Search engines often follow a pipeline architecture, where complex but effective reranking components are used to refine the results of an initial retrieval. Retrieval augmented generation (RAG) is an exciting application of the pipeline architecture, where the final component generates a coherent answer for the users from the retrieved documents. In this demo paper, we describe how such RAG pipelines can be formulated in the declarative PyTerrier architecture, and the advantages of doing so. Our PyTerrier-RAG extension for PyTerrier provides easy access to standard RAG datasets and evaluation measures, state-of-the-art LLM readers, and using PyTerrier's unique operator notation, easy-to-build pipelines. We demonstrate the succinctness of indexing and RAG pipelines on standard datasets (including Natural Questions) and how to build on the larger PyTerrier ecosystem with state-of-the-art sparse, learned-sparse, and dense retrievers, and other neural rankers.
Retrieval-augmented generation (RAG) has seen many empirical successes in recent years by aiding the LLM with external knowledge. However, its theoretical aspect has remained mostly unexplored. In this paper, we propose the first finite-sample generalization bound for RAG in in-context linear regression and derive an exact bias-variance tradeoff. Our framework views the retrieved texts as query-dependent noisy in-context examples and recovers the classical in-context learning (ICL) and standard RAG as the limit cases. Our analysis suggests that an intrinsic ceiling on generalization error exists on RAG as opposed to the ICL. Furthermore, our framework is able to model retrieval both from the training data and from external corpora by introducing uniform and non-uniform RAG noise. In line with our theory, we show the sample efficiency of ICL and RAG empirically with experiments on common QA benchmarks, such as Natural Questions and TriviaQA.
Retrieval-Augmented Generation (RAG), which integrates external knowledge into Large Language Models (LLMs), has proven effective in enabling LLMs to produce more accurate and reliable responses. However, it remains a significant challenge how to effectively integrate external retrieved knowledge with internal parametric knowledge in LLMs. In this work, we propose a novel Self-Selection RAG framework, where the LLM is made to select from pairwise responses generated with internal parametric knowledge solely and with external retrieved knowledge together to achieve enhanced accuracy. To this end, we devise a Self-Selection-RGP method to enhance the capabilities of the LLM in both generating and selecting the correct answer, by training the LLM with Direct Preference Optimization (DPO) over a curated Retrieval Generation Preference (RGP) dataset. Experimental results with two open-source LLMs (i.e., Llama2-13B-Chat and Mistral-7B) well demonstrate the superiority of our approach over other baseline methods on Natural Questions (NQ) and TrivialQA datasets.
Table Question Answering (TQA) involves retrieving relevant tables from a large corpus to answer natural language queries. Traditional dense retrieval models, such as DTR and ColBERT, not only incur high computational costs for large-scale retrieval tasks but also require retraining or fine-tuning on new datasets, limiting their adaptability to evolving domains and knowledge. In this work, we propose $\textbf{CRAFT}$, a cascaded retrieval approach that first uses a sparse retrieval model to filter a subset of candidate tables before applying more computationally expensive dense models and neural re-rankers. Our approach achieves better retrieval performance than state-of-the-art (SOTA) sparse, dense, and hybrid retrievers. We further enhance table representations by generating table descriptions and titles using Gemini Flash 1.5. End-to-end TQA results using various Large Language Models (LLMs) on NQ-Tables, a subset of the Natural Questions Dataset, demonstrate $\textbf{CRAFT}$ effectiveness.
Retrieval-Augmented Generation (RAG) systems, widely used to improve the factual grounding of large language models (LLMs), are increasingly vulnerable to poisoning attacks, where adversaries inject manipulated content into the retriever's corpus. While prior research has predominantly focused on single-attacker settings, real-world scenarios often involve multiple, competing attackers with conflicting objectives. In this work, we introduce PoisonArena, the first benchmark to systematically study and evaluate competing poisoning attacks in RAG. We formalize the multi-attacker threat model, where attackers vie to control the answer to the same query using mutually exclusive misinformation. PoisonArena leverages the Bradley-Terry model to quantify each method's competitive effectiveness in such adversarial environments. Through extensive experiments on the Natural Questions and MS MARCO datasets, we demonstrate that many attack strategies successful in isolation fail under competitive pressure. Our findings highlight the limitations of conventional evaluation metrics like Attack Success Rate (ASR) and F1 score and underscore the need for competitive evaluation to assess real-world attack robustness. PoisonArena provides a standardized framework to benchmark and develop future attack and defense strategies under more realistic, multi-adversary conditions. Project page: https://github.com/yxf203/PoisonArena.
