Item | Notes |
|---|
Updates from MolSSI | BP – We have kind of a plan to re appropriate the MolSSI server for QCArchive - This is a large machine that was previously shared within MolSSI. So that will cause some downtime, and I may migrate the server at the same time as I release the next branch - Looking at week of Dec 5. PB – Compatibility with QCSubmit? BP – DD is still working on that. The nice thing is that since all the interface is in QCSubmit, we can just update that and the rest of the tooling doesn’t have to change. JW – DD’s full efforts right now are dedicated to F@H project - Won’t have time to update QCSubmit until that’s resolved, but should have availability before Dec.
(General) – There will be some time before Dec where the new qcfractal is released but the server isn’t migrated - You’ll need to pin your dependencies in this period.
|
Infrastructure advances | |
Throughput status | OpenFF Protein Capped 3-mer Backbones v1.0 RNA Single Point Dataset v1.0 default: 8864 → 8868 (100 errors) wb97m-d3bj/def2-tzvppd: 3468 → 4209 (280 errors, 130 of which are convergence errors)
SPICE sets with openff-default: almost done PB still needs to debug the MBIS failure for Iodine with openff-default
PB – JC mentioned that we could generate more chemically diverse sets for training. I may prepare sets like this for submission in a few weeks. PB – CC, any sets you’re thinking of? CC – It looks like we’re kinda hitting our compute limit - Things seem to be hitting SCF failures. I’m thinking of some implicit solvent sets using PCM. But that’s still several months away. PB – PCM was kinda slow for us in the past, do you think this will be fast enough? CC – I saw that they recently multi-threaded PCM calcs in psi4 - I’d like to try this out and see how it goes. So I’m envisioning adding a new compute spec to an existing data set. PB – Like, adding implicit solvent to gen2 training data? CC – Yeah, something like that. Then if we retrain valence parameters with that, we’ll get better condensed-phase properties down the line. So that could be an interesting experiment to run.
JW: For the capped 3-mer do you think any of the incomplete data is useful? CC: I’m still waiting on geometric release that may move some of the remaining incomplete forward.
|
User questions/issues | |
Science support needs | |