Oliver Schulz
Oliver Schulz
> Yes, thunks are an emulation of a compiler feature julia doesn't have. ~~Hm, this got me thinking - thunks are really monads, right? [...]~~ Moved to JuliaDiff/ChainRulesCore.jl#373 .
> Issue for adding it back is still open JuliaDiff/ChainRulesCore.jl#7 (comment) > I do worry about the same thing re mem allocs, here Thanks, I added a suggestion there.
@DhairyaLGandhi do we get your blessing on the "plan"?
> at this point the plan might be moot. And we can soon basically just merge something like this PR. More than fine with me - we'll still need to...
FluxML/ZygoteRules.jl#17 has been merged, I'll get this PR into shape as soon as there's new release of ZygoteRules, so CI can run.
Ok, new release of ZygoteRules is out (had overlooked it), I'll jump back on this.
Sorry, bit overloaded but haven't forgotten about this.
Rebasing this now, but several things have changed in Zygote in the mean time ... hope it still works.
CI results don't look too horrible - at least some of the test failures look similar to the ones in #1104, so may be unrelated to this PR. Could an...
I have one test failure that I can't get figured out (how to fix): The "Params nesting" testset results in a `map is not defined on dictionaries`. It's caused by...