Merge FluxML#1549

1549: Docstrings for f32 and f64 r=darsnack a=B-LIE I was challenged to add docstrings to f32 and f64, and have done that. Since I really never have used Git before, I feel on thin ice here, so I guess the challenge was more to push me out on the ice... Co-authored-by: B-LIE <[email protected]>
jtan21at · Mar 27, 2021 · e251327 · e251327
2 parents 5c40716 + 1808c75
commit e251327
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 0 deletions.
diff --git a/docs/src/utilities.md b/docs/src/utilities.md
@@ -40,6 +40,15 @@ Flux.orthogonal
 Flux.sparse_init
 ```
 
+## Changing the type of model parameters
+
+```@docs
+Flux.f64
+Flux.f32
+```
+
+The default `eltype` for models is `Float32` since models are often trained/run on GPUs. The `eltype` of model `m` can be changed to `Float64` by `f64(m)`, or to `Float32` by `f32(m)`.
+
 ## Model Building
 
 Flux provides some utility functions to help you generate models in an automated fashion.

diff --git a/src/functor.jl b/src/functor.jl
@@ -74,7 +74,18 @@ adapt_storage(T::Type{<:Real}, xs::AbstractArray{<:Real}) = convert.(T, xs)
 
 paramtype(T::Type{<:Real}, m) = fmap(x -> adapt(T, x), m)
 
+"""
+    f32(m)
+
+Convert the `eltype` of model's parameters to `Float32`.
+"""
 f32(m) = paramtype(Float32, m)
+
+"""
+    f64(m)
+
+Convert the `eltype` of model's parameters to `Float64`.
+"""
 f64(m) = paramtype(Float64, m)
 
 # Functors for certain Julia data structures