{"slug": "unlocking-lossless-speedups-in-llms-via-discrete-diffusion-5000-tk-s", "title": "Unlocking Lossless Speedups in LLMs via Discrete Diffusion (5000 Tk/S)", "summary": "Researchers from the Institute of Foundation Models, Cornell Tech, Cerebras Systems, and other institutions introduced a discrete diffusion method that achieves lossless speedups in large language models (LLMs), reaching 5,000 tokens per second. The approach enables faster inference without sacrificing output quality, marking a significant advance in LLM efficiency.", "body_md": "# Unlocking Lossless Speedups in LLMs\n\nvia Discrete Diffusion\n\n**Subham Sekhar Sahoo**†,1,\n**Lingjie Chen**†,1,2,\n**Khiem Pham**†,1,3,\n**Jonathan Geuter**†,1,4,\n[Junlin Chen]1,5,\n[Chaitanya Dwivedi](https://scholar.google.com/citations?user=ghpn6JkAAAAJ&hl=en)1,\n[Varad Pimpalkhute](https://nightlessbaron.github.io)1,\n[Yash Akhauri](https://akhauriyash.github.io)1,\n[Alexander Moreno](https://www.linkedin.com/in/alexander-moreno-ab151542/)1,\n[Mikhail Yurochkin](https://moonfolk.github.io)1,\n[Zhenting Wang](https://zhentingwang.github.io)1,\n[Mostafa Elhoushi](https://scholar.google.com/citations?user=y_cwSKAAAAAJ&hl=en)6,\n[Nolan Dey](https://scholar.google.com/citations?user=JHUfMr0AAAAJ&hl=en)6,\n[Shane Bergsma](https://sites.google.com/site/shaneabergsma/)6,\n[Joel Hestness](https://scholar.google.com/citations?user=wkbvCf0AAAAJ&hl=en)6,\n[Hongyi Wang](https://hwang595.github.io)1,5,\n[John Thickstun](https://johnthickstun.com)3,\n[Eric Xing](https://mbzuai.ac.ae/study/faculty/professor-eric-xing/)1,\n[Zhengzhong Liu](https://hunterhector.github.io)1\n1Institute of Foundation Models\n2University of Illinois Urbana-Champaign\n3Cornell Tech\n\n4Harvard University\n5Rutgers University\n6Cerebras Systems\n\n†Core contributors", "url": "https://wpnews.pro/news/unlocking-lossless-speedups-in-llms-via-discrete-diffusion-5000-tk-s", "canonical_source": "https://s-sahoo.com/uno/", "published_at": "2026-09-04 06:16:48+00:00", "updated_at": "2026-09-04 06:52:53.108481+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "generative-ai", "ai-research"], "entities": ["Institute of Foundation Models", "Cornell Tech", "Cerebras Systems", "Subham Sekhar Sahoo", "Lingjie Chen", "Khiem Pham", "Jonathan Geuter"], "alternates": {"html": "https://wpnews.pro/news/unlocking-lossless-speedups-in-llms-via-discrete-diffusion-5000-tk-s", "markdown": "https://wpnews.pro/news/unlocking-lossless-speedups-in-llms-via-discrete-diffusion-5000-tk-s.md", "text": "https://wpnews.pro/news/unlocking-lossless-speedups-in-llms-via-discrete-diffusion-5000-tk-s.txt", "jsonld": "https://wpnews.pro/news/unlocking-lossless-speedups-in-llms-via-discrete-diffusion-5000-tk-s.jsonld"}}