arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.05935cs.CVcs.AIcs.CLcs.LG
收藏

SPHINX-X:面向多模态大语言模型家族的数据与参数扩展

SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models

Shanghai AI Laboratory(上海人工智能实验室);CUHK(香港中文大学);University of California, Los Angeles(加利福尼亚大学洛杉矶分校);Rutgers University(罗格斯大学);Centre for Perceptual and Interactive Intelligence (CPII)(感知与交互智能中心(CPII))
Dongyang Liu, Renrui Zhang, Longtian Qiu, Siyuan Huang, Weifeng Lin, Shitian Zhao, Shijie Geng, Ziyi Lin, Peng Jin, Kaipeng Zhang, Wenqi Shao, Chao Xu, Conghui … 展开作者

Dongyang Liu, Renrui Zhang, Longtian Qiu, Siyuan Huang, Weifeng Lin, Shitian Zhao, Shijie Geng, Ziyi Lin, Peng Jin, Kaipeng Zhang, Wenqi Shao, Chao Xu, Conghui He, Junjun He, Hao Shao, Pan Lu, Hongsheng Li, Yu Qiao, Peng Gao

AI总结:

本研究提出多模态大模型系列SPHINX-X,通过优化框架、构建多领域数据集并基于不同规模基础LLM训练,验证了多模态性能与数据、参数规模的强相关性。

摘要:

我们提出SPHINX-X,这是基于SPHINX开发的一个规模庞大的多模态大语言模型(Multimodality Large Language Model, MLLM)系列。为提升架构与训练效率,我们对SPHINX框架进行了修改:移除冗余视觉编码器、通过跳跃令牌跳过全填充子图像,并将多阶段训练简化为单阶段一体化范式。为充分释放MLLM的潜力,我们构建了一个全面的多领域多模态数据集,涵盖语言、视觉及视觉-语言任务的公开可用资源;我们还通过自行整理的OCR密集型数据集和Set-of-Mark数据集进一步丰富该集合,拓展了多样性与通用性。通过在TinyLlama1.1B、InternLM2-7B、LLaMA2-13B和Mixtral8x7B等不同基础大语言模型上训练,我们得到了一系列参数规模和多语言能力各异的MLLM。综合基准测试表明,多模态性能与数据规模、参数规模之间存在强相关性。代码与模型已发布于https://github.com/Alpha-VLLM/LLaMA2-Accessory。

英文摘要:

We propose SPHINX-X, an extensive Multimodality Large Language Model (MLLM) series developed upon SPHINX. To improve the architecture and training efficiency, we modify the SPHINX framework by removing redundant visual encoders, bypassing fully-padded sub-images with skip tokens, and simplifying multi-stage training into a one-stage all-in-one paradigm. To fully unleash the potential of MLLMs, we assemble a comprehensive multi-domain and multimodal dataset covering publicly available resources in language, vision, and vision-language tasks. We further enrich this collection with our curated OCR intensive and Set-of-Mark datasets, extending the diversity and generality. By training over different base LLMs including TinyLlama1.1B, InternLM2-7B, LLaMA2-13B, and Mixtral8x7B, we obtain a spectrum of MLLMs that vary in parameter size and multilingual capabilities. Comprehensive benchmarking reveals a strong correlation between the multi-modal performance with the data and parameter scales. Code and models are released at https://github.com/Alpha-VLLM/LLaMA2-Accessory

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