Gympedia is a collection of agentic LLM trajectories and distilled skills gathered across the diverse task environments of GEM: A Gym for Agentic LLMs. It pairs raw agent rollouts (per task, per model) with human-/model-readable "skill" write-ups distilled from successful runs, so it can be used for behavior analysis, skill/knowledge distillation, SFT data curation, and agentic RL research.
Note: This release intentionally excludes trajectories collected in synthetic… See the full description on the dataset page:
https://huggingface.co/datasets/Septzzz/Gympedia.
