Accelerate PyTorch models with ONNX Runtime
Project description
The torch-ort packages uses the PyTorch APIs to accelerate PyTorch models using ONNX Runtime.
Dependencies
The torch-ort package depends on the onnxruntime-training package, which depends on specific versions of GPU libraries such as NVIDIA CUDA.
The default command pip install torch-ort
installs the onnxruntime-training version that depends on CUDA 10.2.
If you have a different version of CUDA installed, you can install a different version of onnxruntime-training explicitly:
- CUDA 11.1
pip install onnxruntime-training -f https://download.onnxruntime.ai/onnxruntime_stable_cu111.html
Post-installation step
Once torch-ort is installed, there is a post-installation step:
python -m torch_ort.configure
If this step fails, it is likely due to GPU library version mismatch between onnxruntime-training and your installation. You can check the version of onnxruntime-training by running pip list
. For example:
onnxruntime-training 1.9.0+cu111
Releases
-
1.9.0
Release Notes : https://github.com/pytorch/ort/releases/tag/v1.9.0
-
1.8.1
Release Notes : https://github.com/pytorch/ort/releases/tag/v1.8.1
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distributions
Built Distribution
Hashes for torch_ort-1.17.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 2aa06f0ae7e0a13a5d9b839b89f60e89473cbca4ab6d72fee97c3e91587daed8 |
|
MD5 | 0167c8993d83d2dec057b2b511117a3b |
|
BLAKE2b-256 | 86e1e7126e9e967183b25fc2accc20512a4710ffa85645527b0bb298608b15b4 |