Employing Large Language Models (LLMs) to assess the quality of generated responses has become a widely adopted evaluation method. Specifically, instruct-tuned models and fine-tuned judge models based on open-source LLMs have been reported. While it is known that judge models are vulnerable to certain biases, such as favoring longer answers regardless of content, the specifics of these biases remain under-explored. In this work, we qualitatively identify six types of biases inherent in various judge models. We propose EvalBiasBench as a meta-evaluation collection of hand-crafted test cases for each bias type. Additionally, we present de-biasing dataset construction methods and the associated preference dataset OffsetBias. Experimental results demonstrate that fine-tuning on our dataset significantly enhances the robustness of judge models against biases and improves performance across most evaluation scenarios. We release our datasets and the fine-tuned judge model to public.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
Negating Negatives: Alignment with Human Negative Samples via Distributional Dispreference Optimization
PDF ↗Large language models (LLMs) have revolutionized the role of AI, yet pose potential social risks. To steer LLMs towards human preference, alignment technologies have been introduced and gained increasing attention. Nevertheless, existing methods heavily rely on high-quality positive-negative training pairs, suffering from noisy positive responses that are barely distinguishable from negative ones. Given recent LLMs’ proficiency in generating helpful responses, this work pivots towards a new research question: **can we achieve alignment using solely human-annotated negative samples, preserving helpfulness while reducing harmfulness?** For this purpose, we propose Distributional Dispreference Optimization (D^2O), which maximizes the discrepancy between dispreferred responses and the generated non-negative ones. In this way, D^2O effectively eschews harmful information without incorporating noisy positive samples, while avoiding collapse using self-generated responses as anchors. We demonstrate that D^2O can be regarded as learning a distributional preference model reflecting human dispreference against negative responses, which is theoretically an upper bound of the instance-level DPO. Extensive experiments manifest that our method achieves comparable generation quality and surpasses the latest strong baselines in producing less harmful and more informative responses with better training stability and faster convergence.
Introducing Compiler Semantics into Large Language Models as Programming Language Translators: A Case Study of C to x86 Assembly
PDF ↗Compilers are complex software containing millions of lines of code, taking years to develop. This paper investigates to what extent Large Language Models (LLMs) can replace hand-crafted compilers in translating high-level programming languages to machine instructions, using C to x86 assembly as a case study. We identify two challenges of using LLMs for code translation and introduce two novel data pre-processing techniques to address the challenges: numerical value conversion and training data resampling. While only using a 13B model, our approach achieves a behavioral accuracy of over 91%, outperforming the much larger GPT-4 Turbo model by over 50%. Our results are encouraging, showing that LLMs have the potential to transform how compilation tools are constructed.
This study aims to address the pervasive challenge of quantifying uncertainty in large language models (LLMs) with black-box API access. Conformal Prediction (CP), known for its model-agnostic and distribution-free features, is a desired approach for various LLMs and data distributions. However, existing CP methods for LLMs typically assume access to the logits, which are unavailable for some API-only LLMs. In addition, logits are known to be miscalibrated, potentially leading to degraded CP performance. To tackle these challenges, we introduce a novel CP method that (1) is tailored for API-only LLMs without logit-access; (2) minimizes the size of prediction sets; and (3) ensures a statistical guarantee of the user-defined coverage. The core idea of this approach is to formulate nonconformity measures using both coarse-grained (i.e., sample frequency) and fine-grained uncertainty notions (e.g., semantic similarity). Experimental results on both close-ended and open-ended Question Answering tasks show our approach can mostly outperform the logit-based CP baselines.
In this work, we introduce a novel approach that equips LLM agents with introspection, enhancing consistency and adaptability in solving complex tasks. Our approach prompts LLM agents to decompose a given task into manageable subtasks (i.e., to make a plan), and to continuously introspect upon the suitability and results of their actions. We implement a three-fold introspective intervention: 1) anticipatory reflection on potential failures and alternative remedy before action execution, 2) post-action alignment with subtask objectives and backtracking with remedy to ensure utmost effort in plan execution, and 3) comprehensive review upon plan completion for future strategy refinement. By deploying and experimenting with this methodology—a zero-shot approach—within WebArena for practical tasks in web environments, our agent demonstrates superior performance with a success rate of 23.5% over existing zero-shot methods by 3.5%. The experimental results suggest that our introspection-driven approach not only enhances the agent’s ability to navigate unanticipated challenges through a robust mechanism of plan execution, but also improves efficiency by reducing the number of trials and plan revisions by 45% needed to achieve a task.
