FASTER MOE INFERENCE WITH WARP DECODE

Summary from the TLDR AI newsletter

Cursor's “warp decode” is a kernel design that reorganizes MoE inference around output neurons instead of experts. It achieves ~1.8x higher throughput and improved numerical accuracy on Blackwell GPUs.