FuTransformer: FP8 prefill and INT4 decode for Llama-style inference on Hopper

Sign in to join the discussion — your thebay.events account works here.