python3.6-torch-1.9.0

Torch 1.9.0 for Python 3.6 running on the Nvidia Jetson Tegra. Torch (a.k.a. pytorch) is an optimised tensor library for GPUs and CPUs.

You may also want to check out Qengineering/PyTorch-Jetson-Nano which also provides a wheel for torch.

How to use

If you have pip (which should be run in a suitable virtual environment) this should work. Note you should probably install numpy as shown here:

pip install 'https://github.com/jetson-nano-wheels/python3.6-numpy-1.19.4/releases/download/v0.0.2/numpy-1.19.4-cp36-cp36m-linux_aarch64.whl'
wget \
    'https://nvidia.box.com/shared/static/h1z9sw4bb1ybi0rm3tu8qdj8hs05ljbm.whl' \
    -O /tmp/torch-1.9.0-cp36-cp36m-linux_aarch64.whl \
    && pip install /tmp/torch-1.9.0-cp36-cp36m-linux_aarch64.whl

If the wheel fails to download, or if Nvidia remove it, you may want to search for it online. There seems to be a few copies floating around.

How to contribute

1. Fork this repo
2. Add a branch for your new feature.
3. Run the init.sh script.

Like this:

git clone git@github.com:jetson-nano-wheels/python3.6-torch-1.9.0
cd python3.6-torch-1.9.0
git checkout -b feature-my-fancy-addition
./init.sh

After that, do your edits, commit and push to your repo, and send a pull-request if you like 😃