Lossless LLM compression for efficient GPU inference via dynamic-length float

Posted 15 hours ago by CharlesW
341 points

https://arxiv.org/abs/2504.11651

22 comments

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..