Build a stripped down version of the sampler/solver to see what is reducing our performance if anything

[arik-shurygin]

build a really stripped down version of our infrastructure. No classes, just the sampler, parameters, and ODEs / diffrax solve. To see if there are dramatic improvements to the performance of fitting.

If so, slowly add back onto the model until we understand what exactly is slowing it down and how to best avoid it.

[SamuelBrand1]

Do you have any benchmarks? When did performance degrade, or is this a general idea for performance improvement?

[arik-shurygin]

Do you have any benchmarks? When did performance degrade, or is this a general idea for performance improvement?

General idea, this was sidetracked by the fall virus response + SMH meeting. Will likely get started soon

[arik-shurygin]

being explored on the gpu optim branch here: https://github.com/CDCgov/DynODE/tree/gpu-optim