Peter
Peter
MWE: ```julia julia> Zygote.gradient(randn(3,3)) do x sum(sin.(cat(x; dims=4))) end[1] 3×3×1×1 Array{Float64, 4}: # should be 3 x 3 but get 3 x 3 x 1 x 1 [:, :, 1,...
I test the code on ubuntu16.04 with python3.5 and it run without errors, so I think the dependency require for python2.7 is not necessary.
The parameter `16e3` at `wavfile.write` in model.py should be `int(16e3)` since the `wavfile.wirte` API only support integer sample rate
I make the implementation for the RACE dataset. And in order to make the code can run both on RACE and ROCstories, I make some change to the origin implementation...
It would be convenient to support index padding in `scatter` / `gather` which ignore the index with specific padding value (`0` for example).
Here are some stuff I'm going to rewrite for the new release: - [x] Tokenizer: Define the tokenizer with [TextEncodeBase.jl](https://github.com/chengchingwen/TextEncodeBase.jl) and replace the old `Basic.Vocabulary` with `TextEncodeBase.Vocab`. - [X] Layers:...
mentioned in #108. Currently we don't have an inference api, like the `pipeline` from huggingface transformers. Right now you need to manually load the model/tokenizer, apply them on the input...
mentioned in #9.
Hi, I'm trying to build a simple programming language. Is it possible to build the AST with CombinedParsers?
Is there any ways to compute the backward gradient of a keyword argument? For example, the `var` function can pass a pre-computed mean value as a keyword argument, but right...