Closed rok-cesnovar closed 2 years ago
fyi running a jenkins job with a math PR to check if I have the math fixes right
Okay! Now I want to turn on the partial evaluator to see if my math fixes also fix that
Mkay! So it looks like with those two math PRs we are good for the partial evaluator
@SteveBronder does that mean that we merge the 2 math PRs and we should be good to go for 2.29.2?
Oh and disable AD levels for O1?
Yep!
Is the AD levels optimization a low impact one?
Its an optimization where imo the user could also just move the data only operation to transform data
themselves. imo not a huge one. I think part of the issue is just that we shouldn't run it over functions. The main place it is beneficial is in log_prob
Both Math PRs were just merged in, I cleaned up this PR which will now just disable AD levels optimizations in O1. Running it once more with --O1
to make sure all models compile and will then remove changes to Jenkinsfile and tag for review.
Agreed! See https://github.com/stan-dev/performance-tests-cmdstan/pull/37
Once that is in we just need to add a Jenkins stage to run it.
Thanks Rok! I thought we tested all of these with O1 before but I must have missed something. I will go through these today to see whatsup. My initial guess is that turning off the partial evaluator will fix a lot of these