tanhevg
tanhevg
Hey, sorry, I have moved on to something else at the moment. Will keep this in mind next time I need to build a large database for HHBlits search.
The issue here is that BioJulia handles gap open and extend penalties a bit differently from BioPython. A gap of size `k` costs `gap_open + k * gap_extend` in BioJulia....
From what I've seen in the literature, original papers from the 80's and 90's mostly use the `gap_open + k * gap_extend` notation. One could argue that `gap_open + (k...
Need to clean up README.md from the folders with training and test .bson files; READMEs upset the BSON parser.
Looks like the error is still there in 2.1.0, although #96980 has been merged a few months ago. Should this be re-opened? CPU=Mac m2 Pro torch=2.1.0 OS=Mac OS Sonoma 14.0
Thanks for the update. To clarify, the slow runtime (30 seconds to 1 minute per training step) is without flash attention. I thought the way to speed things up would...
Yes, I am
Which distance function are you using? By definition, a distance function should return a non-negative value for any pair of arguments. That's what the default (`euclidean`) does. If using a...
Ok, this makes sense now: in emulation mode the distances are predicted by the GP. Predictions can indeed turn out negative, for example when the mean is very close to...
Does this problem still occur with the latest version of GpABC (v0.1.1)? Need more details to verify that the issue is with GpABC and not with Flux.