Authors: Yuchen Fan, Xin Zhong, Heng Zhou, Yuchen Zhang, Mingyu Liang, Chengxing Xie, Ermo Hua, Ning Ding, Bowen Zhou
Abstract: Long-Form Question Answering (LFQA) refers to generating in-depth, paragraph-level responses to open-ended questions. Although lots of LFQA methods are developed, evaluating LFQA effectively and efficiently remains challenging due to its high complexity and cost. Therefore, there is no standard benchmark for LFQA evaluation till now. To address this gap, we make the first attempt by proposing a well-constructed, reference-based benchmark named Chinese exAmination for LFQA Evaluation (CALF), aiming to rigorously assess the performance of automatic evaluation metrics for LFQA. The CALF benchmark is derived from Chinese examination questions that have been translated into English. It includes up to 1476 examples consisting of knowledge-intensive and nuanced responses. Our evaluation comprises three different settings to ana lyze the behavior of automatic metrics comprehensively. We conducted extensive experiments on 7 traditional evaluation metrics, 3 prompt-based metrics, and 3 trained evaluation metrics, and tested on agent systems for the LFQA evaluation. The results reveal that none of the current automatic evaluation metrics shows comparable performances with humans, indicating that they cannot capture dense information contained in long-form responses well. In addition, we provide a detailed analysis of the reasons why automatic evaluation metrics fail when evaluating LFQA, offering valuable insights to advance LFQA evaluation systems. Dataset and associated codes can be accessed at our GitHub repository.
Authors: Zhiwen You, Kanyao Han, Haotian Zhu, Bertram Lud\"ascher, Jana Diesner
Abstract: Prompt-based fine-tuning has become an essential method for eliciting information encoded in pre-trained language models for a variety of tasks, including text classification. For multi-class classification tasks, prompt-based fine-tuning under low-resource scenarios has resulted in performance levels comparable to those of fully fine-tuning methods. Previous studies have used crafted prompt templates and verbalizers, mapping from the label terms space to the class space, to solve the classification problem as a masked language modeling task. However, cross-domain and fine-grained prompt-based fine-tuning with an automatically enriched verbalizer remains unexplored, mainly due to the difficulty and costs of manually selecting domain label terms for the verbalizer, which requires humans with domain expertise. To address this challenge, we introduce SciPrompt, a framework designed to automatically retrieve scientific topic-related terms for low-resource text classification tasks. To this end, we select semantically correlated and domain-specific label terms within the context of scientific literature for verbalizer augmentation. Furthermore, we propose a new verbalization strategy that uses correlation scores as additional weights to enhance the prediction performance of the language model during model tuning. Our method outperforms state-of-the-art, prompt-based fine-tuning methods on scientific text classification tasks under few and zero-shot settings, especially in classifying fine-grained and emerging scientific topics.
Authors: Danqing Wang, Jianxin Ma, Fei Fang, Lei Li
Abstract: Despite significant advancements in the reasoning capabilities of Large Language Models (LLMs), the lack of diverse reasoning solutions often makes them trapped in a limited solution search area. In this paper, we propose TypedThinker, a novel framework that enhances LLMs' problem-solving abilities by incorporating multiple reasoning types (deductive, inductive, abductive, and analogical). Our analysis across four benchmarks reveals that different reasoning types uniquely solve distinct sets of problems, highlighting the importance of diverse thinking approaches. TypedThinker addresses two key challenges: selecting appropriate reasoning types for given problems and effectively implementing specific reasoning types. Through self-training on successful experiences, TypedThinker learns an implicit policy for reasoning type selection and application. Experimental results demonstrate significant improvements over baseline models, with accuracy increases of 3.4% for Mistral 7B and 16.7% for LLaMA3 8B across four reasoning benchmarks. Notably, TypedThinker shows effective generalization to new benchmarks and can further enhance the reasoning capability of powerful models like GPT-4o. The code is released at https://github.com/dqwang122/ThinkHub.
Authors: I-Fan Lin, Faegheh Hasibi, Suzan Verberne
Abstract: In this short paper we propose a data augmentation method for intent detection in zero-resource domains. Existing data augmentation methods rely on few labelled examples for each intent category, which can be expensive in settings with many possible intents. We use a two-stage approach: First, we generate utterances for intent labels using an open-source large language model in a zero-shot setting. Second, we develop a smaller sequence-to-sequence model (the Refiner), to improve the generated utterances. The Refiner is fine-tuned on seen domains and then applied to unseen domains. We evaluate our method by training an intent classifier on the generated data, and evaluating it on real (human) data. We find that the Refiner significantly improves the data utility and diversity over the zero-shot LLM baseline for unseen domains and over common baseline approaches. Our results indicate that a two-step approach of a generative LLM in zero-shot setting and a smaller sequence-to-sequence model can provide high-quality data for intent detection.
Authors: Min-Hsuan Yeh, Leitian Tao, Jeffrey Wang, Xuefeng Du, Yixuan Li
Abstract: Most alignment research today focuses on designing new learning algorithms using datasets like Anthropic-HH, assuming human feedback data is inherently reliable. However, little attention has been given to the qualitative unreliability of human feedback and its impact on alignment. To address this gap, we conduct a comprehensive study and provide an in-depth analysis of human feedback data. We assess feedback reliability using a committee of gold reward models, revealing that over 25% of the dataset shows low or no agreement with these models, implying a high degree of unreliability. Through a qualitative analysis, we identify six key sources of unreliability, such as mis-labeling, subjective preferences, differing criteria and thresholds for helpfulness and harmlessness, etc. Lastly, to mitigate unreliability, we propose Source-Aware Cleaning, an automatic data-cleaning method guided by the insight of our qualitative analysis, to significantly improve data quality. Extensive experiments demonstrate that models trained on our cleaned dataset, HH-Clean, substantially outperform those trained on the original dataset. We release HH-Clean to support more reliable LLM alignment evaluation in the future.
Authors: Qian Ruan, Ilia Kuznetsov, Iryna Gurevych
Abstract: Classification is a core NLP task architecture with many potential applications. While large language models (LLMs) have brought substantial advancements in text generation, their potential for enhancing classification tasks remains underexplored. To address this gap, we propose a framework for thoroughly investigating fine-tuning LLMs for classification, including both generation- and encoding-based approaches. We instantiate this framework in edit intent classification (EIC), a challenging and underexplored classification task. Our extensive experiments and systematic comparisons with various training approaches and a representative selection of LLMs yield new insights into their application for EIC. We investigate the generalizability of these findings on five further classification tasks. To demonstrate the proposed methods and address the data shortage for empirical edit analysis, we use our best-performing EIC model to create Re3-Sci2.0, a new large-scale dataset of 1,780 scientific document revisions with over 94k labeled edits. The quality of the dataset is assessed through human evaluation. The new dataset enables an in-depth empirical study of human editing behavior in academic writing. We make our experimental framework, models and data publicly available.
Authors: Xiao Yu, Baolin Peng, Vineeth Vajipey, Hao Cheng, Michel Galley, Jianfeng Gao, Zhou Yu
Abstract: Autonomous agents have demonstrated significant potential in automating complex multistep decision-making tasks. However, even state-of-the-art vision-language models (VLMs), such as GPT-4o, still fall short of human-level performance, particularly in intricate web environments and long-horizon planning tasks. To address these limitations, we introduce Reflective Monte Carlo Tree Search (R-MCTS), a novel test-time algorithm designed to enhance the ability of AI agents, e.g., powered by GPT-4o, to explore decision space on the fly. R-MCTS extends traditional MCTS by 1) incorporating contrastive reflection, allowing agents to learn from past interactions and dynamically improve their search efficiency; and 2) using multi-agent debate to provide reliable state evaluation. Moreover, we improve the agent's performance by fine-tuning GPT-4o through self-learning, using R-MCTS generated tree traversals without any human-provided labels. On the challenging VisualWebArena benchmark, our GPT-4o-based R-MCTS agent achieves a 6% to 30% relative improvement across various tasks compared to the previous state-of-the-art. Additionally, we show that the knowledge gained from test-time search can be effectively transferred back to GPT-4o via fine-tuning. The fine-tuned GPT-4o matches 97% of R-MCTS's performance while reducing compute usage by a factor of four at test time. Furthermore, qualitative results reveal that the fine-tuned GPT-4o model demonstrates the ability to explore the environment, evaluate a state, and backtrack to viable ones when it detects that the current state cannot lead to success. Moreover, our work demonstrates the compute scaling properties in both training - data collection with R-MCTS - and testing time. These results suggest a promising research direction to enhance VLMs' reasoning and planning capabilities for agentic applications via test-time search and self-learning.
Authors: Jonas Gehring, Kunhao Zheng, Jade Copet, Vegard Mella, Taco Cohen, Gabriel Synnaeve
Abstract: Large language models (LLMs) deployed as agents solve user-specified tasks over multiple steps while keeping the required manual engagement to a minimum. Crucially, such LLMs need to ground their generations in any feedback obtained to reliably achieve desired outcomes. We propose an end-to-end reinforcement learning method for teaching models to leverage execution feedback in the realm of code synthesis, where state-of-the-art LLMs struggle to improve code iteratively compared to independent sampling. We benchmark on competitive programming tasks, where we achieve new start-of-the art results with both small (8B parameters) and large (70B) models while reducing the amount of samples required by an order of magnitude. Our analysis of inference-time behavior demonstrates that our method produces LLMs that effectively leverage automatic feedback over multiple steps.
Authors: Michael A. Lepori, Michael Mozer, Asma Ghandeharioun
Abstract: The profound success of transformer-based language models can largely be attributed to their ability to integrate relevant contextual information from an input sequence in order to generate a response or complete a task. However, we know very little about the algorithms that a model employs to implement this capability, nor do we understand their failure modes. For example, given the prompt "John is going fishing, so he walks over to the bank. Can he make an ATM transaction?", a model may incorrectly respond "Yes" if it has not properly contextualized "bank" as a geographical feature, rather than a financial institution. We propose the LLM Race Conditions Hypothesis as an explanation of contextualization errors of this form. This hypothesis identifies dependencies between tokens (e.g., "bank" must be properly contextualized before the final token, "?", integrates information from "bank"), and claims that contextualization errors are a result of violating these dependencies. Using a variety of techniques from mechanistic intepretability, we provide correlational and causal evidence in support of the hypothesis, and suggest inference-time interventions to address it.
Authors: Xiangyu Peng, Congying Xia, Xinyi Yang, Caiming Xiong, Chien-Sheng Wu, Chen Xing
Abstract: Post-training Large Language Models (LLMs) with explicit reasoning trajectories can enhance their reasoning abilities. However, acquiring such high-quality trajectory data typically demands meticulous supervision from humans or superior models, which can be either expensive or license-constrained. In this paper, we explore how far an LLM can improve its reasoning by self-synthesizing reasoning paths as training data without any additional supervision. Existing self-synthesizing methods, such as STaR, suffer from poor generalization to out-of-domain (OOD) reasoning tasks. We hypothesize it is due to that their self-synthesized reasoning paths are too task-specific, lacking general task-agnostic reasoning guidance. To address this, we propose Reasoning Generalist via Self-Improvement (ReGenesis), a method to self-synthesize reasoning paths as post-training data by progressing from abstract to concrete. More specifically, ReGenesis self-synthesizes reasoning paths by converting general reasoning guidelines into task-specific ones, generating reasoning structures, and subsequently transforming these structures into reasoning paths, without the need for human-designed task-specific examples used in existing methods. We show that ReGenesis achieves superior performance on all in-domain and OOD settings tested compared to existing methods. For six OOD tasks specifically, while previous methods exhibited an average performance decrease of approximately 4.6% after post training, ReGenesis delivers around 6.1% performance improvement. We also conduct in-depth analysis of our framework and show ReGenesis is effective across various LLMs and design choices.
Authors: Zecheng Tang, Keyan Zhou, Juntao Li, Baibei Ji, Jianye Hou, Min Zhang
Abstract: Long-context models (LCMs) have made remarkable strides in recent years, offering users great convenience for handling tasks that involve long context, such as document summarization. As the community increasingly prioritizes the faithfulness of generated results, merely ensuring the accuracy of LCM outputs is insufficient, as it is quite challenging for humans to verify the results from the extremely lengthy context. Yet, although some efforts have been made to assess whether LCMs respond truly based on the context, these works either are limited to specific tasks or heavily rely on external evaluation resources like GPT-4.In this work, we introduce L-CiteEval, a comprehensive multi-task benchmark for long-context understanding with citations, aiming to evaluate both the understanding capability and faithfulness of LCMs. L-CiteEval covers 11 tasks from diverse domains, spanning context lengths from 8K to 48K, and provides a fully automated evaluation suite. Through testing with 11 cutting-edge closed-source and open-source LCMs, we find that although these models show minor differences in their generated results, open-source models substantially trail behind their closed-source counterparts in terms of citation accuracy and recall. This suggests that current open-source LCMs are prone to responding based on their inherent knowledge rather than the given context, posing a significant risk to the user experience in practical applications. We also evaluate the RAG approach and observe that RAG can significantly improve the faithfulness of LCMs, albeit with a slight decrease in the generation quality. Furthermore, we discover a correlation between the attention mechanisms of LCMs and the citation generation process.
Authors: Haruka Fukushima, Daniel Plesniak, Daisuke Bekki
Abstract: This paper provides evidence that weak crossover effects differ in nature between matrix and relative clauses. Fukushima et al. (2024) provided similar evidence, showing that, when various non-structural factors were eliminated English speakers never accepted matrix weak crossover cases, but often accepted relative weak crossover ones. Those results were limited, however, by English word order, which lead to uncertainty as to whether this difference was due to the effects of linear precedence or syntactic structure. In this paper, to distinguish between these two possibilities, we conduct an experiment using Japanese, which lacks the word-order confound that English had. We find results that are qualitatively in line with Fukushima et al. (2024) suggesting that the relevant distinction is structural and not based simply on precedence.
Authors: Collin Zhang, Tingwei Zhang, Vitaly Shmatikov
Abstract: Recent work showed that retrieval based on embedding similarity (e.g., for retrieval-augmented generation) is vulnerable to poisoning: an adversary can craft malicious documents that are retrieved in response to broad classes of queries. We demonstrate that previous, HotFlip-based techniques produce documents that are very easy to detect using perplexity filtering. Even if generation is constrained to produce low-perplexity text, the resulting documents are recognized as unnatural by LLMs and can be automatically filtered from the retrieval corpus. We design, implement, and evaluate a new controlled generation technique that combines an adversarial objective (embedding similarity) with a "naturalness" objective based on soft scores computed using an open-source, surrogate LLM. The resulting adversarial documents (1) cannot be automatically detected using perplexity filtering and/or other LLMs, except at the cost of significant false positives in the retrieval corpus, yet (2) achieve similar poisoning efficacy to easily-detectable documents generated using HotFlip, and (3) are significantly more effective than prior methods for energy-guided generation, such as COLD.
Authors: Anwoy Chatterjee, H S V N S Kowndinya Renduchintala, Sumit Bhatia, Tanmoy Chakraborty
Abstract: Despite their remarkable capabilities, Large Language Models (LLMs) are found to be surprisingly sensitive to minor variations in prompts, often generating significantly divergent outputs in response to minor variations in the prompts, such as spelling errors, alteration of wording or the prompt template. However, while assessing the quality of an LLM, the focus often tends to be solely on its performance on downstream tasks, while very little to no attention is paid to prompt sensitivity. To fill this gap, we propose POSIX - a novel PrOmpt Sensitivity IndeX as a reliable measure of prompt sensitivity, thereby offering a more comprehensive evaluation of LLM performance. The key idea behind POSIX is to capture the relative change in loglikelihood of a given response upon replacing the corresponding prompt with a different intent-preserving prompt. We provide thorough empirical evidence demonstrating the efficacy of POSIX in capturing prompt sensitivity and subsequently use it to measure and thereby compare prompt sensitivity of various open-source LLMs. We find that merely increasing the parameter count or instruction tuning does not necessarily reduce prompt sensitivity whereas adding some few-shot exemplars, even just one, almost always leads to significant decrease in prompt sensitivity. We also find that alterations to prompt template lead to the highest sensitivity in the case of MCQtype tasks, whereas paraphrasing results in the highest sensitivity in open-ended generation tasks. The code for reproducing our results is open-sourced at https://github.com/kowndinyarenduchintala/POSIX.
Authors: Pedro Colon-Hernandez, Nanxi Liu, Chelsea Joe, Peter Chin, Claire Yin, Henry Lieberman, Yida Xin, Cynthia Breazeal
Abstract: Generating commonsense assertions within a given story context remains a difficult task for modern language models. Previous research has addressed this problem by aligning commonsense inferences with stories and training language generation models accordingly. One of the challenges is determining which topic or entity in the story should be the focus of an inferred assertion. Prior approaches lack the ability to control specific aspects of the generated assertions. In this work, we introduce "hinting," a data augmentation technique that enhances contextualized commonsense inference. "Hinting" employs a prefix prompting strategy using both hard and soft prompts to guide the inference process. To demonstrate its effectiveness, we apply "hinting" to two contextual commonsense inference datasets: ParaCOMET and GLUCOSE, evaluating its impact on both general and context-specific inference. Furthermore, we evaluate "hinting" by incorporating synonyms and antonyms into the hints. Our results show that "hinting" does not compromise the performance of contextual commonsense inference while offering improved controllability.
Authors: Yinhong Liu, Zhijiang Guo, Tianya Liang, Ehsan Shareghi, Ivan Vuli\'c, Nigel Collier
Abstract: Recent research in Large Language Models (LLMs) has shown promising progress related to LLM alignment with human preferences. LLM-empowered decision-making systems are expected to be predictable, reliable and trustworthy, which implies being free from paradoxes or contradictions that could undermine their credibility and validity. However, LLMs still exhibit inconsistent and biased behaviour when making decisions or judgements. In this work, we focus on studying logical consistency of LLMs as a prerequisite for more reliable and trustworthy systems. Logical consistency ensures that decisions are based on a stable and coherent understanding of the problem, reducing the risk of erratic or contradictory outputs. We first propose a universal framework to quantify the logical consistency via three fundamental proxies: transitivity, commutativity and negation invariance. We then evaluate logical consistency, using the defined measures, of a wide range of LLMs, demonstrating that it can serve as a strong proxy for overall robustness. Additionally, we introduce a data refinement and augmentation technique that enhances the logical consistency of LLMs without sacrificing alignment to human preferences. It augments noisy and sparse pairwise-comparison annotations by estimating a partially or totally ordered preference rankings using rank aggregation methods. Finally, we show that logical consistency impacts the performance of LLM-based logic-dependent algorithms, where LLMs serve as logical operators.
Authors: Wei Cheng, Tianlu Wang, Yanmin Ji, Fan Yang, Keren Tan, Yiyu Zheng
Abstract: While in-context learning with large language models (LLMs) has shown impressive performance, we have discovered a unique miscalibration behavior where both correct and incorrect predictions are assigned the same level of confidence. We refer to this phenomenon as indiscriminate miscalibration. We found that traditional calibration metrics, such as Expected Calibrated Errors (ECEs), are unable to capture this behavior effectively. To address this issue, we propose new metrics to measure the severity of indiscriminate miscalibration. Additionally, we develop a novel in-context comparative inference method to alleviate miscalibrations and improve classification performance. Through extensive experiments on five datasets, we demonstrate that our proposed method can achieve more accurate and calibrated predictions compared to regular zero-shot and few-shot prompting.
Authors: Richard Zhuang, Tianhao Wu, Zhaojin Wen, Andrew Li, Jiantao Jiao, Kannan Ramchandran
Abstract: With hundreds of thousands of language models available on Huggingface today, efficiently evaluating and utilizing these models across various downstream, tasks has become increasingly critical. Many existing methods repeatedly learn task-specific representations of Large Language Models (LLMs), which leads to inefficiencies in both time and computational resources. To address this, we propose EmbedLLM, a framework designed to learn compact vector representations, of LLMs that facilitate downstream applications involving many models, such as model routing. We introduce an encoder-decoder approach for learning such embeddings, along with a systematic framework to evaluate their effectiveness. Empirical results show that EmbedLLM outperforms prior methods in model routing both in accuracy and latency. Additionally, we demonstrate that our method can forecast a model's performance on multiple benchmarks, without incurring additional inference cost. Extensive probing experiments validate that the learned embeddings capture key model characteristics, e.g. whether the model is specialized for coding tasks, even without being explicitly trained on them. We open source our dataset, code and embedder to facilitate further research and application.
Authors: Arthur Amalvy (LIA), Vincent Labatut (LIA)
Abstract: The Novelties corpus is a collection of novels (and parts of novels) annotated for Named Entity Recognition (NER) among other tasks. This document describes the guidelines applied during its annotation. It contains the instructions used by the annotators, as well as a number of examples retrieved from the annotated novels, and illustrating expressions that should be marked as entities as well as expressions that should not.
Authors: \'Oscar Garc\'ia-Sierra, Ana Fern\'andez-Pampill\'on Cesteros, Miguel Ortega-Mart\'in
Abstract: Subword tokenization algorithms used by Large Language Models are significantly more efficient and can independently build the necessary vocabulary of words and subwords without human intervention. However, those subwords do not always align with real morphemes, potentially impacting the models' performance, though it remains uncertain when this might occur. In previous research, we proposed a method to assess the morphological quality of vocabularies, focusing on the overlap between these vocabularies and the morphemes of a given language. Our evaluation method was built on three quality measures, relevance, cohesion, and morphological accuracy, and a procedure for their assessment. By applying this method to vocabularies created by three subword tokenization algorithms, BPE, Wordpiece, and Unigram, we concluded that these vocabularies generally exhibit very low morphological quality. In this article, we apply this evaluation to the tokenizer of BETO, a BERT language model trained on large Spanish corpora. This evaluation, along with our previous results, helped us conclude that its vocabulary has a low morphological quality, and we also found that training the tokenizer in a larger corpus does not improve the morphological quality of the generated vocabulary. Additionally, this evaluation helps clarify the algorithm used by the tokenizer, that is, Wordpiece, given the inconsistencies between the authors' claims and the model's configuration.
Authors: Letian Peng, Chenyang An, Jingbo Shang
Abstract: Language model (LM) decoding is based on the next-token prediction (NTP) probability distribution. For neural LMs (e.g., Transformer-based), NTP distribution is essentially a softmax-regularized dot product between an encoded input context (query) and fixed vocabulary representations (keys). In this paper, we study the effect of the key distribution on the NTP distribution, with a focus on whether the similarity between keys will trigger spurious correlations in NTP. Through knowledge-probing tasks, we show that in the NTP distribution, the few top-ranked tokens are typically accurate. However, the middle-ranked prediction is highly biased towards the tokens that are distributionally (not necessarily semantically) similar to these top ones. For instance, if "P" is predicted as the top-1 token, "A"-"Z" will all be ranked high in NTP, no matter whether they can lead to correct decoding results. This hurts the sampling diversity and makes the sampling of correct, long-tail results hopeless and noisy. We attempt to alleviate this issue via a novel in-context method that iteratively pushes the query representation away from explored regions. Specifically, we include the explored decoding results in the context and prompt the LM to generate something else, which encourages the LM to produce a query representation that has small dot products with explored keys. Experiments on knowledge-probing tasks show that our method leads to efficient navigation away from explored keys to correct new keys. We further extend our method to open-ended and chain-of-thought (for reasoning) generation. Experiment results show that ICN contributes to better generation diversity and improved self-consistency voting performance. Finally, we discuss potential training issues caused by the fixed key space together with the challenges and possible ways to address them in future research.
Authors: Zhe Xu, Kaveh Hassani, Si Zhang, Hanqing Zeng, Michihiro Yasunaga, Limei Wang, Dongqi Fu, Ning Yao, Bo Long, Hanghang Tong
Abstract: Language Models (LMs) are increasingly challenging the dominance of domain-specific models, including Graph Neural Networks (GNNs) and Graph Transformers (GTs), in graph learning tasks. Following this trend, we propose a novel approach that empowers off-the-shelf LMs to achieve performance comparable to state-of-the-art GNNs on node classification tasks, without requiring any architectural modification. By preserving the LM's original architecture, our approach retains a key benefit of LM instruction tuning: the ability to jointly train on diverse datasets, fostering greater flexibility and efficiency. To achieve this, we introduce two key augmentation strategies: (1) Enriching LMs' input using topological and semantic retrieval methods, which provide richer contextual information, and (2) guiding the LMs' classification process through a lightweight GNN classifier that effectively prunes class candidates. Our experiments on real-world datasets show that backbone Flan-T5 models equipped with these augmentation strategies outperform state-of-the-art text-output node classifiers and are comparable to top-performing vector-output node classifiers. By bridging the gap between specialized task-specific node classifiers and general LMs, this work paves the way for more versatile and widely applicable graph learning models. We will open-source the code upon publication.
Authors: Yongsik Seo, Sungwon Song, Ryang Heo, Jieyong Kim, Dongha Lee
Abstract: In the domain of Aspect-Based Sentiment Analysis (ABSA), generative methods have shown promising results and achieved substantial advancements. However, despite these advancements, the tasks of extracting sentiment quadruplets, which capture the nuanced sentiment expressions within a sentence, remain significant challenges. In particular, compound sentences can potentially contain multiple quadruplets, making the extraction task increasingly difficult as sentence complexity grows. To address this issue, we are focusing on simplifying sentence structures to facilitate the easier recognition of these elements and crafting a model that integrates seamlessly with various ABSA tasks. In this paper, we propose Aspect Term Oriented Sentence Splitter (ATOSS), which simplifies compound sentence into simpler and clearer forms, thereby clarifying their structure and intent. As a plug-and-play module, this approach retains the parameters of the ABSA model while making it easier to identify essential intent within input sentences. Extensive experimental results show that utilizing ATOSS outperforms existing methods in both ASQP and ACOS tasks, which are the primary tasks for extracting sentiment quadruplets.
Authors: Rui Meng, Ye Liu, Lifu Tu, Daqing He, Yingbo Zhou, Semih Yavuz
Abstract: Phrases are fundamental linguistic units through which humans convey semantics. This study critically examines the capacity of API-based large language models (LLMs) to comprehend phrase semantics, utilizing three human-annotated datasets. We assess the performance of LLMs in executing phrase semantic reasoning tasks guided by natural language instructions and explore the impact of common prompting techniques, including few-shot demonstrations and Chain-of-Thought reasoning. Our findings reveal that LLMs greatly outperform traditional embedding methods across the datasets; however, they do not show a significant advantage over fine-tuned methods. The effectiveness of advanced prompting strategies shows variability. We conduct detailed error analyses to interpret the limitations faced by LLMs in comprehending phrase semantics. Code and data can be found at https://github.com/memray/llm_phrase_semantics.
Authors: Nathaniel Berger, Stefan Riezler, Miriam Exel, Matthias Huck
Abstract: Preference Optimization (PO) techniques are currently one of the state of the art techniques for fine-tuning large language models (LLMs) on pairwise preference feedback from human annotators. However, in machine translation, this sort of feedback can be difficult to solicit. Additionally, Kreutzer et al. (2018) have shown that, for machine translation, pairwise preferences are less reliable than other forms of human feedback, such as 5-point ratings. We examine post-edits to see if they can be a source of reliable human preferences by construction. In PO, a human annotator is shown sequences $s_1$ and $s_2$ and asked for a preference judgment, %$s_1 > s_2$; while for post-editing, editors \emph{create} $s_1$ and know that it should be better than $s_2$. We attempt to use these implicit preferences for PO and show that it helps the model move towards post-edit-like hypotheses and away from machine translation-like hypotheses. Furthermore, we show that best results are obtained by pre-training the model with supervised fine-tuning (SFT) on post-edits in order to promote post-edit-like hypotheses to the top output ranks.
