Page Comparison

General updates

• MT –

  • Cancel Friday one-on-one. Also I don’t have any questions/comments for Peter.

  • Improved GROMACS export speed based on working with HBaumann. 20k atom protein previously took hours, now is ~30x faster (a couple seconds per thousand atoms). There may even be more optimizations left, but this is already a big win. Somewhat concerned about performance regressions for small molecule boxes/simple liquids. It’s possible that liquid boxes will be a little slower, but this is against the background of a global speedup that offsets it. Last communication with HB was “I’ll take a look”, earlier today. Could be in a release in the next few days, but would like to wait for Hannah’s feedback.

  • JW – Cool, were the effects all to GROMACS export, or were some speedups general?

  • MT – These specific changes were GROMACS only, but the some of the general approaches should be applicable to other engines when I go to optimize them.

  • MT – I’m meeting with IA and MH next week, will communicate pydantic 2 plans to them again. You’re free to join but I don’t expect you to.

  • MT – Meeting with ZB next week.

• JW –

  • I’m putting together a #general blast about recent releases. Do you think the Vanilla installs of openff-toolkit-examples and . I’ll mention Interchange 0.3.28 - Are there other releases from your repos that I’m forgetting?

@here We’re pleased to announce a few recent minor releases:
OpenFF QCSubmit 0.52.0, with opt-in caching of retrieved datasets and records 
OpenFF NAGL 0.4.0, adding support for lookups tables to be used in conjunction with 
OpenFF ForceFields 2024.07.0, including openff-2.2.1-rc1.offxml, featuring 
OpenFF Toolkit 0.16.2, with some minor bugfixes
Interchange 0.3.28 with a LAMMPS export bugfix (and an upcoming release may substantially speed up GROMACS protien exports)

• Possibly related to QC deployment discussion yesterday? Vanilla installs of openff-qcsubmit aren’t resolving to OFFTK 0.16.2, and conda nightly tests are failing on version check (installed 0.16.1 but latest is 0.16.2). I think this is because of the zstandard pin in the feedstock (which was due to an incompatibility between OpenEye and QCF).

  • Bespokefit CI (which should exercise compatibility for this whole stack) is resolving to OFFTK 0.15.2 (old), QCF 0.53 (old), OE 2024.1.1 (latest), and zstandard 0.23 (latest)

  • MT – Would prefer to be presented this info in an issue tracker/async. Appreciate that it can be stressful to push half-baked info in the open but it’d be better to occasionally ask silly things publicly.

  • JW – Ok, I’ll compile thoughts and post for discussion async

Trello

https://trello.com/b/dzvFZnv4/infrastructure