new research: A tiny 109M-parameter model outperforms models up to 72B parameters at image-text alignment, using 42× less training data

🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1xkzony/new_research_a_tiny_109mparameter_model/
🏷️ 分组: Reddit|AI·自托管·VPS
🕒 时间: 2026-11-01 18:24:21 From the LocalLLaMA community on Reddit: new research: A tiny 109M-parameter model outperforms models up to 72B parameters at image…