Page Comparison

General updates

• MT – Roadmap to non-experimental interchange? Experimental flag is going away, though some pushback from folks nervous about public API. Looking for ways to thread the needle. My current thoughts are 1) documenting what I already know 2) emitting lots of warnings 3) raising errors when I don’t think things will work well. I don’t love the idea of spamming warnings here but I don’t know of a good alternative.

  • JW – I think this all makes sense. I’d avoid a “general warning” whenever someone calls “interchange.from_X”, but I think that narrow warnings like “hey, these constrained bonds don’t have force constants” are a great idea. I think the main message we want to send is “we’re 75% sure that everything here is fine, and YOU need to tell us if you have a problem so we can catch the remaining fraction”. All we can realistically do is commit to taking reports seriously and turning around fixes fast.

  • MT – Sounds good.

• JW –

  • Thoughts on IA issues?

    • (Reviewed synchronously)

  • Could you spend a few hours trying to set up GHA testing for evaluator? POSE has instructions for setting up GPU runners for GHA. We’ll need to request access to GPU (“P series”) runners from AWS.

    • MT – Yeah, let’s add this as a ticket. Would like clarity from lead team on long-term plans for evaluator.

    • JW – We recently had an issue where LW got bad results from evaluator, then they went away in a subsequent rerun. Realized Ihad no diea where we’d start debugging. So better/simulation-including CI here could help derisk this a bit. But currently no big plans for evaluator. If new properties need to be added that would start with the science team, and there’s nothing in that pipefor at least 3 months.

    • JW will get Ethan/Swenson to show us how to set up AWS