The easiest option to go on the GCP console and go on AI Platform. Here in the Notebook instances
section click on New instances
and select PyTorch X.X
and Without GCPU
so you have conda installed. Soon all images should be installed with Anaconda.
The second option with require more installation is explain in this separate Github repository: https://github.com/tarrade/proj_custom_ai_platform_notebook. Here you have you will create a custom image of the AI Platform notebook and deploy it on GCP using some automatic pipeline.
When you have your Jupyter Lab instance running locally you will need to install conda environment to have all python packages needed to run the code as describe in this link. Then you will be ready to train your model.