Explicit dcgan mnist (#373)

* update to explicit

* remove Parameters.jl

* don't define device, just call gpu

* version + manifest updates

* maybe say explicit gradients

* unicode confuses my editor

* Update vision/dcgan_mnist/dcgan_mnist.jl

Co-authored-by: Michael Abbott <32575566+mcabbott@users.noreply.github.com>

Co-authored-by: Carlo Lucibello <carlo.lucibello@unibocconi.it>

README.md

@@ -59,7 +59,7 @@ Flux has a [NEWS page](https://github.com/FluxML/Flux.jl/blob/master/NEWS.md) li
 
 To run the old examples, Flux v0.11 can be installed and run on [Julia 1.6, the LTS version](https://julialang.org/downloads/#long_term_support_release).
 Flux v0.12 works on Julia 1.8.
-Flux 0.13 is the latest right now, marked with ☀️.
+Flux v0.13 is the latest right now, marked with ☀️; models upgraded to use  explicit gradients (v0.13.9+) have a `+`.
 
 ## Examples in the Model Zoo
 
@@ -68,7 +68,7 @@ Flux 0.13 is the latest right now, marked with ☀️.
   * [Simple multi-layer perceptron](vision/mlp_mnist) ☀️ v0.13
   * [Simple ConvNet (LeNet)](vision/conv_mnist) ☀️ v0.13
   * [Variational Auto-Encoder](vision/vae_mnist) 🌤 v0.12
-  * [Deep Convolutional Generative Adversarial Networks](vision/dcgan_mnist) ☀️ v0.13
+  * [Deep Convolutional Generative Adversarial Networks](vision/dcgan_mnist) ☀️ v0.13 +
   * [Conditional Deep Convolutional Generative Adversarial Networks](vision/cdcgan_mnist) ☀️ v0.13
   * [Score-Based Generative Modeling (Diffusion Model)](vision/diffusion_mnist) ☀️ v0.13
   * [Spatial Transformer](vision/spatial_transformer) ☀️ v0.13