Dense retrievers utilize pre-trained backbone language models (e.g., BERT, LLaMA) that are fine-tuned via contrastive learning to perform the task of encoding text into sense representations that can be then compared via a shallow similarity operation, e.g. inner product. Recent research has questioned the role of fine-tuning vs. that of pre-training within dense retrievers, specifically arguing that retrieval knowledge is primarily gained during pre-training, meaning knowledge not acquired during pre-training cannot be sub-sequentially acquired via fine-tuning. We revisit this idea here as the claim was only studied in the context of a BERT-based encoder using DPR as representative dense retriever. We extend the previous analysis by testing other representation approaches (comparing the use of CLS tokens with that of mean pooling), backbone architectures (encoder-only BERT vs. decoder-only LLaMA), and additional datasets (MSMARCO in addition to Natural Questions). Our study confirms that in DPR tuning, pre-trained knowledge underpins retrieval performance, with fine-tuning primarily adjusting neuron activation rather than reorganizing knowledge. However, this pattern does not hold universally, such as in mean-pooled (Contriever) and decoder-based (LLaMA) models. We ensure full reproducibility and make our implementation publicly available at https://github.com/ielab/DenseRetriever-Knowledge-Acquisition.
Retrieval Augmented Generation (RAG) frameworks improve the accuracy of large language models (LLMs) by integrating external knowledge from retrieved documents, thereby overcoming the limitations of models' static intrinsic knowledge. However, these systems are susceptible to adversarial attacks that manipulate the retrieval process by introducing documents that are adversarial yet semantically similar to the query. Notably, while these adversarial documents resemble the query, they exhibit weak similarity to benign documents in the retrieval set. Thus, we propose a simple yet effective Graph-based Reranking against Adversarial Document Attacks (GRADA) framework aiming at preserving retrieval quality while significantly reducing the success of adversaries. Our study evaluates the effectiveness of our approach through experiments conducted on five LLMs: GPT-3.5-Turbo, GPT-4o, Llama3.1-8b, Llama3.1-70b, and Qwen2.5-7b. We use three datasets to assess performance, with results from the Natural Questions dataset demonstrating up to an 80% reduction in attack success rates while maintaining minimal loss in accuracy.
Chain-of-thought prompting has demonstrated great success in facilitating the reasoning abilities of large language models. In this work, we explore how these enhanced reasoning abilities can be exploited to improve the robustness of large language models in tasks that are not necessarily reasoning-focused. In particular, we show how a wide range of large language models exhibit significantly improved robustness against reference corruption using a simple method called chain-of-defensive-thought, where only a few exemplars with structured and defensive reasoning are provided as demonstrations. Empirically, the improvements can be astounding, especially given the simplicity and applicability of the method. For example, in the Natural Questions task, the accuracy of GPT-4o degrades from 60% to as low as 3% with standard prompting when 1 out of 10 references provided is corrupted with prompt injection attacks. In contrast, GPT-4o using chain-of-defensive-thought prompting maintains an accuracy of 50%.
We introduce S1-Bench, a novel benchmark designed to evaluate the performance of Large Reasoning Models (LRMs) on simple tasks that favor intuitive system 1 thinking rather than deliberative system 2 reasoning. While LRMs have achieved significant breakthroughs in complex reasoning tasks through explicit chains of thought, their heavy reliance on system 2 thinking may limit their system 1 thinking capabilities. However, there is a lack of an appropriate benchmark for evaluating LRM's system 1 thinking capabilities. To fill this gap, S1-Bench introduces a suite of simple, diverse, and natural questions across multiple domains and languages, specifically designed to assess LRMs' performance on questions more suitable for system 1 . We conduct extensive evaluations across 28 LRMs, revealing their inefficiency, inadequate accuracy, and limited robustness when handling simple questions. Additionally, we observe a gap between their difficulty perception and generation length. Overall, this work paves the way toward dual-system compatibility in the development of LRMs.
