gpu support docs

docs/make.jl

@@ -14,8 +14,8 @@ makedocs(modules=[Flux],
                   "Training Models" =>
                     ["Optimisers" => "training/optimisers.md",
                      "Training" => "training/training.md"],
-                  "Data Munging" =>
-                    ["One-Hot Encoding" => "data/onehot.md"],
+                  "One-Hot Encoding" => "data/onehot.md",
+                  "GPU Support" => "gpu.md",
                   "Contributing & Help" => "contributing.md"])
 
 deploydocs(

docs/src/gpu.md

@@ -0,0 +1,33 @@
+# GPU Support
+
+Support for array operations on other hardware backends, like GPUs, is provided by external packages like [CuArrays](https://github.com/JuliaGPU/CuArrays.jl) and [CLArrays](https://github.com/JuliaGPU/CLArrays.jl). Flux doesn't care what array type you use, so we can just plug these in without any other changes.
+
+For example, we can use `CuArrays` (with the `cu` array converter) to run our [basic example](models/basics.md) on an NVIDIA GPU.
+
+```julia
+using CuArrays
+
+W = cu(rand(2, 5))
+b = cu(rand(2))
+
+predict(x) = W*x .+ b
+loss(x, y) = sum((predict(x) .- y).^2)
+
+x, y = cu(rand(5)), cu(rand(2)) # Dummy data
+loss(x, y) # ~ 3
+```
+
+Note that we convert both the parameters (`W`, `b`) and the data set (`x`, `y`) to cuda arrays. Taking derivatives and training works exactly as before.
+
+If you define a structured model, like a `Dense` layer or `Chain`, you just need to convert the internal parameters. Flux provides `mapparams`, which allows you to alter all parameters of a model at once.
+
+```julia
+d = Dense(10, 5, σ)
+d = mapparams(cu, d)
+d.W # Tracked CuArray
+d(cu(rand(10))) # CuArray output
+
+m = Chain(Dense(10, 5, σ), Dense(5, 2), softmax)
+m = mapparams(cu, m)
+d(cu(rand(10)))
+```