Ultra fast reinforcement learning in accelerator control demonstrated on CERN AWAKE in IPAC'23 - 14th International Particle Accelerator Conference