Generative information retrieval (GenIR) is a promising neural retrieval paradigm that formulates document retrieval as a document identifier (docid) generation task, allowing for end-to-end optimization toward a unified global retrieval objective. However, existing GenIR models suffer from token-level misalignment, where models trained to predict the next token often fail to capture document-level relevance effectively. While reinforcement learning-based methods, such as reinforcement learning from relevance feedback (RLRF), aim to address this misalignment through reward modeling, they introduce significant complexity, requiring the optimization of an auxiliary reward function followed by reinforcement fine-tuning, which is computationally expensive and often unstable. To address these challenges, we propose direct document relevance optimization (DDRO), which aligns token-level docid generation with document-level relevance estimation through direct optimization via pairwise ranking, eliminating the need for explicit reward modeling and reinforcement learning. Experimental results on benchmark datasets, including MS MARCO document and Natural Questions, show that DDRO outperforms reinforcement learning-based methods, achieving a 7.4% improvement in MRR@10 for MS MARCO and a 19.9% improvement for Natural Questions. These findings highlight DDRO's potential to enhance retrieval effectiveness with a simplified optimization approach. By framing alignment as a direct optimization problem, DDRO simplifies the ranking optimization pipeline of GenIR models while offering a viable alternative to reinforcement learning-based methods.
Large Language Models (LLMs) excel across a variety of language tasks yet are constrained by limited input lengths and high computational costs. Existing approaches\textemdash such as relative positional encodings (e.g., RoPE, ALiBi) and sliding window mechanisms\textemdash partially alleviate these issues but often require additional training or suffer from performance degradation with longer inputs. In this paper, we introduce \textbf{\textit{CacheFocus}}, a method that enhances length normalization and reduces inference latency without any further training. Our approach leverages query-independent, offline caching to efficiently reuse a Context KV Cache Store. We address the amplification of abnormal token distributions problem by re-positioning cached keys and introducing Layer-Adaptive Cache Pruning to discard low-relevance caches during pre-filling. Additionally, our Adaptive Positional Allocation Strategy dynamically reassigns cache positions to maximize the use of the available positional encoding range. Experiments on the Natural Questions and TriviaQA datasets demonstrate that CacheFocus outperforms alternative methods even when inputs exceed the $4$K limit of the \texttt{LLaMA-2} model, emphasizing its practical effectiveness for long-context LLMs. Moreover, even with large maximum input length of \texttt{Qwen2}, the performance of CacheFocus shows that it maintains consistent performance even as the number of documents increases, effectively managing long-text generation without degradation.
Large language models (LLMs) exhibit impressive performance across diverse tasks but often struggle to accurately gauge their knowledge boundaries, leading to confident yet incorrect responses. This paper explores leveraging LLMs' internal states to enhance their perception of knowledge boundaries from efficiency and risk perspectives. We investigate whether LLMs can estimate their confidence using internal states before response generation, potentially saving computational resources. Our experiments on datasets like Natural Questions, HotpotQA, and MMLU reveal that LLMs demonstrate significant pre-generation perception, which is further refined post-generation, with perception gaps remaining stable across varying conditions. To mitigate risks in critical domains, we introduce Consistency-based Confidence Calibration ($C^3$), which assesses confidence consistency through question reformulation. $C^3$ significantly improves LLMs' ability to recognize their knowledge gaps, enhancing the unknown perception rate by 5.6\% on NQ and 4.9\% on HotpotQA. Our findings suggest that pre-generation confidence estimation can optimize efficiency, while $C^3$ effectively controls output risks, advancing the reliability of LLMs in practical applications.
