Add Cog Support

[RichardMcSorley]

Adding cog to the project could make it easy to utilize imaginAIry on platforms such as replicate.com or simplify deploying imaginAIry to your own ML infrastructure.

Willing to give it a go on my end, when I have time and hopefully can get access to pushing to replicate soon.

A simple POC with command line arguments as input would be the easiest path forward imo but open to discuss.

[brycedrennan]

I'm not familiar with cog so my initial reaction is "i dont know".

I have deployed imaginairy in containers a few times this is the setup script I'm using currently:

  apt-get update
  apt-get install -y libgl1 libglib2.0-0

  wget https://developer.download.nvidia.com/compute/cuda/repos/debian11/x86_64/cuda-keyring_1.0-1_all.deb
  dpkg -i cuda-keyring_1.0-1_all.deb
  apt-get update
  apt-get install -y cuda=11.5.1-1

  pip install --upgrade pip setuptools wheel pip-tools;

  cat requirements-dev.txt | grep -v "^torch==" | grep -v "^torchvision==" > requirements-filtered.txt;
  cat testing_support/requirements-torch.txt >> requirements-filtered.txt
  pip-sync requirements-filtered.txt;

  pip install --no-deps -e .;
  imagine pizza --steps 2;
  touch /root/.initial-setup-done;

[dalaohuniu]

good thankyou