https://github.com/vvmnnnkv/pytorch-wheels
Pytorch and torchvision wheels for ARM architecture. With docker!
https://github.com/vvmnnnkv/pytorch-wheels
arm armv7 docker pytorch raspberry-pi torchvision
Last synced: about 2 months ago
JSON representation
Pytorch and torchvision wheels for ARM architecture. With docker!
- Host: GitHub
- URL: https://github.com/vvmnnnkv/pytorch-wheels
- Owner: vvmnnnkv
- License: apache-2.0
- Created: 2020-07-08T12:42:09.000Z (almost 6 years ago)
- Default Branch: master
- Last Pushed: 2020-07-08T15:32:56.000Z (almost 6 years ago)
- Last Synced: 2025-01-29T04:36:10.143Z (over 1 year ago)
- Topics: arm, armv7, docker, pytorch, raspberry-pi, torchvision
- Language: Dockerfile
- Homepage:
- Size: 31.9 MB
- Stars: 3
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Compile PyTorch for ARM architecture
Docker has wonderful [buildx plugin](https://docs.docker.com/buildx/working-with-buildx/)
that can build images for multiple platforms using QEMU emulator.
Provided `Dockerfile` can be used to build PyTorch 1.4.1 and Torchvision 0.5.0 for Raspberri Pi on a x64 machine.
It builds for ARMv7 (32-bit) but in theory can be extended to other architectures supported by buildx.
While it is possible to build packages on Raspberri Pi itself, it might take whole night or so.
In contrast building on a decent x64 machine in emulator finishes within 2 hours.
Usage example:
```
docker buildx build --platform=linux/arm/7 -t ptwheels .
# Copy wheel files out of image
docker create -name ptwheels ptwheels echo
docker cp ptwheels:/wheels .
```
These wheels have been tested on Raspberri Pi 3 Model B v1.2 with Raspberry Pi OS (32-bit),
together with [PySyft](https://github.com/openmined/pysyft) library.