MT/BW Sync

• Historical context is that Brent & Lexi have run (what they thought were) the same benchmark inputs and got different results with the force fields they were working on

• Idea is to have canonical datasets and scripts to produce them

  • Lots of ideas/iteration in directory openff-benchmarks

• Brent & Lexi might use different filtering steps to go from QCArchive data to YAMMBS input molecules

Cloud pricing

GHA’s “large” runners are ~$15/hr, so ~$60 for a new benchmarking run. Amazon is something like $3-4/hr. GCP is ~$10 for 4 hours on a 64-core node

• https://cloud.google.com/compute/all-pricing