Gemini Diffusion Preview: 1,400 Tokens/Second Text Generation

Google said its experimental diffusion model sampled about 1,400 tokens per second by refining whole text blocks, and shipped it as a demo rather than a product.