Hacker News new | ask | show | jobs
Cursor: 1.5x Faster Moe Training on Blackwell with MXFP8 Kernels (cursor.com)
2 points by tmm1 300 days ago