Martin Robinson

just a note to say we will definitely need this for the pkpdapp inference. Different outputs of the ODE model will have different noise models, hence the need for different...

After discussion with @jsbrittain, could possibly reuse the jax backend to do jax->hlo->xla->execute in C, using the c api for the xla compiler (https://github.com/openxla/xla/tree/main/xla/examples/axpy). Xla uses mlir anyway I think!

Problem is always sparsity :) openxla is planning to add but not there yet (https://github.com/openxla/stablehlo/blob/main/rfcs/20230210-sparsity.md) Jax has sparse matrix support, so wonder what sort of hlo they write out for...

Jax to hlo example https://jax.readthedocs.io/en/latest/aot.html

I think so, we just have to first add the idaklu solver to the benchmarks so we've got some evidence for making it the default, I've opened #2599 for this

Thanks @jsbrittain, nice catch. You would think the lru_cache docs would put some sort of warning on that decorator

@jsbrittain's solution of wrapping the functions during class construction doesn't seem to have a downside (apart from an extra line of code), sounds like we should do this for all...

I've had a quick look at this. At first I thought it was a bit strange that the windows wheel was even looking for a dll, as far as I...

The solution here might be to avoid the use of the casadi plugin system in pybamm. At the moment we are using it since we use the `casadi.Interpolant` class (https://github.com/pybamm-team/PyBaMM/blob/c0ff64f2b45e3a201105c6de7a7d082e9e30c14a/pybamm/expression_tree/operations/convert_to_casadi.py#L160)...

I'm not sure how to use `interpn_linear` :) so I've asked on the casadi repo: https://github.com/casadi/casadi/discussions/3668