Authors: Tianxiang Chen, Zhentao Tan, Tao Gong, Yue Wu, Qi Chu, Bin Liu, Jieping Ye, Nenghai Yu
Abstract: As a manner to augment pre-trained large language models (LLM), knowledge injection is critical to develop vertical domain large models and has been widely studied. Although most current approaches, including parameter-efficient fine-tuning (PEFT) and block expansion methods, uniformly apply knowledge across all LLM layers, it raises the question: are all layers equally crucial for knowledge injection? We begin by evaluating the importance of each layer in finding the optimal layer range for knowledge injection. Intuitively, the more important layers should play a more critical role in knowledge injection and deserve a denser injection. We observe performance dips in question-answering benchmarks after the removal or expansion of the shallow layers, and the degradation shrinks as the layer gets deeper, indicating that the shallow layers hold the key to knowledge injection. This insight leads us to propose the S strategy, a post-pretraining strategy of selectively enhancing shallow layers while pruning the less effective deep ones. Based on this strategy, we introduce Llama Slayer-8B and Llama Slayer-8B-Instruct. We experimented on the corpus of code $\&$ math and demonstrated the effectiveness of our strategy. Further experiments across different LLM, Mistral-7B, and a legal corpus confirmed the general applicability of the approach, underscoring its wide-ranging efficacy. Our code is available at: \https://github.com/txchen-USTC/Llama-Slayer
Authors: Jingyu Liu, Jiaen Lin, Yong Liu
Abstract: Retrieval-Augmented Generation (RAG) has gained significant popularity in modern Large Language Models (LLMs) due to its effectiveness in introducing new knowledge and reducing hallucinations. However, the deep understanding of RAG remains limited, how does RAG help the reasoning process and can RAG help improve the reasoning capability remains question. While external documents are typically considered as a method to incorporate domain-specific information, they also contain intermediate reasoning results related to the query, this suggests that documents could enhance the reasoning capability of LLMs, which has not been previously explored. In this paper, we investigate this issue in depth and find that while RAG can assist with reasoning, the help is limited. If we conceptualize the reasoning process as a tree with fixed depth, then RAG struggles to assist LLMs in performing deeper reasoning. Additionally, the information in the documents requires preprocessing to filter out noise. We demonstrate that this preprocessing is difficult to achieve simply fine-tuning of the LLM, it often necessitates numerous additional transformer layers to solve the problem. To simplify the problem, we propose DPrompt tuning, which effectively resolves the issue within just limited transformer layers, leading to improved performance.
Authors: Eduard Tulchinskii, Laida Kushnareva, Kristian Kuznetsov, Anastasia Voznyuk, Andrei Andriiainen, Irina Piontkovskaya, Evgeny Burnaev, Serguei Barannikov
Abstract: A standard way to evaluate the abilities of LLM involves presenting a multiple-choice question and selecting the option with the highest logit as the model's predicted answer. However, such a format for evaluating LLMs has limitations, since even if the model knows the correct answer, it may struggle to select the corresponding letter simply due to difficulties in following this rigid format. To address this, we introduce new scores that better capture and reveal model's underlying knowledge: the Query-Key Score (QK-score), derived from the interaction between query and key representations in attention heads, and the Attention Score, based on attention weights. These scores are extracted from specific \textit{select-and-copy} heads, which show consistent performance across popular Multi-Choice Question Answering (MCQA) datasets. Based on these scores, our method improves knowledge extraction, yielding up to 16\% gain for LLaMA2-7B and up to 10\% for larger models on popular MCQA benchmarks. At the same time, the accuracy on a simple synthetic dataset, where the model explicitly knows the right answer, increases by almost 60\%, achieving nearly perfect accuracy, therefore demonstrating the method's efficiency in mitigating MCQA format limitations. To support our claims, we conduct experiments on models ranging from 7 billion to 70 billion parameters in both zero- and few-shot setups.
Authors: Junfeng Fang, Houcheng Jiang, Kun Wang, Yunshan Ma, Xiang Wang, Xiangnan He, Tat-seng Chua
Abstract: Large language models (LLMs) often exhibit hallucinations due to incorrect or outdated knowledge. Hence, model editing methods have emerged to enable targeted knowledge updates. To achieve this, a prevailing paradigm is the locating-then-editing approach, which first locates influential parameters and then edits them by introducing a perturbation. While effective, current studies have demonstrated that this perturbation inevitably disrupt the originally preserved knowledge within LLMs, especially in sequential editing scenarios. To address this, we introduce AlphaEdit, a novel solution that projects perturbation onto the null space of the preserved knowledge before applying it to the parameters. We theoretically prove that this projection ensures the output of post-edited LLMs remains unchanged when queried about the preserved knowledge, thereby mitigating the issue of disruption. Extensive experiments on various LLMs, including LLaMA3, GPT2-XL, and GPT-J, show that AlphaEdit boosts the performance of most locating-then-editing methods by an average of 36.4% with a single line of additional code for projection solely. Our code is available at: https://github.com/jianghoucheng/AlphaEdit.
Authors: Haodong Xie, Rahul Singh Maharjan, Federico Tavella, Angelo Cangelosi
Abstract: Understanding and manipulating concrete and abstract concepts is fundamental to human intelligence. Yet, they remain challenging for artificial agents. This paper introduces a multimodal generative approach to high order abstract concept learning, which integrates visual and categorical linguistic information from concrete ones. Our model initially grounds subordinate level concrete concepts, combines them to form basic level concepts, and finally abstracts to superordinate level concepts via the grounding of basic-level concepts. We evaluate the model language learning ability through language-to-visual and visual-to-language tests with high order abstract concepts. Experimental results demonstrate the proficiency of the model in both language understanding and language naming tasks.
Authors: Junyu Lu, Bo Xu, Xiaokun Zhang, Hongbo Wang, Haohao Zhu, Dongyu Zhang, Liang Yang, Hongfei Lin
Abstract: This paper has been accepted in the NeurIPS 2024 D & B Track. Harmful memes have proliferated on the Chinese Internet, while research on detecting Chinese harmful memes significantly lags behind due to the absence of reliable datasets and effective detectors. To this end, we focus on the comprehensive detection of Chinese harmful memes. We construct ToxiCN MM, the first Chinese harmful meme dataset, which consists of 12,000 samples with fine-grained annotations for various meme types. Additionally, we propose a baseline detector, Multimodal Knowledge Enhancement (MKE), incorporating contextual information of meme content generated by the LLM to enhance the understanding of Chinese memes. During the evaluation phase, we conduct extensive quantitative experiments and qualitative analyses on multiple baselines, including LLMs and our MKE. The experimental results indicate that detecting Chinese harmful memes is challenging for existing models while demonstrating the effectiveness of MKE. The resources for this paper are available at https://github.com/DUT-lujunyu/ToxiCN_MM.
Authors: Genta Indra Winata, David Anugraha, Lucky Susanto, Garry Kuwanto, Derry Tanti Wijaya
Abstract: Understanding the quality of a performance evaluation metric is crucial for ensuring that model outputs align with human preferences. However, it remains unclear how well each metric captures the diverse aspects of these preferences, as metrics often excel in one particular area but not across all dimensions. To address this, it is essential to systematically calibrate metrics to specific aspects of human preference, catering to the unique characteristics of each aspect. We introduce MetaMetrics, a calibrated meta-metric designed to evaluate generation tasks across different modalities in a supervised manner. MetaMetrics optimizes the combination of existing metrics to enhance their alignment with human preferences. Our metric demonstrates flexibility and effectiveness in both language and vision downstream tasks, showing significant benefits across various multilingual and multi-domain scenarios. MetaMetrics aligns closely with human preferences and is highly extendable and easily integrable into any application. This makes MetaMetrics a powerful tool for improving the evaluation of generation tasks, ensuring that metrics are more representative of human judgment across diverse contexts.
Authors: Ido Cohen, Jacob Gidron, Idan Pinto
Abstract: Diacritical marks in the Hebrew language give words their vocalized form. The task of adding diacritical marks to plain Hebrew text is still dominated by a system that relies heavily on human-curated resources. Recent models trained on diacritized Hebrew texts still present a gap in performance. We use a recently developed char-based PLM to narrowly bridge this gap. Presenting MenakBERT, a character level transformer pretrained on Hebrew text and fine-tuned to produce diacritical marks for Hebrew sentences. We continue to show how finetuning a model for diacritizing transfers to a task such as part of speech tagging.
Authors: Ben Fauber
Abstract: We demonstrate that small pretrained foundational generative language models with millions of parameters can learn the latent rules of a process from data associated with the process. Inspired by Stefan Zweig's novella "Schachnovelle," also known as "The Royal Game" in English, we show that 28M and 125M parameter pretrained foundational small language models (SLMs) can be instruction fine-tuned with 1,000-to-1,000,000 examples to learn the rules of chess, propose legal moves, and accurately solve chess problems. We also explore the impact of successive language model fine-tuning epochs on improved outcomes and demonstrate reductions in model hallucinations by increasing the number of instruction fine-tuning examples.
Authors: Minwook Bae, Hyounghun Kim
Abstract: Generating a long story of several thousand words with narrative coherence using Large Language Models (LLMs) has been a challenging task. Previous research has addressed this challenge by proposing different frameworks that create a story plan and generate a long story based on that plan. However, these frameworks have been mainly focusing on maintaining narrative coherence in stories, often overlooking creativity in story planning and the expressiveness of the stories generated from those plans, which are desirable properties to captivate readers' interest. In this paper, we propose Collective Critics for Creative Story Generation framework (CritiCS), which is composed of plan refining stage (CrPlan) and story generation stage (CrText), to integrate a collective revision mechanism that promotes those properties into long-form story generation process. Specifically, in each stage, a group of LLM critics and one leader collaborate to incrementally refine drafts of plan and story throughout multiple rounds. Extensive human evaluation shows that the CritiCS can significantly enhance story creativity and reader engagement, while also maintaining narrative coherence. Furthermore, the design of the framework allows active participation from human writers in any role within the critique process, enabling interactive human-machine collaboration in story writing.
Authors: Mingyang Wang, Lukas Lange, Heike Adel, Jannik Str\"otgen, Hinrich Sch\"utze
Abstract: To ensure large language models contain up-to-date knowledge, they need to be updated regularly. However, model editing is challenging as it might also affect knowledge that is unrelated to the new data. State-of-the-art methods identify parameters associated with specific knowledge and then modify them via direct weight updates. However, these locate-and-edit methods suffer from heavy computational overhead and lack theoretical validation. In contrast, directly fine-tuning the model on requested edits affects the model's behavior on unrelated knowledge, and significantly damages the model's generation fluency and consistency. To address these challenges, we propose SAUL, a streamlined model editing method that uses sentence concatenation with augmented random facts for generation regularization. Evaluations on three model editing benchmarks show that SAUL is a practical and reliable solution for model editing outperforming state-of-the-art methods while maintaining generation quality and reducing computational overhead.
Authors: Takashi Shibuya, Takehito Utsuro
Abstract: Topic modeling analyzes a collection of documents to learn meaningful patterns of words. However, previous topic models consider only the spelling of words and do not take into consideration the homography of words. In this study, we incorporate the Wikipedia knowledge into a neural topic model to make it aware of named entities. We evaluate our method on two datasets, 1) news articles of \textit{New York Times} and 2) the AIDA-CoNLL dataset. Our experiments show that our method improves the performance of neural topic models in generalizability. Moreover, we analyze frequent terms in each topic and the temporal dependencies between topics to demonstrate that our entity-aware topic models can capture the time-series development of topics well.
Authors: Seokhyun An, Hyounghun Kim
Abstract: Instruction tuning-supervised fine-tuning using instruction-response pairs-is a foundational step in transitioning pre-trained Large Language Models (LLMs) into helpful and safe chat assistants. Our hypothesis is that establishing an adequate output space can enable such a transition given the capabilities inherent in pre-trained LLMs. To verify this, we propose Response Tuning (RT), which eliminates the instruction-conditioning step in instruction tuning and solely focuses on response space supervision. Our experiments demonstrate that RT models, trained only using responses, can effectively respond to a wide range of instructions and exhibit helpfulness comparable to that of their instruction-tuned counterparts. Furthermore, we observe that controlling the training response distribution can significantly improve their user preference or elicit target behaviors such as refusing assistance for unsafe queries. Our findings illuminate the role of establishing an adequate output space in alignment, highlighting the potential of the extensive inherent capabilities of pre-trained LLMs.
Authors: Constanza Fierro, Ruchira Dhar, Filippos Stamatiou, Nicolas Garneau, Anders S{\o}gaard
Abstract: Knowledge claims are abundant in the literature on large language models (LLMs); but can we say that GPT-4 truly "knows" the Earth is round? To address this question, we review standard definitions of knowledge in epistemology and we formalize interpretations applicable to LLMs. In doing so, we identify inconsistencies and gaps in how current NLP research conceptualizes knowledge with respect to epistemological frameworks. Additionally, we conduct a survey of 100 professional philosophers and computer scientists to compare their preferences in knowledge definitions and their views on whether LLMs can really be said to know. Finally, we suggest evaluation protocols for testing knowledge in accordance to the most relevant definitions.
Authors: Jihyoung Jang, Taeyoung Kim, Hyounghun Kim
Abstract: Recently introduced dialogue systems have demonstrated high usability. However, they still fall short of reflecting real-world conversation scenarios. Current dialogue systems exhibit an inability to replicate the dynamic, continuous, long-term interactions involving multiple partners. This shortfall arises because there have been limited efforts to account for both aspects of real-world dialogues: deeply layered interactions over the long-term dialogue and widely expanded conversation networks involving multiple participants. As the effort to incorporate these aspects combined, we introduce Mixed-Session Conversation, a dialogue system designed to construct conversations with various partners in a multi-session dialogue setup. We propose a new dataset called MiSC to implement this system. The dialogue episodes of MiSC consist of 6 consecutive sessions, with four speakers (one main speaker and three partners) appearing in each episode. Also, we propose a new dialogue model with a novel memory management mechanism, called Egocentric Memory Enhanced Mixed-Session Conversation Agent (EMMA). EMMA collects and retains memories from the main speaker's perspective during conversations with partners, enabling seamless continuity in subsequent interactions. Extensive human evaluations validate that the dialogues in MiSC demonstrate a seamless conversational flow, even when conversation partners change in each session. EMMA trained with MiSC is also evaluated to maintain high memorability without contradiction throughout the entire conversation.
Authors: Olga Iakovenko, Thomas Hain
Abstract: Code-switching (CS) is the process of speakers interchanging between two or more languages which in the modern world becomes increasingly common. In order to better describe CS speech the Matrix Language Frame (MLF) theory introduces the concept of a Matrix Language, which is the language that provides the grammatical structure for a CS utterance. In this work the MLF theory was used to develop systems for Matrix Language Identity (MLID) determination. The MLID of English/Mandarin and English/Spanish CS text and speech was compared to acoustic language identity (LID), which is a typical way to identify a language in monolingual utterances. MLID predictors from audio show higher correlation with the textual principles than LID in all cases while also outperforming LID in an MLID recognition task based on F1 macro (60\%) and correlation score (0.38). This novel approach has identified that non-English languages (Mandarin and Spanish) are preferred over the English language as the ML contrary to the monolingual choice of LID.
Authors: John X. Morris, Alexander M. Rush
Abstract: Dense document embeddings are central to neural retrieval. The dominant paradigm is to train and construct embeddings by running encoders directly on individual documents. In this work, we argue that these embeddings, while effective, are implicitly out-of-context for targeted use cases of retrieval, and that a contextualized document embedding should take into account both the document and neighboring documents in context - analogous to contextualized word embeddings. We propose two complementary methods for contextualized document embeddings: first, an alternative contrastive learning objective that explicitly incorporates the document neighbors into the intra-batch contextual loss; second, a new contextual architecture that explicitly encodes neighbor document information into the encoded representation. Results show that both methods achieve better performance than biencoders in several settings, with differences especially pronounced out-of-domain. We achieve state-of-the-art results on the MTEB benchmark with no hard negative mining, score distillation, dataset-specific instructions, intra-GPU example-sharing, or extremely large batch sizes. Our method can be applied to improve performance on any contrastive learning dataset and any biencoder.
Authors: Olga Iakovenko, Ivan Bondarenko, Mariya Borovikova, Daniil Vodolazsky
Abstract: This paper presents an overview of rule-based system for automatic accentuation and phonemic transcription of Russian texts for speech connected tasks, such as Automatic Speech Recognition (ASR). Two parts of the developed system, accentuation and transcription, use different approaches to achieve correct phonemic representations of input phrases. Accentuation is based on "Grammatical dictionary of the Russian language" of A.A. Zaliznyak and wiktionary corpus. To distinguish homographs, the accentuation system also utilises morphological information of the sentences based on Recurrent Neural Networks (RNN). Transcription algorithms apply the rules presented in the monograph of B.M. Lobanov and L.I. Tsirulnik "Computer Synthesis and Voice Cloning". The rules described in the present paper are implemented in an open-source module, which can be of use to any scientific study connected to ASR or Speech To Text (STT) tasks. Automatically marked up text annotations of the Russian Voxforge database were used as training data for an acoustic model in CMU Sphinx. The resulting acoustic model was evaluated on cross-validation, mean Word Accuracy being 71.2%. The developed toolkit is written in the Python language and is accessible on GitHub for any researcher interested.
Authors: Junjie Chen, Xiangheng He, Yusuke Miyao, Danushka Bollegala
Abstract: Unsupervised constituency parsers organize phrases within a sentence into a tree-shaped syntactic constituent structure that reflects the organization of sentence semantics. However, the traditional objective of maximizing sentence log-likelihood (LL) does not explicitly account for the close relationship between the constituent structure and the semantics, resulting in a weak correlation between LL values and parsing accuracy. In this paper, we introduce a novel objective for training unsupervised parsers: maximizing the information between constituent structures and sentence semantics (SemInfo). We introduce a bag-of-substrings model to represent the semantics and apply the probability-weighted information metric to estimate the SemInfo. Additionally, we develop a Tree Conditional Random Field (TreeCRF)-based model to apply the SemInfo maximization objective to Probabilistic Context-Free Grammar (PCFG) induction, the state-of-the-art method for unsupervised constituency parsing. Experiments demonstrate that SemInfo correlates more strongly with parsing accuracy than LL. Our algorithm significantly enhances parsing accuracy by an average of 7.85 points across five PCFG variants and in four languages, achieving new state-of-the-art results in three of the four languages.
Authors: Angana Borah, Rada Mihalcea
Abstract: As Large Language Models (LLMs) continue to evolve, they are increasingly being employed in numerous studies to simulate societies and execute diverse social tasks. However, LLMs are susceptible to societal biases due to their exposure to human-generated data. Given that LLMs are being used to gain insights into various societal aspects, it is essential to mitigate these biases. To that end, our study investigates the presence of implicit gender biases in multi-agent LLM interactions and proposes two strategies to mitigate these biases. We begin by creating a dataset of scenarios where implicit gender biases might arise, and subsequently develop a metric to assess the presence of biases. Our empirical analysis reveals that LLMs generate outputs characterized by strong implicit bias associations (>= 50\% of the time). Furthermore, these biases tend to escalate following multi-agent interactions. To mitigate them, we propose two strategies: self-reflection with in-context examples (ICE); and supervised fine-tuning. Our research demonstrates that both methods effectively mitigate implicit biases, with the ensemble of fine-tuning and self-reflection proving to be the most successful.
Authors: Fantine Huot, Reinald Kim Amplayo, Jennimaria Palomaki, Alice Shoshana Jakobovits, Elizabeth Clark, Mirella Lapata
Abstract: Writing compelling fiction is a multifaceted process combining elements such as crafting a plot, developing interesting characters, and using evocative language. While large language models (LLMs) show promise for story writing, they currently rely heavily on intricate prompting, which limits their use. We propose Agents' Room, a generation framework inspired by narrative theory, that decomposes narrative writing into subtasks tackled by specialized agents. To illustrate our method, we introduce Tell Me A Story, a high-quality dataset of complex writing prompts and human-written stories, and a novel evaluation framework designed specifically for assessing long narratives. We show that Agents' Room generates stories that are preferred by expert evaluators over those produced by baseline systems by leveraging collaboration and specialization to decompose the complex story writing task into tractable components. We provide extensive analysis with automated and human-based metrics of the generated output.
Authors: Mesay Gemeda Yigezu, Melkamu Abay Mersha, Girma Yohannis Bade, Jugal Kalita, Olga Kolesnikova, Alexander Gelbukh
Abstract: The proliferation of fake news has emerged as a significant threat to the integrity of information dissemination, particularly on social media platforms. Misinformation can spread quickly due to the ease of creating and disseminating content, affecting public opinion and sociopolitical events. Identifying false information is therefore essential to reducing its negative consequences and maintaining the reliability of online news sources. Traditional approaches to fake news detection often rely solely on content-based features, overlooking the crucial role of social context in shaping the perception and propagation of news articles. In this paper, we propose a comprehensive approach that integrates social context-based features with news content features to enhance the accuracy of fake news detection in under-resourced languages. We perform several experiments utilizing a variety of methodologies, including traditional machine learning, neural networks, ensemble learning, and transfer learning. Assessment of the outcomes of the experiments shows that the ensemble learning approach has the highest accuracy, achieving a 0.99 F1 score. Additionally, when compared with monolingual models, the fine-tuned model with the target language outperformed others, achieving a 0.94 F1 score. We analyze the functioning of the models, considering the important features that contribute to model performance, using explainable AI techniques.
Authors: Akhilesh Aravapalli, Mounika Marreddy, Subba Reddy Oota, Radhika Mamidi, Manish Gupta
Abstract: Transformer-based models have revolutionized the field of natural language processing. To understand why they perform so well and to assess their reliability, several studies have focused on questions such as: Which linguistic properties are encoded by these models, and to what extent? How robust are these models in encoding linguistic properties when faced with perturbations in the input text? However, these studies have mainly focused on BERT and the English language. In this paper, we investigate similar questions regarding encoding capability and robustness for 8 linguistic properties across 13 different perturbations in 6 Indic languages, using 9 multilingual Transformer models (7 universal and 2 Indic-specific). To conduct this study, we introduce a novel multilingual benchmark dataset, IndicSentEval, containing approximately $\sim$47K sentences. Surprisingly, our probing analysis of surface, syntactic, and semantic properties reveals that while almost all multilingual models demonstrate consistent encoding performance for English, they show mixed results for Indic languages. As expected, Indic-specific multilingual models capture linguistic properties in Indic languages better than universal models. Intriguingly, universal models broadly exhibit better robustness compared to Indic-specific models, particularly under perturbations such as dropping both nouns and verbs, dropping only verbs, or keeping only nouns. Overall, this study provides valuable insights into probing and perturbation-specific strengths and weaknesses of popular multilingual Transformer-based models for different Indic languages. We make our code and dataset publicly available [https://tinyurl.com/IndicSentEval}].
Authors: Tianxiang Hu, Pei Zhang, Baosong Yang, Jun Xie, Derek F. Wong, Rui Wang
Abstract: Achieving consistent high-quality machine translation (MT) across diverse domains remains a significant challenge, primarily due to the limited and imbalanced parallel training data available in various domains. While large language models (LLMs) have demonstrated impressive general understanding and generation abilities, their potential in multi-domain MT is under-explored. We establish a comprehensive benchmark for multi-domain translation, featuring 25 German$\Leftrightarrow$English and 22 Chinese$\Leftrightarrow$English test sets respectively covering 15 domains. Our evaluation of prominent LLMs reveals a discernible performance gap against traditional MT systems, highlighting domain overfitting and catastrophic forgetting issues after fine-tuning on domain-limited corpora. To mitigate this, we propose a domain Chain of Thought (CoT) fine-tuning technique that utilizes the intrinsic multi-domain intelligence of LLMs to improve translation performance. This method inspires the LLM to perceive domain information from the source text, which then serves as a helpful hint to guide the translation process. Despite being trained on a small dataset of four domains, our CoT fine-tune approach achieves notable enhancements in translation accuracy and domain robustness than traditional fine-tuning, as evidenced by an average 1.53 BLEU score increase in over 20 German$\rightarrow$English distinct out-of-domain tests.
Authors: Shijie Chen, Bernal Jim\'enez Guti\'errez, Yu Su
Abstract: Information retrieval (IR) systems have played a vital role in modern digital life and have cemented their continued usefulness in this new era of generative AI via retrieval-augmented generation. With strong language processing capabilities and remarkable versatility, large language models (LLMs) have become popular choices for zero-shot re-ranking in IR systems. So far, LLM-based re-ranking methods rely on strong generative capabilities, which restricts their use to either specialized or powerful proprietary models. Given these restrictions, we ask: is autoregressive generation necessary and optimal for LLMs to perform re-ranking? We hypothesize that there are abundant signals relevant to re-ranking within LLMs that might not be used to their full potential via generation. To more directly leverage such signals, we propose in-context re-ranking (ICR), a novel method that leverages the change in attention pattern caused by the search query for accurate and efficient re-ranking. To mitigate the intrinsic biases in LLMs, we propose a calibration method using a content-free query. Due to the absence of generation, ICR only requires two ($O(1)$) forward passes to re-rank $N$ documents, making it substantially more efficient than generative re-ranking methods that require at least $O(N)$ forward passes. Our novel design also enables ICR to be applied to any LLM without specialized training while guaranteeing a well-formed ranking. Extensive experiments with two popular open-weight LLMs on standard single-hop and multi-hop information retrieval benchmarks show that ICR outperforms RankGPT while cutting the latency by more than 60% in practice. Through detailed analyses, we show that ICR's performance is specially strong on tasks that require more complex re-ranking signals. Our findings call for further exploration on novel ways of utilizing open-weight LLMs beyond text generation.
Authors: Ali Satvaty, Suzan Verberne, Fatih Turkmen
Abstract: While recent research increasingly showcases the remarkable capabilities of Large Language Models (LLMs), it's vital to confront their hidden pitfalls. Among these challenges, the issue of memorization stands out, posing significant ethical and legal risks. In this paper, we presents a Systematization of Knowledge (SoK) on the topic of memorization in LLMs. Memorization is the effect that a model tends to store and reproduce phrases or passages from the training data and has been shown to be the fundamental issue to various privacy and security attacks against LLMs. We begin by providing an overview of the literature on the memorization, exploring it across five key dimensions: intentionality, degree, retrievability, abstraction, and transparency. Next, we discuss the metrics and methods used to measure memorization, followed by an analysis of the factors that contribute to memorization phenomenon. We then examine how memorization manifests itself in specific model architectures and explore strategies for mitigating these effects. We conclude our overview by identifying potential research topics for the near future: to develop methods for balancing performance and privacy in LLMs, and the analysis of memorization in specific contexts, including conversational agents, retrieval-augmented generation, multilingual language models, and diffusion language models.
