[FR] iminuit + jax benchmark

[mmikhasenko]

I notice interesting studies in the iminuit package. Just dropping the links here, might be linked somewhere in README, or used for an inspiration when writing future benchmarks.

• RooFit vs numba JIT
  https://iminuit.readthedocs.io/en/latest/benchmark.html

• Jax JIT wiwth gradient is doing slightly worse than numba JIT without grad
  https://iminuit.readthedocs.io/en/latest/notebooks/automatic_differentiation.html#Fit-of-a-gaussian-model-to-a-histogram

[redeboer]

Nice, thanks! Indeed good to keep in mind for general fitting benchmarks

Some remarks with regard to TW.

• Numba is not yet working too well as backend in TW. Probably doing something wrong there with JIT
• For autodiff with large amplitude models, there's some work going on on this here: DOC: investigate fit with autodiff gradient compwa.github.io#195