-
### Description
Enclosed are two reproducers that will reach the same assertion failure in the same line of code, however, the stack trace is different for each of them.
One stack trace contains the…
-
- Should we support primitives beyond pushforwards / pullbacks ?
- Requires making assumptions on argument types
- Should we support higher order derivatives?
- Requires mixing backends, ergo …
-
I am on Julia 1.8.5 and Enzyme main (16818fd3d39d0583915aee38595b54a7fcce6b58). The following work:
```julia
using Enzyme
autodiff(Reverse, x -> x * 2.0, Active, Active(Float64(1.0))) # ((2.0,),)
…
-
Hey! I have been playing with this package to implement AD for [WaterLily.jl](https://github.com/WaterLily-jl/WaterLily.jl). The ForwardDiff backend works as expected, and it provides the same result …
b-fg updated
2 months ago
-
I am on Julia 1.8.5 and Enzyme main (73285ce). The following works:
```julia
using Enzyme
f1(x) = sum([ifelse(i > 0, i, zero(i)) for i in x])
x = randn(5)
dx = zero(x)
autodiff(Reverse, f1…
-
Copying over and summarizing some discussion from Slack:
@wsmoses:
> 1. Type-stable deep learning libraries. So far all the DL library code I've seen people try to AD with Enzyme is very very type…
-
See https://github.com/gdalle/ImplicitDifferentiation.jl/issues/71#issuecomment-1662287726 by @thorek1
-
MWE:
```julia
using CUDA, Enzyme, Random
rng = MersenneTwister(1234)
m = 32
n = 16
Z = cu(randn(rng, Float32, (n,m)))
𝒯 = 2.0
Δτ = 0.1
ca_init = cu([zeros(1) ; ones(m)])
function f!(ċȧ…
-
Opening this to track progress in taking gradients through `CuArray` broadcasting. With Enzyme main (a68bf83) and CUDA v5.3.4:
```julia
using Enzyme, CUDA
f(x, y) = sum(x .+ y)
x = CuArray(rand(5)…
-
### Description
When using both `custom_vjp` and `custom_vmap` on the same function (applied in either order), I get the following error:
```
File "/build/work/88c64dd2a6f04c75e55857743c6570e5c61…