Authors: Somesh Singh, Yaman K Singla, Harini SI, Balaji Krishnamurthy
Abstract: LLMs are increasingly being used in workflows involving generating content to be consumed by humans (e.g., marketing) and also in directly interacting with humans (e.g., through chatbots). The development of such systems that are capable of generating verifiably persuasive messages presents both opportunities and challenges for society. On the one hand, such systems could positively impact domains like advertising and social good, such as addressing drug addiction, and on the other, they could be misused for spreading misinformation and shaping political opinions. To channel LLMs' impact on society, we need to develop systems to measure and benchmark their persuasiveness. With this motivation, we introduce PersuasionBench and PersuasionArena, the first large-scale benchmark and arena containing a battery of tasks to measure the persuasion ability of generative models automatically. We investigate to what extent LLMs know and leverage linguistic patterns that can help them generate more persuasive language. Our findings indicate that the persuasiveness of LLMs correlates positively with model size, but smaller models can also be made to have a higher persuasiveness than much larger models. Notably, targeted training using synthetic and natural datasets significantly enhances smaller models' persuasive capabilities, challenging scale-dependent assumptions. Our findings carry key implications for both model developers and policymakers. For instance, while the EU AI Act and California's SB-1047 aim to regulate AI models based on the number of floating point operations, we demonstrate that simple metrics like this alone fail to capture the full scope of AI's societal impact. We invite the community to explore and contribute to PersuasionArena and PersuasionBench, available at https://bit.ly/measure-persuasion, to advance our understanding of AI-driven persuasion and its societal implications.
Authors: Sarah Masud, Sahajpreet Singh, Viktor Hangya, Alexander Fraser, Tanmoy Chakraborty
Abstract: For subjective tasks such as hate detection, where people perceive hate differently, the Large Language Model's (LLM) ability to represent diverse groups is unclear. By including additional context in prompts, we comprehensively analyze LLM's sensitivity to geographical priming, persona attributes, and numerical information to assess how well the needs of various groups are reflected. Our findings on two LLMs, five languages, and six datasets reveal that mimicking persona-based attributes leads to annotation variability. Meanwhile, incorporating geographical signals leads to better regional alignment. We also find that the LLMs are sensitive to numerical anchors, indicating the ability to leverage community-based flagging efforts and exposure to adversaries. Our work provides preliminary guidelines and highlights the nuances of applying LLMs in culturally sensitive cases.
Authors: Tianyu Gao, Alexander Wettig, Howard Yen, Danqi Chen
Abstract: We study continued training and supervised fine-tuning (SFT) of a language model (LM) to make effective use of long-context information. We first establish a reliable evaluation protocol to guide model development -- Instead of perplexity or simple needle-in-a-haystack (NIAH) tests, we use a broad set of long-context tasks, and we evaluate models after SFT with instruction data as this better reveals long-context abilities. Supported by our robust evaluations, we run thorough experiments to decide the data mix for continued pre-training, the instruction tuning dataset, and many other design choices. We find that (1) code repositories and books are excellent sources of long data, but it is crucial to combine them with high-quality short data; (2) training with a sequence length beyond the evaluation length boosts long-context performance; (3) for SFT, using only short instruction datasets yields strong performance on long-context tasks. Our final model, ProLong-8B, which is initialized from Llama-3 and trained on 40B tokens, demonstrates state-of-the-art long-context performance among similarly sized models at a length of 128K. ProLong outperforms Llama-3.18B-Instruct on the majority of long-context tasks despite having seen only 5% as many tokens during long-context training. Additionally, ProLong can effectively process up to 512K tokens, one of the longest context windows of publicly available LMs.
Authors: Craig Messner, Tom Lippincott
Abstract: We present a dataset of 19th century American literary orthovariant tokens with a novel layer of human-annotated dialect group tags designed to serve as the basis for computational experiments exploring literarily meaningful orthographic variation. We perform an initial broad set of experiments over this dataset using both token (BERT) and character (CANINE)-level contextual language models. We find indications that the "dialect effect" produced by intentional orthographic variation employs multiple linguistic channels, and that these channels are able to be surfaced to varied degrees given particular language modelling assumptions. Specifically, we find evidence showing that choice of tokenization scheme meaningfully impact the type of orthographic information a model is able to surface.
Authors: Yu Ying Chiu, Liwei Jiang, Bill Yuchen Lin, Chan Young Park, Shuyue Stella Li, Sahithya Ravi, Mehar Bhatia, Maria Antoniak, Yulia Tsvetkov, Vered Shwartz, Yejin Choi
Abstract: To make large language models (LLMs) more helpful across diverse cultures, it is essential to have effective cultural knowledge benchmarks to measure and track our progress. Effective benchmarks need to be robust, diverse, and challenging. We introduce CulturalBench: a set of 1,227 human-written and human-verified questions for effectively assessing LLMs' cultural knowledge, covering 45 global regions including the underrepresented ones like Bangladesh, Zimbabwe, and Peru. Questions - each verified by five independent annotators - span 17 diverse topics ranging from food preferences to greeting etiquettes. We evaluate models on two setups: CulturalBench-Easy and CulturalBench-Hard which share the same questions but asked differently. We find that LLMs are sensitive to such difference in setups (e.g., GPT-4o with 27.3% difference). Compared to human performance (92.6% accuracy), CulturalBench-Hard is more challenging for frontier LLMs with the best performing model (GPT-4o) at only 61.5% and the worst (Llama3-8b) at 21.4%. Moreover, we find that LLMs often struggle with tricky questions that have multiple correct answers (e.g., What utensils do the Chinese usually use?), revealing a tendency to converge to a single answer. Our results also indicate that OpenAI GPT-4o substantially outperform other proprietary and open source models in questions related to all but one region (Oceania). Nonetheless, all models consistently underperform on questions related to South America and the Middle East.
Authors: William Held, Ella Li, Michael Ryan, Weiyan Shi, Yanzhe Zhang, Diyi Yang
Abstract: Voice assistants, such as Siri and Google Assistant, typically model audio and text separately, resulting in lost speech information and increased complexity. Recent efforts to address this with end-to-end Speech Large Language Models (LLMs) trained with supervised finetuning (SFT) have led to models ``forgetting" capabilities from text-only LLMs. Our work proposes an alternative paradigm for training Speech LLMs without instruction data, using the response of a text-only LLM to transcripts as self-supervision. Importantly, this process can be performed without annotated responses. We show that our Distilled Voice Assistant (DiVA) generalizes to Spoken Question Answering, Classification, and Translation. Furthermore, we show that DiVA better meets user preferences, achieving a 72\% win rate compared with state-of-the-art models like Qwen 2 Audio, despite using $>$100x less training compute.
Authors: Yu Ying Chiu, Liwei Jiang, Yejin Choi
Abstract: As we increasingly seek guidance from LLMs for decision-making in daily life, many of these decisions are not clear-cut and depend significantly on the personal values and ethical standards of the users. We present DailyDilemmas, a dataset of 1,360 moral dilemmas encountered in everyday life. Each dilemma includes two possible actions and with each action, the affected parties and human values invoked. Based on these dilemmas, we consolidated a set of human values across everyday topics e.g., interpersonal relationships, workplace, and environmental issues. We evaluated LLMs on these dilemmas to determine what action they will take and the values represented by these actions. Then, we analyzed these values through the lens of five popular theories inspired by sociology, psychology and philosophy. These theories are: World Value Survey, Moral Foundation Theory, Maslow's Hierarchy of Needs, Aristotle's Virtues, and Plutchik Wheel of Emotion. We find that LLMs are most aligned with the self-expression over survival values in terms of World Value Survey, care over loyalty in Moral Foundation Theory. Interestingly, we find large preferences differences in models for some core values such as truthfulness e.g., Mixtral-8x7B model tends to neglect it by 9.7% while GPT-4-turbo model tends to select it by 9.4%. We also study the recent guidance released by OpenAI (ModelSpec), and Anthropic (Constitutional AI) to understand how their released principles reflect their actual value prioritization when facing nuanced moral reasoning in daily-life settings. We find that end users cannot effectively steer such prioritization using system prompts.
Authors: Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Ruibin Yuan, Xueqi Cheng
Abstract: As Large Language Models (LLMs) grow increasingly powerful, ensuring their safety and alignment with human values remains a critical challenge. Ideally, LLMs should provide informative responses while avoiding the disclosure of harmful or sensitive information. However, current alignment approaches, which rely heavily on refusal strategies, such as training models to completely reject harmful prompts or applying coarse filters are limited by their binary nature. These methods either fully deny access to information or grant it without sufficient nuance, leading to overly cautious responses or failures to detect subtle harmful content. For example, LLMs may refuse to provide basic, public information about medication due to misuse concerns. Moreover, these refusal-based methods struggle to handle mixed-content scenarios and lack the ability to adapt to context-dependent sensitivities, which can result in over-censorship of benign content. To overcome these challenges, we introduce HiddenGuard, a novel framework for fine-grained, safe generation in LLMs. HiddenGuard incorporates Prism (rePresentation Router for In-Stream Moderation), which operates alongside the LLM to enable real-time, token-level detection and redaction of harmful content by leveraging intermediate hidden states. This fine-grained approach allows for more nuanced, context-aware moderation, enabling the model to generate informative responses while selectively redacting or replacing sensitive information, rather than outright refusal. We also contribute a comprehensive dataset with token-level fine-grained annotations of potentially harmful information across diverse contexts. Our experiments demonstrate that HiddenGuard achieves over 90% in F1 score for detecting and redacting harmful content while preserving the overall utility and informativeness of the model's responses.
Authors: Mario Giulianelli, Luca Malagutti, Juan Luis Gastaldi, Brian DuSell, Tim Vieira, Ryan Cotterell
Abstract: Language models are widely used in computational psycholinguistics to test theories that relate the negative log probability (the surprisal) of a region of interest (a substring of characters) under a language model to its cognitive cost experienced by readers, as operationalized, for example, by gaze duration on the region. However, the application of modern language models to psycholinguistic studies is complicated by the practice of using tokenization as an intermediate step in training a model. Doing so results in a language model over token strings rather than one over character strings. Vexingly, regions of interest are generally misaligned with these token strings. The paper argues that token-level language models should be (approximately) marginalized into character-level language models before they are used in psycholinguistic studies to compute the surprisal of a region of interest; then, the marginalized character-level language model can be used to compute the surprisal of an arbitrary character substring, which we term a focal area, that the experimenter may wish to use as a predictor. Our proposal of marginalizing a token-level model into a character-level one solves this misalignment issue independently of the tokenization scheme. Empirically, we discover various focal areas whose surprisal is a better psychometric predictor than the surprisal of the region of interest itself.
Authors: Howard Yen, Tianyu Gao, Minmin Hou, Ke Ding, Daniel Fleischer, Peter Izasak, Moshe Wasserblat, Danqi Chen
Abstract: There have been many benchmarks for evaluating long-context language models (LCLMs), but developers often rely on synthetic tasks like needle-in-a-haystack (NIAH) or arbitrary subsets of tasks. It remains unclear whether they translate to the diverse downstream applications of LCLMs, and the inconsistency further complicates model comparison. We investigate the underlying reasons behind current practices and find that existing benchmarks often provide noisy signals due to low coverage of applications, insufficient lengths, unreliable metrics, and incompatibility with base models. In this work, we present HELMET (How to Evaluate Long-context Models Effectively and Thoroughly), a comprehensive benchmark encompassing seven diverse, application-centric categories. We also address many issues in previous benchmarks by adding controllable lengths up to 128k tokens, model-based evaluation for reliable metrics, and few-shot prompting for robustly evaluating base models. Consequently, we demonstrate that HELMET offers more reliable and consistent rankings of frontier LCLMs. Through a comprehensive study of 51 LCLMs, we find that (1) synthetic tasks like NIAH are not good predictors of downstream performance; (2) the diverse categories in HELMET exhibit distinct trends and low correlation with each other; and (3) while most LCLMs achieve perfect NIAH scores, open-source models significantly lag behind closed ones when the task requires full-context reasoning or following complex instructions -- the gap widens with increased lengths. Finally, we recommend using our RAG tasks for fast model development, as they are easy to run and more predictive of other downstream performance; ultimately, we advocate for a holistic evaluation across diverse tasks.
Authors: Yaniv Leviathan, Matan Kalman, Yossi Matias
Abstract: Unneeded elements in the attention's context degrade performance. We introduce Selective Attention, a simple parameter-free change to the standard attention mechanism which reduces attention to unneeded elements. Selective attention improves language modeling performance in a variety of model sizes and context lengths. For example, a range of transformers trained with the language modeling objective on C4 with selective attention perform equivalently to standard transformers with ~2X more heads and parameters in their attention modules. Selective attention also allows decreasing the size of the attention's context buffer, leading to meaningful reductions in the memory and compute requirements during inference. For example, transformers with 100M parameters trained on C4 with context sizes of 512, 1,024, and 2,048 need 16X, 25X, and 47X less memory for their attention module, respectively, when equipped with selective attention, as those without selective attention, with the same validation perplexity.
Authors: Hadas Orgad, Michael Toker, Zorik Gekhman, Roi Reichart, Idan Szpektor, Hadas Kotek, Yonatan Belinkov
Abstract: Large language models (LLMs) often produce errors, including factual inaccuracies, biases, and reasoning failures, collectively referred to as "hallucinations". Recent studies have demonstrated that LLMs' internal states encode information regarding the truthfulness of their outputs, and that this information can be utilized to detect errors. In this work, we show that the internal representations of LLMs encode much more information about truthfulness than previously recognized. We first discover that the truthfulness information is concentrated in specific tokens, and leveraging this property significantly enhances error detection performance. Yet, we show that such error detectors fail to generalize across datasets, implying that -- contrary to prior claims -- truthfulness encoding is not universal but rather multifaceted. Next, we show that internal representations can also be used for predicting the types of errors the model is likely to make, facilitating the development of tailored mitigation strategies. Lastly, we reveal a discrepancy between LLMs' internal encoding and external behavior: they may encode the correct answer, yet consistently generate an incorrect one. Taken together, these insights deepen our understanding of LLM errors from the model's internal perspective, which can guide future research on enhancing error analysis and mitigation.
Authors: Zixuan Li, Jing Xiong, Fanghua Ye, Chuanyang Zheng, Xun Wu, Jianqiao Lu, Zhongwei Wan, Xiaodan Liang, Chengming Li, Zhenan Sun, Lingpeng Kong, Ngai Wong
Abstract: We present UncertaintyRAG, a novel approach for long-context Retrieval-Augmented Generation (RAG) that utilizes Signal-to-Noise Ratio (SNR)-based span uncertainty to estimate similarity between text chunks. This span uncertainty enhances model calibration, improving robustness and mitigating semantic inconsistencies introduced by random chunking. Leveraging this insight, we propose an efficient unsupervised learning technique to train the retrieval model, alongside an effective data sampling and scaling strategy. UncertaintyRAG outperforms baselines by 2.03% on LLaMA-2-7B, achieving state-of-the-art results while using only 4% of the training data compared to other advanced open-source retrieval models under distribution shift settings. Our method demonstrates strong calibration through span uncertainty, leading to improved generalization and robustness in long-context RAG tasks. Additionally, UncertaintyRAG provides a lightweight retrieval model that can be integrated into any large language model with varying context window lengths, without the need for fine-tuning, showcasing the flexibility of our approach.
Authors: Ryan C. Barron, Ves Grantcharov, Selma Wanna, Maksim E. Eren, Manish Bhattarai, Nicholas Solovyev, George Tompkins, Charles Nicholas, Kim {\O}. Rasmussen, Cynthia Matuszek, Boian S. Alexandrov
Abstract: Large Language Models (LLMs) are pre-trained on large-scale corpora and excel in numerous general natural language processing (NLP) tasks, such as question answering (QA). Despite their advanced language capabilities, when it comes to domain-specific and knowledge-intensive tasks, LLMs suffer from hallucinations, knowledge cut-offs, and lack of knowledge attributions. Additionally, fine tuning LLMs' intrinsic knowledge to highly specific domains is an expensive and time consuming process. The retrieval-augmented generation (RAG) process has recently emerged as a method capable of optimization of LLM responses, by referencing them to a predetermined ontology. It was shown that using a Knowledge Graph (KG) ontology for RAG improves the QA accuracy, by taking into account relevant sub-graphs that preserve the information in a structured manner. In this paper, we introduce SMART-SLIC, a highly domain-specific LLM framework, that integrates RAG with KG and a vector store (VS) that store factual domain specific information. Importantly, to avoid hallucinations in the KG, we build these highly domain-specific KGs and VSs without the use of LLMs, but via NLP, data mining, and nonnegative tensor factorization with automatic model selection. Pairing our RAG with a domain-specific: (i) KG (containing structured information), and (ii) VS (containing unstructured information) enables the development of domain-specific chat-bots that attribute the source of information, mitigate hallucinations, lessen the need for fine-tuning, and excel in highly domain-specific question answering tasks. We pair SMART-SLIC with chain-of-thought prompting agents. The framework is designed to be generalizable to adapt to any specific or specialized domain. In this paper, we demonstrate the question answering capabilities of our framework on a corpus of scientific publications on malware analysis and anomaly detection.
Authors: Rohin Manvi, Anikait Singh, Stefano Ermon
Abstract: Inference-time computation is a powerful paradigm to enhance the performance of large language models (LLMs), with Best-of-N sampling being a widely used technique. However, this method is computationally expensive, requiring both (1) an external reward model and (2) the generation of multiple samples. In this work, we introduce a new generative self-evaluation scheme designed to adaptively reduce the number of generated samples while maintaining or even improving performance. We use a generative reward model formulation, allowing the LLM to predict mid-generation the probability that restarting the generation will yield a better response. These predictions are obtained without an external reward model and can be used to decide whether or not to generate more samples, prune unpromising samples early on, or to pick the best sample. This capability is very inexpensive as it involves generating a single predefined token. Trained using a dataset constructed with real unfiltered LMSYS user prompts, Llama 3.1 8B's win rate against GPT-4 on AlpacaEval increases from 21% to 34% with 16 samples and math performance on GSM8K improves from 84% to 91%. By sampling only when the LLM determines that it is beneficial to do so and adaptively adjusting temperature annealing, we demonstrate that 74% of the improvement from using 16 samples can be achieved with only 1.2 samples on average. We further demonstrate that 50-75% of samples can be pruned early in generation with minimal degradation in performance. Overall, our methods enable more efficient and scalable compute utilization during inference for LLMs.
Authors: Jaewoo Lee, Joonho Ko, Jinheon Baek, Soyeong Jeong, Sung Ju Hwang
Abstract: Information Retrieval (IR) methods aim to identify relevant documents in response to a given query, which have gained remarkable attention due to their successful application in various natural language tasks. However, existing approaches typically consider only the textual information within the documents, which overlooks the fact that documents can contain multiple modalities, including texts, images, and tables. Further, they often segment each long document into multiple discrete passages for embedding, preventing them from capturing the overall document context and interactions between paragraphs. We argue that these two limitations lead to suboptimal document representations for retrieval. In this work, to address them, we aim to produce more comprehensive and nuanced document representations by holistically embedding documents interleaved with different modalities. Specifically, we achieve this by leveraging the capability of recent vision-language models that enable the processing and integration of text, images, and tables into a unified format and representation. Moreover, to mitigate the information loss from segmenting documents into passages, instead of representing and retrieving passages individually, we further merge the representations of segmented passages into one single document representation, while we additionally introduce a reranking strategy to decouple and identify the relevant passage within the document if necessary. Then, through extensive experiments on diverse information retrieval scenarios considering both the textual and multimodal queries, we show that our approach substantially outperforms relevant baselines, thanks to the consideration of the multimodal information interleaved within the documents in a unified way.
Authors: Jiayi Ye, Yanbo Wang, Yue Huang, Dongping Chen, Qihui Zhang, Nuno Moniz, Tian Gao, Werner Geyer, Chao Huang, Pin-Yu Chen, Nitesh V Chawla, Xiangliang Zhang
Abstract: LLM-as-a-Judge has been widely utilized as an evaluation method in various benchmarks and served as supervised rewards in model training. However, despite their excellence in many domains, potential issues are under-explored, undermining their reliability and the scope of their utility. Therefore, we identify 12 key potential biases and propose a new automated bias quantification framework-CALM-which systematically quantifies and analyzes each type of bias in LLM-as-a-Judge by using automated and principle-guided modification. Our experiments cover multiple popular language models, and the results indicate that while advanced models have achieved commendable overall performance, significant biases persist in certain specific tasks. Empirical results suggest that there remains room for improvement in the reliability of LLM-as-a-Judge. Moreover, we also discuss the explicit and implicit influence of these biases and give some suggestions for the reliable application of LLM-as-a-Judge. Our work highlights the need for stakeholders to address these issues and remind users to exercise caution in LLM-as-a-Judge applications.
Authors: Lei Xu, Mohammed Asad Karim, Saket Dingliwal, Aparna Elangovan
Abstract: Large language models (LLMs) can generate fluent summaries across domains using prompting techniques, reducing the need to train models for summarization applications. However, crafting effective prompts that guide LLMs to generate summaries with the appropriate level of detail and writing style remains a challenge. In this paper, we explore the use of salient information extracted from the source document to enhance summarization prompts. We show that adding keyphrases in prompts can improve ROUGE F1 and recall, making the generated summaries more similar to the reference and more complete. The number of keyphrases can control the precision-recall trade-off. Furthermore, our analysis reveals that incorporating phrase-level salient information is superior to word- or sentence-level. However, the impact on hallucination is not universally positive across LLMs. To conduct this analysis, we introduce Keyphrase Signal Extractor (SigExt), a lightweight model that can be finetuned to extract salient keyphrases. By using SigExt, we achieve consistent ROUGE improvements across datasets and open-weight and proprietary LLMs without any LLM customization. Our findings provide insights into leveraging salient information in building prompt-based summarization systems.
Authors: Haolan Liu, Jishen Zhao
Abstract: Despite a widespread success in various applications, large language models (LLMs) often stumble when tackling basic physical reasoning or executing robotics tasks, due to a lack of direct experience with the physical nuances of the real world. To address these issues, we propose a Grounding Large language model with Imperfect world MOdel (GLIMO), which utilizes proxy world models such as simulators to collect and synthesize trining data. GLIMO incorporates an LLM agent-based data generator to automatically create high-quality and diverse instruction datasets. The generator includes an iterative self-refining module for temporally consistent experience sampling, a diverse set of question-answering instruction seeds, and a retrieval-augmented generation module for reflecting on prior experiences. Comprehensive experiments show that our approach improve the performance of strong open-source LLMs like LLaMA-3 with a performance boost of 2.04 $\times$, 1.54 $\times$, and 1.82 $\times$ across three different benchmarks, respectively. The performance is able to compete with or surpass their larger counterparts such as GPT-4.
Authors: Yekun Chai, Haoran Sun, Huang Fang, Shuohuan Wang, Yu Sun, Hua Wu
Abstract: Reinforcement learning from human feedback (RLHF) has demonstrated effectiveness in aligning large language models (LLMs) with human preferences. However, token-level RLHF suffers from the credit assignment problem over long sequences, where delayed rewards make it challenging for the model to discern which actions contributed to successful outcomes. This hinders learning efficiency and slows convergence. In this paper, we propose MA-RLHF, a simple yet effective RLHF framework that incorporates macro actions -- sequences of tokens or higher-level language constructs -- into the learning process. By operating at this higher level of abstraction, our approach reduces the temporal distance between actions and rewards, facilitating faster and more accurate credit assignment. This results in more stable policy gradient estimates and enhances learning efficiency within each episode, all without increasing computational complexity during training or inference. We validate our approach through extensive experiments across various model sizes and tasks, including text summarization, dialogue generation, question answering, and program synthesis. Our method achieves substantial performance improvements over standard RLHF, with performance gains of up to 30% in text summarization and code generation, 18% in dialogue, and 8% in question answering tasks. Notably, our approach reaches parity with vanilla RLHF 1.7x to 2x faster in terms of training time and continues to outperform it with further training. We will make our code and data publicly available at https://github.com/ernie-research/MA-RLHF .
Authors: Franck Signe Talla, Herve Jegou, Edouard Grave
Abstract: We address the problem of extending a pretrained large language model to a new domain that was not seen at training time, like adding a language for which the original model has seen no or little training data. Popular solutions like fine-tuning or low-rank adaptation are successful at domain adaptation, but formally they do not add any extra capacity and degrade the performance in the original domain. Our paper analyzes this extension problem under three angles: data, architecture and training procedure, which are advantageously considered jointly. In particular, we improve adapters and make it possible to learn an entire new language while ensuring that the output of the neural network is almost unchanged in the original domain. For this purpose, we modify the new residual blocks in a way that leads each new residual block to output near-zeros in the original domain. This solution of neutral residues, which borrows architectural components from mixture of experts, is effective: with only 20% extra learnable weights compared to an original model trained on English, we get results that are significantly better than concurrent approaches (fine-tuning, low-rank or vanilla adapters) in terms of the trade-off between learning a new language and not forgetting English.
Authors: Han He, Qianchu Liu, Lei Xu, Chaitanya Shivade, Yi Zhang, Sundararajan Srinivasan, Katrin Kirchhoff
Abstract: Large language models (LLMs) can generate fluent summaries across domains using prompting techniques, reducing the need to train models for summarization applications. However, crafting effective prompts that guide LLMs to generate summaries with the appropriate level of detail and writing style remains a challenge. In this paper, we explore the use of salient information extracted from the source document to enhance summarization prompts. We show that adding keyphrases in prompts can improve ROUGE F1 and recall, making the generated summaries more similar to the reference and more complete. The number of keyphrases can control the precision-recall trade-off. Furthermore, our analysis reveals that incorporating phrase-level salient information is superior to word- or sentence-level. However, the impact on hallucination is not universally positive across LLMs. To conduct this analysis, we introduce Keyphrase Signal Extractor (CriSPO), a lightweight model that can be finetuned to extract salient keyphrases. By using CriSPO, we achieve consistent ROUGE improvements across datasets and open-weight and proprietary LLMs without any LLM customization. Our findings provide insights into leveraging salient information in building prompt-based summarization systems.
Authors: Jifan Zhang, Robert Nowak
Abstract: Creating specialized large language models requires vast amounts of clean, special purpose data for training and fine-tuning. With only a handful of existing large-scale, domain-specific datasets, creation of new datasets is required in most applications. This requires the development of new application-specific filtering of web-scale data. Filtering with a high-performance, general-purpose LLM such as GPT-4o can be highly effective, but this is extremely expensive at web-scale. This paper proposes SIEVE, a lightweight alternative that matches GPT-4o accuracy at a fraction of the cost. SIEVE can perform up to 500 filtering operations for the cost of one GPT-4o filtering call. The key to SIEVE is a seamless integration of GPT-4o and lightweight T5 models, using active learning to fine-tune T5 in the background with a small number of calls to GPT-4o. Once trained, it performs as well as GPT-4o at a tiny fraction of the cost. We experimentally validate SIEVE on the OpenWebText dataset, using five highly customized filter tasks targeting high quality and domain-specific content. Our results demonstrate the effectiveness and efficiency of our method in curating large, high-quality datasets for language model training at a substantially lower cost (1%) than existing techniques. To further validate SIEVE, experiments show that SIEVE and GPT-4o achieve similar accuracy, with human evaluators preferring SIEVE's filtering results to those of GPT-4o.
Authors: Milan Straka
Abstract: We present CorPipe 24, the winning entry to the CRAC 2024 Shared Task on Multilingual Coreference Resolution. In this third iteration of the shared task, a novel objective is to also predict empty nodes needed for zero coreference mentions (while the empty nodes were given on input in previous years). This way, coreference resolution can be performed on raw text. We evaluate two model variants: a~two-stage approach (where the empty nodes are predicted first using a pretrained encoder model and then processed together with sentence words by another pretrained model) and a single-stage approach (where a single pretrained encoder model generates empty nodes, coreference mentions, and coreference links jointly). In both settings, CorPipe surpasses other participants by a large margin of 3.9 and 2.8 percent points, respectively. The source code and the trained model are available at https://github.com/ufal/crac2024-corpipe .
