Skip to content

Topic

GPU throughput optimization

Engineering work that raises tokens processed per second on given hardware, including reduced-precision arithmetic, kernel selection and padding tensor shapes to hardware-friendly sizes.

Current clusters