NVIDIA / modulus

Open-source deep-learning framework for building, training, and fine-tuning deep learning models using state-of-the-art Physics-ML methods
https://developer.nvidia.com/modulus
Apache License 2.0
797 stars 174 forks source link

0.6.0 docker patch #471

Closed ktangsali closed 2 months ago

ktangsali commented 2 months ago

Modulus Pull Request

Description

With different base containers, we have different requirements for the wheels. This PR allows a mechanism to control this using a build-arg.

Also tested the build using just the defaults (in the absence of pre-built wheels)

Closes: https://github.com/NVIDIA/modulus/issues/453

Checklist

Dependencies

ktangsali commented 2 months ago

/blossom-ci

ktangsali commented 2 months ago

This looks good to me. Just to confirm, this PR should now allow local docker builds without having the cached wheels right?

Yes, that is correct. There is one caveat, if the base container is changed such that the fallbacks don't work (say has unsupported CUDA version, etc.), then that might require some additional handling, but other than that, it should work fine. If you run into issues, please let me know. I tested this dockerfile without any cached wheels (just defaults, i.e. make container-deploy) and that worked.