Authors: Rohit Gandikota, Sheridan Feucht, Samuel Marks, David Bau
Abstract: Concept erasure in language models has traditionally lacked a comprehensive evaluation framework, leading to incomplete assessments of effectiveness of erasure methods. We propose an evaluation paradigm centered on three critical criteria: innocence (complete knowledge removal), seamlessness (maintaining conditional fluent generation), and specificity (preserving unrelated task performance). Our evaluation metrics naturally motivate the development of Erasure of Language Memory (ELM), a new method designed to address all three dimensions. ELM employs targeted low-rank updates to alter output distributions for erased concepts while preserving overall model capabilities including fluency when prompted for an erased concept. We demonstrate ELM's efficacy on biosecurity, cybersecurity, and literary domain erasure tasks. Comparative analysis shows that ELM achieves superior performance across our proposed metrics, including near-random scores on erased topic assessments, generation fluency, maintained accuracy on unrelated benchmarks, and robustness under adversarial attacks. Our code, data, and trained models are available at https://elm.baulab.info
URLs: https://elm.baulab.info
Authors: Melkamu Mersha, Khang Lam, Joseph Wood, Ali AlShami, Jugal Kalita
Abstract: Artificial intelligence models encounter significant challenges due to their black-box nature, particularly in safety-critical domains such as healthcare, finance, and autonomous vehicles. Explainable Artificial Intelligence (XAI) addresses these challenges by providing explanations for how these models make decisions and predictions, ensuring transparency, accountability, and fairness. Existing studies have examined the fundamental concepts of XAI, its general principles, and the scope of XAI techniques. However, there remains a gap in the literature as there are no comprehensive reviews that delve into the detailed mathematical representations, design methodologies of XAI models, and other associated aspects. This paper provides a comprehensive literature review encompassing common terminologies and definitions, the need for XAI, beneficiaries of XAI, a taxonomy of XAI methods, and the application of XAI methods in different application areas. The survey is aimed at XAI researchers, XAI practitioners, AI model developers, and XAI beneficiaries who are interested in enhancing the trustworthiness, transparency, accountability, and fairness of their AI models.
Authors: Fiifi Dawson, Zainab Mosunmola, Sahil Pocker, Raj Abhijit Dandekar, Rajat Dandekar, Sreedath Panat
Abstract: Although LLMs have been extremely effective in a large number of complex tasks, their understanding and functionality for regional languages and cultures are not well studied. In this paper, we explore the ability of various LLMs to comprehend the cultural aspects of two regional languages: Malayalam (state of Kerala, India) and Yoruba (West Africa). Using Hofstede's six cultural dimensions: Power Distance (PDI), Individualism (IDV), Motivation towards Achievement and Success (MAS), Uncertainty Avoidance (UAV), Long Term Orientation (LTO), and Indulgence (IVR), we quantify the cultural awareness of LLM-based responses. We demonstrate that although LLMs show a high cultural similarity for English, they fail to capture the cultural nuances across these 6 metrics for Malayalam and Yoruba. We also highlight the need for large-scale regional language LLM training with culturally enriched datasets. This will have huge implications for enhancing the user experience of chat-based LLMs and also improving the validity of large-scale LLM agent-based market research.
Authors: Qian Niu, Keyu Chen, Ming Li, Pohsun Feng, Ziqian Bi, Junyu Liu, Benji Peng
Abstract: Large Language Models (LLMs) have rapidly evolved from text-based systems to multimodal platforms, significantly impacting various sectors including healthcare. This comprehensive review explores the progression of LLMs to Multimodal Large Language Models (MLLMs) and their growing influence in medical practice. We examine the current landscape of MLLMs in healthcare, analyzing their applications across clinical decision support, medical imaging, patient engagement, and research. The review highlights the unique capabilities of MLLMs in integrating diverse data types, such as text, images, and audio, to provide more comprehensive insights into patient health. We also address the challenges facing MLLM implementation, including data limitations, technical hurdles, and ethical considerations. By identifying key research gaps, this paper aims to guide future investigations in areas such as dataset development, modality alignment methods, and the establishment of ethical guidelines. As MLLMs continue to shape the future of healthcare, understanding their potential and limitations is crucial for their responsible and effective integration into medical practice.
Authors: Alexander Wuttke, Matthias A{\ss}enmacher, Christopher Klamm, Max M. Lang, Quirin W\"urschinger, Frauke Kreuter
Abstract: Traditional methods for eliciting people's opinions face a trade-off between depth and scale: structured surveys enable large-scale data collection but limit respondents' ability to express unanticipated thoughts in their own words, while conversational interviews provide deeper insights but are resource-intensive. This study explores the potential of replacing human interviewers with large language models (LLMs) to conduct scalable conversational interviews. Our goal is to assess the performance of AI Conversational Interviewing and to identify opportunities for improvement in a controlled environment. We conducted a small-scale, in-depth study with university students who were randomly assigned to be interviewed by either AI or human interviewers, both employing identical questionnaires on political topics. Various quantitative and qualitative measures assessed interviewer adherence to guidelines, response quality, participant engagement, and overall interview efficacy. The findings indicate the viability of AI Conversational Interviewing in producing quality data comparable to traditional methods, with the added benefit of scalability. Based on our experiences, we present specific recommendations for effective implementation.
Authors: Hui Yi Leong, Yi Fan Gao, Shuai Ji, Bora Kalaycioglu, Uktu Pamuksuz
Abstract: The increasing administrative burden of medical documentation, particularly through Electronic Health Records (EHR), significantly reduces the time available for direct patient care and contributes to physician burnout. To address this issue, we propose MediNotes, an advanced generative AI framework designed to automate the creation of SOAP (Subjective, Objective, Assessment, Plan) notes from medical conversations. MediNotes integrates Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and Automatic Speech Recognition (ASR) to capture and process both text and voice inputs in real time or from recorded audio, generating structured and contextually accurate medical notes. The framework also incorporates advanced techniques like Quantized Low-Rank Adaptation (QLoRA) and Parameter-Efficient Fine-Tuning (PEFT) for efficient model fine-tuning in resource-constrained environments. Additionally, MediNotes offers a query-based retrieval system, allowing healthcare providers and patients to access relevant medical information quickly and accurately. Evaluations using the ACI-BENCH dataset demonstrate that MediNotes significantly improves the accuracy, efficiency, and usability of automated medical documentation, offering a robust solution to reduce the administrative burden on healthcare professionals while improving the quality of clinical workflows.
Authors: Jaehoon Oh, Seungjun Shin, Dokwan Oh
Abstract: Massive activations, which manifest in specific feature dimensions of hidden states, introduce a significant bias in large language models (LLMs), leading to an overemphasis on the corresponding token. In this paper, we identify that massive activations originate not from the hidden state but from the intermediate state of a feed-forward network module in an early layer. Expanding on the previous observation that massive activations occur only in specific feature dimensions, we dive deep into the weights that cause massive activations. Specifically, we define top-$k$ massive weights as the weights that contribute to the dimensions with the top-$k$ magnitudes in the intermediate state. When these massive weights are set to zero, the functionality of LLMs is entirely disrupted. However, when all weights except for massive weights are set to zero, it results in a relatively minor performance drop, even though a much larger number of weights are set to zero. This implies that during the pre-training process, learning is dominantly focused on massive weights. Building on this observation, we propose a simple plug-and-play method called MacDrop (massive weights curriculum dropout), to rely less on massive weights during parameter-efficient fine-tuning. This method applies dropout to the pre-trained massive weights, starting with a high dropout probability and gradually decreasing it as fine-tuning progresses. Through experiments, we demonstrate that MacDrop generally improves performance across zero-shot downstream tasks and generation tasks.
Authors: Yibo Zhong, Haoxiang Jiang, Lincan Li, Ryumei Nakada, Tianci Liu, Linjun Zhang, Huaxiu Yao, Haoyu Wang
Abstract: Fine-tuning pre-trained models is crucial for adapting large models to downstream tasks, often delivering state-of-the-art performance. However, fine-tuning all model parameters is resource-intensive and laborious, leading to the emergence of parameter-efficient fine-tuning (PEFT) methods. One widely adopted PEFT technique, Low-Rank Adaptation (LoRA), freezes the pre-trained model weights and introduces two low-rank matrices whose ranks are significantly smaller than the dimensions of the original weight matrices. This enables efficient fine-tuning by adjusting only a small number of parameters. Despite its efficiency, LoRA approximates weight updates using low-rank decomposition, which struggles to capture complex, non-linear components and efficient optimization trajectories. As a result, LoRA-based methods often exhibit a significant performance gap compared to full fine-tuning. Closing this gap requires higher ranks, which increases the number of parameters. To address these limitations, we propose a nonlinear parameter-efficient adaptation method (NEAT). NEAT introduces a lightweight neural network that takes pre-trained weights as input and learns a nonlinear transformation to approximate cumulative weight updates. These updates can be interpreted as functions of the corresponding pre-trained weights. The nonlinear approximation directly models the cumulative updates, effectively capturing complex and non-linear structures in the weight updates. Our theoretical analysis demonstrates taht NEAT can be more efficient than LoRA while having equal or greater expressivity. Extensive evaluations across four benchmarks and over twenty datasets demonstrate that NEAT significantly outperforms baselines in both vision and text tasks.
Authors: Liang Chen, Sinan Tan, Zefan Cai, Weichu Xie, Haozhe Zhao, Yichi Zhang, Junyang Lin, Jinze Bai, Tianyu Liu, Baobao Chang
Abstract: This work tackles the information loss bottleneck of vector-quantization (VQ) autoregressive image generation by introducing a novel model architecture called the 2-Dimensional Autoregression (DnD) Transformer. The DnD-Transformer predicts more codes for an image by introducing a new autoregression direction, \textit{model depth}, along with the sequence length direction. Compared to traditional 1D autoregression and previous work utilizing similar 2D image decomposition such as RQ-Transformer, the DnD-Transformer is an end-to-end model that can generate higher quality images with the same backbone model size and sequence length, opening a new optimization perspective for autoregressive image generation. Furthermore, our experiments reveal that the DnD-Transformer's potential extends beyond generating natural images. It can even generate images with rich text and graphical elements in a self-supervised manner, demonstrating an understanding of these combined modalities. This has not been previously demonstrated for popular vision generative models such as diffusion models, showing a spark of vision-language intelligence when trained solely on images. Code, datasets and models are open at https://github.com/chenllliang/DnD-Transformer.
Authors: Mohammadreza Pourreza, Hailong Li, Ruoxi Sun, Yeounoh Chung, Shayan Talaei, Gaurav Tarlok Kakkar, Yu Gan, Amin Saberi, Fatma Ozcan, Sercan O. Arik
Abstract: In tackling the challenges of large language model (LLM) performance for Text-to-SQL tasks, we introduce CHASE-SQL, a new framework that employs innovative strategies, using test-time compute in multi-agent modeling to improve candidate generation and selection. CHASE-SQL leverages LLMs' intrinsic knowledge to generate diverse and high-quality SQL candidates using different LLM generators with: (1) a divide-and-conquer method that decomposes complex queries into manageable sub-queries in a single LLM call; (2) chain-of-thought reasoning based on query execution plans, reflecting the steps a database engine takes during execution; and (3) a unique instance-aware synthetic example generation technique, which offers specific few-shot demonstrations tailored to test questions.To identify the best candidate, a selection agent is employed to rank the candidates through pairwise comparisons with a fine-tuned binary-candidates selection LLM. This selection approach has been demonstrated to be more robust over alternatives. The proposed generators-selector framework not only enhances the quality and diversity of SQL queries but also outperforms previous methods. Overall, our proposed CHASE-SQL achieves the state-of-the-art execution accuracy of 73.0% and 73.01% on the test set and development set of the notable BIRD Text-to-SQL dataset benchmark, rendering CHASE-SQL the top submission of the leaderboard (at the time of paper submission).
Authors: Yanxin Shen, Pulin Kirin Zhang
Abstract: Financial sentiment analysis (FSA) is crucial for evaluating market sentiment and making well-informed financial decisions. The advent of large language models (LLMs) such as BERT and its financial variant, FinBERT, has notably enhanced sentiment analysis capabilities. This paper investigates the application of LLMs and FinBERT for FSA, comparing their performance on news articles, financial reports and company announcements. The study emphasizes the advantages of prompt engineering with zero-shot and few-shot strategy to improve sentiment classification accuracy. Experimental results indicate that GPT-4o, with few-shot examples of financial texts, can be as competent as a well fine-tuned FinBERT in this specialized field.
Authors: Bolun (Namir), Xia, Mohammed J. Zaki, Aparna Gupta
Abstract: The advent of large language models (LLMs) has initiated much research into their various financial applications. However, in applying LLMs on long documents, semantic relations are not explicitly incorporated, and a full or arbitrarily sparse attention operation is employed. In recent years, progress has been made in Abstract Meaning Representation (AMR), which is a graph-based representation of text to preserve its semantic relations. Since AMR can represent semantic relationships at a deeper level, it can be beneficially utilized by graph neural networks (GNNs) for constructing effective document-level graph representations built upon LLM embeddings to predict target metrics in the financial domain. We propose FLAG: Financial Long document classification via AMR-based GNN, an AMR graph based framework to generate document-level embeddings for long financial document classification. We construct document-level graphs from sentence-level AMR graphs, endow them with specialized LLM word embeddings in the financial domain, apply a deep learning mechanism that utilizes a GNN, and examine the efficacy of our AMR-based approach in predicting labeled target data from long financial documents. Extensive experiments are conducted on a dataset of quarterly earnings calls transcripts of companies in various sectors of the economy, as well as on a corpus of more recent earnings calls of companies in the S&P 1500 Composite Index. We find that our AMR-based approach outperforms fine-tuning LLMs directly on text in predicting stock price movement trends at different time horizons in both datasets. Our work also outperforms previous work utilizing document graphs and GNNs for text classification.
Authors: Yuan Zhou, Peng Zhang, Mengya Song, Alice Zheng, Yiwen Lu, Zhiheng Liu, Yong Chen, Zhaohan Xi
Abstract: Large language models (LLMs) have demonstrated remarkable progress in healthcare. However, a significant gap remains regarding LLMs' professionalism in domain-specific clinical practices, limiting their application in real-world diagnostics. In this work, we introduce ZODIAC, an LLM-powered framework with cardiologist-level professionalism designed to engage LLMs in cardiological diagnostics. ZODIAC assists cardiologists by extracting clinically relevant characteristics from patient data, detecting significant arrhythmias, and generating preliminary reports for the review and refinement by cardiologists. To achieve cardiologist-level professionalism, ZODIAC is built on a multi-agent collaboration framework, enabling the processing of patient data across multiple modalities. Each LLM agent is fine-tuned using real-world patient data adjudicated by cardiologists, reinforcing the model's professionalism. ZODIAC undergoes rigorous clinical validation with independent cardiologists, evaluated across eight metrics that measure clinical effectiveness and address security concerns. Results show that ZODIAC outperforms industry-leading models, including OpenAI's GPT-4o, Meta's Llama-3.1-405B, and Google's Gemini-pro, as well as medical-specialist LLMs like Microsoft's BioGPT. ZODIAC demonstrates the transformative potential of specialized LLMs in healthcare by delivering domain-specific solutions that meet the stringent demands of medical practice. Notably, ZODIAC has been successfully integrated into electrocardiography (ECG) devices, exemplifying the growing trend of embedding LLMs into Software-as-Medical-Device (SaMD).
Authors: Mahshid Dehghani, Amirahmad Shafiee, Ali Shafiei, Neda Fallah, Farahmand Alizadeh, Mohammad Mehdi Gholinejad, Hamid Behroozi, Jafar Habibi, Ehsaneddin Asgari
Abstract: Existing 3D facial emotion modeling have been constrained by limited emotion classes and insufficient datasets. This paper introduces "Emo3D", an extensive "Text-Image-Expression dataset" spanning a wide spectrum of human emotions, each paired with images and 3D blendshapes. Leveraging Large Language Models (LLMs), we generate a diverse array of textual descriptions, facilitating the capture of a broad spectrum of emotional expressions. Using this unique dataset, we conduct a comprehensive evaluation of language-based models' fine-tuning and vision-language models like Contranstive Language Image Pretraining (CLIP) for 3D facial expression synthesis. We also introduce a new evaluation metric for this task to more directly measure the conveyed emotion. Our new evaluation metric, Emo3D, demonstrates its superiority over Mean Squared Error (MSE) metrics in assessing visual-text alignment and semantic richness in 3D facial expressions associated with human emotions. "Emo3D" has great applications in animation design, virtual reality, and emotional human-computer interaction.
Authors: Sreyan Ghosh, Sonal Kumar, Zhifeng Kong, Rafael Valle, Bryan Catanzaro, Dinesh Manocha
Abstract: We present Synthio, a novel approach for augmenting small-scale audio classification datasets with synthetic data. Our goal is to improve audio classification accuracy with limited labeled data. Traditional data augmentation techniques, which apply artificial transformations (e.g., adding random noise or masking segments), struggle to create data that captures the true diversity present in real-world audios. To address this shortcoming, we propose to augment the dataset with synthetic audio generated from text-to-audio (T2A) diffusion models. However, synthesizing effective augmentations is challenging because not only should the generated data be acoustically consistent with the underlying small-scale dataset, but they should also have sufficient compositional diversity. To overcome the first challenge, we align the generations of the T2A model with the small-scale dataset using preference optimization. This ensures that the acoustic characteristics of the generated data remain consistent with the small-scale dataset. To address the second challenge, we propose a novel caption generation technique that leverages the reasoning capabilities of Large Language Models to (1) generate diverse and meaningful audio captions and (2) iteratively refine their quality. The generated captions are then used to prompt the aligned T2A model. We extensively evaluate Synthio on ten datasets and four simulated limited-data settings. Results indicate our method consistently outperforms all baselines by 0.1%-39% using a T2A model trained only on weakly-captioned AudioSet.
Authors: Zefang Liu, Yinzhu Quan
Abstract: Temporal point processes (TPPs) are widely used to model the timing and occurrence of events in domains such as social networks, transportation systems, and e-commerce. In this paper, we introduce TPP-LLM, a novel framework that integrates large language models (LLMs) with TPPs to capture both the semantic and temporal aspects of event sequences. Unlike traditional methods that rely on categorical event type representations, TPP-LLM directly utilizes the textual descriptions of event types, enabling the model to capture rich semantic information embedded in the text. While LLMs excel at understanding event semantics, they are less adept at capturing temporal patterns. To address this, TPP-LLM incorporates temporal embeddings and employs parameter-efficient fine-tuning (PEFT) methods to effectively learn temporal dynamics without extensive retraining. This approach improves both predictive accuracy and computational efficiency. Experimental results across diverse real-world datasets demonstrate that TPP-LLM outperforms state-of-the-art baselines in sequence modeling and event prediction, highlighting the benefits of combining LLMs with TPPs.
Authors: Christopher Ackerman, Nina Panickssery
Abstract: It has been reported that LLMs can recognize their own writing. As this has potential implications for AI safety, yet is relatively understudied, we investigate the phenomenon, seeking to establish whether it robustly occurs at the behavioral level, how the observed behavior is achieved, and whether it can be controlled. First, we find that the Llama3-8b-Instruct chat model - but not the base Llama3-8b model - can reliably distinguish its own outputs from those of humans, and present evidence that the chat model is likely using its experience with its own outputs, acquired during post-training, to succeed at the writing recognition task. Second, we identify a vector in the residual stream of the model that is differentially activated when the model makes a correct self-written-text recognition judgment, show that the vector activates in response to information relevant to self-authorship, present evidence that the vector is related to the concept of "self" in the model, and demonstrate that the vector is causally related to the model's ability to perceive and assert self-authorship. Finally, we show that the vector can be used to control both the model's behavior and its perception, steering the model to claim or disclaim authorship by applying the vector to the model's output as it generates it, and steering the model to believe or disbelieve it wrote arbitrary texts by applying the vector to them as the model reads them.
Authors: Sara Ghazanfari, Alexandre Araujo, Prashanth Krishnamurthy, Siddharth Garg, Farshad Khorrami
Abstract: Multi-modal Large Language Models (MLLMs) have recently exhibited impressive general-purpose capabilities by leveraging vision foundation models to encode the core concepts of images into representations. These are then combined with instructions and processed by the language model to generate high-quality responses. Despite significant progress in enhancing the language component, challenges persist in optimally fusing visual encodings within the language model for task-specific adaptability. Recent research has focused on improving this fusion through modality adaptation modules but at the cost of significantly increased model complexity and training data needs. In this paper, we propose EMMA (Efficient Multi-Modal Adaptation), a lightweight cross-modality module designed to efficiently fuse visual and textual encodings, generating instruction-aware visual representations for the language model. Our key contributions include: (1) an efficient early fusion mechanism that integrates vision and language representations with minimal added parameters (less than 0.2% increase in model size), (2) an in-depth interpretability analysis that sheds light on the internal mechanisms of the proposed method; (3) comprehensive experiments that demonstrate notable improvements on both specialized and general benchmarks for MLLMs. Empirical results show that EMMA boosts performance across multiple tasks by up to 9.3% while significantly improving robustness against hallucinations. Our code is available at https://github.com/SaraGhazanfari/EMMA
Authors: Dara Bahri, John Wieting, Dana Alon, Donald Metzler
Abstract: Watermarking has recently emerged as an effective strategy for detecting the outputs of large language models (LLMs). Most existing schemes require \emph{white-box} access to the model's next-token probability distribution, which is typically not accessible to downstream users of an LLM API. In this work, we propose a principled watermarking scheme that requires only the ability to sample sequences from the LLM (i.e. \emph{black-box} access), boasts a \emph{distortion-free} property, and can be chained or nested using multiple secret keys. We provide performance guarantees, demonstrate how it can be leveraged when white-box access is available, and show when it can outperform existing white-box schemes via comprehensive experiments.
Authors: Amogh Mannekote, Adam Davies, Jina Kang, Kristy Elizabeth Boyer
Abstract: Simulating learner actions helps stress-test open-ended interactive learning environments and prototype new adaptations before deployment. While recent studies show the promise of using large language models (LLMs) for simulating human behavior, such approaches have not gone beyond rudimentary proof-of-concept stages due to key limitations. First, LLMs are highly sensitive to minor prompt variations, raising doubts about their ability to generalize to new scenarios without extensive prompt engineering. Moreover, apparently successful outcomes can often be unreliable, either because domain experts unintentionally guide LLMs to produce expected results, leading to self-fulfilling prophecies; or because the LLM has encountered highly similar scenarios in its training data, meaning that models may not be simulating behavior so much as regurgitating memorized content. To address these challenges, we propose Hyp-Mix, a simulation authoring framework that allows experts to develop and evaluate simulations by combining testable hypotheses about learner behavior. Testing this framework in a physics learning environment, we found that GPT-4 Turbo maintains calibrated behavior even as the underlying learner model changes, providing the first evidence that LLMs can be used to simulate realistic behaviors in open-ended interactive learning environments, a necessary prerequisite for useful LLM behavioral simulation.
Authors: Manh Pham, Aaqib Saeed, Dong Ma
Abstract: Accurate interpretation of Electrocardiogram (ECG) signals is pivotal for diagnosing cardiovascular diseases. Integrating ECG signals with their accompanying textual reports holds immense potential to enhance clinical diagnostics through the combination of physiological data and qualitative insights. However, this integration faces significant challenges due to inherent modality disparities and the scarcity of labeled data for robust cross-modal learning. To address these obstacles, we propose C-MELT, a novel framework that pre-trains ECG and text data using a contrastive masked auto-encoder architecture. C-MELT uniquely combines the strengths of generative with enhanced discriminative capabilities to achieve robust cross-modal representations. This is accomplished through masked modality modeling, specialized loss functions, and an improved negative sampling strategy tailored for cross-modal alignment. Extensive experiments on five public datasets across diverse downstream tasks demonstrate that C-MELT significantly outperforms existing methods, achieving 15% and 2% increases in linear probing and zero-shot performance over state-of-the-art models, respectively. These results highlight the effectiveness of C-MELT, underscoring its potential to advance automated clinical diagnostics through multi-modal representations.
Authors: Wanpeng Zhang, Zilong Xie, Yicheng Feng, Yijiang Li, Xingrun Xing, Sipeng Zheng, Zongqing Lu
Abstract: Multimodal Large Language Models have made significant strides in integrating visual and textual information, yet they often struggle with effectively aligning these modalities. We introduce a novel image tokenizer that bridges this gap by applying the principle of Byte-Pair Encoding (BPE) to visual data. Unlike conventional approaches that rely on separate visual encoders, our method directly incorporates structural prior information into image tokens, mirroring the successful tokenization strategies used in text-only Large Language Models. This innovative approach enables Transformer models to more effectively learn and reason across modalities. Through theoretical analysis and extensive experiments, we demonstrate that our BPE Image Tokenizer significantly enhances MLLMs' multimodal understanding capabilities, even with limited training data. Our method not only improves performance across various benchmarks but also shows promising scalability, potentially paving the way for more efficient and capable multimodal foundation models.
Authors: Tonghe Zhuang, Zhicheng Lin
Abstract: Computer programming (coding) is indispensable for researchers across disciplines, yet it remains challenging to learn and time-consuming to carry out. Generative AI, particularly large language models (LLMs), has the potential to transform coding into intuitive conversations, but best practices and effective workflows are only emerging. We dissect AI-based coding through three key lenses: the nature and role of LLMs in coding (why), six types of coding assistance they provide (what), and a five-step workflow in action with practical implementation strategies (how). Additionally, we address the limitations and future outlook of AI in coding. By offering actionable insights, this framework helps to guide researchers in effectively leveraging AI to enhance coding practices and education, accelerating scientific progress.
Authors: Mansi Sakarvadia, Aswathy Ajith, Arham Khan, Nathaniel Hudson, Caleb Geniesse, Kyle Chard, Yaoqing Yang, Ian Foster, Michael W. Mahoney
Abstract: Language models (LMs) can "memorize" information, i.e., encode training data in their weights in such a way that inference-time queries can lead to verbatim regurgitation of that data. This ability to extract training data can be problematic, for example, when data are private or sensitive. In this work, we investigate methods to mitigate memorization: three regularizer-based, three finetuning-based, and eleven machine unlearning-based methods, with five of the latter being new methods that we introduce. We also introduce TinyMem, a suite of small, computationally-efficient LMs for the rapid development and evaluation of memorization-mitigation methods. We demonstrate that the mitigation methods that we develop using TinyMem can successfully be applied to production-grade LMs, and we determine via experiment that: regularizer-based mitigation methods are slow and ineffective at curbing memorization; fine-tuning-based methods are effective at curbing memorization, but overly expensive, especially for retaining higher accuracies; and unlearning-based methods are faster and more effective, allowing for the precise localization and removal of memorized information from LM weights prior to inference. We show, in particular, that our proposed unlearning method BalancedSubnet outperforms other mitigation methods at removing memorized information while preserving performance on target tasks.
Authors: Yucheng Chu, Hang Li, Kaiqi Yang, Harry Shomer, Hui Liu, Yasemin Copur-Gencturk, Jiliang Tang
Abstract: Open-ended short-answer questions (SAGs) have been widely recognized as a powerful tool for providing deeper insights into learners' responses in the context of learning analytics (LA). However, SAGs often present challenges in practice due to the high grading workload and concerns about inconsistent assessments. With recent advancements in natural language processing (NLP), automatic short-answer grading (ASAG) offers a promising solution to these challenges. Despite this, current ASAG algorithms are often limited in generalizability and tend to be tailored to specific questions. In this paper, we propose a unified multi-agent ASAG framework, GradeOpt, which leverages large language models (LLMs) as graders for SAGs. More importantly, GradeOpt incorporates two additional LLM-based agents - the reflector and the refiner - into the multi-agent system. This enables GradeOpt to automatically optimize the original grading guidelines by performing self-reflection on its errors. Through experiments on a challenging ASAG task, namely the grading of pedagogical content knowledge (PCK) and content knowledge (CK) questions, GradeOpt demonstrates superior performance in grading accuracy and behavior alignment with human graders compared to representative baselines. Finally, comprehensive ablation studies confirm the effectiveness of the individual components designed in GradeOpt.
Authors: Hongkang Li, Meng Wang, Songtao Lu, Xiaodong Cui, Pin-Yu Chen
Abstract: Chain-of-Thought (CoT) is an efficient prompting method that enables the reasoning ability of large language models by augmenting the query using multiple examples with multiple intermediate steps. Despite the empirical success, the theoretical understanding of how to train a Transformer to achieve the CoT ability remains less explored. This is primarily due to the technical challenges involved in analyzing the nonconvex optimization on nonlinear attention models. To the best of our knowledge, this work provides the first theoretical study of training Transformers with nonlinear attention to obtain the CoT generalization capability so that the resulting model can inference on unseen tasks when the input is augmented by examples of the new task. We first quantify the required training samples and iterations to train a Transformer model towards CoT ability. We then prove the success of its CoT generalization on unseen tasks with distribution-shifted testing data. Moreover, we theoretically characterize the conditions for an accurate reasoning output by CoT even when the provided reasoning examples contain noises and are not always accurate. In contrast, in-context learning (ICL), which can be viewed as one-step CoT without intermediate steps, may fail to provide an accurate output when CoT does. These theoretical findings are justified through experiments.
Authors: Adrian Chan, Anupam Mijar, Mehreen Saeed, Chau-Wai Wong, Akram Khater
Abstract: Arabic handwritten text recognition (HTR) is challenging, especially for historical texts, due to diverse writing styles and the intrinsic features of Arabic script. Additionally, Arabic handwriting datasets are smaller compared to English ones, making it difficult to train generalizable Arabic HTR models. To address these challenges, we propose HATFormer, a transformer-based encoder-decoder architecture that builds on a state-of-the-art English HTR model. By leveraging the transformer's attention mechanism, HATFormer captures spatial contextual information to address the intrinsic challenges of Arabic script through differentiating cursive characters, decomposing visual representations, and identifying diacritics. Our customization to historical handwritten Arabic includes an image processor for effective ViT information preprocessing, a text tokenizer for compact Arabic text representation, and a training pipeline that accounts for a limited amount of historic Arabic handwriting data. HATFormer achieves a character error rate (CER) of 8.6% on the largest public historical handwritten Arabic dataset, with a 51% improvement over the best baseline in the literature. HATFormer also attains a comparable CER of 4.2% on the largest private non-historical dataset. Our work demonstrates the feasibility of adapting an English HTR method to a low-resource language with complex, language-specific challenges, contributing to advancements in document digitization, information retrieval, and cultural preservation.
Authors: Weixi Tong, Tianyi Zhang
Abstract: Large Language Models (LLMs) have shown promising performance in code generation. However, how to reliably evaluate code generated by LLMs remains an unresolved problem. This paper presents CodeJudge, a code evaluation framework that leverages LLMs to evaluate the semantic correctness of generated code without the need for test cases. We investigate different ways to guide the LLM in performing "slow thinking" to arrive at an in-depth and reliable evaluation. We experimented with four LLMs as evaluators on four code generation datasets and five programming languages. The results show that CodeJudge significantly outperformed existing methods in most settings. Furthermore, compared with a SOTA GPT-3.5-based code evaluation method, CodeJudge achieved better results even when using a much smaller model, Llama-3-8B-Instruct. Our code and datasets are available on GitHub https://github.com/VichyTong/CodeJudge.
Authors: Yifan Zhang, Ge Zhang, Yue Wu, Kangping Xu, Quanquan Gu
Abstract: Modeling human preferences is crucial for aligning foundation models with human values. Traditional reward modeling methods, such as the Bradley-Terry (BT) reward model, fall short in expressiveness, particularly in addressing intransitive preferences. Although supervised pair preference models (PairPM) can express general preferences, their implementation is highly ad-hoc and cannot guarantee a consistent preference probability of compared pairs. Additionally, they impose high computational costs due to their quadratic query complexity when comparing multiple responses. In this paper, we introduce preference representation learning, an approach that embeds responses into a latent space to capture intricate preference structures efficiently, achieving linear query complexity. Additionally, we propose preference score-based General Preference Optimization (GPO), which generalizes reward-based reinforcement learning from human feedback. Experimental results show that our General Preference representation model (GPM) outperforms the BT reward model on the RewardBench benchmark with a margin of up to 5.6% and effectively models cyclic preferences where any BT reward model behaves like a random guess. Furthermore, evaluations on downstream tasks such as AlpacaEval2.0 and MT-Bench, following the language model post-training with GPO and our general preference model, reveal substantial performance improvements with margins up to 9.3%. These findings indicate that our method may enhance the alignment of foundation models with nuanced human values. The code is available at https://github.com/general-preference/general-preference-model.
URLs: https://github.com/general-preference/general-preference-model.
Authors: Huimu Yu, Xing Wu, Weidong Yin, Debing Zhang, Songlin Hu
Abstract: Large language models (LLMs) have made significant progress in natural language understanding and generation, driven by scalable pretraining and advanced finetuning. However, enhancing reasoning abilities in LLMs, particularly via reinforcement learning from human feedback (RLHF), remains challenging due to the scarcity of high-quality preference data, which is labor-intensive to annotate and crucial for reward model (RM) finetuning. To alleviate this issue, we introduce CodePMP, a scalable preference model pretraining (PMP) pipeline that utilizes a large corpus of synthesized code-preference pairs from publicly available high-quality source code. CodePMP improves RM finetuning efficiency by pretraining preference models on large-scale synthesized code-preference pairs. We evaluate CodePMP on mathematical reasoning tasks (GSM8K, MATH) and logical reasoning tasks (ReClor, LogiQA2.0), consistently showing significant improvements in reasoning performance of LLMs and highlighting the importance of scalable preference model pretraining for efficient reward modeling.
Authors: Haopeng Geng, Daisuke Saito, Nobuaki Minematsu
Abstract: Utterances by L2 speakers can be unintelligible due to mispronunciation and improper prosody. In computer-aided language learning systems, textual feedback is often provided using a speech recognition engine. However, an ideal form of feedback for L2 speakers should be so fine-grained that it enables them to detect and diagnose unintelligible parts of L2 speakers' utterances. Inspired by language teachers who correct students' pronunciation through a voice-to-voice process, this pilot study utilizes a unique semi-parallel dataset composed of non-native speakers' (L2) reading aloud, shadowing of native speakers (L1) and their script-shadowing utterances. We explore the technical possibility of replicating the process of an L1 speaker's shadowing L2 speech using Voice Conversion techniques, to create a virtual shadower system. Experimental results demonstrate the feasibility of the VC system in simulating L1's shadowing behavior. The output of the virtual shadower system shows a reasonable similarity to the real L1 shadowing utterances in both linguistic and acoustic aspects.
Authors: Tianchi Xie, Jiangning Zhu, Guozu Ma, Minzhi Lin, Wei Chen, Weikai Yang, Shixia Liu
Abstract: Sample selection improves the efficiency and effectiveness of machine learning models by providing informative and representative samples. Typically, samples can be modeled as a sample graph, where nodes are samples and edges represent their similarities. Most existing methods are based on local information, such as the training difficulty of samples, thereby overlooking global information, such as connectivity patterns. This oversight can result in suboptimal selection because global information is crucial for ensuring that the selected samples well represent the structural properties of the graph. To address this issue, we employ structural entropy to quantify global information and losslessly decompose it from the whole graph to individual nodes using the Shapley value. Based on the decomposition, we present $\textbf{S}$tructural-$\textbf{E}$ntropy-based sample $\textbf{S}$election ($\textbf{SES}$), a method that integrates both global and local information to select informative and representative samples. SES begins by constructing a $k$NN-graph among samples based on their similarities. It then measures sample importance by combining structural entropy (global metric) with training difficulty (local metric). Finally, SES applies importance-biased blue noise sampling to select a set of diverse and representative samples. Comprehensive experiments on three learning scenarios -- supervised learning, active learning, and continual learning -- clearly demonstrate the effectiveness of our method.
Authors: James Vo
Abstract: Second-order optimization methods offer notable advantages in training deep neural networks by utilizing curvature information to achieve faster convergence. However, traditional second-order techniques are computationally prohibitive, primarily due to the large matrix inversions and high memory demands they require. While adaptive trust-region methods have been developed to mitigate these issues, their performance is often hindered by conservative estimates of key parameters, such as the Lipschitz constant of the Hessian, resulting in suboptimal outcomes. In this paper, we introduce SecondOrderAdaptiveAdam (SOAA), a novel optimization algorithm designed to overcome these limitations. SOAA approximates the Fisher information matrix using a diagonal representation, reducing computational complexity from \(O(n^{2})\) to \(O(n)\), thereby making it suitable for large-scale deep learning models, including large language models (LLMs). Additionally, the algorithm integrates an adaptive trust-region mechanism that dynamically adjusts the trust region size based on observed loss reduction, ensuring both robust convergence and computational efficiency. We empirically demonstrate that SOAA achieves faster and more stable convergence compared to first-order optimizers, such as Adam, under similar computational constraints. However, the diagonal approximation of the Fisher information matrix may be less effective in capturing higher-order interactions between gradients, suggesting potential areas for further refinement and future research.
Authors: Guobin Shen, Dongcheng Zhao, Yiting Dong, Xiang He, Yi Zeng
Abstract: As large language models (LLMs) become integral to various applications, ensuring both their safety and utility is paramount. Jailbreak attacks, which manipulate LLMs into generating harmful content, pose significant challenges to this balance. Existing defenses, such as prompt engineering and safety fine-tuning, often introduce computational overhead, increase inference latency, and lack runtime flexibility. Moreover, overly restrictive safety measures can degrade model utility by causing refusals of benign queries. In this paper, we introduce Jailbreak Antidote, a method that enables real-time adjustment of LLM safety preferences by manipulating a sparse subset of the model's internal states during inference. By shifting the model's hidden representations along a safety direction with varying strengths, we achieve flexible control over the safety-utility balance without additional token overhead or inference delays. Our analysis reveals that safety-related information in LLMs is sparsely distributed; adjusting approximately 5% of the internal state is as effective as modifying the entire state. Extensive experiments on nine LLMs (ranging from 2 billion to 72 billion parameters), evaluated against ten jailbreak attack methods and compared with six defense strategies, validate the effectiveness and efficiency of our approach. By directly manipulating internal states during reasoning, Jailbreak Antidote offers a lightweight, scalable solution that enhances LLM safety while preserving utility, opening new possibilities for real-time safety mechanisms in widely-deployed AI systems.
Authors: Guodong Du, Junlin Lee, Jing Li, Runhua Jiang, Yifei Guo, Shuyang Yu, Hanting Liu, Sim Kuan Goh, Ho-Kin Tang, Daojing He, Min Zhang
Abstract: While fine-tuning pretrained models has become common practice, these models often underperform outside their specific domains. Recently developed model merging techniques enable the direct integration of multiple models, each fine-tuned for distinct tasks, into a single model. This strategy promotes multitasking capabilities without requiring retraining on the original datasets. However, existing methods fall short in addressing potential conflicts and complex correlations between tasks, especially in parameter-level adjustments, posing a challenge in effectively balancing parameter competition across various tasks. This paper introduces an innovative technique named PCB-Merging (Parameter Competition Balancing), a lightweight and training-free technique that adjusts the coefficients of each parameter for effective model merging. PCB-Merging employs intra-balancing to gauge parameter significance within individual tasks and inter-balancing to assess parameter similarities across different tasks. Parameters with low importance scores are dropped, and the remaining ones are rescaled to form the final merged model. We assessed our approach in diverse merging scenarios, including cross-task, cross-domain, and cross-training configurations, as well as out-of-domain generalization. The experimental results reveal that our approach achieves substantial performance enhancements across multiple modalities, domains, model sizes, number of tasks, fine-tuning forms, and large language models, outperforming existing model merging methods. The code is publicly available at: \url{https://github.com/duguodong7/pcb-merging}.
Authors: Ma{\l}gorzata {\L}azuka, Andreea Anghel, Thomas Parnell
Abstract: As Large Language Models (LLMs) are rapidly growing in popularity, LLM inference services must be able to serve requests from thousands of users while satisfying performance requirements. The performance of an LLM inference service is largely determined by the hardware onto which it is deployed, but understanding of which hardware will deliver on performance requirements remains challenging. In this work we present LLM-Pilot - a first-of-its-kind system for characterizing and predicting performance of LLM inference services. LLM-Pilot performs benchmarking of LLM inference services, under a realistic workload, across a variety of GPUs, and optimizes the service configuration for each considered GPU to maximize performance. Finally, using this characterization data, LLM-Pilot learns a predictive model, which can be used to recommend the most cost-effective hardware for a previously unseen LLM. Compared to existing methods, LLM-Pilot can deliver on performance requirements 33% more frequently, whilst reducing costs by 60% on average.
Authors: Tuo An, Yunjiao Zhou, Han Zou, Jianfei Yang
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities across textual and visual domains but often generate outputs that violate physical laws, revealing a gap in their understanding of the physical world. Inspired by human cognition, where perception is fundamental to reasoning, we explore augmenting LLMs with enhanced perception abilities using Internet of Things (IoT) sensor data and pertinent knowledge for IoT task reasoning in the physical world. In this work, we systematically study LLMs capability to address real-world IoT tasks by augmenting their perception and knowledge base, and then propose a unified framework, IoT-LLM, to enhance such capability. In IoT-LLM, we customize three steps for LLMs: preprocessing IoT data into formats amenable to LLMs, activating their commonsense knowledge through chain-of-thought prompting and specialized role definitions, and expanding their understanding via IoT-oriented retrieval-augmented generation based on in-context learning. To evaluate the performance, We design a new benchmark with five real-world IoT tasks with different data types and reasoning difficulties and provide the benchmarking results on six open-source and close-source LLMs. Experimental results demonstrate the limitations of existing LLMs with naive textual inputs that cannot perform these tasks effectively. We show that IoT-LLM significantly enhances the performance of IoT tasks reasoning of LLM, such as GPT-4, achieving an average improvement of 65% across various tasks against previous methods. The results also showcase LLMs ability to comprehend IoT data and the physical law behind data by providing a reasoning process. Limitations of our work are claimed to inspire future research in this new era.
Authors: Gurucharan Marthi Krishna Kumar, Aman Chadha, Janine Mendola, Amir Shmuel
Abstract: Large Language Models (LLMs), known for their versatility in textual data, are increasingly being explored for their potential to enhance medical image segmentation, a crucial task for accurate diagnostic imaging. This study explores enhancing Vision Transformers (ViTs) for medical image segmentation by integrating pre-trained LLM transformer blocks. Our approach, which incorporates a frozen LLM transformer block into the encoder of a ViT-based model, leads to substantial improvements in segmentation performance across various medical imaging modalities. We propose a Hybrid Attention Mechanism that combines global and local feature learning with a Multi-Scale Fusion Block for aggregating features across different scales. The enhanced model shows significant performance gains, including an average Dice score increase from 0.74 to 0.79 and improvements in accuracy, precision, and the Jaccard Index. These results demonstrate the effectiveness of LLM-based transformers in refining medical image segmentation, highlighting their potential to significantly boost model accuracy and robustness. The source code and our implementation are available at: https://bit.ly/3zf2CVs
URLs: https://bit.ly/3zf2CVs
Authors: Xuchen Li, Shiyu Hu, Xiaokun Feng, Dailing Zhang, Meiqi Wu, Jing Zhang, Kaiqi Huang
Abstract: Visual language tracking (VLT) has emerged as a cutting-edge research area, harnessing linguistic data to enhance algorithms with multi-modal inputs and broadening the scope of traditional single object tracking (SOT) to encompass video understanding applications. Despite this, most VLT benchmarks still depend on succinct, human-annotated text descriptions for each video. These descriptions often fall short in capturing the nuances of video content dynamics and lack stylistic variety in language, constrained by their uniform level of detail and a fixed annotation frequency. As a result, algorithms tend to default to a "memorize the answer" strategy, diverging from the core objective of achieving a deeper understanding of video content. Fortunately, the emergence of large language models (LLMs) has enabled the generation of diverse text. This work utilizes LLMs to generate varied semantic annotations (in terms of text lengths and granularities) for representative SOT benchmarks, thereby establishing a novel multi-modal benchmark. Specifically, we (1) propose a new visual language tracking benchmark with diverse texts, named DTVLT, based on five prominent VLT and SOT benchmarks, including three sub-tasks: short-term tracking, long-term tracking, and global instance tracking. (2) We offer four granularity texts in our benchmark, considering the extent and density of semantic information. We expect this multi-granular generation strategy to foster a favorable environment for VLT and video understanding research. (3) We conduct comprehensive experimental analyses on DTVLT, evaluating the impact of diverse text on tracking performance and hope the identified performance bottlenecks of existing algorithms can support further research in VLT and video understanding. The proposed benchmark, experimental results and toolkit will be released gradually on http://videocube.aitestunion.com/.
Authors: Simin Fan, David Grangier, Pierre Ablin
Abstract: The composition of training data mixtures is critical for effectively training large language models (LLMs), as it directly impacts their performance on downstream tasks. Our goal is to identify an optimal data mixture to specialize an LLM for a specific task with access to only a few examples. Traditional approaches to this problem include ad-hoc reweighting methods, importance sampling, and gradient alignment techniques. This paper focuses on gradient alignment and introduces Dynamic Gradient Alignment (DGA), a scalable online gradient alignment algorithm. DGA dynamically estimates the pre-training data mixture on which the models' gradients align as well as possible with those of the model on the specific task. DGA is the first gradient alignment approach that incurs minimal overhead compared to standard pre-training and outputs a competitive model, eliminating the need for retraining the model. Experimentally, we demonstrate significant improvements over importance sampling in two key scenarios: (i) when the pre-training set is small and importance sampling overfits due to limited data; and (ii) when there is insufficient specialized data, trapping importance sampling on narrow pockets of data. Our findings underscore the effectiveness of gradient alignment methods in optimizing training data mixtures, particularly in data-constrained environments, and offer a practical solution for enhancing LLM performance on specific tasks with limited data availability.
Authors: Weikang Yuan, Junjie Cao, Zhuoren Jiang, Yangyang Kang, Jun Lin, Kaisong Song, tianqianjin lin, Pengwei Yan, Changlong Sun, Xiaozhong Liu
Abstract: Large Language Models (LLMs) could struggle to fully understand legal theories and perform complex legal reasoning tasks. In this study, we introduce a challenging task (confusing charge prediction) to better evaluate LLMs' understanding of legal theories and reasoning capabilities. We also propose a novel framework: Multi-Agent framework for improving complex Legal Reasoning capability (MALR). MALR employs non-parametric learning, encouraging LLMs to automatically decompose complex legal tasks and mimic human learning process to extract insights from legal rules, helping LLMs better understand legal theories and enhance their legal reasoning abilities. Extensive experiments on multiple real-world datasets demonstrate that the proposed framework effectively addresses complex reasoning issues in practical scenarios, paving the way for more reliable applications in the legal domain.
Authors: Zixiang Wang, Yinghao Zhu, Huiya Zhao, Xiaochen Zheng, Tianlong Wang, Wen Tang, Yasha Wang, Chengwei Pan, Ewen M. Harrison, Junyi Gao, Liantao Ma
Abstract: We introduce ColaCare, a framework that enhances Electronic Health Record (EHR) modeling through multi-agent collaboration driven by Large Language Models (LLMs). Our approach seamlessly integrates domain-specific expert models with LLMs to bridge the gap between structured EHR data and text-based reasoning. Inspired by clinical consultations, ColaCare employs two types of agents: DoctorAgent and MetaAgent, which collaboratively analyze patient data. Expert models process and generate predictions from numerical EHR data, while LLM agents produce reasoning references and decision-making reports within the collaborative consultation framework. We additionally incorporate the Merck Manual of Diagnosis and Therapy (MSD) medical guideline within a retrieval-augmented generation (RAG) module for authoritative evidence support. Extensive experiments conducted on four distinct EHR datasets demonstrate ColaCare's superior performance in mortality prediction tasks, underscoring its potential to revolutionize clinical decision support systems and advance personalized precision medicine. The code, complete prompt templates, more case studies, etc. are publicly available at the anonymous link: https://colacare.netlify.app.
Authors: Olga Yakovenko, Ivan Bondarenko
Abstract: For many Automatic Speech Recognition (ASR) tasks audio features as spectrograms show better results than Mel-frequency Cepstral Coefficients (MFCC), but in practice they are hard to use due to a complex dimensionality of a feature space. The following paper presents an alternative approach towards generating compressed spectrogram representation, based on Convolutional Variational Autoencoders (VAE). A Convolutional VAE model was trained on a subsample of the LibriSpeech dataset to reconstruct short fragments of audio spectrograms (25 ms) from a 13-dimensional embedding. The trained model for a 40-dimensional (300 ms) embedding was used to generate features for corpus of spoken commands on the GoogleSpeechCommands dataset. Using the generated features an ASR system was built and compared to the model with MFCC features.
Authors: Mor Ventura, Michael Toker, Nitay Calderon, Zorik Gekhman, Yonatan Bitton, Roi Reichart
Abstract: Will a Visual Language Model (VLM)-based bot warn us about slipping if it detects a wet floor? Recent VLMs have demonstrated impressive capabilities, yet their ability to infer outcomes and causes remains underexplored. To address this, we introduce NL-Eye, a benchmark designed to assess VLMs' visual abductive reasoning skills. NL-Eye adapts the abductive Natural Language Inference (NLI) task to the visual domain, requiring models to evaluate the plausibility of hypothesis images based on a premise image and explain their decisions. NL-Eye consists of 350 carefully curated triplet examples (1,050 images) spanning diverse reasoning categories: physical, functional, logical, emotional, cultural, and social. The data curation process involved two steps - writing textual descriptions and generating images using text-to-image models, both requiring substantial human involvement to ensure high-quality and challenging scenes. Our experiments show that VLMs struggle significantly on NL-Eye, often performing at random baseline levels, while humans excel in both plausibility prediction and explanation quality. This demonstrates a deficiency in the abductive reasoning capabilities of modern VLMs. NL-Eye represents a crucial step toward developing VLMs capable of robust multimodal reasoning for real-world applications, including accident-prevention bots and generated video verification.
Authors: Song Li, Yang Tan, Song Ke, Liang Hong, Bingxin Zhou
Abstract: Immunogenicity prediction is a central topic in reverse vaccinology for finding candidate vaccines that can trigger protective immune responses. Existing approaches typically rely on highly compressed features and simple model architectures, leading to limited prediction accuracy and poor generalizability. To address these challenges, we introduce ProVaccine, a novel deep learning solution with a dual attention mechanism that integrates pre-trained latent vector representations of protein sequences and structures. We also compile the most comprehensive immunogenicity dataset to date, encompassing over 9,500 antigen sequences, structures, and immunogenicity labels from bacteria, viruses, and tumors. Extensive experiments demonstrate that ProVaccine outperforms existing methods across a wide range of evaluation metrics. Furthermore, we establish a post-hoc validation protocol to assess the practical significance of deep learning models in tackling vaccine design challenges. Our work provides an effective tool for vaccine design and sets valuable benchmarks for future research.
Authors: Yihong Dong, Ge Li, Yongding Tao, Xue Jiang, Kechi Zhang, Jia Li, Jing Su, Jun Zhang, Jingjing Xu
Abstract: Despite the remarkable success achieved by neural networks, particularly those represented by MLP and Transformer, we reveal that they exhibit potential flaws in the modeling and reasoning of periodicity, i.e., they tend to memorize the periodic data rather than genuinely understanding the underlying principles of periodicity. However, periodicity is a crucial trait in various forms of reasoning and generalization, underpinning predictability across natural and engineered systems through recurring patterns in observations. In this paper, we propose FAN, a novel network architecture based on Fourier Analysis, which empowers the ability to efficiently model and reason about periodic phenomena. By introducing Fourier Series, the periodicity is naturally integrated into the structure and computational processes of the neural network, thus achieving a more accurate expression and prediction of periodic patterns. As a promising substitute to multi-layer perceptron (MLP), FAN can seamlessly replace MLP in various models with fewer parameters and FLOPs. Through extensive experiments, we demonstrate the effectiveness of FAN in modeling and reasoning about periodic functions, and the superiority and generalizability of FAN across a range of real-world tasks, including symbolic formula representation, time series forecasting, and language modeling.
Authors: Tianyi Xiong, Xiyao Wang, Dong Guo, Qinghao Ye, Haoqi Fan, Quanquan Gu, Heng Huang, Chunyuan Li
Abstract: We introduce LLaVA-Critic, the first open-source large multimodal model (LMM) designed as a generalist evaluator to assess performance across a wide range of multimodal tasks. LLaVA-Critic is trained using a high-quality critic instruction-following dataset that incorporates diverse evaluation criteria and scenarios. Our experiments demonstrate the model's effectiveness in two key areas: (1) LMM-as-a-Judge, where LLaVA-Critic provides reliable evaluation scores, performing on par with or surpassing GPT models on multiple evaluation benchmarks; and (2) Preference Learning, where it generates reward signals for preference learning, enhancing model alignment capabilities. This work underscores the potential of open-source LMMs in self-critique and evaluation, setting the stage for future research into scalable, superhuman alignment feedback mechanisms for LMMs.
Authors: Yuanhan Zhang, Jinming Wu, Wei Li, Bo Li, Zejun Ma, Ziwei Liu, Chunyuan Li
Abstract: The development of video large multimodal models (LMMs) has been hindered by the difficulty of curating large amounts of high-quality raw data from the web. To address this, we propose an alternative approach by creating a high-quality synthetic dataset specifically for video instruction-following, namely LLaVA-Video-178K. This dataset includes key tasks such as detailed captioning, open-ended question-answering (QA), and multiple-choice QA. By training on this dataset, in combination with existing visual instruction tuning data, we introduce LLaVA-Video, a new video LMM. Our experiments demonstrate that LLaVA-Video achieves strong performance across various video benchmarks, highlighting the effectiveness of our dataset. We plan to release the dataset, its generation pipeline, and the model checkpoints.
Authors: Oussama Zekri, Ambroise Odonnat, Abdelhakim Benechehab, Linus Bleistein, Nicolas Boull\'e, Ievgen Redko
Abstract: Large language models (LLMs) have proven to be remarkably efficient, both across a wide range of natural language processing tasks and well beyond them. However, a comprehensive theoretical analysis of the origins of their impressive performance remains elusive. In this paper, we approach this challenging task by drawing an equivalence between generic autoregressive language models with vocabulary of size $T$ and context window of size $K$ and Markov chains defined on a finite state space of size $\mathcal{O}(T^K)$. We derive several surprising findings related to the existence of a stationary distribution of Markov chains that capture the inference power of LLMs, their speed of convergence to it, and the influence of the temperature on the latter. We then prove pre-training and in-context generalization bounds and show how the drawn equivalence allows us to enrich their interpretation. Finally, we illustrate our theoretical guarantees with experiments on several recent LLMs to highlight how they capture the behavior observed in practice.
