Skip to content

Issue with density fitting in Skala in GPU4PySCF #135

Description

Hey devs,

I was using Skala on colab to test out some code on small molecules, but there seems to be a bug with respect to density fitting, which in short is here : AttributeError: 'Mole' object has no attribute 'shape'. This seems to show up on other GPUs I have used too.

The detailed setup for this is given in this google colab link : https://colab.research.google.com/drive/1BJ-owZ2zNFse7mIcpho7LlzWpKW2lnhU?usp=sharing

Gemini provided a monkey patch which worked :

import gpu4pyscf.df.df as df_module

_original_get_jk = df_module.DF.get_jk

def _patched_get_jk(self, *args, **kwargs):
    # If the first argument lacks a 'shape' attribute, it is the mistakenly passed 'mol' object.
    if len(args) > 0 and not hasattr(args[0], 'shape'):
        args = args[1:]
    return _original_get_jk(self, *args, **kwargs)

# Overwrite the buggy class method
df_module.DF.get_jk = _patched_get_jk

(again the detailed report is in the colab link).

But I am reporting here because if it is a bug with Skala, then perhaps it could be patched out from the source code.

On an unrelated note, in the Skala paper on arXive, it is mentioned that gridlevel=3 was used for all calculations (section D.4). I was wondering would gridlevel=4 be more appropriate for obtaining single-point energies for Transition sates with 60-100 atoms, or would it be negligible in difference? If this is not the right forum to ask this please let me know, and please look the bug reported.

Regards,
Neil

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions