C Sample Usage Guide【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/geDirectory Structure├── CMakeLists.txt cmake configuration file ├── README.md Sample usage guide ├── config │ ├── model_generator.py Script to generate models required for test cases │ ├── add_func_multi.json Configuration file for multi-func used in test cases │ ├── add_func_multi_control.json Configuration file for multi-func used in test cases │ ├── add_func.json Configuration file for add FunctionPp used in test cases │ ├── add_graph.json Configuration file for add GraphPp used in test cases │ ├── data_flow_deploy_info.json Configuration file to specify node deployment location in test cases │ └── invoke_func.json Configuration file for udf calling nn used in test cases ├── node_builder.h Common methods to construct FunctionPp and GraphPp ├── sample_base.cpp This sample demonstrates basic DataFlow API graph building, including construction and execution of UDF, GraphPp, and UDF executing NN inference types of nodes ├── sample_timebatch.cpp This sample demonstrates TimeBatch usage method ├── sample_countbatch.cpp This sample demonstrates CountBatch usage method ├── sample_tensorflow.cpp This sample demonstrates TensorFlow graph construction DataFlow node method ├── sample_multifunc.cpp This sample demonstrates multi-func calling method ├── sample_exception.cpp This sample demonstrates enabling exception reporting method └── sample_perf.cpp This sample tests Feed and Fetch interface performanceEnvironment RequirementsReference Environment Preparation to download and install driver/firmware/CANN software packages;Model generation script model_generator.py in config directory depends on tensorflow, need to install through pip3 install tensorflow.Program Compilation# Execute tensorflow original model generation script in config directory: python3 config/model_generator.py # After execution completes, generate add.pb model in config directory source {HOME}/Ascend/cann/set_env.sh # {HOME}/Ascend is CANN software package installation directory, replace according to actual installation path. mkdir build cd build cmake .. make -j 64 cd ../outputRun Samplesnuma_config.json file configuration reference in the following text: numa_config field description and sample# Optional export ASCEND_GLOBAL_LOG_LEVEL3 #0 debug 1 info 2 warn 3 error Default error level if not set # Required source {HOME}/Ascend/cann/set_env.sh # {HOME}/Ascend is CANN software package installation directory, replace according to actual installation path. export RESOURCE_CONFIG_PATHxxx/xxx/xxx/numa_config.json ./sample_base ./sample_timebatch ./sample_countbatch ./sample_tensorflow ./sample_multifunc ./sample_exception ./sample_perf # unset this environment variable to prevent affecting non-dflow test cases unset RESOURCE_CONFIG_PATH【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