Authors: Zhaowei Wang, Hongming Zhang, Tianqing Fang, Ye Tian, Yue Yang, Kaixin Ma, Xiaoman Pan, Yangqiu Song, Dong Yu
Abstract: Object navigation in unknown environments is crucial for deploying embodied agents in real-world applications. While we have witnessed huge progress due to large-scale scene datasets, faster simulators, and stronger models, previous studies mainly focus on limited scene types and target objects. In this paper, we study a new task of navigating to diverse target objects in a large number of scene types. To benchmark the problem, we present a large-scale scene dataset, DivScene, which contains 4,614 scenes across 81 different types. With the dataset, we build an end-to-end embodied agent, NatVLM, by fine-tuning a Large Vision Language Model (LVLM) through imitation learning. The LVLM is trained to take previous observations from the environment and generate the next actions. We also introduce CoT explanation traces of the action prediction for better performance when tuning LVLMs. Our extensive experiments find that we can build a performant LVLM-based agent through imitation learning on the shortest paths constructed by a BFS planner without any human supervision. Our agent achieves a success rate that surpasses GPT-4o by over 20%. Meanwhile, we carry out various analyses showing the generalization ability of our agent.
Authors: Zhibin Lan, Liqiang Niu, Fandong Meng, Wenbo Li, Jie Zhou, Jinsong Su
Abstract: Recently, when dealing with high-resolution images, dominant LMMs usually divide them into multiple local images and one global image, which will lead to a large number of visual tokens. In this work, we introduce AVG-LLaVA, an LMM that can adaptively select the appropriate visual granularity based on the input image and instruction. This approach not only reduces the number of visual tokens and speeds up inference, but also improves the overall model performance. Specifically, we introduce the following modules based on LLaVA-NeXT: (a) a visual granularity scaler that includes multiple pooling layers to obtain visual tokens with different granularities; (b) a visual granularity router, which includes a Transformer layer, an MLP layer, and a voter layer, used to select the appropriate visual granularity based on the image and instruction. Furthermore, we propose RGLF, a novel training paradigm that aims at aligning the granularity predicted by the router with the preferences of the LMM, without the need for additional manually annotated data. Extensive experiments and analysis show that AVG-LLaVA achieves superior performance across 11 benchmarks, as well as significantly reduces the number of visual tokens and speeds up inference (e.g., an 85.3% reduction in visual tokens and a 2.53$\times$ increase in inference speed on the AI2D benchmark).
Authors: Ulyana Piterbarg, Lerrel Pinto, Rob Fergus
Abstract: Software engineers mainly write code by editing existing programs. In contrast, large language models (LLMs) autoregressively synthesize programs in a single pass. One explanation for this is the scarcity of open-sourced edit data. While high-quality instruction data for code synthesis is already scarce, high-quality edit data is even scarcer. To fill this gap, we develop a synthetic data generation algorithm called LintSeq. This algorithm refactors existing code into a sequence of code edits by using a linter to procedurally sample across the error-free insertions that can be used to sequentially write programs. It outputs edit sequences as text strings consisting of consecutive program diffs. To test LintSeq, we use it to refactor a dataset of instruction + program pairs into instruction + program-diff-sequence tuples. Then, we instruction finetune a series of smaller LLMs ranging from 2.6B to 14B parameters on both the re-factored and original versions of this dataset, comparing zero-shot performance on code synthesis benchmarks. We show that during repeated sampling, edit sequence finetuned models produce more diverse programs than baselines. This results in better inference-time scaling for benchmark coverage as a function of samples, i.e. the fraction of problems "pass@k" solved by any attempt given "k" tries. For example, on HumanEval pass@50, small LLMs finetuned on synthetic edit sequences are competitive with GPT-4 and outperform models finetuned on the baseline dataset by +20% (+/-3%) in absolute score. Finally, we also pretrain our own tiny LMs for code understanding. We show that finetuning tiny models on synthetic code edits results in state-of-the-art code synthesis for the on-device model class. Our 150M parameter edit sequence LM matches or outperforms code models with twice as many parameters, both with and without repeated sampling, including Codex and AlphaCode.
Authors: Jianrui Zhang, Mu Cai, Yong Jae Lee
Abstract: There has been growing sentiment recently that modern large multimodal models (LMMs) have addressed most of the key challenges related to short video comprehension. As a result, both academia and industry are gradually shifting their attention towards the more complex challenges posed by understanding long-form videos. However, is this really the case? Our studies indicate that LMMs still lack many fundamental reasoning capabilities even when dealing with short videos. We introduce Vinoground, a temporal counterfactual LMM evaluation benchmark encompassing 1000 short and natural video-caption pairs. We demonstrate that existing LMMs severely struggle to distinguish temporal differences between different actions and object transformations. For example, the best model GPT-4o only obtains ~50% on our text and video scores, showing a large gap compared to the human baseline of ~90%. All open-source multimodal models and CLIP-based models perform much worse, producing mostly random chance performance. Through this work, we shed light onto the fact that temporal reasoning in short videos is a problem yet to be fully solved. The dataset and evaluation code are available at https://vinoground.github.io.
Authors: Gabriele Merlin, Mariya Toneva
Abstract: Pretrained language models have been shown to significantly predict brain recordings of people comprehending language. Recent work suggests that the prediction of the next word is a key mechanism that contributes to this alignment. What is not yet understood is whether prediction of the next word is necessary for this observed alignment or simply sufficient, and whether there are other shared mechanisms or information that are similarly important. In this work, we take a step towards understanding the reasons for brain alignment via two simple perturbations in popular pretrained language models. These perturbations help us design contrasts that can control for different types of information. By contrasting the brain alignment of these differently perturbed models, we show that improvements in alignment with brain recordings are due to more than improvements in next-word prediction and word-level information.
Authors: Abdullatif K\"oksal, Timo Schick, Anna Korhonen, Hinrich Sch\"utze
Abstract: Instruction tuning enables language models to more effectively generalize and better follow user intent. However, obtaining instruction data is costly and challenging. Prior work employs methods such as expensive human annotation, crowd-sourced datasets with alignment issues, and generating noisy examples via LLMs. We introduce the LongForm-C dataset, which is created by reverse instructions. We generate instructions via LLMs for human-written corpus examples using reverse instructions. First we select a diverse set of human-written documents from corpora such as C4 and Wikipedia; then we generate instructions for these documents via LLMs. This approach provides a cheaper and cleaner instruction-tuning dataset with natural output and one suitable for long text generation. Our models outperform 10x larger language models without instruction tuning on tasks such as story/recipe generation and long-form question answering. Moreover, LongForm models outperform prior instruction-tuned models such as FLAN-T5 and Alpaca by a large margin, and improve language understanding capabilities further. We publicly release our data and models: https://github.com/akoksal/LongForm.
Authors: Zhengliang Liu, Zihao Wu, Mengxuan Hu, Bokai Zhao, Lin Zhao, Tianyi Zhang, Haixing Dai, Xianyan Chen, Ye Shen, Sheng Li, Quanzheng Li, Xiang Li, Brian Murray, Tianming Liu, Andrea Sikora
Abstract: In this study, we introduce PharmacyGPT, a novel framework to assess the capabilities of large language models (LLMs) such as ChatGPT and GPT-4 in emulating the role of clinical pharmacists. Our methodology encompasses the utilization of LLMs to generate comprehensible patient clusters, formulate medication plans, and forecast patient outcomes. We conduct our investigation using real data acquired from the intensive care unit (ICU) at the University of North Carolina Chapel Hill (UNC) Hospital. Our analysis offers valuable insights into the potential applications and limitations of LLMs in the field of clinical pharmacy, with implications for both patient care and the development of future AI-driven healthcare solutions. By evaluating the performance of PharmacyGPT, we aim to contribute to the ongoing discourse surrounding the integration of artificial intelligence in healthcare settings, ultimately promoting the responsible and efficacious use of such technologies.
Authors: Ildik\'o Pil\'an, Laurent Pr\'evot, Hendrik Buschmeier, Pierre Lison
Abstract: Scripted dialogues such as movie and TV subtitles constitute a widespread source of training data for conversational NLP models. However, there are notable linguistic differences between these dialogues and spontaneous interactions, especially regarding the occurrence of communicative feedback such as backchannels, acknowledgments, or clarification requests. This paper presents a quantitative analysis of such feedback phenomena in both subtitles and spontaneous conversations. Based on conversational data spanning eight languages and multiple genres, we extract lexical statistics, classifications from a dialogue act tagger, expert annotations and labels derived from a fine-tuned Large Language Model (LLM). Our main empirical findings are that (1) communicative feedback is markedly less frequent in subtitles than in spontaneous dialogues and (2) subtitles contain a higher proportion of negative feedback. We also show that dialogues generated by standard LLMs lie much closer to scripted dialogues than spontaneous interactions in terms of communicative feedback.
Authors: Wanyun Cui, Qianle Wang
Abstract: Instructions augmentation is a crucial step for unleashing the full potential of large language models (LLMs) in downstream tasks. Existing Self-Instruct methods primarily simulate new instructions from a few initial instructions with in-context learning. However, our study identifies a critical flaw in this approach: even with GPT4o, Self-Instruct cannot generate complex instructions of length $\ge 100$, which is necessary in complex tasks such as code completion. To address this issue, our key insight is that fine-tuning open source LLMs with only ten examples can produce complex instructions that maintain distributional consistency for complex reasoning tasks. We introduce Ada-Instruct, an adaptive instruction generator developed through fine-tuning. We empirically validated Ada-Instruct's efficacy across different applications. The results highlight Ada-Instruct's capacity to generate long, intricate, and distributionally consistent instructions.
Authors: Yuxiang Zhou, Jiazheng Li, Yanzheng Xiang, Hanqi Yan, Lin Gui, Yulan He
Abstract: Understanding in-context learning (ICL) capability that enables large language models (LLMs) to excel in proficiency through demonstration examples is of utmost importance. This importance stems not only from the better utilization of this capability across various tasks, but also from the proactive identification and mitigation of potential risks, including concerns regarding truthfulness, bias, and toxicity, that may arise alongside the capability. In this paper, we present a thorough survey on the interpretation and analysis of in-context learning. First, we provide a concise introduction to the background and definition of in-context learning. Then, we give an overview of advancements from two perspectives: 1) a theoretical perspective, emphasizing studies on mechanistic interpretability and delving into the mathematical foundations behind ICL; and 2) an empirical perspective, concerning studies that empirically analyze factors associated with ICL. We conclude by highlighting the challenges encountered and suggesting potential avenues for future research. We believe that our work establishes the basis for further exploration into the interpretation of in-context learning. Additionally, we have created a repository containing the resources referenced in our survey.
Authors: Wenhao Yu, Hongming Zhang, Xiaoman Pan, Kaixin Ma, Hongwei Wang, Dong Yu
Abstract: Retrieval-augmented language models (RALMs) represent a substantial advancement in the capabilities of large language models, notably in reducing factual hallucination by leveraging external knowledge sources. However, the reliability of the retrieved information is not always guaranteed. The retrieval of irrelevant data can lead to misguided responses, and potentially causing the model to overlook its inherent knowledge, even when it possesses adequate information to address the query. Moreover, standard RALMs often struggle to assess whether they possess adequate knowledge, both intrinsic and retrieved, to provide an accurate answer. In situations where knowledge is lacking, these systems should ideally respond with "unknown" when the answer is unattainable. In response to these challenges, we introduces Chain-of-Noting (CoN), a novel approach aimed at improving the robustness of RALMs in facing noisy, irrelevant documents and in handling unknown scenarios. The core idea of CoN is to generate sequential reading notes for retrieved documents, enabling a thorough evaluation of their relevance to the given question and integrating this information to formulate the final answer. We employed ChatGPT to create training data for CoN, which was subsequently trained on an LLaMa-2 7B model. Our experiments across four open-domain QA benchmarks show that RALMs equipped with CoN significantly outperform standard RALMs. Notably, CoN achieves an average improvement of +7.9 in EM score given entirely noisy retrieved documents and +10.5 in rejection rates for real-time questions that fall outside the pre-training knowledge scope.
Authors: Yiqing Xie, Sheng Zhang, Hao Cheng, Pengfei Liu, Zelalem Gero, Cliff Wong, Tristan Naumann, Hoifung Poon, Carolyn Rose
Abstract: Medical text generation aims to assist with administrative work and highlight salient information to support decision-making. To reflect the specific requirements of medical text, in this paper, we propose a set of metrics to evaluate the completeness, conciseness, and attribution of the generated text at a fine-grained level. The metrics can be computed by various types of evaluators including instruction-following (both proprietary and open-source) and supervised entailment models. We demonstrate the effectiveness of the resulting framework, DocLens, with three evaluators on three tasks: clinical note generation, radiology report summarization, and patient question summarization. A comprehensive human study shows that DocLens exhibits substantially higher agreement with the judgments of medical experts than existing metrics. The results also highlight the need to improve open-source evaluators and suggest potential directions.
Authors: Jiaju Chen, Yuxuan Lu, Shao Zhang, Bingsheng Yao, Yuanzhe Dong, Ying Xu, Yunyao Li, Qianwen Wang, Dakuo Wang, Yuling Sun
Abstract: Interactive story reading is a common parent-child activity, where parents expect to teach both language skills and real-world knowledge beyond the story. While increasing storytelling and reading systems have been developed for this activity, they often fail to infuse real-world knowledge into the conversation. This limitation can be attributed to the existing question-answering (QA) datasets used for children's education, upon which the systems are built, failing to capture the nuances of how education experts think when conducting interactive story reading activities. To bridge this gap, we design an annotation framework, empowered by existing knowledge graph to capture experts' annotations and thinking process, and leverage this framework to construct StorySparkQA dataset, which comprises 5,868 expert-annotated QA pairs with real-world knowledge. We conduct automated and human expert evaluations across various QA pair generation settings to demonstrate that our StorySparkQA can effectively support models in generating QA pairs that target real-world knowledge beyond story content. StorySparkQA is available at https://huggingface.co/datasets/NEU-HAI/StorySparkQA.
Authors: Bowen Jin, Gang Liu, Chi Han, Meng Jiang, Heng Ji, Jiawei Han
Abstract: Large language models (LLMs), such as GPT4 and LLaMA, are creating significant advancements in natural language processing, due to their strong text encoding/decoding ability and newly found emergent capability (e.g., reasoning). While LLMs are mainly designed to process pure texts, there are many real-world scenarios where text data is associated with rich structure information in the form of graphs (e.g., academic networks, and e-commerce networks) or scenarios where graph data is paired with rich textual information (e.g., molecules with descriptions). Besides, although LLMs have shown their pure text-based reasoning ability, it is underexplored whether such ability can be generalized to graphs (i.e., graph-based reasoning). In this paper, we provide a systematic review of scenarios and techniques related to large language models on graphs. We first summarize potential scenarios of adopting LLMs on graphs into three categories, namely pure graphs, text-attributed graphs, and text-paired graphs. We then discuss detailed techniques for utilizing LLMs on graphs, including LLM as Predictor, LLM as Encoder, and LLM as Aligner, and compare the advantages and disadvantages of different schools of models. Furthermore, we discuss the real-world applications of such methods and summarize open-source codes and benchmark datasets. Finally, we conclude with potential future research directions in this fast-growing field. The related source can be found at https://github.com/PeterGriffinJin/Awesome-Language-Model-on-Graphs.
URLs: https://github.com/PeterGriffinJin/Awesome-Language-Model-on-Graphs.
Authors: Tannon Kew, Florian Schottmann, Rico Sennrich
Abstract: The vast majority of today's large language models (LLMs) are English-centric, having been pretrained predominantly on English text. Yet, in order to meet user expectations, models need to be able to respond appropriately in multiple languages once deployed in downstream applications. This requires strong cross-lingual transfer abilities. In this work, we investigate the minimal amount of multilinguality required during finetuning to elicit cross-lingual generalisation in English-centric LLMs. In experiments across four LLMs, we find that multilingual instruction tuning with as few as two to three languages is both necessary and sufficient to elicit effective cross-lingual generalisation, with the limiting factor being the degree to which a target language is seen during pretraining. Evaluations on five different tasks further reveal that multilingual instruction tuning is most beneficial for generative tasks that assume input/output language agreement, such as in chat settings, while being of less importance for highly structured classification-style tasks. Our code and data is available at https://github.com/ZurichNLP/multilingual-instruction-tuning.
URLs: https://github.com/ZurichNLP/multilingual-instruction-tuning.
Authors: Xianming Li, Jing Li
Abstract: Social media data exhibits severe redundancy caused by its noisy nature. It leads to increased training time and model bias in its processing. To address this issue, we propose a novel Generative Deduplication framework for social media data selection by removing semantically duplicate data. While related work involves data selection in task-specific training, our model acts as an efficient pre-processing method to universally enhance social media NLP pipelines. Specifically, we train a generative model via self-supervised learning to predict a keyword to capture the semantics of noisy social media text for deduplication. Meanwhile, time-dimensional Gaussian noise is added to improve training complexity and avoid learning trivial features. Extensive experiments suggest that our model can better reduce training samples while improving performance than baselines. The results show our model's potential to broadly advance social media language understanding in effectiveness and efficiency.
Authors: Yotam Wolf, Noam Wies, Dorin Shteyman, Binyamin Rothberg, Yoav Levine, Amnon Shashua
Abstract: Language model alignment has become an important component of AI safety, allowing safe interactions between humans and language models, by enhancing desired behaviors and inhibiting undesired ones. It is often done by tuning the model or inserting preset aligning prompts. Recently, representation engineering, a method which alters the model's behavior via changing its representations post-training, was shown to be effective in aligning LLMs (Zou et al., 2023a). Representation engineering yields gains in alignment oriented tasks such as resistance to adversarial attacks and reduction of social biases, but was also shown to cause a decrease in the ability of the model to perform basic tasks. In this paper we study the tradeoff between the increase in alignment and decrease in helpfulness of the model. We propose a theoretical framework which provides bounds for these two quantities, and demonstrate their relevance empirically. First, we find that under the conditions of our framework, alignment can be guaranteed with representation engineering, and at the same time that helpfulness is harmed in the process. Second, we show that helpfulness is harmed quadratically with the norm of the representation engineering vector, while the alignment increases linearly with it, indicating a regime in which it is efficient to use representation engineering. We validate our findings empirically, and chart the boundaries to the usefulness of representation engineering for alignment.
Authors: Tao Yuan, Xuefei Ning, Dong Zhou, Zhijie Yang, Shiyao Li, Minghui Zhuang, Zheyue Tan, Zhuyu Yao, Dahua Lin, Boxun Li, Guohao Dai, Shengen Yan, Yu Wang
Abstract: State-of-the-art large language models (LLMs) are now claiming remarkable supported context lengths of 256k or even more. In contrast, the average context lengths of mainstream benchmarks are insufficient (5k-21k), and they suffer from potential knowledge leakage and inaccurate metrics, resulting in biased evaluation. This paper introduces LV-Eval, a challenging long-context benchmark with five length levels (16k, 32k, 64k, 128k, and 256k) reaching up to 256k words. LV-Eval features two main tasks, single-hop QA and multi-hop QA, comprising 11 bilingual datasets. The design of LV-Eval has incorporated three key techniques, namely confusing facts insertion, keyword and phrase replacement, and keyword-recall-based metric design. The advantages of LV-Eval include controllable evaluation across different context lengths, challenging test instances with confusing facts, mitigated knowledge leakage, and more objective evaluations. We evaluate 15 LLMs on LV-Eval and conduct ablation studies on the benchmarking techniques. The results reveal that: (i) Moonshot-v1 and recent large-scale open-source models, such as Qwen-2.5-72B and Llama-3.1-70B, achieve the highest performance on LV-Eval, particularly at lengths below 64k. (ii) Models exhibit distinct score trends. For example, GLM-4-9B-128k, Yi-6B-200k, and Llama3-8B-1M exhibit a relatively gentle degradation of performance, but their absolute performances may not necessarily be higher than those of LLMs with shorter context lengths. (iii) LLMs' performances can significantly degrade in the presence of confusing information, especially in the pressure test of "needle in a haystack". (iv) Issues related to knowledge leakage and inaccurate metrics introduce bias in evaluation, and these concerns are alleviated in LV-Eval. All datasets and evaluation codes are released at: https://github.com/infinigence/LVEval.
Authors: Matthew Renze, Erhan Guven
Abstract: In this research study, we empirically investigate the effect of sampling temperature on the performance of Large Language Models (LLMs) on various problem-solving tasks. We created a multiple-choice question-and-answer (MCQA) exam by randomly sampling problems from standard LLM benchmarks. Then, we used nine popular LLMs with five prompt-engineering techniques to solve the MCQA problems while increasing the sampling temperature from 0.0 to 1.6. Despite anecdotal reports to the contrary, our empirical results indicate that changes in temperature from 0.0 to 1.0 do not have a statistically significant impact on LLM performance for problem-solving tasks. In addition, these results appear to generalize across LLMs, prompt-engineering techniques, and problem domains. All code, data, and supplemental materials are available on GitHub at: https://github.com/matthewrenze/jhu-llm-temperature
Authors: Yongchao Chen, Jacob Arkin, Yilun Hao, Yang Zhang, Nicholas Roy, Chuchu Fan
Abstract: Prompt optimization aims to find the best prompt to a large language model (LLM) for a given task. LLMs have been successfully used to help find and improve prompt candidates for single-step tasks. However, realistic tasks for agents are multi-step and introduce new challenges: (1) Prompt content is likely to be more extensive and complex, making it more difficult for LLMs to analyze errors, (2) the impact of an individual step is difficult to evaluate, and (3) different people may have varied preferences about task execution. While humans struggle to optimize prompts, they are good at providing feedback about LLM outputs; we therefore introduce a new LLM-driven discrete prompt optimization framework PRompt Optimization in Multi-Step Tasks (PROMST) that incorporates human-designed feedback rules to automatically offer direct suggestions for improvement. We also use an extra learned heuristic model that predicts prompt performance to efficiently sample from prompt candidates. This approach significantly outperforms both human-engineered prompts and several other prompt optimization methods across 11 representative multi-step tasks (an average 10.6\%-29.3\% improvement to current best methods on five LLMs respectively). We believe our work can serve as a benchmark for automatic prompt optimization for LLM-driven multi-step tasks. Datasets and Codes are available at https://github.com/yongchao98/PROMST. Project Page is available at https://yongchao98.github.io/MIT-REALM-PROMST.
URLs: https://github.com/yongchao98/PROMST., https://yongchao98.github.io/MIT-REALM-PROMST.
Authors: Branislav Pecher, Ivan Srba, Maria Bielikova
Abstract: While learning with limited labelled data can improve performance when the labels are lacking, it is also sensitive to the effects of uncontrolled randomness introduced by so-called randomness factors (e.g., varying order of data). We propose a method to systematically investigate the effects of randomness factors while taking the interactions between them into consideration. To measure the true effects of an individual randomness factor, our method mitigates the effects of other factors and observes how the performance varies across multiple runs. Applying our method to multiple randomness factors across in-context learning and fine-tuning approaches on 7 representative text classification tasks and meta-learning on 3 tasks, we show that: 1) disregarding interactions between randomness factors in existing works caused inconsistent findings due to incorrect attribution of the effects of randomness factors, such as disproving the consistent sensitivity of in-context learning to sample order even with random sample selection; and 2) besides mutual interactions, the effects of randomness factors, especially sample order, are also dependent on more systematic choices unexplored in existing works, such as number of classes, samples per class or choice of prompt format.
Authors: Prakamya Mishra, Zonghai Yao, Parth Vashisht, Feiyun Ouyang, Beining Wang, Vidhi Dhaval Mody, Hong Yu
Abstract: Large Language Models (LLMs) such as GPT & Llama have demonstrated significant achievements in summarization tasks but struggle with factual inaccuracies, a critical issue in clinical NLP applications where errors could lead to serious consequences. To counter the high costs and limited availability of expert-annotated data for factual alignment, this study introduces an innovative pipeline that utilizes >100B parameter GPT variants like GPT-3.5 & GPT-4 to act as synthetic experts to generate high-quality synthetics feedback aimed at enhancing factual consistency in clinical note summarization. Our research primarily focuses on edit feedback generated by these synthetic feedback experts without additional human annotations, mirroring and optimizing the practical scenario in which medical professionals refine AI system outputs. Although such 100B+ parameter GPT variants have proven to demonstrate expertise in various clinical NLP tasks, such as the Medical Licensing Examination, there is scant research on their capacity to act as synthetic feedback experts and deliver expert-level edit feedback for improving the generation quality of weaker (<10B parameter) LLMs like GPT-2 (1.5B) & Llama 2 (7B) in clinical domain. So in this work, we leverage 100B+ GPT variants to act as synthetic feedback experts offering expert-level edit feedback, that is used to reduce hallucinations and align weaker (<10B parameter) LLMs with medical facts using two distinct alignment algorithms (DPO & SALT), endeavoring to narrow the divide between AI-generated content and factual accuracy. This highlights the substantial potential of LLM-based synthetic edits in enhancing the alignment of clinical factuality.
Authors: Domenic Rosati, Jan Wehner, Kai Williams, {\L}ukasz Bartoszcze, Jan Batzner, Hassan Sajjad, Frank Rudzicz
Abstract: Large Language Models (LLMs) are often trained with safety guards intended to prevent harmful text generation. However, such safety training can be removed by fine-tuning the LLM on harmful datasets. While this emerging threat (harmful fine-tuning attacks) has been characterized by previous work, there is little understanding of how we should proceed in constructing and validating defenses against these attacks especially in the case where defenders would not have control of the fine-tuning process. We introduce a formal framework based on the training budget of an attacker which we call "Immunization" conditions. Using a formal characterisation of the harmful fine-tuning problem, we provide a thorough description of what a successful defense must comprise of and establish a set of guidelines on how rigorous defense research that gives us confidence should proceed.
Authors: Rares Dolga, Marius Cobzarenco, David Barber
Abstract: The time complexity of the standard attention mechanism in transformers scales quadratically with sequence length. We propose a probabilistic framework for attention, enabling us to derive a novel low-rank linear re-parameterisation of both bidirectional and causal cases, based on defining a latent variable model. Our method can be seamlessly integrated as a drop-in replacement for the standard attention mechanism. Additionally, this framework provides a natural extension for combining local standard attention with our global linear attention. This approach allows us to extend the context length of existing large pre-trained models with only a few additional training steps. The resulting ``Latte Transformer'' achieves performance comparable to standard attention and other state-of-the-art models, while maintaining linear time and memory complexity, along with constant-time next-token prediction during inference.
Authors: Giuseppe Attanasio, Beatrice Savoldi, Dennis Fucci, Dirk Hovy
Abstract: Current automatic speech recognition (ASR) models are designed to be used across many languages and tasks without substantial changes. However, this broad language coverage hides performance gaps within languages, for example, across genders. Our study systematically evaluates the performance of two widely used multilingual ASR models on three datasets, encompassing 19 languages from eight language families and two speaking conditions. Our findings reveal clear gender disparities, with the advantaged group varying across languages and models. Surprisingly, those gaps are not explained by acoustic or lexical properties. However, probing internal model states reveals a correlation with gendered performance gap. That is, the easier it is to distinguish speaker gender in a language using probes, the more the gap reduces, favoring female speakers. Our results show that gender disparities persist even in state-of-the-art models. Our findings have implications for the improvement of multilingual ASR systems, underscoring the importance of accessibility to training data and nuanced evaluation to predict and mitigate gender gaps. We release all code and artifacts at https://github.com/g8a9/multilingual-asr-gender-gap.
