frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

quantized model export

David Fan
2021-12-09 13:29:09 -08:00
parent b02476b8f7
commit a41847567c
1 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
PyTorch-ONNX-exporter.md

@ -15,6 +15,8 @@ Documentation for developing the PyTorch-ONNX exporter (`torch.onnx`).
* [Adding tests](#adding-tests)
* [Links](#links)
* [Relevant parts of PyTorch repo](#relevant-parts-of-pytorch-repo)
* [Features](#features)
* [Quantized model export](#quantized-model-export)
# Development process
@ -151,3 +153,10 @@ An example of adding unit tests for a new symbolic function: [Add binary_cross_e
* More Python tests: [test/jit/test_onnx_export.py](https://github.com/pytorch/pytorch/tree/onnx_ms_1/test/jit/test_onnx_export.py)
* Python code: [torch/onnx/](https://github.com/pytorch/pytorch/tree/onnx_ms_1/torch/onnx)
* C++ code: [torch/csrc/jit/passes/onnx/](https://github.com/pytorch/pytorch/tree/onnx_ms_1/torch/csrc/jit/passes/onnx)
# Features
## Quantized model export
To support quantized model export, we need to unpack the quantized tensor inputs and the PackedParam weights (https://github.com/pytorch/pytorch/pull/69232). We construct through `TupleConstruct` to have a 1-to-1 input mapping,
so that we can use `replaceAllUsesWith` API for its successors. In addition, we support quantized namespace export, and the developers can add more symbolics for quantized operators conveniently in the current framework.