This paper proposes a novel approach to develop an open-domain and long-form Over-The-Top (OTT) Question-Answering (QA) dataset, DragonVerseQA, specifically oriented to the fantasy universe of "House of the Dragon" and "Game Of Thrones" TV series. Most existing QA datasets focus on short, fact-based answers sourced almost solely from Wikipedia articles, devoid of depth and contextual richness for sophisticated narrative understanding. We curate a dataset that combines full episode summaries sourced from HBO and fandom wiki websites, user reviews from sources like IMDb and Rotten Tomatoes, and high-quality, open-domain, legally admissible sources, and structured data from repositories like WikiData into one dataset. The dataset provides a multi-dimensional context, reflecting complex character dynamics and plot developments from these varied sources. That means, on equal footing, only after heavy data preprocessing and filtering methods will meaningful, non-spam unbiased reviews be available in this enriched dataset. The comprehensive insights are given through the long-form answers generated from this enriched context. This is what makes this valuable dataset for improving conversational AI, narrative analysis, sentiment analysis, summarization techniques, and relation extraction. A comparative analysis with state-of-the-art QA datasets such as SQuAD 2.0, TriviaQA, and Natural Questions brings to light the unique advantages of our dataset in terms of contextual complexity and answer length. Detailed reviews add layers to audience sentiment and narrative interpretation, raising the bar for domain-specific QA with a new quality benchmark. Our work also allows a deeper understanding of entertainment-industry content and opens the door to more knowledgeable and creative AI-driven interactions within digital media environments.
This study aims to optimize the existing retrieval-augmented generation model (RAG) by introducing a graph structure to improve the performance of the model in dealing with complex knowledge reasoning tasks. The traditional RAG model has the problem of insufficient processing efficiency when facing complex graph structure information (such as knowledge graphs, hierarchical relationships, etc.), which affects the quality and consistency of the generated results. This study proposes a scheme to process graph structure data by combining graph neural network (GNN), so that the model can capture the complex relationship between entities, thereby improving the knowledge consistency and reasoning ability of the generated text. The experiment used the Natural Questions (NQ) dataset and compared it with multiple existing generation models. The results show that the graph-based RAG model proposed in this paper is superior to the traditional generation model in terms of quality, knowledge consistency, and reasoning ability, especially when dealing with tasks that require multi-dimensional reasoning. Through the combination of the enhancement of the retrieval module and the graph neural network, the model in this study can better handle complex knowledge background information and has broad potential value in multiple practical application scenarios.
Retrieval-augmented generation (RAG) has become integral to large language models (LLMs), particularly for conversational AI systems where user questions may reference knowledge beyond the LLMs' training cutoff. However, many natural user questions lack well-defined answers, either due to limited domain knowledge or because the retrieval system returns documents that are relevant in appearance but uninformative in content. In such cases, LLMs often produce hallucinated answers without flagging them. While recent work has largely focused on questions with false premises, we study out-of-scope questions, where the retrieved document appears semantically similar to the question but lacks the necessary information to answer it. In this paper, we propose a guided hallucination-based approach ELOQ to automatically generate a diverse set of out-of-scope questions from post-cutoff documents, followed by human verification to ensure quality. We use this dataset to evaluate several LLMs on their ability to detect out-of-scope questions and generate appropriate responses. Finally, we introduce an improved detection method that enhances the reliability of LLM-based question-answering systems in handling out-of-scope questions.
This paper introduces a novel task to assess the faithfulness of large language models (LLMs) using local perturbations and self-explanations. Many LLMs often require additional context to answer certain questions correctly. For this purpose, we propose a new efficient alternative explainability technique, inspired by the commonly used leave-one-out approach. Using this approach, we identify the sufficient and necessary parts for the LLM to generate correct answers, serving as explanations. We propose a metric for assessing faithfulness that compares these crucial parts with the self-explanations of the model. Using the Natural Questions dataset, we validate our approach, demonstrating its effectiveness in explaining model decisions and assessing faithfulness.
