Adding NVMe SSDs to Enable and Accelerate 100B Model Fine-Tuning on a Single GPU

Also discussed here: https://news.ycombinator.com/item?id=39691144