DALL-E: PyTorch discrete VAE implementation
Official PyTorch package implementing the discrete VAE component for image tokenization used in OpenAI's DALL-E system.
Learn more about DALL-E
DALL-E is a PyTorch implementation of the discrete Variational Autoencoder (dVAE) component from OpenAI's DALL-E image generation system. The package uses discrete latent representations to encode and decode images, converting continuous image data into discrete tokens that can be processed by transformer models. The implementation focuses specifically on the VAE architecture that handles image tokenization and reconstruction. This component is commonly used for research into discrete image representations and as a foundation for building text-to-image generation systems.
Official Implementation
Direct release from OpenAI providing the exact discrete VAE architecture used in the original DALL-E research paper.
Discrete Tokenization
Converts images into discrete token representations suitable for processing by transformer-based language models.
Modular Design
Provides only the VAE component, allowing researchers to integrate it with custom transformer architectures for text-to-image tasks.
Related Repositories
Discover similar tools and frameworks used by developers
Triton
Domain-specific language and compiler for writing GPU deep learning primitives with higher productivity than CUDA.
Ray
Unified framework for scaling AI and Python applications from laptops to clusters with distributed runtime.
Model Context Protocol Servers
Reference implementations for LLM tool and data integration.
Weights & Biases
ML experiment tracking platform with logging, visualization, and model versioning.
Wan2.2
Open-source diffusion framework for multi-modal video generation.