swimbird-switchable-reasoning

【效率与协作】Enable multimodal models to dynamically switch between text and vision reasoning modes, allocating computation based on perceived difficulty and image resolution, achieving strong performance on both vision-dense and text-heavy benchmarks.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准