Added diffusers demo with batching & some misc improvements to doc (#3224)

* added diffusers demo

* guide updates

* changelog

* update demo

* notebook

CHANGELOG.md

@@ -81,6 +81,7 @@ By [@dawoodkhan82](https://github.com/dawoodkhan82) in [PR 3165](https://github.
 * Keep pnginfo metadata for gallery by [@wfng92](https://github.com/wfng92) in [PR 3150](https://github.com/gradio-app/gradio/pull/3150)
 * Add a section on how to run a Gradio app locally [@osanseviero](https://github.com/osanseviero) in [PR 3170](https://github.com/gradio-app/gradio/pull/3170)
 * Fixed typos in gradio events function documentation by [@vidalmaxime](https://github.com/vidalmaxime) in [PR 3168](https://github.com/gradio-app/gradio/pull/3168)
+* Added an example using Gradio's batch mode with the diffusers library by [@abidlabs](https://github.com/abidlabs) in [PR 3224](https://github.com/gradio-app/gradio/pull/3224) 
 
 ## Testing and Infrastructure Changes:
 No changes to highlight.

demo/diffusers_with_batching/requirements.txt

@@ -0,0 +1,3 @@
+torch
+transformers
+diffusers

demo/diffusers_with_batching/run.ipynb

@@ -0,0 +1 @@
+{"cells": [{"cell_type": "markdown", "id": 302934307671667531413257853548643485645, "metadata": {}, "source": ["# Gradio Demo: diffusers_with_batching"]}, {"cell_type": "code", "execution_count": null, "id": 272996653310673477252411125948039410165, "metadata": {}, "outputs": [], "source": ["!pip install -q gradio torch transformers diffusers"]}, {"cell_type": "code", "execution_count": null, "id": 288918539441861185822528903084949547379, "metadata": {}, "outputs": [], "source": ["import torch\n", "from diffusers import DiffusionPipeline\n", "import gradio as gr\n", "\n", "generator = DiffusionPipeline.from_pretrained(\"CompVis/ldm-text2im-large-256\")\n", "# move to GPU if available\n", "if torch.cuda.is_available():\n", "    generator = generator.to(\"cuda\")\n", "\n", "def generate(prompts):\n", "  images = generator(list(prompts)).images\n", "  return [images]\n", "\n", "demo = gr.Interface(generate, \n", "             \"textbox\", \n", "             \"image\", \n", "             batch=True, \n", "             max_batch_size=4  # Set the batch size based on your CPU/GPU memory\n", ").queue()\n", "\n", "if __name__ == \"__main__\":\n", "    demo.launch()\n"]}], "metadata": {}, "nbformat": 4, "nbformat_minor": 5}

demo/diffusers_with_batching/run.py

@@ -0,0 +1,22 @@
+import torch
+from diffusers import DiffusionPipeline
+import gradio as gr
+
+generator = DiffusionPipeline.from_pretrained("CompVis/ldm-text2im-large-256")
+# move to GPU if available
+if torch.cuda.is_available():
+    generator = generator.to("cuda")
+
+def generate(prompts):
+  images = generator(list(prompts)).images
+  return [images]
+
+demo = gr.Interface(generate, 
+             "textbox", 
+             "image", 
+             batch=True, 
+             max_batch_size=4  # Set the batch size based on your CPU/GPU memory
+).queue()
+
+if __name__ == "__main__":
+    demo.launch()

guides/01_getting-started/02_key-features.md

@@ -258,6 +258,8 @@ demo.launch()
 
 In the example above, 16 requests could be processed in parallel (for a total inference
 time of 5 seconds), instead of each request being processed separately (for a total
-inference time of 80 seconds).
+inference time of 80 seconds). Many Hugging Face `transformers` and `diffusers` models
+work very naturally with Gradio's batch mode: here's [an example demo using diffusers to
+generate images in batches](https://github.com/gradio-app/gradio/blob/main/demo/diffusers_with_batching/run.py)
 
-Supplying a generator into Gradio **requires** you to enable queuing in the underlying Interface or Blocks (see the queuing section above).
+Note: using batch functions with Gradio **requires** you to enable queuing in the underlying Interface or Blocks (see the queuing section above).

guides/03_building-with-blocks/02_controlling-layout.md

@@ -57,8 +57,10 @@ Both Components and Layout elements have a `visible` argument that can set initi
 $code_blocks_form
 $demo_blocks_form
 
+## Variable Number of Outputs
+
 By adjusting the visibility of components in a dynamic way, it is possible to create 
-machine learning demos that support *variable numbers of outputs*. Here's a simple example
+demos with Gradio that support a *variable numbers of outputs*. Here's a very simple example
 where the number of output textboxes is controlled by an input slider:
 
 $code_variable_outputs