pytorch-aarch64
PyTorch, torchvision, torchaudio and torchtext wheels (whl) and docker image for aarch64 / ARMv8 / ARM64 devices
中文版 (for Gitee) | GitHub | Web
Install
Run:
pip install torch -f https://torch.maku.ml/whl/stable.html
Add torchvision
, torchaudio
, torchtext
and other packages if you need.
Note: this command installs the latest version. For choosing a specific version, please check the Custom Builds section.
To pick the whl
files manually, check the releases.
Docker
docker run -it kumatea/pytorch
To pull the image, run docker pull kumatea/pytorch
.
To check all available tags, click here.
FastAI
FastAI is a great open-source high-level deep learning framework based on PyTorch.
It recommends installing by conda
, but there is no Anaconda builds for aarch64
.
So, install fastai
by:
pip install fastai -f https://torch.maku.ml/whl/stable.html
torch
and torchvision
will be installed as dependencies automatically.
Similarly, fastbook
could be installed by:
pip install fastbook -f https://torch.maku.ml/whl/stable.html
Custom Builds
torch |
torchvision |
torchaudio |
torchtext |
Status | python |
---|---|---|---|---|---|
master nightly |
master nightly |
master nightly |
master nightly |
>=3.6 |
|
1.7.1 |
0.8.2 |
0.7.2 |
0.8.1 |
>=3.6 |
|
1.7.0 |
0.8.1 0.8.0 |
0.7.0 |
0.8.0 |
>=3.6 |
|
1.6.0 [i] |
0.7.0 |
0.6.0 |
0.7.0 |
>=3.6 |
|
1.5.1 |
0.6.1 |
0.5.0 |
0.6.0 |
>=3.5 |
|
1.5.0 |
0.6.0 |
0.5.0 |
0.6.0 |
>=3.5 |
|
1.4.1 1.4.0 |
0.5.0 |
0.4.0 |
0.5.0 |
==2.7 , >=3.5 , <=3.8 |
|
1.3.1 |
0.4.2 |
==2.7 , >=3.5 , <=3.7 |
|||
1.3.0 |
0.4.1 |
==2.7 , >=3.5 , <=3.7 |
|||
1.2.0 |
0.4.0 |
==2.7 , >=3.5 , <=3.7 |
|||
1.1.0 |
0.3.0 |
==2.7 , >=3.5 , <=3.7 |
|||
<=1.0.1 |
0.2.2 |
==2.7 , >=3.5 , <=3.7 |
Corresponding Versions
- Corresponding
torch
andtorchvision
versions - Corresponding
torch
andtorchaudio
versions - Corresponding
torch
andtorchtext
versions
Official PyTorch CI Builds
You might not be able to see the statuses.
ZUUL /
openlabtesting
uses HTTP API to fetch its CI statuses, but GitHub Page are forced HTTPS.If so, you will need to visit this page via GitHub.
py ver |
3.6 | 3.7 | 3.8 | 3.9 |
---|---|---|---|---|
master | ||||
1.8.0 | ||||
1.7.1 | ||||
1.7.0 | ||||
1.6.0 | ||||
1.5.0 | ||||
1.4.0 |
More Info
FAQ
- Q: Does this run on Raspberry Pi?
A: Yes, if the architecture of the SoC isaarch64
. It should run on all ARMv8 chips.
- Q: Does this support CUDA / CUDNN?
A: No. Check here for more information.
- Q: Does this run on Nvidia Jetson?
A: Yes, but extremely slow. Each Nvidia Jetson boards contains an Nvidia GPU, but this project only build cpu wheels. To better make use of your hardware, build it yourself.
About PyTorch v1.6.0
A fatal bug is encountered and this patch is applied while building PyTorch v1.6.0. The patch has been merged into mainstream in later versions.
About TorchText
The latest torchtext
version you can install from PyPI (pip
) is 0.6.0
, namely torchtext-0.6.0-py3-none-any.whl
,
is the last official version that is available for all devices.
From that on, the newer version are only built for x86_64
/ amd64
, like other projects.
However, for torchtext
, version 0.6.0
is deprecated but significantly more recent than the dropped versions of other projects.
CUDA / CUDNN Support
Since the building environment (as below) does not contain an Nvidia GPU, the wheels could not be built with cuda support.
If you need it, please use an Nvidia Jetson board to run the building code.
Building Environment
SoC: Allwinner H6 (quad-core A53)
Architecture: ARMv8 / ARM64 /
aarch64
OS: Debian Buster
GCC: v8.3.0
Virtualization: Docker
Performance
Test date: 2020-12-25
Less execution time is better
Platform | Specs | Task | Avg. Time | Version |
---|---|---|---|---|
aarch64 |
ARM Cortex-A53 | Image Prediction | 41,264.514 ms | 1.7.1 / 3.8.5 |
aarch64 |
QUALCOMM Snapdragon 845 | Image Prediction | 9,763.317 ms | 1.7.1 / 3.8.5 |
amd64 |
INTEL Core i7-6500U | Image Prediction | 374.274 ms | 1.7.1+cpu / 3.8.6 |
Google Colab | INTEL Xeon ??? / NVIDIA Tesla T4 | Image Prediction | 314.650 ms | 1.7.1+cu101 / 3.6.9 |
Kaggle | INTEL Xeon ??? / NVIDIA Tesla P100 | Image Prediction | 307.503 ms | 1.7.1+cu110 / 3.7.6 |
Note:
- This test was done by using a same Cat or Dog model, to predict 10 random animal images (while same for each group).
- The latest version of PyTorch was manually installed on all platforms, but driver and Python remained default.