reinforcement-learning

Minimal LLM post-training experiments on a single 8GB GPU, covering SFT, DPO, and GRPO with a focus on forgetting, seed variance, and RL behavior.

A benchmark page for frontier models racing to write a StarCraft: Brood War bot in C++ using BWAPI and OpenBW, climbing tiers of increasingly strong Protoss opponents.

Meta AI research on RL-XAR, a method for training language models toward expert-level writing using expert-aligned rubrics and reinforcement learning. The page explains the problem, the rubric-learning loop, and experiments on paper sections, story writing, and Wikipedia-style text.

A small post-trained Qwen2.5-0.5B-Instruct model tuned to write in Gen Z slang, with training and inference notebooks plus dataset files in the repo.

Market-derived RL environments for training agents on quant trading and long-horizon planning under adversarial noise.

Agora-1 is a multi-agent world model from Odyssey that simulates shared real-time environments for up to four participants, human or AI, with a focus on gaming, robotics, reinforcement learning, and foundation model research.

A paper page about FrogNano, a 4B coding agent trained with RL on roughly 1,500 SWE environments using synthetic tasks and no distillation from a larger model.

A blog post describing a small reinforcement-learning agent trained with PPO to play and beat a Pokelike/Pokerogue-style game, including the input representation, model architecture, and training approach.

An educational article explaining world models, latent states, dynamics learning, and planning for agents, with examples from gridworld, Dreamer, and MuZero.

arXiv paper on self-speculating LLM agents that predict their next tool call to hide tool latency and improve next-call Hit@1 while preserving task success.