Mike J Innes
Mike J Innes
I believe that line mirrors [this one](https://github.com/JuliaGPU/CuArrays.jl/blob/eac33f8f383f473c31bd879611fad15ee56671bb/src/dnn/rnn.jl#L106) which does happen in the forward pass (we just do it again in the backwards pass because we don't have access to the...
If `h` is a vector then `dh` should also be a vector; I'm not sure if we handle that correctly. However, it seems likely to be independent of the current...
I think it's fine to consolidate our docs into CuArrays. It's easy to find then, and there's not much reason to look for it unless you are interested in doing...
You may just need to add NNlib master, and make sure CuArrays is up to date with master as well.
bors try
Looks like you need to update the manifest to include `SpecialFunctions`. You should see a diff in git and can just push that.
@vchuravy there probably isn't much in it, so if the lead maintainers of this package strongly prefer Cassette then I imagine it'd be OK to port it over. Though as...
Good stuff! So you're right that the core issue is the inconsistency in the way unions are emitted (as objects) vs type bindings (as expressions). However, I think the right...
This is unfortunately due to Base's default parsing/printing of lambdas: ```julia julia> :(x->1) :(x->begin #= REPL[1]:1 =# 1 end) ``` Would be great to fix, but unfortunately might require a...
Alternatively, since `Expr(:->, :x, 1)` seems to do the right thing, maybe we can just tweak CSTParser to not insert the block / line number.