arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.06252cs.LGcs.AIcs.CL
收藏

Transformer-Squared:自适应大语言模型

Transformer-Squared: Self-adaptive LLMs

Sakana AI(萨卡纳人工智能公司);Institute of Science Tokyo(东京科学大学)

Qi Sun, Edoardo Cetin, Yujin Tang

AI总结:

本文提出Transformer-Squared,一种通过选择性调整权重奇异分量并利用强化学习训练专家向量动态混合的自适应框架,以高效实时适应新任务,优于LoRA并支持多模态。

摘要:

自适应大语言模型(LLMs)旨在解决传统微调方法所面临的挑战,这些方法通常计算密集且在处理多样化任务时能力静态。我们提出了Transformer-Squared,一种新颖的自适应框架,通过选择性地仅调整其权重矩阵的奇异分量,实时适应未见过的任务。在推理过程中,Transformer-Squared采用两遍机制:首先,调度系统识别任务属性,然后,使用强化学习训练的任务特定“专家”向量被动态混合,以获得针对输入提示的目标行为。我们的方法在参数更少、效率更高的情况下,持续优于如LoRA等普遍方法。此外,Transformer-Squared在不同LLM架构和模态(包括视觉-语言任务)中展示了多功能性。Transformer-Squared代表了重大进步,提供了一种可扩展、高效的解决方案,以增强LLM的适应性和任务特定性能,为真正动态、自组织的AI系统铺平道路。

英文摘要:

Self-adaptive large language models (LLMs) aim to solve the challenges posed by traditional fine-tuning methods, which are often computationally intensive and static in their ability to handle diverse tasks. We introduce Transformer-Squared, a novel self-adaptation framework that adapts LLMs for unseen tasks in real-time by selectively adjusting only the singular components of their weight matrices. During inference, Transformer-Squared employs a two-pass mechanism: first, a dispatch system identifies the task properties, and then task-specific 'expert' vectors, trained using reinforcement learning, are dynamically mixed to obtain targeted behavior for the incoming prompt. Our method consistently outperforms ubiquitous approaches such as LoRA, with fewer parameters and greater efficiency. Furthermore, Transformer-Squared demonstrates versatility across different LLM architectures and modalities, including vision-language tasks. Transformer-Squared represents a significant leap forward, offering a scalable, efficient solution for enhancing the adaptability and task-specific performance of LLMs, paving the way for truly dynamic, self-organizing AI systems.

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。

↑