Page Comparison

Project updates

BW

• Past week

  • Wrapping up PRs (QCSubmit fix, etcyammbs)

  • Resubmitted hessian dataset

  • Monitoring QCA workers

    • MLPepper ran in 2 days

    • Lipid maps is going slowly as it’s not split by size – the resource requirements vary. Also, scaling deployment > 60 workers means everything crashes

  • yammbs PR

  Benchmarks

  • PR openers upload:

    • a yaml file with config

    • Optionally, a FF file

  • Things we could review:

    • CSV files that get committed to repo

      • Human input: check they got uploaded

      • LW: can script generate plots for you?

      • BW: that’s possible

    • Zenodo submission (which is now on production OpenFF)

      • Note: YDS doesn’t publish entry. To review Zenodo, you need to be able to log into Zenodo.

      • Someone needs to review and hit publish, to include DOI, to include in PR.

      • BW: picturing DOI as a line in a README

  AMI: is the idea to review before the run starts or after it ends?

  • BW: afterwards – there’s not a whole lot to review before the run. Just the yaml file and maybe an FF.

  • AMI: there’s not that much an external person could review afterwards either, other than checking the CSVs are there. Don’t really see downside to requiring a review

  • BW: just worried about best practices. Also need write permissions to trigger bot.

  Next week

  • More benchmarks

  • QCA workers

    • Hessian

    • Lipidmaps

    • Lexie’s PRs

  • Benchmarks

    • Parsley

      • Added retry logic to Zenodo, and backups of GH artifacts

  • Protein fits

    • Used Chapin’s env, kicked off a fit but crashed. Rerunning now. Also updating code to modern (e.g. CachedPortalClient). Pushed updates to fork.

• Next week

  • More benchmarks – Sage 2.0 up first

  • QCA workers (expecting Hessian datasets to finish up this week)

  • Split up second lipidmaps dataset

    • Separate PRs into the same directory

    • Considering tagging individual records separately

  • Organometallics dataset

    • Talked about limitations of organometallics in toolkit

  • yammbs PR(s)

  • YDS

    • Update scripts with plots

    • Publish Zenodo entry and merge PR

    • Which FFs?

      • 1.3.1

      • 2.0

      • 2.1

      • 2.2.1

      • experimental FFs

  • Protein fits

Project

updates

AMI

• Past week

• mamba issues

• DDX errors

  • looked into functional group breakdown

  • Also resubmitted same dataset without diffuse functions. No errors.

• NAGL2 optimization dataset planning

  • Decided to do geom opts at usual level of theory

  • BP: can do 2.5-5 TB dataset. “Let’s see how well we can handle it”. “no problem” to do up to 1 TB.

  • AMI: 5 confs/mol would be ~2.5 TB. 5 TB would be 10 conformers. 2 confs/mol would be 1 TB.

  • LW: aim for 5 conf/mol

• Follow up on standards document

Next week

• NAGL2 optimization dataset

• Hessian DSs

• (Freeze phosphate angle refit?)

• (NAGL2 testing dataset--examine exisiting coverage and maybe setup opts)

• Finish standardsHessian DS submission

• Standards document

• Wrote up NAGL2 error analysis

• Analysis of diffuse vs non-diffuse functional errors

• Next week

  • NAGL2 training optimization set

  • Clean up HPC3

    • Zenodo or OpenFF Google Drive

  • Integrate S data fit into Sage 2.3.0 repo?

    • Options:

      • Have repo reflect most recent

      • Have different bits of work in different folders

      • Have different branches

  • NAGL2 Benchmark set curation

  • Protein fit (Maybe approach 3?)

    • Approach III: training to FF data only

LW

• Past week:

  • Working with various Kubernetes issues

  • Troubleshooting packmol

  • Collaborations and adminProtein FF tasks – validating AMBER FF

  • Evaluator on NRP – improving GPU compute

  • Debugging Interchange boxes

• Next week:

  Protein projects

  • More protein FF tasks

  • Improving Evaluator infrastructure and running POC fit

  • Interchange equilibration times