Adrian Seyboldt

Using this works for me: ``` warnings.filterwarnings( "ignore", message='Cannot cache compiled function "numba_funcified_fgraph"', category=NumbaWarning, ) warnings.filterwarnings( "ignore", message='Cannot cache compiled function "store_core_outputs"', category=NumbaWarning, ) ```

The original int is I think really weird, to a point of just being wrong. Pointer offsets shouldn't be a int, I guess that leads to compiler errors in newer...

Hi @sstadick. This looks awesome! I haven't really been using this for quite some time (and mostly forgot it existed to tell you the truth), so I'll need a bit...

I don't think the difference is due to caching, just to make sure I gave it slightly different models, and that didn't change the timings. I'm also seeing the pretty...

Seems the hessian code is behind a compile time flag right now, and that was off in the code above. To make sure I just deleted all references to the...

I was just running into missing symbol issues with models involving sundials, and that made be check the variables in the makefile. Things there don't seem right to me, and...

It seems the difference I'm observing is because there is a `model_header.hpp.gch` header cache file in the `cmdstan` stan directory, but not in the corresponding directory of bridgestan. If I...

Looking at it again, I think the code that creates the precompiled header is only in the cmdstan repo, not in the stan or stan_math repo, so it just isn't...

Ah, I see your point... The file size of the precompiled header made me wonder what it could be that makes it so big, so I had a quick look...

If I understand this correctly, the reason that the precompiled header helps so much is because without it the compiler has to parse ~300_000 lines of code in ~3000 different...