Flappy describes Kittens Virtual Machine for packing GPU workloads into one kernel

According to kushal1t, Flappy's Kittens Virtual Machine (KVM) launches a megakernel that serves as a persistent runtime with its own instruction-based programming model. The post claims it increases GPU utilization by packing independent workloads into the same kernel and achieves state-of-the-art speeds without hand-tuned kernels. It gives no benchmarks or release terms.