Mixture-of-Kittens: An MoE training megakernel for NVL72

Post Log inSign up Post Cursor @cursor_ai We're open-sourcing Mixture-of-Kittens (MoK), our MoE training megakernel for NVL72s. It fuses all Mixture-of-Experts communication and computation into a single, fully deterministic kernel, and runs up to 2.37x faster than the strongest public baselines. 4:00 PM · Aug 4, 2026157.8KViews Cursor @cursor_ai 3h MoK now powers training across tens of thousands of GPUs at Cursor. In production, it raised end-to-end training throughput by 1.41x over our previous DeepEP-based stack. 16K Cursor @cursor_ai 3h Our hope is that this lowers the barrier to AI research, so more labs can train models efficiently. Here's how we built it: Mixture-of-Kittens: our open-source MoE megakernel for NVL72s · Cursor From cursor.com 15K Lee Robinson @leerob 3h The real mixture of kittens 7.3K