💵 GFusion: A New Diffusion LLM from GigaChat
The GigaChat (Sber) team has introduced GFusion—a diffusion language model (dLLM) that accelerates generation by 70% by using parallel restoration of text blocks instead of character-by-character output.
🌍 This is a path toward a manifold increase in LLM throughput.
👤 Interaction with AI will become instantaneous: instead of the model "typing" out answers, it will output entire paragraphs at once.
Source 1: https://habr.com/ru/companies/sberbank/articles/1054690/ Source 2: https://github.com/sgl-project/sglang/pull/29776