Individual updates | DN Last week I updated website, corrected an ad board note about potentially embargoing new FFs (miscommunicated to include software) Worked with OpenFE to make project page Talked with Virginia Burger - She said that her company was talking to NVidia soon about GPU optimizations. Wants to join together to offer advice to nvidia. JW – I don’t know mucha bout GPUs of what we could tell them, but it couldn’t hurt to start a dialogue CC – Think about future functional forms - Polarizability/DEXP? They may be able to optimize for that.
MT openmmforcefields is updated (again) to 0.11.2, this time with tests that ensure current and backwards compatibility with the OpenFF stack. Let me know if there are issues.
Minor progress on units/models packages - reach out if interested Tried refactoring GROMACS export, unfinished Workshop prep
CC Capped 3-mer backbones QC dataset has completed 12/54 TorsionDrives up from 6 last week | Github link macro |
|---|
| link | https://github.com/openforcefield/qca-dataset-submission/pull/294 |
|---|
|
Found a problem with torsion assignment for delocalized charges due to SMIRKS strings in our FFs. Issue here. Protein parameter fits are progressing slowly. Working with DM on getting set up on UCI cluster. Many ForceBalance workers die due to idle timeout. Consistently, only three workers run simultaneously. Is this a problem with my job setup? On my cluster, only 3 of my jobs actually get running, regardless of how many I submit. PB – Does the output indicate that the workers can’t communicate with the head node? I sometimes see things with ports. CC – Maybe, but the jobs just say “idle timeout” because they weren’t assigned any tasks PB – Does JSetiadi run ForceBalance jobs on TSCC? CC – He uses a smaller Gilson lab cluster, but we assess that it’s too small for the protein FF fits. PB – Ok, then the top suspect is worker communication with the head node. There may also be an option where you increase the time limit… let me check what I did.
JW – Toolkit 0.11.1 release - Forbids loading radicals as OFFMols. Helping with followup workshop prep Preparing tea times Good ad board meeting - Got OK on spending infrastructure time on non-openff software, software reliability Led F@H meeting. Seems like PL dataset reformatting is going to miss target, but we don’t actually need it until December at the soonest.
|