How to Generate Text in One Step
one-step-lm.github.io · 3,983 words · saved by 1 readers
Flow Map Language Models: One-step Language Modeling via Continuous Denoising.
How to Generate Text in One Step ← Back to project page How to Generate Text in One Step Discrete diffusion language models promise fast parallel text generation, but they fall apart with few sampling steps. We show how continuous flows fix this, enabling coherent, high-quality text generation in a single forward pass of the model. Based on Flow Map Language Models: One-step Language Modeling via Continuous Denoising by Chanhyuk Lee, Jaehoon Yoo, Manan Agarwal, Sheel Shah, Jerry Huang, Aditi Raghunathan, Seunghoon Hong, Nicholas M. Boffi † , and Jinwoo Kim † . Authors Jinwoo
saved by
related reading
- What are Diffusion Models? | Lil'Loglilianweng.github.io
- Kuleshov Group | How to Build a Diffusion Language Modelkuleshov-group.github.io
- 6d13e085b79d454da5910e4ca82a3d9d-Paper-Conference.pdfproceedings.neurips.cc
- 2409.02908arxiv.org
- What are Diffusion Models?lilianweng.github.io
- Large Language Diffusion Modelsarxiv.org
- Learning the integral of a diffusion model – Sander Dielemansander.ai
- Esoteric Language Modelsarxiv.org
- Flow Matching for Generative Modelingalphaxiv.org
- 2503.09573arxiv.org
- ⭐️ Diffusion Modelsandrewkchan.dev
- Generative Modeling via Driftingarxiv.org