Closed Red-Portal closed 3 months ago
Well the fact the MWE's are getting harder/less frequent means we're close to them all being done!
Will look at next
@Red-Portal fwiw the issue stems from estimate_energy_with_samples apparently not being type inferrable by julia.
something along the lines of
Core.Compiler.return_type(estimate_energy_with_samples, Tuple{TestNormal{Fill{Float32, 1, Tuple{Base.OneTo{Int64}}}, PDMat{Float32, Matrix{Float32}}}, Matrix{Float32}})
is spitting out a return of Any
Oops sounds like it's a problem on my side. Let me take a look.
@Red-Portal so this passes for me on Enzyme#main (though I had to move the runtimeActivity to right after using Enzyme).
what is the error log?
@Red-Portal gentle ping here, what is your environment/how precisely did you run this (since I was unable to repro)
@wsmoses Sorry for missing this! Yeah, I just tried it again, and it works. Thanks for the ping.
Hi, thanks for the quick patches! Looking at the errors, it seems we're not quite there yet, and reproducing the errors in a MWE is getting harder. But without further due, here's a new issue:
This results in a segfault. Quite bizzarely,
logpdf
ormean
alone don't trigger an error, but the combination does.I suspect this is the cause of another non-segfault error caught in the full tests (this is not from the MWE above):
but hard to tell for me. Let me know if they seem related.