Avik Pal
We capture: 1. ForwardDiff.gradient 2. ForwardDiff.jacobian 3. Zygote.gradient 4. Zygote.jacobian after https://github.com/LuxDL/Lux.jl/pull/598. We should capture the DI jacobian, gradient, and, most importantly `pullback` calls to augment them with the faster...
DynamicExpressions supports ChainRules starting v0.17 https://github.com/SymbolicML/DynamicExpressions.jl/pull/71. We can remove parts of our code with `CRC.rrule_via_ad`. We still need to define a rule because we do an in-place node update. Additionally...
Currently if we see an array as a kwarg to `@compact` we put it in trainable mode. But consider an UDE as a motivating example: ```julia @compact(; model, p_true) do...
Started initial setup in #572 - [x] Generate a minimal set of benchmarks - [x] Some basic ones are present in https://github.com/FluxML/Flux.jl/tree/master/perf - [x] Benchmark different AD backends ~-- use...
See https://docs.open-mpi.org/en/main/man-openmpi/man3/MPIX_Query_rocm_support.3.html#mpix-query-rocm-support Seems to be available from v5 of OpenMPI
- [ ] Add more serious workloads - [ ] Add MIRK Methods
With the latest changes to nonlinear solve, several algorithms compute VJPs and they don't use $J^T v$ rather they use AD to differentiate the internal nonlinear problem, this causes significant...
This needs to be handled similarly to MIRK, but someone needs to go in and do it.
Opening an issue so that I remember to write it once #155 is finished. The combination of options for BVP Jacobians are quite a lot so even I forget which...