ROLV

ROLV is a sparse compute primitive that accelerates MoE and dense AI inference on any hardware — NVIDIA, AMD, Intel, TPU, Apple Silicon. 20.7× faster throughput and 177× faster time-to-first-token on real Llama 4 Maverick weights, hash-verified. No model retraining or hardware changes required.