AI-Workshop-HuggingFace-as-a-platform-for-AI-services

Emoji VAE Workshop Project

This repository contains all the code and resources used for the AI Workshop session on Hugging Face as a Platform for AI Services, presented by Mykhailo Shtopko.

Project overview

In this project:

• Built a Variational Autoencoder (VAE) for emoji data and integrate it with the Hugging Face,
• Prepared a custom emoji dataset and uploaded it to Hugging Face Hub,
• Developed a Gradio Space with:
  • Encoding & Reconstruction: Upload an image and see its reconstruction,
  • Generation from Noise: Sample new images from latent space,
  • Interpolation: Create smooth morphing GIFs between two images.

Project structure

• /space/: Gradio Space implementation (encoding, generation, interpolation)
• emoji_vae.ipynb: Notebook for training & testing the VAE
• emoji_dataset_processing.ipynb: Dataset preparation & Hugging Face upload script s

Quickstart

1. Firstly, clone this repo:

git clone https://github.com/BioMikeUkr/AI-Workshop-HuggingFace.git

2. Preparing the dataset https://colab.research.google.com/github/BioMikeUkr/AI-Workshop-HuggingFace/blob/main/emoji_dataset_processing.ipynb

3. Building the model https://colab.research.google.com/github/BioMikeUkr/AI-Workshop-HuggingFace/blob/main/emoji_vae.ipynb

4. To create the space, first create a new Space on Hugging Face, then drag and drop the contents of the space/ directory into the upload interface.

Hugging Face links

• Model: BioMike/emoji_vae
• Dataset: BioMike/emoji-vae-dataset
• Space: EmojiVAE