相关阅读
Research
Garry Tan: Yes, this result cost millions of dollars.Research
Rohan Paul: Uno shows a simple way to speed up existing LLMs without changing their output distribution: keep the original model in charge, and use diffusion only to draft multipl…Research
Rohan Paul: – https://arxiv.org/abs/2609.04010 Title: "Unlocking Lossless Speedups in LLMs via Discrete Diffusion"