arXiv 2409.201812024-11-20cs.CL

参考可信解码：一种面向大语言模型的免训练增强范式

Reference Trustable Decoding: A Training-Free Augmentation Paradigm for Large Language Models

National Engineering Research Center for Multimedia Software（国家多媒体软件工程技术研究中心）；School of Computer Science, Wuhan University（武汉大学计算机学院）；Department of Computer Science and Engineering, Shanghai Jiao Tong University（上海交通大学计算机科学与工程系）

Luohe Shi, Yao Yao, Zuchao Li, Lefei Zhang, Hai Zhao

备注：: Accepted by the Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS 2024)

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出参考可信解码（RTD），一种免训练的LLM增强范式，通过从训练示例构建参考数据存储并在推理时灵活选择参考来优化词表分布，使模型以低推理成本快速适应下游任务，且可与传统方法正交并用。

摘要：

大语言模型（LLMs）近年来快速发展，展现出令人瞩目的能力。上下文学习（In-Context Learning，ICL）和参数高效微调（Parameter-Efficient Fine-Tuning，PEFT）是目前将LLMs增强到下游任务的两种主流方法。ICL通常通过人工构造或搭建检索增强生成（Retrieval-Augmented Generation，RAG）系统来构建少样本学习场景，在不改变模型参数的情况下帮助模型快速掌握领域知识或问答模式。然而，这种方法存在权衡代价，例如推理速度变慢和空间占用增加。PEFT通过极少的参数修改帮助模型适应任务，但即使涉及的参数量很小，训练过程仍然对硬件有较高要求。为了解决这些挑战，我们提出了参考可信解码（Reference Trustable Decoding，RTD），这是一种让模型无需微调即可快速适应新任务、同时保持低推理成本的范式。RTD从提供的训练示例中构建参考数据存储，并根据输入灵活选择合适参考来优化LLM的最终词表分布，从而产生更可信的响应，使模型能够以低成本适应下游任务。在不同基准上对多种LLMs进行的实验评估表明，RTD为将模型增强到下游任务建立了一种新范式。此外，我们的方法与传统方法表现出很强的正交性，可以同时使用。我们的代码可在 https://github.com/ShiLuohe/ReferenceTrustableDecoding 获取。

英文摘要：

Large language models (LLMs) have rapidly advanced and demonstrated impressive capabilities. In-Context Learning (ICL) and Parameter-Efficient Fine-Tuning (PEFT) are currently two mainstream methods for augmenting LLMs to downstream tasks. ICL typically constructs a few-shot learning scenario, either manually or by setting up a Retrieval-Augmented Generation (RAG) system, helping models quickly grasp domain knowledge or question-answering patterns without changing model parameters. However, this approach involves trade-offs, such as slower inference speed and increased space occupancy. PEFT assists the model in adapting to tasks through minimal parameter modifications, but the training process still demands high hardware requirements, even with a small number of parameters involved. To address these challenges, we propose Reference Trustable Decoding (RTD), a paradigm that allows models to quickly adapt to new tasks without fine-tuning, maintaining low inference costs. RTD constructs a reference datastore from the provided training examples and optimizes the LLM's final vocabulary distribution by flexibly selecting suitable references based on the input, resulting in more trustable responses and enabling the model to adapt to downstream tasks at a low cost. Experimental evaluations on various LLMs using different benchmarks demonstrate that RTD establishes a new paradigm for augmenting models to downstream tasks. Furthermore, our method exhibits strong orthogonality with traditional methods, allowing for concurrent usage. Our code can be found at https://github.com/ShiLuohe/ReferenceTrustableDecoding

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。