Graph Engine Personal Developer Toolchain 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/geGE Developer Toolchain is an automated script toolchain in graph engine designed for individual developers.Currently supports a series of common developer functions including container-based development environment preparation, automatic download/installation/configuration of build dependencies, code formatting, compilation, testing, code coverage checking, document generation, etc.PrerequisitesBelow are the manual prerequisites needed for using GE developer toolchain;The following are verified best performance recommended configurations:Operating System, choose one of the following:Native Linux operating system, such as ubuntu;Windows operating system, recommended to install WSL ubuntu system, strongly suggest downloading code directly inside WSL, do not mount volumes (poor build performance)!MAC OS;Docker installation:Docker installed successfully, and relevant mirror sources configured correctly, can normally download external images.OS-supported command line tools: Native Linux or WSL shell;Available but not recommended configurations:Install docker directly in windows, use Linux-like bash (Cygwin, minGW etc.) to execute GE toolchain; (Using this method can also execute all GE toolchain operations, but due to Windows and container heterogeneous kernel file access limitations, build speed will be relatively slow)Quick StartGE toolchain corresponding scripts are under scripts, can execute according to following flow:Enter scripts directory:$ cd ./scriptsExecutege envto automatically download container environment, and login to environment$ ./ge.sh envDownload and install external libraries required for building$ ge update(Note: After entering container,gecommand already automatically registered into system, so container内 doesnt need to write script full name)Execute testing, default executes unit test cases,ge testwill automatically trigger build$ ge testDetailed UsageUnder scripts directory, run ./ge.sh -h to view all subcommand集合.$ ./ge.sh -h Usage: ge COMMANDS Run ge commands Commands: env Prepare docker env config Config dependencies server update Update dependencies format Format code lint Static verify build Build code test Run test of UT/ST cov Run Coverage docs Generate documents clean CleanEach subcommand built into script represents an independent function; each subcommand also provides secondary parameters for flexible specification of execution方式.Each subcommand can view supported configurable parameters through-h.For example queryingenvsubcommand parameters, can use following command:$ ./ge.sh env -hEach subcommand has a default behavior when不带参数.ge envThis command用于 prepares container environment used for building and testing, specifically contains following parameters:$ ./ge.sh env -h Usage: ge env [OPTIONS] Prepare for docker env for build and test Options: -b, --build Build docker image -p, --pull Pull docker image -e, --enter Enter container -r, --reset Reset container -h, --helpParameter detailed explanation:-b -- build: Generate required running container image based on scripts/env/Dockerfile;-p -- pull: Pull required container image from locally configured container central repository;-e -- enter: Login to container runtime environment前提 existing local container image;-r -- reset: Delete local running container image environment;Default: Pull corresponding container image from central container repository, run instance and login.ge configConfigure external dependency server, specific parameters如下:$ ge config -h Usage: ge config [OPTIONS] update server config for ge, you need input all config info (ip, user, password) Options: -i, --ip Config ip config -u, --user Config user name -p, --password Config password -h, --help Example: ge config -iip-adress -uusername -ppassword (Need add escape character \ before special charactor $、#、!)Parameter detailed explanation:-i, --ip: Configure dependency library server IP address;-u, --usr: Configure dependency library server username;-p, --password: Configure dependency library address;Default: Print help information.ge updateInstall external dependency libraries required for graph engine building, specific parameters如下:$ ge update -h Usage: ge update [OPTIONS] update dependencies of build and test Options: -p, --public Download dependencies from community -d, --download Download dependencies -i, --install Install dependencies -c, --clear Clear dependencies -h, --helpParameter detailed explanation:-p, --public: Download and install dependency libraries from community;-d, --download: Download external dependency libraries needed for building;-i, --install: Install external dependency packages to corresponding locations;-c, --clear: Clear downloaded external dependency packages;Default: Download external dependency libraries according to scripts/update/deps_config.sh configuration and install to corresponding directories. (Note: Please ensure server address, username, password in scripts/update/server_config.sh already configured)ge formatUse clang-format for code formatting, specific parameters如下:$ ge format -h Options: -a format of all files -c format of the files changed compared to last commit, default case -l format of the files changed in last commit -h Print usageParameter detailed explanation:-a: Format all code;-c: Only format code modified this time;-l: Format code from last commit;Default: Format code modified this time.ge buildExecute build (Note: Calls original build.sh script, under modification...);ge testBuild and run test cases, currently supports following parameters:$ ge test -h Usage: ge test [OPTIONS] Options: -u, --unit Run unit Test -c, --component Run component Test -h, --helpParameter detailed explanation:-u, --unit: Execute unit testing-c, --component: Execute component testingDefault: Execute unit testing.ge covExecute code coverage checking, supports full coverage and increment coverage functions, this command需要 already ran test cases, currently supports following parameters:$ ge cov -h Usage: ge cov [OPTIONS] Options: -a, --all Full coverage -i, --increment Increment coverage -d, --directory Coverage of directory -h, --helpParameter detailed explanation:-a, --all: Execute full coverage rate statistics;-i, --increment: Execute increment coverage checking, default analyzes uncommitted modified code coverage (if new git untracked files exist, need先 git add to add in);-d, --directory: Code path for increment coverage checking, supports passing path parameters;Default: Execute increment coverage checking;Below command demonstrates how to check all codes increment coverage under ge directory:$ ge cov -dgege docsDoxygen document generation, includes code logic and physical structure and relationships, convenient for reading and understanding code; currently supports following parameters:$ ge docs -h Usage: ge docs [OPTIONS] Options: -b, --brief Build brief docs -a, --all Build all docs -h, --helpParameter detailed explanation:-b, --brief: Generate brief documents, ignore部分 relationship diagram generation, fast speed;-a, --all: Generate full documents, includes various code relationship diagrams, relatively slow speed;Default: Generate full code documents.ge cleanClean various downloaded or generated intermediate files, currently supported parameters如下:$ ge clean -h Usage: ge clean [OPTIONS] Options: -b, --build Clean build dir -d, --docs Clean generate docs -i, --install Clean dependenices -a, --all Clean all -h, --helpParameter detailed explanation:-b, --build: Clean generated compilation build temporary files;-d, --docs: Clean generated document temporary files;-i, --install: Clean installed dependency files;-a, --all: Clean all downloaded and generated temporary files;Default: Clean temporary files generated by compilation build.Follow usToolchain functionality还在 continuously improving, please submit issue if problems, thank you!【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