2025.12.10 Cole Lab Check-In (Aus) Meeting Notes

SMEE/ Descent Discussion

• DC: I heard there’s a plan to talk to Simon about forking smee/descent or adding maintainer

• LW: 2.4 will be fit on SMEE – definitely valence and potentially nonbonded

• LW: Planning light-weight package on top of smee and decsent to include best parts of evaluator

  • https://docs.google.com/presentation/d/1KD7oXa0NFHcKYYsvsHpL7Ti0gNQ-5Oq6X8CawG7PhwY/edit?usp=sharing

• LW: Moving towards SMEE and descent already in back to school projects. Slowly progressing.

Bespoke fitting progress

FC

DC: Don’t bother with fancier comparison of ensembles

LW: Have you checked reproducibility of benchmarks on different hardware? Relevant message:

FYI, for people running yammbs on machines where multiple different node architectures can run the job (e.g. UCI or another cluster), I found that hardware differences can result in different benchmarking results, so you may want to ensure you run on the same hardware for easiest comparison. I plotted the differences between hardware (and uploaded the results of investigative runs) in the repo directory below, as well as shared scripts demonstrating how to request a particular node or constraint in SLURM.https://github.com/lilyminium/protein-alpha-workbook/tree/main/benchmarks/split-hardware-ff

• FC: No