Here we provide a pure-text version of GameQA, encompassing some appropriate games. (See
https://github.com/tongjingqi/Code2Logic/issues/2)
This is the first work, to the best of our knowledge, that leverages game code to synthesize multimodal reasoning data for training VLMs. Furthermore, when trained with a GRPO strategy solely on GameQA (synthesized via our proposed Code2Logic approach), multiple… See the full description on the dataset page:
https://huggingface.co/datasets/OpenMOSS-Team/GameQA-text.
