Andrei Zhabinski
Andrei Zhabinski
Not that I expect this issue to be resolved anytime soon, but I thought it would be nice to write down ideas to have something to start with. In a...
Alternative version of #62. The main difference is that function `safenext()` and `complete()` are replaced with one another iterator `PrefetchIterator` which constantly keeps 1 element in cache, making it available...
From https://github.com/FluxML/NNlib.jl/pull/434#issuecomment-1235674312 ```julia function prod2(xs::Vector) p = one(eltype(xs)) for x in xs p = p * x p == 0 && break # exit early once you know the answer...
During work on tensor differentiation, I've found that derivative `dC/dA` of matrix multiplication `C = A * B` may be expressed as: ``` dCdA[i, j, m, n] = (i ==...
**Is your feature request related to a problem?** No **Which solution do you suggest?** Add CPU, GPU and memory to `resources` and `limits` section. Also, in case of GPU: *...
```julia using Zygote using CUDA f(A, I) = sum(A[I]) A = rand(4) I = [1, 3, 1] # CPU - everything is OK Zygote.gradient(f, A, I) # ==> ([2.0, 0.0,...
## Activation functions - [x] relu - [x] elu (CPU only) - [x] leakyrelu - [x] tanh - [x] sigmoid - [x] logsigmoid - [x] softplus - [x] softsign -...
with something like: ``` X = rand(...) Y = rand(...) Ze = zeros(...) Zb = zeros(...) eval(ein_expr) eval(buf_expr) @assert Ze == Zb ``` sizes of variables should be determined automatically...
Just fixed a bug in `eliminate_common` - one of many graph transformation subroutines that are deadly hard to debug. To discover issues with such utils we can add self-consistency checks....