Hi, I am trying to add one model to sotabench. However, our repo (https://github.com/google-research/big_transfer) is a little non-standard because we have three implementations, each in their subfolder. Hence, a top-level
requirements.txt does not make sense in our case.
I thought of installing the requirements in the
sotabench_setup.sh file instead using
pip install .... But it seems like you run that sh file before entering the virtualenv, which makes this fail. I think it would be more useful to run it inside the virtualenv, especially if someone needs to run custom steps like Cython builds, which is your motivation in the first place.
Second comment is that for this to work, the build process should not fail on missing
requirements.txt but instead just skip running the command.
Screenshot of the problems: https://imgur.com/NVbL9v2