An optional docker container creation for each.

[Anindyadeep]

Every benchmark is right now structured like this:

• venv/
• benchmark.py
• bench.sh
• setup.sh
• requirements.txt

However, there are certain repos, which requires different cuda versions or different version of python etc etc. So just venv creation might not be enough. We should have a docker container for each or some of them (initially) so that running one benchmark (which heavily depends on some system packages or drivers) should not harm the running of other benchmarks.

This has some added benefits too, as proposed on issue #102. Where inference endpoints can come from docker containers and it can be useful for K8s deployment (correct me here if I am inaccurate here 🙂 )

[Anindyadeep]

cc: @nsosio @filopedraz