GE (Graph Engine)【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/geLatest News[2026/03] Continuously enhanced graph optimization capabilities, improved unit test coverage, and fixed compilation issues in multiple edge cases. Optimized documentation structure to enhance developer experience.[2026/02] Optimized auto-fusion and broadcast scenario support, added BF16 data type support. Enhanced HostCPU engine capabilities, optimized session creation and destruction lock mechanism. Fixed bugs in external weight and thread loading scenarios. Supported Reduce axis splitting Store address conflict penalty to improve operator fusion effects.[2026/01] GE project first release, open-sourced graph compiler and executor, supporting PyTorch and TensorFlow frontend integration and ONNX and PB model format parsing and compilation.OverviewGE (Graph Engine) is a graph compiler and executor for Ascend, providing computation graph optimization, multi-stream parallelism, memory reuse, and model sinking technologies to accelerate model execution efficiency and reduce model memory footprint.GE provides friendly integration capabilities for PyTorch and TensorFlow frontends, and simultaneously supports parsing and compilation of mainstream model formats such as onnx and pb. See Ascend Community Documentation - Graph Mode Development Guide.Quick StartTo quickly experience GEs working methods and basic development process, refer to the following documentation:Build Verification: Introduces the complete build process of components and test case execution flow.Quick Start: Using ResNet50 model as an example, introduces how to use the ATC tool for model conversion and execute inference on Ascend AI processors.Quick Start-LLM: Using Qwen model as an example, introduces how to use the ATC tool for LLM model conversion and implement LLM model loading, execution, and result retrieval.DocumentationTo learn how to use GE for model compilation and execution, refer to the Graph Mode Development Guide, technical articles, and other content: GE Reference MaterialsTo gain a deep understanding of GEs internal design, architecture mechanisms, and development processes, refer to the following documentation:GE Architecture Documentation: Introduces core components, execution flows, optimization mechanisms, and other internal principles.Contributing Guide: Explains how to submit Issues, Pull Requests, and code standards.AI Agent Support: Introduces some default skills used in the repository and using agents to assist the development process.Ecosystem IntegrationThe following projects have integrated GE as an inference or graph mode backend:TorchAir: Integrates GE into PyTorch graph mode. LinkTFA (TensorFlow Adapter): Uses GE as TensorFlow backend. LinkJittorInfer: Large model C inference framework based on Ascend chips. LinkTriton GE Backend: GEs Triton Inference Server backend. LinkNote:The above list shows known and publicly disclosed integration cases,not a complete list. If you are using GE, welcome to supplement through Issues or Pull Requests, and we will continuously update relevant information.Other InformationSecurity StatementLicenseContact Us【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