Sagemaker Local mode with cached pip installs

[danielfoguelman]

Describe the feature you'd like
In local mode I would like to be able to cache the pip installs made for preprocess, train and evaluate steps. This would make local development faster as I wouldn't need to re install torch for instance that takes a long time.

How would this feature be used? Please describe.

In local development we need the containers to have dependencies installed, docker-compose caches steps if they are in the compose yml file. If they are not, then the installation gets redone every time.

Describe alternatives you've considered

Provide examples on how to add a requirements file that get's cached in docker containers between local executions.

Additional context
Add any other context or screenshots about the feature request here.

[ArlindNocaj]

I think we could just allow to add a set of parameters to the created docker-compose file, that way we can add caching by adding the right folders into the container.