We provide our own OS image, TensorML 18.04, for machine learning.
It comes preinstalled with the most popular Python libraries: Jupyter Notebook, Jupyter Lab, TensorFlow, Keras, PyTorch, MxNet, Numpy, Pandas, NVIDIA drivers, CUDA, cuDNN, CUDA Toolkit, Docker, and NVIDIA Docker2.
To get started:
- 1.SSH into your server with the IP and username in the "Metadata section"
jupyter lab --ip=0.0.0.0for a Jupyter Lab or
jupyter notebook --ip=0.0.0.0for a Jupyter Notebook
- 3.Connect to your Jupyter Lab or Notebook by typing the IP:8888 in your browser. You'll need to copy the token shown in your SSH session to login.
- 4.You should see a default Jupyter notebook. Run it. Reboot if there is an error doing so. If NVIDIA drivers update to a new version during the first boot, you'll have to reboot to have them working properly.
Note: If you receive a "successful NUMA node read"... verbose message, you can silence it by running:
for a in /sys/bus/pci/devices/*; do echo 0 | sudo tee -a $a/numa_node; done
That's it! You're all set to immediately start ML-related work!