Specialized lexicons are collections of words with associated constraints such as special definitions, specific roles, and intended target audiences. These constraints are necessary for content generation and documentation tasks (e.g., writing technical manuals or children’s reading materials), where the goal is to reduce the ambiguity of text content and increase its overall readability for a specific group of audience. Understanding how large language models can capture these constraints can help researchers build better, more impactful tools for wider use beyond the NLP community. Towards this end, we introduce SpeciaLex, a benchmark for evaluating a language model’s ability to follow specialized lexicon-based constraints across 18 diverse subtasks with 1,785 test instances covering core tasks of Checking, Identification, Rewriting, and Open Generation. We present an empirical evaluation of 15 open and closed-source LLMs and discuss insights on how factors such as model scale, openness, setup, and recency affect performance upon evaluating with the benchmark.
Large Language Models (LLMs) demonstrate robust capabilities across various fields, leading to a paradigm shift in LLM-enhanced Recommender System (RS). Research to date focuses on point-wise and pair-wise recommendation paradigms, which are inefficient for LLM-based recommenders due to high computational costs. However, existing list-wise approaches also fall short in ranking tasks due to misalignment between ranking objectives and next-token prediction. Moreover, these LLM-based methods struggle to effectively address the order relation among candidates, particularly given the scale of ratings. To address these challenges, this paper introduces the large language model framework with Aligned Listwise Ranking Objectives (ALRO). ALRO is designed to bridge the gap between the capabilities of LLMs and the nuanced requirements of ranking tasks. Specifically, ALRO employs explicit feedback in a listwise manner by introducing soft lambda loss, a customized adaptation of lambda loss designed for optimizing order relations. This mechanism provides more accurate optimization goals, enhancing the ranking process. Additionally, ALRO incorporates a permutation-sensitive learning mechanism that addresses position bias, a prevalent issue in generative models, without imposing additional computational burdens during inference. Our evaluative studies reveal that ALRO outperforms both existing embedding-based recommendation methods and LLM-based recommendation baselines.
Due to the excellent capacities of large language models (LLMs), it becomes feasible to develop LLM-based agents for reliable user simulation. Considering the scarcity and limit (e.g., privacy issues) of real user data, in this paper, we conduct large-scale user simulations for the web search scenario to improve the analysis and modeling of user search behavior. Specially, we propose BASES, a novel user simulation framework with LLM-based agents, designed to facilitate comprehensive simulations of web search user behaviors. Our simulation framework can generate unique user profiles at scale, which subsequently leads to diverse search behaviors. To demonstrate the effectiveness of BASES, we conduct evaluation experiments based on two human benchmarks in both Chinese and English, demonstrating that BASES can effectively simulate large-scale human-like search behaviors. To further accommodate the research on web search, we develop WARRIORS, a new large-scale dataset encompassing web search user behaviors, including both Chinese and English versions, which can greatly bolster research in the field of information retrieval.
As Large Language Models (LLMs) and Retrieval Augmentation Generation (RAG) techniques have evolved, query rewriting has been widely incorporated into the RAG system for downstream tasks like open-domain QA to enhance document retrieval by reformulating queries. Many works have attempted to improve query rewriting in smaller models to avoid rewriting with costly LLMs, and the most common method is to employ reinforcement learning for feedback training. However, current methods require annotations (labeled relevant documents or downstream answers) or predesigned rewards for feedback, lack generalization, and fail to utilize signals tailored for query rewriting. In this paper, we propose RaFe, a framework for training query rewriting models. By leveraging reranker, RaFe provides ranking feedback aligned well with the rewriting objectives without needing signals from annotations and supports both online and offline training models. Experimental results demonstrate that with a general and publicly available reranker, RaFe can effectively steer the training for rewrite models.
Current long-context benchmarks primarily focus on retrieval-based tests, requiring Large Language Models (LLMs) to locate specific information within extensive input contexts, such as the needle-in-a-haystack (NIAH) benchmark. Long-context generation refers to the ability of a language model to generate coherent and contextually accurate text that spans across lengthy passages or documents. While recent studies show strong performance on NIAH and other retrieval-based long-context benchmarks, there is a significant lack of benchmarks for evaluating long-context generation capabilities. To bridge this gap and offer a comprehensive assessment, we introduce a synthetic benchmark, LongGenBench, which allows for flexible configurations of customized generation context lengths. LongGenBench advances beyond traditional benchmarks by redesigning the format of questions and necessitating that LLMs respond with a single, cohesive long-context answer. Upon extensive evaluation using LongGenBench, we observe that: (1) both API accessed and open source models exhibit performance degradation in long-context generation scenarios, ranging from 1.2% to 47.1%; (2) different series of LLMs exhibit varying trends of performance degradation, with the Gemini-1.5-Flash model showing the least degradation among API accessed models, and the Qwen2 series exhibiting the least degradation in LongGenBench among open source models.
