Updated PyTorch ONNX exporter (markdown)

Gary Miguel
2022-05-17 12:53:06 -07:00
parent b2f9618de2
commit 712b94e705

@ -4,24 +4,22 @@ Documentation for developing the PyTorch-ONNX exporter (`torch.onnx`).
<!-- TOC generated with https://github.com/ekalinin/github-markdown-toc --> <!-- TOC generated with https://github.com/ekalinin/github-markdown-toc -->
- [Table of Contents](#table-of-contents) * [Development process](#development-process)
- [Development process](#development-process) * [Environment setup](#environment-setup)
- [Environment setup](#environment-setup) * [Fork PyTorch](#fork-pytorch)
- [Fork PyTorch](#fork-pytorch) * [Build PyTorch](#build-pytorch)
- [Build PyTorch](#build-pytorch) * [Optional build tips](#optional-build-tips)
- [Optional build tips](#optional-build-tips) * [Install additional dependencies](#install-additional-dependencies)
- [Install additional dependencies](#install-additional-dependencies) * [ONNX and ONNX Runtime](#onnx-and-onnx-runtime)
- [ONNX Runtime](#onnx-runtime) * [TorchVision](#torchvision)
- [ONNX](#onnx) * [Sanity check](#sanity-check)
- [TorchVision](#torchvision) * [VS Code](#vs-code)
- [Sanity check](#sanity-check) * [Pull requests](#pull-requests)
- [VS Code](#vs-code) * [Tests](#tests)
- [Pull requests](#pull-requests) * [Links](#links)
- [Tests](#tests) * [Relevant parts of PyTorch repo](#relevant-parts-of-pytorch-repo)
- [Links](#links) * [Features](#features)
- [Relevant parts of PyTorch repo](#relevant-parts-of-pytorch-repo) * [Quantized model export](#quantized-model-export)
- [Features](#features)
- [Quantized model export](#quantized-model-export)
# Development process # Development process
@ -101,7 +99,7 @@ The ONNX tests depend on torchvision.
This is tricky because TorchVision depends on PyTorch, but we don't want our This is tricky because TorchVision depends on PyTorch, but we don't want our
package manager to install PyTorch, we want to use our locally built one. package manager to install PyTorch, we want to use our locally built one.
The best solution I've found is to install torchvision with pip (so that The best solution I've found is to install torchvision with pip (so that
conda doesn't try to manage it) without any deps (so that pip doesn't install conda doesn't try to manage it) without any deps (so that pip doesn't install
pytorch and the locally built version is used). pytorch and the locally built version is used).
```sh ```sh