On this day: May 29

Real measurements showing hardware-native FP4 has not yet beaten community quantization. The format race is unresolved

A head-on rebuttal of the FP8 training narrative. Documented that reports that it works coexist with reports that it cannot be reproduced

Introduced a non-production license that publishes weights for research and testing only, with commercial use requiring a separate license

Optimizes directly on preference pairs without a reward model or PPO, popularizing open-source alignment

A format that removed FP16's scaling burden. Later the de facto default training precision for the TPU, Ampere and Hopper generations