Authors: Sheikh Shafayat, Eunsu Kim, Juhyun Oh, Alice Oh
Abstract: Evaluating the factuality of long-form large language model (LLM)-generated text is an important challenge. Recently there has been a surge of interest in factuality evaluation for English, but little is known about the factuality evaluation of multilingual LLMs, specially when it comes to long-form generation. %This paper systematically evaluates multilingual LLMs' factual accuracy across languages and geographic regions. We introduce a simple pipeline for multilingual factuality evaluation, by applying FActScore (Min et al., 2023) for diverse languages. In addition to evaluating multilingual factual generation, we evaluate the factual accuracy of long-form text generation in topics that reflect regional diversity. We also examine the feasibility of running the FActScore pipeline using non-English Wikipedia and provide comprehensive guidelines on multilingual factual evaluation for regionally diverse topics.
Authors: Hossein Aboutalebi, Hwanjun Song, Yusheng Xie, Arshit Gupta, Justin Sun, Hang Su, Igor Shalyminov, Nikolaos Pappas, Siffi Singh, Saab Mansour
Abstract: Development of multimodal interactive systems is hindered by the lack of rich, multimodal (text, images) conversational data, which is needed in large quantities for LLMs. Previous approaches augment textual dialogues with retrieved images, posing privacy, diversity, and quality constraints. In this work, we introduce Multimodal Augmented Generative Images Dialogues (MAGID), a framework to augment text-only dialogues with diverse and high-quality images. Subsequently, a diffusion model is applied to craft corresponding images, ensuring alignment with the identified text. Finally, MAGID incorporates an innovative feedback loop between an image description generation module (textual LLM) and image quality modules (addressing aesthetics, image-text matching, and safety), that work in tandem to generate high-quality and multi-modal dialogues. We compare MAGID to other SOTA baselines on three dialogue datasets, using automated and human evaluation. Our results show that MAGID is comparable to or better than baselines, with significant improvements in human evaluation, especially against retrieval baselines where the image database is small.
Authors: Mitodru Niyogi, Arnab Bhattacharya
Abstract: In this paper, we present Paramanu-Ayn, a collection of legal language models trained exclusively on Indian legal case documents. This 97-million-parameter Auto-Regressive (AR) decoder-only model was pretrained from scratch with a context size of 8192 on a single GPU for just 185 hours, achieving an efficient MFU of 41.35. We also developed a legal domain specialized BPE tokenizer. We evaluated our model using perplexity and zero-shot tasks: case judgment prediction with explanation and abstractive case summarization. Paramanu-Ayn outperformed Llama-2 7B and Gemini-Pro in case judgment prediction with explanation task on test accuracy by nearly 2 percentage points, despite being 72 times smaller. In zero-shot abstractive summarization, it surpassed decoder-only LLMs generating fixed-length summaries (5000 tokens) by over 10 percentage points in BLEU and METEOR metrics, and by nearly 4 percentage points in BERTScore. Further evaluations on zero-shot commonsense and mathematical benchmarks showed that Paramanu-Ayn excelled despite being trained exclusively on legal documents, outperforming Llama-1, Llama-2, and Falcon on AGIEVAL-AQuA-RAT and AGIEVAL-SAT-Math tasks. We also instruction-tuned our model on 10,763 diverse legal tasks, including legal clause generation, legal drafting, case summarization, etc. The Paramanu-Ayn-instruct model scored above 8 out of 10 in clarity, relevance, completeness, and legal reasoning metrics by GPT-3.5-Turbo. We found that our models, were able to learn drafting knowledge and generalize to draft legal contracts and legal clauses with limited instruction-tuning. Hence, we conclude that for a strong domain-specialized generative language model (such as legal), domain specialized pretraining from scratch is more cost effective, environmentally friendly, and remains competitive with larger models or even better than adapting LLMs for legal domain tasks.
Authors: Juan De Gregorio, Ra\'ul Toral, David S\'anchez
Abstract: Languages are grouped into families that share common linguistic traits. While this approach has been successful in understanding genetic relations between diverse languages, more analyses are needed to accurately quantify their relatedness, especially in less studied linguistic levels such as syntax. Here, we explore linguistic distances using series of parts of speech (POS) extracted from the Universal Dependencies dataset. Within an information-theoretic framework, we show that employing POS trigrams maximizes the possibility of capturing syntactic variations while being at the same time compatible with the amount of available data. Linguistic connections are then established by assessing pairwise distances based on the POS distributions. Intriguingly, our analysis reveals definite clusters that correspond to well known language families and groups, with exceptions explained by distinct morphological typologies. Furthermore, we obtain a significant correlation between language similarity and geographic distance, which underscores the influence of spatial proximity on language kinships.
Authors: Yi Ren, Shangmin Guo, Linlu Qiu, Bailin Wang, Danica J. Sutherland
Abstract: With the widespread adoption of Large Language Models (LLMs), the prevalence of iterative interactions among these models is anticipated to increase. Notably, recent advancements in multi-round self-improving methods allow LLMs to generate new examples for training subsequent models. At the same time, multi-agent LLM systems, involving automated interactions among agents, are also increasing in prominence. Thus, in both short and long terms, LLMs may actively engage in an evolutionary process. We draw parallels between the behavior of LLMs and the evolution of human culture, as the latter has been extensively studied by cognitive scientists for decades. Our approach involves leveraging Iterated Learning (IL), a Bayesian framework that elucidates how subtle biases are magnified during human cultural evolution, to explain some behaviors of LLMs. This paper outlines key characteristics of agents' behavior in the Bayesian-IL framework, including predictions that are supported by experimental verification with various LLMs. This theoretical framework could help to more effectively predict and guide the evolution of LLMs in desired directions.
Authors: Bowen Jin, Chulin Xie, Jiawei Zhang, Kashob Kumar Roy, Yu Zhang, Zheng Li, Ruirui Li, Xianfeng Tang, Suhang Wang, Yu Meng, Jiawei Han
Abstract: Large language models (LLMs), while exhibiting exceptional performance, suffer from hallucinations, especially on knowledge-intensive tasks. Existing works propose to augment LLMs with individual text units retrieved from external knowledge corpora to alleviate the issue. However, in many domains, texts are interconnected (e.g., academic papers in a bibliographic graph are linked by citations and co-authorships) which form a (text-attributed) graph. The knowledge in such graphs is encoded not only in single texts/nodes but also in their associated connections. To facilitate the research of augmenting LLMs with graphs, we manually construct a Graph Reasoning Benchmark dataset called GRBench, containing 1,740 questions that can be answered with the knowledge from 10 domain graphs. Then, we propose a simple and effective framework called Graph Chain-of-thought (Graph-CoT) to augment LLMs with graphs by encouraging LLMs to reason on the graph iteratively. Each Graph-CoT iteration consists of three sub-steps: LLM reasoning, LLM-graph interaction, and graph execution. We conduct systematic experiments with three LLM backbones on GRBench, where Graph-CoT outperforms the baselines consistently. The code is available at https://github.com/PeterGriffinJin/Graph-CoT.
Authors: Yekun Chai, Qingyi Liu, Shuohuan Wang, Yu Sun, Qiwei Peng, Hua Wu
Abstract: Amidst the rapid advancements in generative language models, the investigation of how training data shapes the performance of GPT models is still emerging. This paper presents GPTfluence, a novel approach that leverages a featurized simulation to assess the impact of training examples on the training dynamics of GPT models. Our approach not only traces the influence of individual training instances on performance trajectories, such as loss and other key metrics, on targeted test points but also enables a comprehensive comparison with existing methods across various training scenarios in GPT models, ranging from 14 million to 2.8 billion parameters, across a range of downstream tasks. Contrary to earlier methods that struggle with generalization to new data, GPTfluence introduces a parameterized simulation of training dynamics, demonstrating robust generalization capabilities to unseen training data. This adaptability is evident across both fine-tuning and instruction-tuning scenarios, spanning tasks in natural language understanding and generation. We make our code and data publicly available at https://github.com/ernie-research/gptfluence.
Authors: Bin Wang, Chengwei Wei, Zhengyuan Liu, Geyu Lin, Nancy F. Chen
Abstract: As the rapidly advancing domain of natural language processing (NLP), large language models (LLMs) have emerged as powerful tools for interpreting human commands and generating text across various tasks. Nonetheless, the resilience of LLMs to handle text containing inherent errors, stemming from human interactions and collaborative systems, has not been thoroughly explored. Our study investigates the resilience of LLMs against five common types of disruptions including 1) ASR (Automatic Speech Recognition) errors, 2) OCR (Optical Character Recognition) errors, 3) grammatical mistakes, 4) typographical errors, and 5) distractive content. We aim to investigate how these models react by deliberately embedding these errors into instructions. Our findings reveal that while some LLMs show a degree of resistance to certain types of noise, their overall performance significantly suffers. This emphasizes the importance of further investigation into enhancing model resilience. In response to the observed decline in performance, our study also evaluates a "re-pass" strategy, designed to purify the instructions of noise before the LLMs process them. Our analysis indicates that correcting noisy instructions, particularly for open-source LLMs, presents significant challenges.
Authors: Hyeonbin Hwang, Doyoung Kim, Seungone Kim, Seonghyeon Ye, Minjoon Seo
Abstract: Training on large amounts of rationales (i.e., CoT Fine-tuning) is effective at improving the reasoning capabilities of large language models (LLMs). However, acquiring human-authored rationales or augmenting rationales from proprietary models is costly and not scalable. In this paper, we study the problem of whether LLMs could self-improve their reasoning capabilities. To this end, we propose Self-Explore, where the LLM is tasked to explore the first wrong step (i.e., the first pit) within the rationale and use such signals as fine-grained rewards for further improvement. On the GSM8K and MATH test set, Self-Explore achieves 11.57% and 2.89% improvement on average across three LLMs compared to supervised fine-tuning (SFT). Our code is available at https://github.com/hbin0701/Self-Explore.
Authors: Yekun Chai, Qingyi Liu, Jingwu Xiao, Shuohuan Wang, Yu Sun, Hua Wu
Abstract: The integration of visual and textual information represents a promising direction in the advancement of language models. In this paper, we explore the dual modality of language--both visual and textual--within an autoregressive framework, pre-trained on both document images and texts. Our method employs a multimodal training strategy, utilizing visual data through next patch prediction with a regression head and/or textual data through next token prediction with a classification head. We focus on understanding the interaction between these two modalities and their combined impact on model performance. Our extensive evaluation across a wide range of benchmarks shows that incorporating both visual and textual data significantly improves the performance of pixel-based language models. Remarkably, we find that a unidirectional pixel-based model trained solely on visual data can achieve comparable results to state-of-the-art bidirectional models on several language understanding tasks. This work uncovers the untapped potential of integrating visual and textual modalities for more effective language modeling. We release our code, data, and model checkpoints at \url{https://github.com/ernie-research/pixelgpt}.
Authors: Yating Wu, Ritika Mangla, Alexandros G. Dimakis, Greg Durrett, Junyi Jessy Li
Abstract: Inquisitive questions -- open-ended, curiosity-driven questions people ask as they read -- are an integral part of discourse processing (Kehler and Rohde, 2017; Onea, 2016) and comprehension (Prince, 2004). Recent work in NLP has taken advantage of question generation capabilities of LLMs to enhance a wide range of applications. But the space of inquisitive questions is vast: many questions can be evoked from a given context. So which of those should be prioritized to find answers? Linguistic theories, unfortunately, have not yet provided an answer to this question. This paper presents QSALIENCE, a salience predictor of inquisitive questions. QSALIENCE is instruction-tuned over our dataset of linguist-annotated salience scores of 1,766 (context, question) pairs. A question scores high on salience if answering it would greatly enhance the understanding of the text (Van Rooy, 2003). We show that highly salient questions are empirically more likely to be answered in the same article, bridging potential questions (Onea, 2016) with Questions Under Discussion (Roberts, 2012). We further validate our findings by showing that answering salient questions is an indicator of summarization quality in news.
Authors: Yao Xu, Shizhu He, Jiabei Chen, Zihao Wang, Yangqiu Song, Hanghang Tong, Guang Liu, Kang Liu, Jun Zhao
Abstract: To address the issues of insufficient knowledge and hallucination in Large Language Models (LLMs), numerous studies have explored integrating LLMs with Knowledge Graphs (KGs). However, these methods are typically evaluated on conventional Knowledge Graph Question Answering (KGQA) with complete KGs, where all factual triples required for each question are entirely covered by the given KG. In such cases, LLMs primarily act as an agent to find answer entities within the KG, rather than effectively integrating the internal knowledge of LLMs and external knowledge sources such as KGs. In fact, KGs are often incomplete to cover all the knowledge required to answer questions. To simulate these real-world scenarios and evaluate the ability of LLMs to integrate internal and external knowledge, we propose leveraging LLMs for QA under Incomplete Knowledge Graph (IKGQA), where the provided KG lacks some of the factual triples for each question, and construct corresponding datasets. To handle IKGQA, we propose a training-free method called Generate-on-Graph (GoG), which can generate new factual triples while exploring KGs. Specifically, GoG performs reasoning through a Thinking-Searching-Generating framework, which treats LLM as both Agent and KG in IKGQA. Experimental results on two datasets demonstrate that our GoG outperforms all previous methods.
Authors: Constanza Fierro, Jiaang Li, Anders S{\o}gaard
Abstract: The purpose of instruction tuning is enabling zero-shot performance, but instruction tuning has also been shown to improve chain-of-thought reasoning and value alignment (Si et al., 2023). Here we consider the impact on $\textit{consistency}$, i.e., the sensitivity of language models to small perturbations in the input. We compare 10 instruction-tuned LLaMA models to the original LLaMA-7b model and show that almost across-the-board they become more consistent, both in terms of their representations and their predictions in zero-shot and downstream tasks. We explain these improvements through mechanistic analyses of factual recall.
Authors: Xiaoming Liu, Chen Liu, Zhaohan Zhang, Chengzhengxu Li, Longtian Wang, Yu Lan, Chao Shen
Abstract: Large language models have shown their ability to become effective few-shot learners with prompting, revolutionizing the paradigm of learning with data scarcity. However, this approach largely depends on the quality of prompt initialization, and always exhibits large variability among different runs. Such property makes prompt tuning highly unreliable and vulnerable to poorly constructed prompts, which limits its extension to more real-world applications. To tackle this issue, we propose to treat the hard prompt and soft prompt as separate inputs to mitigate noise brought by the prompt initialization. Furthermore, we optimize soft prompts with contrastive learning for utilizing class-aware information in the training process to maintain model performance. Experimental results demonstrate that \sysname outperforms state-of-the-art methods by 6.97% in accuracy and reduces the standard deviation by 1.92 on average. Furthermore, extensive experiments underscore its robustness and stability across 8 datasets covering various tasks. Codes are available at https://github.com/lccc0528/Stable/tree/main.
Authors: Yuanhao Yue, Chengyu Wang, Jun Huang, Peng Wang
Abstract: Instruction tuning aims to align large language models (LLMs) with open-domain instructions and human-preferred responses. While several studies have explored autonomous approaches to distilling and annotating instructions from powerful proprietary LLMs, such as ChatGPT, they often neglect the impact of the distributions and characteristics of tasks, together with the varying difficulty of instructions in training sets. This oversight can lead to imbalanced knowledge capabilities and poor generalization powers of student LLMs. To address these challenges, we introduce Task-Aware Curriculum Planning for Instruction Refinement (TAPIR), a multi-round distillation framework that utilizes an oracle LLM to select instructions that are difficult for a student LLM to follow. To balance the student's capabilities, task distributions in training sets are adjusted with responses automatically refined according to their corresponding tasks. In addition, by incorporating curriculum planning, our approach systematically escalates the difficulty levels of tasks, progressively enhancing the student LLM's capabilities. We rigorously evaluate TAPIR using several widely recognized benchmarks (such as AlpacaEval 2.0, MT-Bench, etc.) and multiple student LLMs. Empirical results demonstrate that student LLMs, trained with our method and less training data, outperform larger instruction-tuned models and strong distillation baselines.
Authors: Zhenyu Wu, Qingkai Zeng, Zhihan Zhang, Zhaoxuan Tan, Chao Shen, Meng Jiang
Abstract: Intrinsic self-correct was a method that instructed large language models (LLMs) to verify and correct their responses without external feedback. Unfortunately, the study concluded that the LLMs could not self-correct reasoning yet. We find that a simple yet effective verification method can unleash inherent capabilities of the LLMs. That is to mask a key condition in the question, add the current response to construct a verification question, and predict the condition to verify the response. The condition can be an entity in an open-domain question or a numeric value in a math question, which requires minimal effort (via prompting) to identify. We propose an iterative verify-then-correct framework to progressively identify and correct (probably) false responses, named ProCo. We conduct experiments on three reasoning tasks. On average, ProCo, with GPT-3.5-Turbo as the backend LLM, yields $+6.8$ exact match on four open-domain question answering datasets, $+14.1$ accuracy on three arithmetic reasoning datasets, and $+9.6$ accuracy on a commonsense reasoning dataset, compared to Self-Correct. Our implementation is made publicly available at https://wzy6642.github.io/proco.github.io/.
Authors: Zhengyan Shi, Adam X. Yang, Bin Wu, Laurence Aitchison, Emine Yilmaz, Aldo Lipani
Abstract: Instruction tuning plays a crucial role in shaping the outputs of language models (LMs) to desired styles. In this work, we propose a simple yet effective method, Instruction Modelling (IM), which trains LMs by applying a loss function to the instruction and prompt part rather than solely to the output part. Through experiments across 21 diverse benchmarks, we show that, in many scenarios, IM can effectively improve the LM performance on both NLP tasks (e.g., MMLU, TruthfulQA, and HumanEval) and open-ended generation benchmarks (e.g., MT-Bench and AlpacaEval). Remarkably, in the most advantageous case, IM boosts model performance on AlpacaEval 1.0 by over 100%. We identify two key factors influencing the effectiveness of IM: (1) The ratio between instruction length and output length in the training data; and (2) The number of training examples. We observe that IM is especially beneficial when trained on datasets with lengthy instructions paired with brief outputs, or under the Superficial Alignment Hypothesis (SAH) where a small amount of training examples are used for instruction tuning. Further analysis substantiates our hypothesis that our improvement can be attributed to reduced overfitting to instruction tuning datasets. It is worth noting that we are not proposing \ours as a replacement for current fine-tuning processes. Instead, our work aims to provide practical guidance for instruction tuning LMs, especially in low-resource scenarios.
Authors: Tianhao Zhang, Zhexiao Lin, Zhecheng Sheng, Chen Jiang, Dongyeop Kang
Abstract: Generative models have gained significant prominence in Natural Language Processing (NLP), especially in tackling the complex task of modeling and evaluating long text sequences. This task is crucial for advancing various downstream applications, such as text generation and machine translation. Recent methods that utilize stochastic processes to capture the intrinsic dynamics of sequences have shown superior performance in generative modeling. However, the accurate encoding of both temporal and structural dependencies from text datasets, as well as leveraging this encoded information for sequence evaluation, remains an open area of research. In this paper, we propose a novel approach to learn the stochastic dynamics of long text sequences, utilizing a negative log-likelihood-based encoder that outperforms contrastive learning methods. We also introduce a likelihood-based evaluation metric for long-text assessment, which measures sequence coherence and can be applied to downstream tasks such as Human-AI discrimination. Our encoder preserves sequence coherence effectively and performs robustly on out-of-domain datasets. Additionally, the proposed evaluation metric captures both temporal and structural information comprehensively. Theoretical analysis demonstrates the superiority of our metric in sequence evaluation, and experimental results highlight its flexibility and exceptional performance across a variety of tasks, showcasing its utility in diverse NLP applications.
Authors: Eshaan Agarwal, Joykirat Singh, Vivek Dani, Raghav Magazine, Tanuja Ganu, Akshay Nambi
Abstract: Large language models (LLMs) have transformed AI across diverse domains, with prompting being central to their success in guiding model outputs. However, manual prompt engineering is both labor-intensive and domain-specific, necessitating the need for automated solutions. We introduce PromptWizard, a novel, fully automated framework for discrete prompt optimization, utilizing a self-evolving, self-adapting mechanism. Through a feedback-driven critique and synthesis process, PromptWizard achieves an effective balance between exploration and exploitation, iteratively refining both prompt instructions and in-context examples to generate human-readable, task-specific prompts. This guided approach systematically improves prompt quality, resulting in superior performance across 45 tasks. PromptWizard excels even with limited training data, smaller LLMs, and various LLM architectures. Additionally, our cost analysis reveals a substantial reduction in API calls, token usage, and overall cost, demonstrating PromptWizard's efficiency, scalability, and advantages over existing prompt optimization strategies.
Authors: Zefan Cai, Yichi Zhang, Bofei Gao, Yuliang Liu, Tianyu Liu, Keming Lu, Wayne Xiong, Yue Dong, Baobao Chang, Junjie Hu, Wen Xiao
Abstract: In this study, we investigate whether attention-based information flow inside large language models (LLMs) is aggregated through noticeable patterns for long context processing. Our observations reveal that LLMs aggregate information through Pyramidal Information Funneling where attention is scattering widely in lower layers, progressively consolidating within specific contexts, and ultimately focusing on critical tokens (a.k.a massive activation or attention sink) in higher layers. Motivated by these insights, we developed PyramidKV, a novel and effective KV cache compression method. This approach dynamically adjusts the KV cache size across different layers, allocating more cache in lower layers and less in higher ones, diverging from traditional methods that maintain a uniform KV cache size. Our experimental evaluations, utilizing the LongBench benchmark, show that PyramidKV matches the performance of models with a full KV cache while retaining only 12% of the KV cache, thus significantly reducing memory usage. In scenarios emphasizing memory efficiency, where only 0.7% of the KV cache is maintained, PyramidKV surpasses other KV cache compression techniques, achieving up to a 20.5 absolute accuracy improvement on TREC dataset. In the Needle-in-a-Haystack experiment, PyramidKV outperforms competing methods in maintaining long-context comprehension in LLMs; notably, retaining just 128 KV cache entries enables the LLAMA-3-70B model to achieve 100% Acc. performance, matching that of a full KV cache.
Authors: Baixuan Xu, Weiqi Wang, Haochen Shi, Wenxuan Ding, Huihao Jing, Tianqing Fang, Jiaxin Bai, Long Chen, Yangqiu Song
Abstract: Improving user experience and providing personalized search results in E-commerce platforms heavily rely on understanding purchase intention. However, existing methods for acquiring large-scale intentions bank on distilling large language models with human annotation for verification. Such an approach tends to generate product-centric intentions, overlook valuable visual information from product images, and incurs high costs for scalability. To address these issues, we introduce MIND, a multimodal framework that allows Large Vision-Language Models (LVLMs) to infer purchase intentions from multimodal product metadata and prioritize human-centric ones. Using Amazon Review data, we apply MIND and create a multimodal intention knowledge base, which contains 1,264,441 million intentions derived from 126,142 co-buy shopping records across 107,215 products. Extensive human evaluations demonstrate the high plausibility and typicality of our obtained intentions and validate the effectiveness of our distillation framework and filtering mechanism. Additional experiments reveal that our obtained intentions significantly enhance large language models in two intention comprehension tasks.
Authors: Bolei Ma, Xinpeng Wang, Tiancheng Hu, Anna-Carolina Haensch, Michael A. Hedderich, Barbara Plank, Frauke Kreuter
Abstract: Recent advances in Large Language Models (LLMs) have sparked wide interest in validating and comprehending the human-like cognitive-behavioral traits LLMs may capture and convey. These cognitive-behavioral traits include typically Attitudes, Opinions, Values (AOVs). However, measuring AOVs embedded within LLMs remains opaque, and different evaluation methods may yield different results. This has led to a lack of clarity on how different studies are related to each other and how they can be interpreted. This paper aims to bridge this gap by providing a comprehensive overview of recent works on the evaluation of AOVs in LLMs. Moreover, we survey related approaches in different stages of the evaluation pipeline in these works. By doing so, we address the potential and challenges with respect to understanding the model, human-AI alignment, and downstream application in social sciences. Finally, we provide practical insights into evaluation methods, model enhancement, and interdisciplinary collaboration, thereby contributing to the evolving landscape of evaluating AOVs in LLMs.
Authors: Siyuan Qi, Bangcheng Yang, Kailin Jiang, Xiaobo Wang, Jiaqi Li, Yifan Zhong, Yaodong Yang, Zilong Zheng
Abstract: In scenarios where language models must incorporate new information efficiently without extensive retraining, traditional fine-tuning methods are prone to overfitting, degraded generalization, and unnatural language generation. To address these limitations, we introduce Consistent In-Context Editing (ICE), a novel approach leveraging the model's in-context learning capability to optimize toward a contextual distribution rather than a one-hot target. ICE introduces a simple yet effective optimization framework for the model to internalize new knowledge by aligning its output distributions with and without additional context. This method enhances the robustness and effectiveness of gradient-based tuning methods, preventing overfitting and preserving the model's integrity. We analyze ICE across four critical aspects of knowledge editing: accuracy, locality, generalization, and linguistic quality, demonstrating its advantages. Experimental results confirm the effectiveness of ICE and demonstrate its potential for continual editing, ensuring that the integrity of the model is preserved while updating information.
Authors: Leonardo Bertolazzi, Albert Gatt, Raffaella Bernardi
Abstract: The reasoning abilities of Large Language Models (LLMs) are becoming a central focus of study in NLP. In this paper, we consider the case of syllogistic reasoning, an area of deductive reasoning studied extensively in logic and cognitive psychology. Previous research has shown that pre-trained LLMs exhibit reasoning biases, such as $\textit{content effects}$, avoid answering that $\textit{no conclusion follows}$, display human-like difficulties, and struggle with multi-step reasoning. We contribute to this research line by systematically investigating the effects of chain-of-thought reasoning, in-context learning (ICL), and supervised fine-tuning (SFT) on syllogistic reasoning, considering syllogisms with conclusions that support or violate world knowledge, as well as ones with multiple premises. Crucially, we go beyond the standard focus on accuracy, with an in-depth analysis of the conclusions generated by the models. Our results suggest that the behavior of pre-trained LLMs can be explained by heuristics studied in cognitive science and that both ICL and SFT improve model performance on valid inferences, although only the latter mitigates most reasoning biases without harming model consistency.
Authors: Yekun Chai, Yewei Fang, Qiwei Peng, Xuhong Li
Abstract: Language models typically tokenize raw text into sequences of subword identifiers from a predefined vocabulary, a process inherently sensitive to typographical errors, length variations, and largely oblivious to the internal structure of tokens--issues we term the curse of tokenization. In this study, we delve into these drawbacks and demonstrate that large language models (LLMs) remain susceptible to these problems. This study systematically investigates these challenges and their impact on LLMs through three critical research questions: (1) complex problem solving, (2) token structure probing, and (3) resilience to typographical variation. Our findings reveal that scaling model parameters can mitigate the issue of tokenization; however, LLMs still suffer from biases induced by typos and other text format variations. Our experiments show that subword regularization such as BPE-dropout can mitigate this issue. We release our evaluation code and data at https://github.com/FloatAI/TKEval.
