General updates | MT – Thinking about career progression - Doesn’t seem like there’s much further to go at OMSF. I’m thinking of starting a job search in the next few months. I don’t see it as my role to my a replacement, but I’m open to having some role in that process. Interchange vsites - see post in #interchange. I have a longer writeup. Originated from a report from MO. She helped me debug a lot, ended up being a niche issue. Hard to know if it affected anyone in the real world. LW has (AFAIK) just been using OpenMM and wouldn’t have been affected. Could have been caught by testing with molecules shuffled into different order, thinking about how to extend this into other tests. I think I’m kinda behind on stuff - HBaumann reported some systems taking a long time to export. I isolated some things and fixed them. But some others slipped out. It’s hard to isolate and profile things effectively. Running stuff as scripts vs. notebooks gives big random time differences. I think there’s maybe something very slow when SimpleMolecules are involved. Topology.chemically_identical_molecules with a bunch of simplemolecules may be slow. Maybe a networkx thing, a comparison thing, an indexing thing. Overall not sure. JW – Noticed super slow imports - Nagl? Might be loading pytorch or compiling models or something. MT – Maybe, but very complex. Could also be something with upstreams. Just nagl and deps should only add maybe 2 secs of import time, not a minute. I think I’m running into something different, since I’m measuring things that scale with number of molecules. Some smoke around memory usage, tried to profile and didn’t get far (even starting around 4-5k waters). I think slow imports are very setup-dependent (hardware, conda env complexity, jupyter magic, etc). MT – Also, even though some reports came FROM openfe, I don’t see this as work I’m doing FOR openfe. These are general issues that affect all users.
JW – Using act to reproduce forcebalance build issues - kinda cool, thought I’d share. I can emulate linux-x86 on a m1 mac (can share the command later) LW and I are meeting this afternoon (3 PM PT) to see if we can get evaluator running on NRP. Not totally relevant to you but you’re welcome to join if curious. Not ready to do AWS runners yet - KCJ and DS are working out org policies/permissions. He’ll reach out to me when ready to proceed. Nothing relevant to report from PEastman meeting - Mostly discussed PDB loading.
|