Skip to content

Conversation

@mariabilichenk0
Copy link
Contributor

This PR introduces a number of fixes and enables stress prediction with NequIP and Allegro:

  • fix indexing of elements in cp2k to correspond to the indexing of torch model

  • update NequIP/Allegro models & TEST_FILES from NequIP==0.5.6 to NequIP==0.6.0

  • change parsing of dtype from custom subroutine to torch_model_read_metadata, see allegro/nequip: run with float32 or float64, check if TF32 is used, freeze model #2786

  • add kinds comparison cp2k torch

  • add prediction of stress and add stress to regtest for nequip water-bulk-dp.inp

update neq/allegro models & TEST_FILES from NequIP==0.5.6 to NequIP==0.6.0

change parsing of dtype from custom subroutine to torch_model_read_metadata

add kinds comparison cp2k torch

add prediction of stress and add stress to regtest for nequip water-bulk-dp.inp
@marci73 marci73 merged commit 519c511 into cp2k:master May 24, 2024
@fstein93
Copy link
Contributor

This PR introduced a bug in parallel builds (wrong results). Can you check what is going on there?

MGraml pushed a commit to MGraml/cp2k that referenced this pull request Jun 12, 2024
)

Co-authored-by: Maria Bilichenko <mbilichenko@tcopt12.chem.uzh.ch>
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.

3 participants