Update app.py

app.py

@@ -119,8 +119,25 @@ def plot_sample(mode):
     
     return make_subplot_latent(test_images, quant)
 
-import gradio as gr
-radio = gr.Radio(choices=['Reconstruction','Latent Representation'])
-out =  gr.Plot()
-
-gr.Interface(plot_sample, radio, out).launch()
+demo = gr.Blocks()
+
+with demo:
+    gr.Markdown("# Vector-Quantized Variational Autoencoders (VQ-VAE)")
+    gr.Markdown("""This space is to demonstrate the use of VQ-VAEs. Similar to tradiitonal VAEs, VQ-VAEs try to create a useful latent representation.
+     However, VQ-VAEs latent space is **discrete** rather than continuous. Below, we can view how well this model compresses and reconstructs MNIST digits, but more importantly, we can see a
+     discretized latent representation. These discrete representations can then be paired with a network like PixelCNN to generate novel images.
+     
+     VQ-VAEs are one of the tools used by DALL-E and are some of the only models that perform on par with VAEs but with a discrete latent space.""")
+
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                radio = gr.Radio(choices=['Reconstruction','Latent Representation'])
+            with gr.Row():
+                button = gr.Button('Run')
+        with gr.Column():
+            out =  gr.Plot()
+
+    button.click(plot_sample, radio, out)
+
+demo.launch()