saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-30 20:18:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

eager: fix build against latest PyTorch master (#8745)

Browse source 
Improve README as well.
This commit is contained in:
Aaron Bockover 2021-08-18 14:27:21 -04:00 committed by GitHub
parent cb67fca738
commit b2813656f5
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 14 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
orttraining/orttraining/eager/README.md
View file

@ -1,11 +1,17 @@
# ONNX Runtime Eager Mode Support for PyTorch
## Build Instrcutions
* Until the ONNX Runtime Eager Mode backend is merged in PyTorch upstream ([PR #58248](https://github.com/pytorch/pytorch/pull/58248)), build PyTorch from [abock/pytorch](https://github.com/abock/pytorch/tree/dev/abock/ort-backend) (branch: `dev/abock/ort-backend`):
* Build Pytorch with commit: 0834a368181d18c8fd2429614fafca50fb412ce1
* Install the pytorch build to your current environment
* Build onnxruntime, make sure you enable training and python build together with eager mode:
```bash
$ git clone https://github.com/abock/pytorch -b dev/abock/ort-backend --recursive
$ cd pytorch
$ python setup.py develop --user
```
```bash
./build.sh --enable_training --enalbe_pybind --build_eager_mode
```
* If not build as noted above, ensure the eager-mode enabled PyTorch build is installed in the current environment.
* Build ONNX Runtime, making sure you enable training and python build together with eager mode:
```bash
./build.sh --cmake_generator=Ninja --enable_training --enable_pybind --build_eager_mode
```

5
orttraining/orttraining/eager/ort_tensor.h
View file

@ -13,10 +13,7 @@ class ORTTensorImpl final : public c10::TensorImpl {
public:
explicit ORTTensorImpl(OrtValue tensor, const at::TensorOptions& options)
: c10::TensorImpl(
c10::DispatchKeySet {
c10::DispatchKey::ORT,
c10::DispatchKey::AutogradORT
},
c10::DispatchKeySet{c10::DispatchKey::ORT},
options.dtype(),
options.device()) {
set_tensor(tensor);