Install Pytorch

On AGX

Source: https://forums.developer.nvidia.com/t/pytorch-for-jetson-nano-version-1-4-0-now-available/72048

note

I followed this page to install and my Jetpack version is v4.3

I am going to install PyTorch v1.3.0

Commands

1. Download the whl file.

   wget https://nvidia.box.com/shared/static/phqe92v26cbhqjohwtvxorrwnmrnfx1o.whl -O torch-1.3.0-cp36-cp36m-linux_aarch64.whl

2. Install this whl file

   sudo -H python3 -m pip install numpy torch-1.3.0-cp36-cp36m-linux_aarch64.whl

3. Install torchvision; so git clone it for my vision.

   They provided us a list.

   PyTorch v1.0 - torchvision v0.2.2
   PyTorch v1.1 - torchvision v0.3.0
   PyTorch v1.2 - torchvision v0.4.0
   PyTorch v1.3 - torchvision v0.4.2
   PyTorch v1.4 - torchvision v0.5.0

   So we can know that the version should be installed by v0.4.2

   sudo apt-get install libjpeg-dev zlib1g-dev
   git clone --branch v0.4.2 https://github.com/pytorch/vision torchvision
   cd torchvision

4. Install it.

   sudo -H python3 setup.py install

5. Install pillow

   sudo -H python3 -m pip install 'pillow<7'

Check

```
>>> import torch
>>> print(torch.__version__)
>>> print('CUDA available: ' + str(torch.cuda.is_available()))
>>> print('cuDNN version: ' + str(torch.backends.cudnn.version()))
>>> a = torch.cuda.FloatTensor(2).zero_()
>>> print('Tensor a = ' + str(a))
>>> b = torch.randn(2).cuda()
>>> print('Tensor b = ' + str(b))
>>> c = a + b
>>> print('Tensor c = ' + str(c))
>>> import torchvision
>>> print(torchvision.__version__)
```

---

On Desktop

We can directly follow the commands from Pytorch official website.

https://pytorch.org/get-started/locally/#mac-installation

Installation

Environmnet setting:

OS: Ubuntu 18.04 CDUA: 10.0 cuDNN: 7.6. Version table:

torch	torchvision
1.5.0	0.6.0
1.4.0	0.5.0
1.3.1	0.4.2

Install torch and torchvision:

python3 -m pip install torch==1.4.0 torchvision==0.5.0 -f https://download.pytorch.org/whl/torch_stable.html  

Note : you can also decide the specific cuda version and choose to install the cpu or gpu version. Please follow the instructions of official website.

For example, my CUDA version is 10.0.

Command:

pip3 install torch==1.4.0+cu100 torchvision==0.5.0+cu100 -f https://download.pytorch.org/whl/torch_stable.html

Uninstall

sudo -H python3 -m pip uninstall torch torchvision 

Check

Use python3 to check. The code from here.

from __future__ import print_function
import torch
x = torch.rand(5, 3)
print(x)

Check the gpu whether it works or not.

import torch
torch.cuda.is_available()

Should be return True