Temporal Cognitive Tree: A Hierarchical Modeling Approach for Event Temporal Relation Extraction
PDF ↗Understanding and analyzing event temporal relations is a crucial task in Natural Language Processing (NLP). This task, known as Event Temporal Relation Extraction (ETRE), aims to identify and extract temporal connections between events in text. Recent studies focus on locating the relative position of event pairs on the timeline by designing logical expressions or auxiliary tasks to predict their temporal occurrence. Despite these advances, this modeling approach neglects the multidimensional information in temporal relation and the hierarchical process of reasoning. In this study, we propose a novel hierarchical modeling approach for this task by introducing a Temporal Cognitive Tree (TCT) that mimics human logical reasoning. Additionally, we also design a integrated model incorporating prompt optimization and deductive reasoning to exploit multidimensional supervised information. Extensive experiments on TB-Dense and MATRES datasets demonstrate that our approach outperforms existing methods.
Towards enhancing the chain-of-thought (CoT) reasoning of large language models (LLMs), much existing work has revealed the effectiveness of straightforward learning on annotated/generated CoT paths. However, there is less evidence yet that reasoning capabilities can be enhanced through a reverse learning process, i.e., learning from potential mistakes in reasoning. To investigate whether LLMs can learn from mistakes, we construct mistake-correction datasets, using GPT-4 to identify and correct the mistakes in inaccurate CoTs. With these mistake-correction datasets, we fine-tune open-source LLMs and arrive at the following conclusions. (1) LLMs can indeed learn from mistakes to enhance their CoT reasoning performances. (2) Compared to CoT data, the mistake-correction data provides additional knowledge on the explanations and reasons for the potential mistakes in CoTs, which consistently contributes to the effectiveness of learning from mistakes. (3) Evolution techniques, especially the correction-centric evolution we introduced, can further enhance the effectiveness of learning from mistakes.
Existing script event prediction task forcasts the subsequent event based on an event script chain. However, the evolution of historical events are more complicated in real world scenarios and the limited information provided by the event script chain also make it difficult to accurately predict subsequent events. This paper introduces a Causality Graph Event Prediction(CGEP) task that forecasting consequential event based on an Event Causality Graph (ECG). We propose a Semantic Enhanced Distance-sensitive Graph Prompt Learning (SeDGPL) Model for the CGEP task. In SeDGPL, (1) we design a Distance-sensitive Graph Linearization (DsGL) module to reformulate the ECG into a graph prompt template as the input of a PLM; (2) propose an Event-Enriched Causality Encoding (EeCE) module to integrate both event contextual semantic and graph schema information; (3) propose a Semantic Contrast Event Prediction (ScEP) module to enhance the event representation among numerous candidate events and predict consequential event following prompt learning paradigm. Experiment results validate our argument our proposed SeDGPL model outperforms the advanced competitors for the CGEP task.
Preserving Pre-trained Representation Space: On Effectiveness of Prefix-tuning for Large Multi-modal Models
PDF ↗Recently, we have observed that Large Multi-modal Models (LMMs) are revolutionizing the way machines interact with the world, unlocking new possibilities across various multi-modal applications. To adapt LMMs for downstream tasks, parameter-efficient fine-tuning (PEFT) which only trains additional prefix tokens or modules, has gained popularity. Nevertheless, there has been little analysis of how PEFT works in LMMs. In this paper, we delve into the strengths and weaknesses of each tuning strategy, shifting the focus from the efficiency typically associated with these approaches. We first discover that model parameter tuning methods such as LoRA and Adapters, distort the feature representation space learned during pre-training, limiting the full utilization of pre-trained knowledge. We also demonstrate that prefix-tuning excels at preserving the representation space, despite of its lower performance on downstream tasks. These findings suggest a simple two-step PEFT strategy called Prefix-Tuned PEFT (PT-PEFT), which successively performs prefix-tuning and then other PEFT (i.e., Adapter, LoRA), combines the benefits of both. Experimental results show that PT-PEFT not only improves performance in image captioning and visual question answering compared to vanilla PEFT methods but also helps preserve the representation space of the four pre-trained models.
