ai

Improving the lower bound for the unit distance problem

The modded nanogpt speedrun, but in JAX and on TPUs

Theoretical properties of optimizers on a toy problem, and some intuition

Deriving RoPE the proper way

Quantizing LLMs for inference