Environment configuration

Using GPUs will make things run faster.

• We can access the server by using SSH: ssh -L 8888:localhost:8888 [your_x_account]@mltgpu.flov.gu.se -p 62266

ssh tells the computer to connect remotely to the server.

-L 8888:localhost:8888 allows us to connect using jupyter notebooks, remove it if you don't want to do that.

-p 62266 tells the server to give your access through port 62266.

• Pay attention to PyTorch-GPU compatibility!

To set up a virtual environment on the server is suggested:

pip install virtualenv

python -m venv nlp # Create a virtual environment folder nlp under the current path

conda activate nlp # activate the virtual environment

pip install numpy jupyter # install necessary packages in your own virtual env

pip install torch torchvision torchaudio # install pytorch & cuda (make sure compatibility, avoid calling from external env)

conda deactivate # exit virtual env

• Check memory and current jobs in jupyter notebook:

!nvidia-smi

!ps -aux | grep python