🥑 Dalle-Mini Poster App ⚡️

This app is a research poster demo of min-dalle (PyTorch port of Dalle-Mini). It showcases a notebook, a blog, and a model demo where you generate images from text prompts.

Getting started

Installation

With Lightning CLI (This method is not activated yet)

lightning install app lightning/dalle_mini_poster

From GitHub

You can clone the app repo and follow the steps below to install the app.

git clone https://github.com/lightning-AI/LAI-dalle-mini-poster-App.git
cd LAI-dalle-mini-poster-App
pip install -r requirements.txt
pip install -e .

Once you have installed the app, you can goto the LAI-dalle-mini-poster-App folder and run lightning run app app.py --cloud from terminal. This will launch the Dalle app in your default browser with tabs containing blog, Training logs, and Model Demo.

You can control the number of generated images using OUTPUT_IMAGES environment variable. To generate 4 images you can do lightning run app app.py --env OUTPUT_IMAGES=4 --cloud

Example

# update app.py at the root of the repo
import lightning as L

poster_dir = "resources"
blog = "https://wandb.ai/dalle-mini/dalle-mini/reports/DALL-E-Mini-Explained-with-Demo--Vmlldzo4NjIxODA"
github = "https://github.com/borisdayma/dalle-mini"
wandb = "https://wandb.ai/dalle-mini/dalle-mini/reports/DALL-E-Mega-Training-Journal--VmlldzoxODMxMDI2"
tabs = ["Poster", "Blog", "Notebook Viewer", "Training Logs", "Demo: Generate images from a text prompt"]

app = L.LightningApp(
    ResearchApp(
        poster_dir=poster_dir,
        blog=blog,
        training_log_url=wandb,
        # notebook_path="resources/DALL·E_mini_Inference_pipeline.ipynb",
        launch_gradio=True,
        tab_order=tabs,
        launch_jupyter_lab=False,  # don't launch for public app, can expose to security vulnerability
    )
)

FAQs

1. How to pull from the latest template code? Answer

Acknowledgement

Credits to Boris Dayma for this awesome work Dalle Mini and Brett Kuprel for min-dalle.