Skip to content

Atom-chunked evaluation of Skala 1.1#85

Merged
tvogels merged 8 commits into
mainfrom
klaas/chunked-skala
Jun 29, 2026
Merged

Atom-chunked evaluation of Skala 1.1#85
tvogels merged 8 commits into
mainfrom
klaas/chunked-skala

Conversation

@kjh-giesbertz

@kjh-giesbertz kjh-giesbertz commented Jun 29, 2026

Copy link
Copy Markdown
Contributor

This PR does two things:

  • adds evaluation of Skala 1.1 (exc, vxc, fxc) chunked by atom-grid. AO computation, feature computation and model evaluation are now chunked together, lowering peak memory.
    • For performance: Exc and vxc the atom chunks are packed together in larger chunks while still fitting in memory, for second order it is faster to run per atom chunk.
    • When on gpu: a helper routine sets the maximum chunk size specific for skala 1.1 based on number of AOs, grid size and order of the derivative of Exc. When on cpu: the available memory needs to set wchi goes via the usual max_memory parameter in PySCF
  • Updates nuclear gradients to work with gpu4pyscf 1.6. With this update gpu4pyscf 1.5 is not supported anymore.

This does two things:
- adds evaluation of Skala 1.1 (exc, vxc, fxc) chunked by atom. Ao
  computation, feature computation and model evaluation are now chunked
together, lowering peak memory.
- updates nuclear gradients to work with gpu4pyscf 1.6.
@kjh-giesbertz kjh-giesbertz requested a review from tvogels June 29, 2026 11:32
@kjh-giesbertz kjh-giesbertz self-assigned this Jun 29, 2026
Comment thread environment-gpu.yml Outdated
Comment thread environment-cpu.yml Outdated

@awvwgk awvwgk left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Looks good to me now.

@awvwgk awvwgk linked an issue Jun 29, 2026 that may be closed by this pull request
@tvogels tvogels merged commit 029975d into main Jun 29, 2026
32 checks passed
@tvogels tvogels deleted the klaas/chunked-skala branch June 29, 2026 14:14
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.

Newer versions of GPU4PySCF break the nuclear gradient implementation

3 participants