sectionYPositions = computeSectionYPositions($el), 10)"
x-init="setTimeout(() => sectionYPositions = computeSectionYPositions($el), 10)"
>
On This Page
Reinforcement learning on top of Owl.
The package has three libraries:
owl-rllib (Owl_rllib): MDPs, tabular methods and dynamic programming; classic control environments; DQN, REINFORCE, A2C and PPO.owl-rllib.llm (Owl_rllib_llm): RL for language models on a toy model (REINFORCE, RLOO, token-level PPO, GRPO and its variants).owl-rllib.bridges (Owl_rllib_bridges): Gymnasium and VizDoom environments, run in a Python child process.Owl_rllib.Ndenv (array observations), Owl_rllib.Env (tabular), Owl_rllib.Vec_env; Owl_rllib.Cartpole and the other classic control tasks.Owl_rllib.Dp, Owl_rllib.Tabular, Owl_rllib.Dqn, Owl_rllib.Reinforce, Owl_rllib.A2c, Owl_rllib.Ppo.Owl_rllib.Neural, Owl_rllib.Policy.Owl_rllib.Rng.Most modules are functors over Owl's single (S) or double (D) precision, with S and D instances provided. The programs in examples/ show each part in use.