Next-generation Autoregressive Visual Generative Model

This project proposes to develop and evaluate a next-generation autoregressive visual generative model that combines the strengths of autoregressive transformers, large language models (LLMs), and diffusion models. 

The aim is to push the boundaries of high-fidelity and controllable image generation, addressing both global structure and fine-grained visual detail. 

EuroHPC’s computational infrastructure will be instrumental in scaling up training, performing multi-path evaluation, and executing large-scale ablation studies