Dr.ICL:Demonstration-Retrieved In-context Learning  

作　　者：Man Luo Xin Xu Zhuyun Dai Panupong Pasupat Mehran Kazemi Chitta Baral Vaiva Imbrasaite Vincent Y Zhao 

机构地区：[1]Ringgold Standard Institution,Arizona State University,Tempe,Arizona 85287,USA [2]Google LLC Ringgold Standard Institution,Mountain View,California 94043,USA

出　　处：《Data Intelligence》2024年第4期909-922,共14页数据智能(英文)

摘　　要：In-context learning(ICL), which teaches a large language model(LLM) to perform a task with few-shot demonstrations rather than adjusting the model parameters, has emerged as a strong paradigm for using LLMs. While early studies primarily used a fixed or random set of demonstrations for all test queries, recent research suggests that retrieving semantically similar demonstrations to the input from a pool of available demonstrations results in better performance. This work expands the applicability of retrieval-based ICL approaches along several dimensions. We extend the success of retrieval-based ICL to instructionfinetuned LLMs as well as Chain-of-Thought(CoT) prompting. While the prior work utilizes general Large Language Models(LLMs), such as GPT-3, we find that retrieved demonstrations also enhance instructionfinetuned LLMs. This insight implies that training data, despite being exposed during the fine-tuning phase, can still be effectively used through retrieval and in-context demonstrations during testing, resulting in superior outcomes when compared to utilizing no demonstrations or selecting them at random. For CoT, when the demonstrations contain reasoning chains, we get improvements by retrieving based on such chains. Finally, we train a task-specific demonstration retriever that outperforms off-the-shelf retrievers.

关 键 词：Information retrieval In-context learning Large language models Retrieval augmented generation Large language model reasoning 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.3[自动化与计算机技术—控制科学与工程]