Sergio Sánchez Ramírez
Sergio Sánchez Ramírez
The fix has been identified and seems like it will appear in Julia ~1.10.6~ 1.10.7 (didn't get in the cut by a matter of a couple of days) Should we...
yeah, don't worry. we can later add it to the `gh-pages` (although i think we need an automated way). let's start by adding some benchmarks. @starsfordummies @jofrevalles @Todorbsc what do...
@jofrevalles that sounds great! i would say that feel free to push to this PR so we can see the result
I managed to get a MWE ```julia using Reactant using Reactant.TracedUtils: broadcast_to_size using Enzyme Reactant.MLIR.IR.DUMP_MLIR_ALWAYS[] = true function f(A, θ) M = copy(A) @allowscalar begin c = cos(θ / 2)...
mmm what's your idea for how to implement this? like adding a MLIR type per structured tensor type? or maybe adding an attribute for marking ops that support a specific...
i guess that just adding `CondaPkg.toml` suffices? https://github.com/JuliaPy/PythonPlot.jl/blob/main/CondaPkg.toml my hope is that this works when PythonCall/CondaPkg is a weak dep
i guess that it'll still work? like CondaPkg.toml will only be checked when CondaPkg is loaded (or precompiled?)