Skip to content

Conversation

@JessieGould
Copy link

@JessieGould JessieGould commented Dec 10, 2025

closes: #590

@JessieGould JessieGould reopened this Dec 10, 2025
@JessieGould JessieGould marked this pull request as draft December 10, 2025 17:18
Copy link
Member

@ElliottKasoar ElliottKasoar left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add a list of things left to do e.g. if you add a space in between the square brackets below, it'll create a checkbox

  • [] Update README

The developer guide may be slightly outdated in terms of what it suggests in terms of documenting a few model, but the main places will be the README and the user guide in docs.

We also need single point tests

# Set before loading model to avoid type mismatches
torch.set_default_dtype(torch.float32)

# Default to MATPES-r2SCAN model
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This doesn't seem to really be setting a default in the same way as this would normally mean, i.e. that this is automatically downloaded/used, not just a default string?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think AlphaNet doesn't call [torch.set_default_dtype()] internally, it sets [self.precision = torch.float32] but doesn't change the global default dtype?

@JessieGould
Copy link
Author

JessieGould commented Dec 11, 2025

Progress;

  • was working with single local model ckpt and json for tests ✓
  • changed to work for all models using github repos for ckpt and json ✗
  • not all model ckpt and json same style ie. some are figshare ✗
  • reworked to work with single model again ✗
  • It doesn't appear to have a built-in function to auto-download from github model like eg. pet-mad has ✗
  • so need to rework logic ...

To do list

  • singlepoint tests
  • write "from github download" function
  • remove "local" model from test models
  • create way to auto download model for tests ( call GH function)
  • direct user to model for manual download
  • numpy 2.0 conflict - PR submitted
  • numpy 2.0 conflict - PR accepted by alphanet?
  • if alphanet accepts PR then should resolve conflicts with fairchem and UMA (test) and if so needs to be added to all too
  • change logic for current implementation
  • review all comments from Elliott above

Final check

  • dependancies (check)
  • conflicts fairchem and uma check
  • architechture (check)
  • calculator (check
  • tests (check)
  • skip extras working as expected?
  • re-test for conflicts between packages

Documenting new model (check dev guide):

  • readme
  • userguide in docs

@JessieGould JessieGould marked this pull request as ready for review December 16, 2025 16:31
@JessieGould JessieGould marked this pull request as draft December 19, 2025 09:52
@JessieGould JessieGould changed the title 590 alphanet closes: #590 add alphanet Dec 19, 2025
@JessieGould JessieGould changed the title closes: #590 add alphanet 590 alphanet Dec 19, 2025
@JessieGould JessieGould marked this pull request as ready for review December 19, 2025 10:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Add AlphaNet

3 participants