arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2309.16058cs.LGcs.CLcs.CV

AnyMAL:高效且可扩展的任意模态增强语言模型

AnyMAL: An Efficient and Scalable Any-Modality Augmented Language Model

FAIR, Meta(元宇宙公司基础人工智能研究院);Meta Reality Labs(元宇宙公司现实实验室)
Seungwhan Moon, Andrea Madotto, Zhaojiang Lin, Tushar Nagarajan, Matt Smith, Shashank Jain, Chun-Fu Yeh, Prakash Murugesan, Peyman Heidari, Yue Liu, Kavya Srine… 展开作者

Seungwhan Moon, Andrea Madotto, Zhaojiang Lin, Tushar Nagarajan, Matt Smith, Shashank Jain, Chun-Fu Yeh, Prakash Murugesan, Peyman Heidari, Yue Liu, Kavya Srinet, Babak Damavandi, Anuj Kumar

AI总结:

本文提出统一的任意模态增强语言模型 AnyMAL,通过预训练对齐器将图像、视频、音频、IMU 等信号映射到 LLM 文本空间,并结合多模态指令微调,在多项任务上达到最先进性能。

摘要:

我们提出了任意模态增强语言模型(Any-Modality Augmented Language Model,AnyMAL),这是一个统一模型,能够对多种输入模态信号(即文本、图像、视频、音频和 IMU 运动传感器)进行推理,并生成文本响应。AnyMAL 继承了包括 LLaMA-2(70B)在内的最先进 LLM 强大的文本推理能力,并通过预训练的对齐器模块将特定模态信号转换到联合文本空间。为进一步增强多模态 LLM 的能力,我们使用人工收集的多模态指令集对模型进行微调,该指令集涵盖简单问答之外的多样主题和任务。我们开展了包含人工评估和自动评估的综合实证分析,并证明其在各类多模态任务上取得最先进性能。

英文摘要:

We present Any-Modality Augmented Language Model (AnyMAL), a unified model that reasons over diverse input modality signals (i.e. text, image, video, audio, IMU motion sensor), and generates textual responses. AnyMAL inherits the powerful text-based reasoning abilities of the state-of-the-art LLMs including LLaMA-2 (70B), and converts modality-specific signals to the joint textual space through a pre-trained aligner module. To further strengthen the multimodal LLM's capabilities, we fine-tune the model with a multimodal instruction set manually collected to cover diverse topics and tasks beyond simple QAs. We conduct comprehensive empirical analysis comprising both human and automatic evaluations, and demonstrate state-of-the-art performance on various multimodal tasks.

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。

↑