Closed JosephMontoya-TRI closed 1 month ago
Thanks for reporting this. I will do a release soon. I am open to ideas on how to do versioning of models. In general, we support the latest version. Otherwise, we will spend way too much time versioning models and keep track of old models.
Perfect, thank you for the release!
Yeah, flexible versioning seems like a headache - and presumably coupling updates to the main branch with the release will prevent the issue since that's where the models are being loaded from in the code as far as I can tell.
If you really wanted to have a more strict loading of the same version as the installed code, my impulse would be to load the model by accessing the git url with the appropriate tag that corresponds to that currently installed version, which presumably would keep the model consistent with the version in the distributed case, but I'm not sure how that would affect the development or testing case. It also wouldn't work with prior versions, but could work with the versions released from now on.
Email (Optional)
No response
Version
1.0.0
Which OS(es) are you using?
What happened?
With a pip-installed matgl, loading models with new versions in the git repo (i.e. M3GNet-MP-2021.2.8-PES) throws errors related to e.g. new output params.
Could pin models to load from a tag that matches current version.
Code snippet
Log output
Code of Conduct