Retrieval-augmented generation (RAG) improves Large Language Models (LLMs) by incorporating external information into the response generation process. However, how context-faithful LLMs are and what factors influence LLMs' context-faithfulness remain largely unexplored. In this study, we investigate the impact of memory strength and evidence presentation on LLMs' receptiveness to external evidence. We introduce a method to quantify the memory strength of LLMs by measuring the divergence in LLMs' responses to different paraphrases of the same question, which is not considered by previous works. We also generate evidence in various styles to evaluate the effects of evidence in different styles. Two datasets are used for evaluation: Natural Questions (NQ) with popular questions and popQA featuring long-tail questions. Our results show that for questions with high memory strength, LLMs are more likely to rely on internal memory, particularly for larger LLMs such as GPT-4. On the other hand, presenting paraphrased evidence significantly increases LLMs' receptiveness compared to simple repetition or adding details.
Open Domain Question Answering (ODQA) has been advancing rapidly in recent times, driven by significant developments in dense passage retrieval and pretrained language models. Current models typically incorporate the FiD framework, which is composed by a neural retriever alongside an encoder-decoder neural reader. In the answer generation process, the retriever will retrieve numerous passages (around 100 for instance), each of which is then individually encoded by the encoder. Subsequently, the decoder makes predictions based on these encoded passages. Nevertheless, this framework can be relatively time-consuming, particularly due to the extensive length of the gathered passages. To address this, we introduce FastFiD in this paper, a novel approach that executes sentence selection on the encoded passages. This aids in retaining valuable sentences while reducing the context length required for generating answers. Experiments on three commonly used datasets (Natural Questions, TriviaQA and ASQA) demonstrate that our method can enhance the inference speed by 2.3X-5.7X, while simultaneously maintaining the model's performance. Moreover, an in-depth analysis of the model's attention reveals that the selected sentences indeed hold a substantial contribution towards the final answer. The codes are publicly available at https://github.com/thunlp/FastFiD.
Solving mathematical problems requires advanced reasoning abilities and presents notable challenges for large language models. Previous works usually synthesize data from proprietary models to augment existing datasets, followed by instruction tuning to achieve top-tier results. However, our analysis of these datasets reveals severe biases towards easy queries, with frequent failures to generate any correct response for the most challenging queries. Hypothesizing that difficult queries are crucial to learn complex reasoning, we propose Difficulty-Aware Rejection Tuning (DART), a method that allocates difficult queries more trials during the synthesis phase, enabling more extensive training on difficult samples. Utilizing DART, we have created new datasets for mathematical problem-solving that focus more on difficult queries and are substantially smaller than previous ones. Remarkably, our synthesis process solely relies on a 7B-sized open-weight model, without reliance on the commonly used proprietary GPT-4. We fine-tune various base models on our datasets ranging from 7B to 70B in size, resulting in a series of strong models called DART-MATH. In comprehensive in-domain and out-of-domain evaluation on 6 mathematical benchmarks, DART-MATH outperforms vanilla rejection tuning significantly, being superior or comparable to previous arts, despite using much smaller datasets and no proprietary models. Furthermore, our results position our synthetic datasets as the most effective and cost-efficient publicly available resources for advancing mathematical problem-solving.
Recent advancements in Large Language Models have transformed ML/AI development, necessitating a reevaluation of AutoML principles for the Retrieval-Augmented Generation (RAG) systems. To address the challenges of hyper-parameter optimization and online adaptation in RAG, we propose the AutoRAG-HP framework, which formulates the hyper-parameter tuning as an online multi-armed bandit (MAB) problem and introduces a novel two-level Hierarchical MAB (Hier-MAB) method for efficient exploration of large search spaces. We conduct extensive experiments on tuning hyper-parameters, such as top-k retrieved documents, prompt compression ratio, and embedding methods, using the ALCE-ASQA and Natural Questions datasets. Our evaluation from jointly optimization all three hyper-parameters demonstrate that MAB-based online learning methods can achieve Recall@5 $\approx 0.8$ for scenarios with prominent gradients in search space, using only $\sim20\%$ of the LLM API calls required by the Grid Search approach. Additionally, the proposed Hier-MAB approach outperforms other baselines in more challenging optimization scenarios. The code will be made available at https://aka.ms/autorag.