NCX load model failed. [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[sample_process.cpp:442]Ncx execute failed. [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[sample_process.cpp:297]Ncx model infer failed. [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[quantize_api.cpp:268]sample process failed [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[quantize_api.cpp:314]Calibration Inference failed! [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[quantize_api.cpp:439]Do Calibration failed. [ERROR][ncx_interface.cpp:168] nca close session: amct ncx lib has not been init [ERROR][ncx_interface.cpp:135] nca finalize: amct ncx lib has not been init [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[sample_process.cpp:518]NCX finalize failed. [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[sample_process.cpp:460]NCX destroy resource failed. [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[quantize_api.cpp:184]sample process destroy resource failed! [ERROR] AMCT(1372344,acl_test_tool):2025-09-12-08:26:47[inner_graph_calibration.cpp:73]Failed to finalize CalibrationInitializer.解决方案确保已经创建设置分组创建分组的方法为调用DCMI提供的dcmi_create_capability_group接口创建分组。【免费下载链接】geGEGraph Engine是面向昇腾的图编译器和执行器提供了计算图优化、多流并行、内存复用和模型下沉等技术手段加速模型执行效率减少模型内存占用。 GE 提供对 PyTorch、TensorFlow 前端的友好接入能力并同时支持 onnx、pb 等主流模型格式的解析与编译。项目地址: https://gitcode.com/cann/ge创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考