Jaxcalibur | How Jaxcalibur works: an 8.5M-parameter transformer trained with self-play reinforcement learning, on a JAX reimplementation of the Pokémon battle engine, with pUCT search adapted for simultaneous moves and hidden information.
jaxcalibur.github.io · saved by 4 readers
How Jaxcalibur works: an 8.5M-parameter transformer trained with self-play reinforcement learning, on a JAX reimplementation of the Pokémon battle engine, with pUCT search adapted for simultaneous moves and hidden information.