Fusion Pass Examples
【免费下载链接】geGE(Graph 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
This directory provides examples for implementing custom fusion passes by inheriting GE-provided classes and overriding their methods:
| Example | Example Link |
|---|---|
| MatMul+Add fused to GEMM custom pass example | README |
| Move ReLU after Concat before Concat custom pass example | README |
| Move ReLU after Concat before Concat custom pass example (Python version) | README |
| Modify Conv operator data format custom pass example | README |
| Modify Conv operator data format custom pass example (Python version) | README |
Development Guide
If developing pattern-based fusion passes, recommended reading:
- Fusion Pattern Pass Mechanism
- Python Fusion Pass Development Guide
- C++ Fusion Pass Development Guide
【免费下载链接】geGE(Graph 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