Legal judgment prediction is essential for enhancing judicial efficiency. In this work, we identify that existing large language models (LLMs) underperform in this domain due to challenges in understanding case complexities and distinguishing between similar charges. To adapt LLMs for effective legal judgment prediction, we introduce the Ask-Discriminate-Predict (ADAPT) reasoning framework inspired by human judicial reasoning. ADAPT involves decomposing case facts, discriminating among potential charges, and predicting the final judgment. We further enhance LLMs through fine-tuning with multi-task synthetic trajectories to improve legal judgment prediction accuracy and efficiency under our ADAPT framework. Extensive experiments conducted on two widely-used datasets demonstrate the superior performance of our framework in legal judgment prediction, particularly when dealing with complex and confusing charges.
MMCode: Benchmarking Multimodal Large Language Models for Code Generation with Visually Rich Programming Problems
PDF ↗Programming often involves converting detailed and complex specifications into code, a process during which developers typically utilize visual aids to more effectively convey concepts. While recent developments in Large Multimodal Models have demonstrated remarkable abilities in visual reasoning and mathematical tasks, there is little work on investigating whether these models can effectively interpret visual elements for code generation. To this end, we present MMCode, the first multi-modal coding dataset for evaluating algorithmic problem-solving skills in visually rich contexts. MMCode contains 3,548 questions and 6,620 images collected from real-world programming challenges harvested from 10 code competition websites, presenting significant challenges due to the extreme demand for reasoning abilities. Our experiment results show that current state-of-the-art models struggle to solve these problems. The results highlight the lack of powerful vision-code models, and we hope MMCode can serve as an inspiration for future works in this domain. The data and code are publicly available.
RAG-Studio: Towards In-Domain Adaptation of Retrieval Augmented Generation Through Self-Alignment
PDF ↗Retrieval-Augmented Generation (RAG) has proven to be an effective paradigm for enhancing the quality of text generation by integrating large language models (LLMs) with external knowledge. However, an off-the-shelf RAG system, which relies on generally pre-trained LLMs and retrievers, often falls short in specialized domains and applications. In this paper, we introduce RAG-Studio, an efficient self-aligned training framework to adapt general RAG models to specific domains solely through synthetic data, eliminating the need for expensive human-labeled in-domain data. RAG-Studio accepts a specialized domain corpus, a general LLM, and a general retriever, then autonomously generates contrastive training data for both the LLM and retriever through self-alignment. We fine-tune them to work cohesively as an integrated and effective domain-specific RAG system, where the LLM is adapted to incorporate new domain knowledge and become robust to noisy contexts, and the retriever learns to better align with the LLM’s preferences, providing more useful information and minimizing the risk of misleading the LLM. Extensive experiments across diverse in-domain question-answering datasets spanning the biomedical, finance, law, and computing domains, show that RAG-Studio attains state-of-the-art performance, consistently outperforming the use of human-annotated data for fine-tuning.
Recently, there has been a growing interest in knowledge editing for Large Language Models (LLMs). Current approaches and evaluations merely explore the instance-level editing, while whether LLMs possess the capability to modify concepts remains unclear. This paper pioneers the investigation of editing conceptual knowledge for LLMs, by constructing a novel benchmark dataset ConceptEdit and establishing a suite of new metrics for evaluation. The experimental results reveal that, although existing editing methods can efficiently modify concept-level definition to some extent, they also have the potential to distort the related instantial knowledge in LLMs, leading to poor performance. We anticipate this work can inspire further progress in understanding LLMs.
Supertagging is an essential task in Categorical grammar parsing and is crucial for dissecting sentence structures. Our research explores the capacity of Large Language Models (LLMs) in supertagging for both Combinatory Categorial Grammar (CCG) and Lambek Categorial Grammar (LCG). We also present a simple method that significantly boosts LLMs, enabling them to outperform LSTM and encoder-based models and achieve state-of-the-art performance. This advancement highlights LLMs’ potential in classification tasks, showcasing their adaptability beyond generative capabilities. Our findings demonstrate the evolving utility of LLMs in natural language processing, particularly in complex tasks like supertagging.
Parameter-efficient fine-tuning (PEFT) methods are increasingly used with pre-trained language models (PLMs) for continual learning (CL). These methods typically involve training a PEFT module for each new task and employing similarity-based selection to route modules during inference. However, they face two major limitations: 1) interference during module training with already learned modules and 2) suboptimal routing when composing modules. In this paper, we present L2R, a method that isolates the training of new PEFT modules to ensure their task specialization. L2R then learns to compose the learned modules by training a network of routers that leverages a small memory containing examples of previously seen tasks. We evaluate our method in two CL setups using various benchmarks. Our results demonstrate that L2R provides an effective composition of PEFT modules, leading to improved generalization and performance compared to other methods.