Authors: Marko Sterbentz, Cameron Barrie, Shubham Shahi, Abhratanu Dutta, Donna Hooshmand, Harper Pack, Kristian J. Hammond
Abstract: Large language models (LLMs) are capable of producing documents, and retrieval augmented generation (RAG) has shown itself to be a powerful method for improving accuracy without sacrificing fluency. However, not all information can be retrieved from text. We propose an approach that uses the analysis of structured data to generate fact sets that are used to guide generation in much the same way that retrieved documents are used in RAG. This analytics augmented generation (AAG) approach supports the ability to utilize standard analytic techniques to generate facts that are then converted to text and passed to an LLM. We present a neurosymbolic platform, Satyrn, that leverages AAG to produce accurate, fluent, and coherent reports grounded in large scale databases. In our experiments, we find that Satyrn generates reports in which over 86% of claims are accurate while maintaining high levels of fluency and coherence, even when using smaller language models such as Mistral-7B, as compared to GPT-4 Code Interpreter in which just 57% of claims are accurate.
Authors: Hawon Jeong, ChaeHun Park, Jimin Hong, Jaegul Choo
Abstract: Pairwise evaluation using large language models (LLMs) is widely adopted for evaluating generated outputs. However, the reliability of LLM evaluators is often compromised by their biased preferences, such as favoring verbosity and an authoritative tone. In this work, we find that the evaluation setup itself can significantly amplify these biases, where pairwise evaluators exhibit more undesirable tendencies than pointwise evaluators. Our analysis further reveals that even when pairwise evaluators make incorrect judgments, they can still accurately identify shortcomings in low-quality outputs. As a simple remedy, we also propose incorporating pointwise reasoning into pairwise evaluation. Experimental results show that our method improves the performance of pairwise evaluators on adversarial samples across various models. We hope our findings encourage further exploration into the reliability of LLM evaluators.
Authors: Minseok Choi, Kyunghyun Min, Jaegul Choo
Abstract: Pretrained language models memorize vast amounts of information, including private and copyrighted data, raising significant safety concerns. Retraining these models after excluding sensitive data is prohibitively expensive, making machine unlearning a viable, cost-effective alternative. Previous research has focused on machine unlearning for monolingual models, but we find that unlearning in one language does not necessarily transfer to others. This vulnerability makes models susceptible to low-resource language attacks, where sensitive information remains accessible in less dominant languages. This paper presents a pioneering approach to machine unlearning for multilingual language models, selectively erasing information across different languages while maintaining overall performance. Specifically, our method employs an adaptive unlearning scheme that assigns language-dependent weights to address different language performances of multilingual language models. Empirical results demonstrate the effectiveness of our framework compared to existing unlearning baselines, setting a new standard for secure and adaptable multilingual language models.
Authors: Kentaro Mitsui, Koh Mitsuda, Toshiaki Wakatsuki, Yukiya Hono, Kei Sawada
Abstract: Multimodal language models that process both text and speech have a potential for applications in spoken dialogue systems. However, current models face two major challenges in response generation latency: (1) generating a spoken response requires the prior generation of a written response, and (2) speech sequences are significantly longer than text sequences. This study addresses these issues by extending the input and output sequences of the language model to support the parallel generation of text and speech. Our experiments on spoken question answering tasks demonstrate that our approach improves latency while maintaining the quality of response content. Additionally, we show that latency can be further reduced by generating speech in multiple sequences. Demo samples are available at https://rinnakk.github.io/research/publications/PSLM.
Authors: Branislav Pecher, Jan Cegin, Robert Belanec, Jakub Simko, Ivan Srba, Maria Bielikova
Abstract: While fine-tuning of pre-trained language models generally helps to overcome the lack of labelled training samples, it also displays model performance instability. This instability mainly originates from randomness in initialisation or data shuffling. To address this, researchers either modify the training process or augment the available samples, which typically results in increased computational costs. We propose a new mitigation strategy, called Delayed Ensemble with Noisy Interpolation (DENI), that leverages the strengths of ensembling, noise regularisation and model interpolation, while retaining computational efficiency. We compare DENI with 9 representative mitigation strategies across 3 models, 4 tuning strategies and 7 text classification datasets. We show that: 1) DENI outperforms the best performing mitigation strategy (Ensemble), while using only a fraction of its cost; 2) the mitigation strategies are beneficial for parameter-efficient fine-tuning (PEFT) methods, outperforming full fine-tuning in specific cases; and 3) combining DENI with data augmentation often leads to even more effective instability mitigation.
Authors: Qinyuan Cheng, Xiaonan Li, Shimin Li, Qin Zhu, Zhangyue Yin, Yunfan Shao, Linyang Li, Tianxiang Sun, Hang Yan, Xipeng Qiu
Abstract: In Retrieval-Augmented Generation (RAG), retrieval is not always helpful and applying it to every instruction is sub-optimal. Therefore, determining whether to retrieve is crucial for RAG, which is usually referred to as Active Retrieval. However, existing active retrieval methods face two challenges: 1. They usually rely on a single criterion, which struggles with handling various types of instructions. 2. They depend on specialized and highly differentiated procedures, and thus combining them makes the RAG system more complicated and leads to higher response latency. To address these challenges, we propose Unified Active Retrieval (UAR). UAR contains four orthogonal criteria and casts them into plug-and-play classification tasks, which achieves multifaceted retrieval timing judgements with negligible extra inference cost. We further introduce the Unified Active Retrieval Criteria (UAR-Criteria), designed to process diverse active retrieval scenarios through a standardized procedure. Experiments on four representative types of user instructions show that UAR significantly outperforms existing work on the retrieval timing judgement and the performance of downstream tasks, which shows the effectiveness of UAR and its helpfulness to downstream tasks.
Authors: Zhipeng Chen, Kun Zhou, Wayne Xin Zhao, Jingyuan Wang, Ji-Rong Wen
Abstract: Large language models (LLMs) are still struggling in aligning with human preference in complex tasks and scenarios. They are prone to overfit into the unexpected patterns or superficial styles in the training data. We conduct an empirical study that only selects the top-10\% most updated parameters in LLMs for alignment training, and see improvements in the convergence process and final performance. It indicates the existence of redundant neurons in LLMs for alignment training. To reduce its influence, we propose a low-redundant alignment method named \textbf{ALLO}, focusing on optimizing the most related neurons with the most useful supervised signals. Concretely, we first identify the neurons that are related to the human preference data by a gradient-based strategy, then identify the alignment-related key tokens by reward models for computing loss. Besides, we also decompose the alignment process into the forgetting and learning stages, where we first forget the tokens with unaligned knowledge and then learn aligned knowledge, by updating different ratios of neurons, respectively. Experimental results on 10 datasets have shown the effectiveness of ALLO. Our code and data are available at \url{https://github.com/RUCAIBox/ALLO}.
Authors: Yidan Sun, Jianfei Yu, Boyang Li
Abstract: Story video-text alignment, a core task in computational story understanding, aims to align video clips with corresponding sentences in their descriptions. However, progress on the task has been held back by the scarcity of manually annotated video-text correspondence and the heavy concentration on English narrations of Hollywood movies. To address these issues, in this paper, we construct a large-scale multilingual video story dataset named Multilingual Synopses of Movie Narratives (M-SYMON), containing 13,166 movie summary videos from 7 languages, as well as manual annotation of fine-grained video-text correspondences for 101.5 hours of video. Training on the human annotated data from SyMoN outperforms the SOTA methods by 15.7 and 16.2 percentage points on Clip Accuracy and Sentence IoU scores, respectively, demonstrating the effectiveness of the annotations. As benchmarks for future research, we create 6 baseline approaches with different multilingual training strategies, compare their performance in both intra-lingual and cross-lingual setups, exemplifying the challenges of multilingual video-text alignment. The dataset is released at: https://github.com/insundaycathy/M-SyMoN
Authors: Jind\v{r}ich Libovick\'y, Jind\v{r}ich Helcl
Abstract: We present three innovations in tokenization and subword segmentation. First, we propose to use unsupervised morphological analysis with Morfessor as pre-tokenization. Second, we present an algebraic method for obtaining subword embeddings grounded in a word embedding space. Based on that, we design a novel subword segmentation algorithm that uses the embeddings, ensuring that the procedure considers lexical meaning. Third, we introduce an efficient segmentation algorithm based on a subword bigram model that can be initialized with the lexically aware segmentation method to avoid using Morfessor and large embedding tables at inference time. We evaluate the proposed approaches using two intrinsic metrics and measure their performance on two downstream tasks: part-of-speech tagging and machine translation. Our experiments show significant improvements in the morphological plausibility of the segmentation when evaluated using segmentation precision on morpheme boundaries and improved R\'enyi efficiency in 8 languages. Although the proposed tokenization methods do not have a large impact on automatic translation quality, we observe consistent performance gains in the arguably more morphological task of part-of-speech tagging.
Authors: Jirui Qi, Gabriele Sarti, Raquel Fern\'andez, Arianna Bisazza
Abstract: Ensuring the verifiability of model answers is a fundamental challenge for retrieval-augmented generation (RAG) in the question answering (QA) domain. Recently, self-citation prompting was proposed to make large language models (LLMs) generate citations to supporting documents along with their answers. However, self-citing LLMs often struggle to match the required format, refer to non-existent sources, and fail to faithfully reflect LLMs' context usage throughout the generation. In this work, we present MIRAGE --Model Internals-based RAG Explanations -- a plug-and-play approach using model internals for faithful answer attribution in RAG applications. MIRAGE detects context-sensitive answer tokens and pairs them with retrieved documents contributing to their prediction via saliency methods. We evaluate our proposed approach on a multilingual extractive QA dataset, finding high agreement with human answer attribution. On open-ended QA, MIRAGE achieves citation quality and efficiency comparable to self-citation while also allowing for a finer-grained control of attribution parameters. Our qualitative evaluation highlights the faithfulness of MIRAGE's attributions and underscores the promising application of model internals for RAG answer attribution.
Authors: Jiangshu Du, Yibo Wang, Wenting Zhao, Zhongfen Deng, Shuaiqi Liu, Renze Lou, Henry Peng Zou, Pranav Narayanan Venkit, Nan Zhang, Mukund Srinath, Haoran Ranran Zhang, Vipul Gupta, Yinghui Li, Tao Li, Fei Wang, Qin Liu, Tianlin Liu, Pengzhi Gao, Congying Xia, Chen Xing, Jiayang Cheng, Zhaowei Wang, Ying Su, Raj Sanjay Shah, Ruohao Guo, Jing Gu, Haoran Li, Kangda Wei, Zihao Wang, Lu Cheng, Surangika Ranathunga, Meng Fang, Jie Fu, Fei Liu, Ruihong Huang, Eduardo Blanco, Yixin Cao, Rui Zhang, Philip S. Yu, Wenpeng Yin
Abstract: This work is motivated by two key trends. On one hand, large language models (LLMs) have shown remarkable versatility in various generative tasks such as writing, drawing, and question answering, significantly reducing the time required for many routine tasks. On the other hand, researchers, whose work is not only time-consuming but also highly expertise-demanding, face increasing challenges as they have to spend more time reading, writing, and reviewing papers. This raises the question: how can LLMs potentially assist researchers in alleviating their heavy workload? This study focuses on the topic of LLMs assist NLP Researchers, particularly examining the effectiveness of LLM in assisting paper (meta-)reviewing and its recognizability. To address this, we constructed the ReviewCritique dataset, which includes two types of information: (i) NLP papers (initial submissions rather than camera-ready) with both human-written and LLM-generated reviews, and (ii) each review comes with "deficiency" labels and corresponding explanations for individual segments, annotated by experts. Using ReviewCritique, this study explores two threads of research questions: (i) "LLMs as Reviewers", how do reviews generated by LLMs compare with those written by humans in terms of quality and distinguishability? (ii) "LLMs as Metareviewers", how effectively can LLMs identify potential issues, such as Deficient or unprofessional review segments, within individual paper reviews? To our knowledge, this is the first work to provide such a comprehensive analysis.
Authors: Markus Frohmann, Igor Sterner, Ivan Vuli\'c, Benjamin Minixhofer, Markus Schedl
Abstract: Segmenting text into sentences plays an early and crucial role in many NLP systems. This is commonly achieved by using rule-based or statistical methods relying on lexical features such as punctuation. Although some recent works no longer exclusively rely on punctuation, we find that no prior method achieves all of (i) robustness to missing punctuation, (ii) effective adaptability to new domains, and (iii) high efficiency. We introduce a new model - Segment any Text (SaT) - to solve this problem. To enhance robustness, we propose a new pretraining scheme that ensures less reliance on punctuation. To address adaptability, we introduce an extra stage of parameter-efficient fine-tuning, establishing state-of-the-art performance in distinct domains such as verses from lyrics and legal documents. Along the way, we introduce architectural modifications that result in a threefold gain in speed over the previous state of the art and solve spurious reliance on context far in the future. Finally, we introduce a variant of our model with fine-tuning on a diverse, multilingual mixture of sentence-segmented data, acting as a drop-in replacement and enhancement for existing segmentation tools. Overall, our contributions provide a universal approach for segmenting any text. Our method outperforms all baselines - including strong LLMs - across 8 corpora spanning diverse domains and languages, especially in practically relevant situations where text is poorly formatted. Our models and code, including documentation, are available at https://github.com/segment-any-text/wtpsplit under the MIT license.
Authors: Minzheng Wang, Longze Chen, Cheng Fu, Shengyi Liao, Xinghua Zhang, Bingli Wu, Haiyang Yu, Nan Xu, Lei Zhang, Run Luo, Yunshui Li, Min Yang, Fei Huang, Yongbin Li
Abstract: Long-context modeling capabilities have garnered widespread attention, leading to the emergence of Large Language Models (LLMs) with ultra-context windows. Meanwhile, benchmarks for evaluating long-context LLMs are gradually catching up. However, existing benchmarks employ irrelevant noise texts to artificially extend the length of test cases, diverging from the real-world scenarios of long-context applications. To bridge this gap, we propose a novel long-context benchmark, Loong, aligning with realistic scenarios through extended multi-document question answering (QA). Unlike typical document QA, in Loong's test cases, each document is relevant to the final answer, ignoring any document will lead to the failure of the answer. Furthermore, Loong introduces four types of tasks with a range of context lengths: Spotlight Locating, Comparison, Clustering, and Chain of Reasoning, to facilitate a more realistic and comprehensive evaluation of long-context understanding. Extensive experiments indicate that existing long-context language models still exhibit considerable potential for enhancement. Retrieval augmented generation (RAG) achieves poor performance, demonstrating that Loong can reliably assess the model's long-context modeling capabilities.
Authors: Hyunjong Ok, Jegwang Ryu, Jaeho Lee
Abstract: How can small-scale large language models (LLMs) efficiently utilize the supervision of LLMs to improve their generative quality? This question has been well studied in scenarios where there is no restriction on the number of LLM supervisions one can use, giving birth to many decoding algorithms that utilize supervision without further training. However, it is still unclear what is an effective strategy under the $\textit{limited supervision}$ scenario, where we assume that no more than a few tokens can be generated by LLMs. To this end, we develop an algorithm to effectively aggregate the small-scale LLM and LLM predictions on initial tokens so that the generated tokens can more accurately condition the subsequent token generation by small-scale LLM only. Critically, we find that it is essential to adaptively overtrust or disregard the LLM prediction based on the confidence of the small-scale LLM. Through our experiments on a wide range of models and datasets, we demonstrate that our method provides a consistent improvement over conventional decoding strategies. $\small$ $\textbf{Code:}$ https://github.com/HJ-Ok/DecLimSup
Authors: Akshay Chaturvedi, Kate Thompson, Nicholas Asher
Abstract: When engaging in collaborative tasks, humans efficiently exploit the semantic structure of a conversation to optimize verbal and nonverbal interactions. But in recent "language to code" or "language to action" models, this information is lacking. We show how incorporating the prior discourse and nonlinguistic context of a conversation situated in a nonlinguistic environment can improve the "language to action" component of such interactions. We finetune an LLM to predict actions based on prior context; our model, Nebula, doubles the net-action F1 score over the baseline on this task of Jayannavar et al.(2020). We also investigate our model's ability to construct shapes and understand location descriptions using a synthetic dataset
Authors: Kate Thompson, Akshay Chaturvedi, Julie Hunter, Nicholas Asher
Abstract: This paper provides the first discourse parsing experiments with a large language model(LLM) finetuned on corpora annotated in the style of SDRT (Segmented Discourse Representation Theory Asher, 1993; Asher and Lascarides, 2003). The result is a discourse parser, Llamipa (Llama Incremental Parser), that leverages discourse context, leading to substantial performance gains over approaches that use encoder-only models to provide local, context-sensitive representations of discourse units. Furthermore, it can process discourse data incrementally, which is essential for the eventual use of discourse information in downstream tasks.
Authors: Xuanming Zhang, Anthony Diaz, Zixun Chen, Qingyang Wu, Kun Qian, Erik Voss, Zhou Yu
Abstract: Coherence in writing, an aspect that second-language (L2) English learners often struggle with, is crucial in assessing L2 English writing. Existing automated writing evaluation systems primarily use basic surface linguistic features to detect coherence in writing. However, little effort has been made to correct the detected incoherence, which could significantly benefit L2 language learners seeking to improve their writing. To bridge this gap, we introduce DECOR, a novel benchmark that includes expert annotations for detecting incoherence in L2 English writing, identifying the underlying reasons, and rewriting the incoherent sentences. To our knowledge, DECOR is the first coherence assessment dataset specifically designed for improving L2 English writing, featuring pairs of original incoherent sentences alongside their expert-rewritten counterparts. Additionally, we fine-tuned models to automatically detect and rewrite incoherence in student essays. We find that incorporating specific reasons for incoherence during fine-tuning consistently improves the quality of the rewrites, achieving a result that is favored in both automatic and human evaluations.
Authors: Xinyi Chen, Baohao Liao, Jirui Qi, Panagiotis Eustratiadis, Christof Monz, Arianna Bisazza, Maarten de Rijke
Abstract: Following multiple instructions is a crucial ability for large language models (LLMs). Evaluating this ability comes with significant challenges: (i) limited coherence between multiple instructions, (ii) positional bias where the order of instructions affects model performance, and (iii) a lack of objectively verifiable tasks. To address these issues, we introduce a benchmark designed to evaluate models' abilities to follow multiple instructions through sequential instruction following (SIFo) tasks. In SIFo, the successful completion of multiple instructions is verifiable by examining only the final instruction. Our benchmark evaluates instruction following using four tasks (text modification, question answering, mathematics, and security rules), each assessing different aspects of sequential instruction following. Our evaluation of popular LLMs, both closed-source and open-source, shows that more recent and larger models significantly outperform their older and smaller counterparts on the SIFo tasks, validating the benchmark's effectiveness. All models struggle with following sequences of instructions, hinting at an important lack of robustness of today's language models.
Authors: Sheridan Feucht, David Atkinson, Byron Wallace, David Bau
Abstract: LLMs process text as sequences of tokens that roughly correspond to words, where less common words are represented by multiple tokens. However, individual tokens are often semantically unrelated to the meanings of the words/concepts they comprise. For example, Llama-2-7b's tokenizer splits the word "northeastern" into the tokens ['_n', 'ort', 'he', 'astern'], none of which correspond to semantically meaningful units like "north" or "east." Similarly, the overall meanings of named entities like "Neil Young" and multi-word expressions like "break a leg" cannot be directly inferred from their constituent tokens. Mechanistically, how do LLMs convert such arbitrary groups of tokens into useful higher-level representations? In this work, we find that last token representations of named entities and multi-token words exhibit a pronounced "erasure" effect, where information about previous and current tokens is rapidly forgotten in early layers. Using this observation, we propose a method to "read out" the implicit vocabulary of an autoregressive LLM by examining differences in token representations across layers, and present results of this method for Llama-2-7b and Llama-3-8B. To our knowledge, this is the first attempt to probe the implicit vocabulary of an LLM.
Authors: Qiyuan Zhang, Fuyuan Lyu, Xue Liu, Chen Ma
Abstract: Comprehensively understanding and accurately predicting the performance of large language models across diverse downstream tasks has emerged as a pivotal challenge in NLP research. The pioneering scaling law on downstream works demonstrated intrinsic similarities within model families and utilized such similarities for performance prediction. However, they tend to overlook the similarities between model families and only consider design factors listed in the original scaling law. To overcome these limitations, we introduce a novel framework, Collaborative Performance Prediction (CPP), which significantly enhances prediction accuracy by leveraging the historical performance of various models on downstream tasks and other design factors for both model and task. We also collect a collaborative data sourced from online platforms containing both historical performance and additional design factors. With the support of the collaborative data, CPP not only surpasses traditional scaling laws in predicting the performance of scaled LLMs but also facilitates a detailed analysis of factor importance, an area previously overlooked.
Authors: Guojun Wu, Shay B. Cohen, Rico Sennrich
Abstract: We introduce a dataset comprising commercial machine translations, gathered weekly over six years across 12 translation directions. Since human A/B testing is commonly used, we assume commercial systems improve over time, which enables us to evaluate machine translation (MT) metrics based on their preference for more recent translations. Our study not only confirms several prior findings, such as the advantage of neural metrics over non-neural ones, but also explores the debated issue of how MT quality affects metric reliability--an investigation that smaller datasets in previous research could not sufficiently explore. Overall, our research demonstrates the dataset's value as a testbed for metric evaluation. We release our code at https://github.com/gjwubyron/Evo
Authors: Md Tahmid Rahman Laskar, Sawsan Alqahtani, M Saiful Bari, Mizanur Rahman, Mohammad Abdullah Matin Khan, Haidar Khan, Israt Jahan, Amran Bhuiyan, Chee Wei Tan, Md Rizwan Parvez, Enamul Hoque, Shafiq Joty, Jimmy Huang
Abstract: Large Language Models (LLMs) have recently gained significant attention due to their remarkable capabilities in performing diverse tasks across various domains. However, a thorough evaluation of these models is crucial before deploying them in real-world applications to ensure they produce reliable performance. Despite the well-established importance of evaluating LLMs in the community, the complexity of the evaluation process has led to varied evaluation setups, causing inconsistencies in findings and interpretations. To address this, we systematically review the primary challenges and limitations causing these inconsistencies and unreliable evaluations in various steps of LLM evaluation. Based on our critical review, we present our perspectives and recommendations to ensure LLM evaluations are reproducible, reliable, and robust.
Authors: Samee Arif, Abdul Hameed Azeemi, Agha Ali Raza, Awais Athar
Abstract: In this paper, we compare general-purpose models, GPT-4-Turbo and Llama-3-8b, with special-purpose models--XLM-Roberta-large, mT5-large, and Llama-3-8b--that have been fine-tuned on specific tasks. We focus on seven classification and seven generation tasks to evaluate the performance of these models on Urdu language. Urdu has 70 million native speakers, yet it remains underrepresented in Natural Language Processing (NLP). Despite the frequent advancements in Large Language Models (LLMs), their performance in low-resource languages, including Urdu, still needs to be explored. We also conduct a human evaluation for the generation tasks and compare the results with the evaluations performed by GPT-4-Turbo, Llama-3-8b and Claude 3.5 Sonnet. We find that special-purpose models consistently outperform general-purpose models across various tasks. We also find that the evaluation done by GPT-4-Turbo for generation tasks aligns more closely with human evaluation compared to the evaluation the evaluation done by Llama-3-8b. This paper contributes to the NLP community by providing insights into the effectiveness of general and specific-purpose LLMs for low-resource languages.
Authors: Xiao Yu, Qingyang Wu, Yu Li, Zhou Yu
Abstract: Alignment is a crucial step to enhance the instruction-following and conversational abilities of language models. Despite many recent work proposing new algorithms, datasets, and training pipelines, there is a lack of comprehensive studies measuring the impact of various design choices throughout the whole training process. We first conduct a rigorous analysis over a three-stage training pipeline consisting of supervised fine-tuning, offline preference learning, and online preference learning. We have found that using techniques like sequence packing, loss masking in SFT, increasing the preference dataset size in DPO, and online DPO training can significantly improve the performance of language models. We then train from Gemma-2b-base and LLama-3-8b-base, and find that our best models exceed the performance of the official instruct models tuned with closed-source data and algorithms. Our code and models can be found at \url{https://github.com/Columbia-NLP-Lab/LionAlignment}.
Authors: Yung-Sung Chuang, Linlu Qiu, Cheng-Yu Hsieh, Ranjay Krishna, Yoon Kim, James Glass
Abstract: When asked to summarize articles or answer questions given a passage, large language models (LLMs) can hallucinate details and respond with unsubstantiated answers that are inaccurate with respect to the input context. This paper describes a simple approach for detecting such contextual hallucinations. We hypothesize that contextual hallucinations are related to the extent to which an LLM attends to information in the provided context versus its own generations. Based on this intuition, we propose a simple hallucination detection model whose input features are given by the ratio of attention weights on the context versus newly generated tokens (for each attention head). We find that a linear classifier based on these lookback ratio features is as effective as a richer detector that utilizes the entire hidden states of an LLM or a text-based entailment model. The lookback ratio-based detector -- Lookback Lens -- is found to transfer across tasks and even models, allowing a detector that is trained on a 7B model to be applied (without retraining) to a larger 13B model. We further apply this detector to mitigate contextual hallucinations, and find that a simple classifier-guided decoding approach is able to reduce the amount of hallucination, for example by 9.6% in the XSum summarization task.
Authors: Alexandre Matton, Tom Sherborne, Dennis Aumiller, Elena Tommasone, Milad Alizadeh, Jingyi He, Raymond Ma, Maxime Voisin, Ellen Gilsenan-McMahon, Matthias Gall\'e
Abstract: In this paper, we consider contamination by code generation test sets, in particular in their use in modern large language models. We discuss three possible sources of such contamination and show findings supporting each of them: (i) direct data leakage, (ii) indirect data leakage through the use of synthetic data and (iii) overfitting to evaluation sets during model selection. To address this, we release Less Basic Python Problems (LBPP): an uncontaminated new benchmark of 161 prompts with their associated Python solutions. LBPP is released at https://huggingface.co/datasets/CohereForAI/lbpp .
Authors: Kaitlyn Zhou, Jena D. Hwang, Xiang Ren, Nouha Dziri, Dan Jurafsky, Maarten Sap
Abstract: The ability to communicate uncertainty, risk, and limitation is crucial for the safety of large language models. However, current evaluations of these abilities rely on simple calibration, asking whether the language generated by the model matches appropriate probabilities. Instead, evaluation of this aspect of LLM communication should focus on the behaviors of their human interlocutors: how much do they rely on what the LLM says? Here we introduce an interaction-centered evaluation framework called Rel-A.I. (pronounced "rely"}) that measures whether humans rely on LLM generations. We use this framework to study how reliance is affected by contextual features of the interaction (e.g, the knowledge domain that is being discussed), or the use of greetings communicating warmth or competence (e.g., "I'm happy to help!"). We find that contextual characteristics significantly affect human reliance behavior. For example, people rely 10% more on LMs when responding to questions involving calculations and rely 30% more on LMs that are perceived as more competent. Our results show that calibration and language quality alone are insufficient in evaluating the risks of human-LM interactions, and illustrate the need to consider features of the interactional context.
Authors: Maksym Andriushchenko, Nicolas Flammarion
Abstract: Refusal training is widely used to prevent LLMs from generating harmful, undesirable, or illegal outputs. We reveal a curious generalization gap in the current refusal training approaches: simply reformulating a harmful request in the past tense (e.g., "How to make a Molotov cocktail?" to "How did people make a Molotov cocktail?") is often sufficient to jailbreak many state-of-the-art LLMs. We systematically evaluate this method on Llama-3 8B, Claude-3.5 Sonnet, GPT-3.5 Turbo, Gemma-2 9B, Phi-3-Mini, GPT-4o mini, GPT-4o, o1-mini, o1-preview, and R2D2 models using GPT-3.5 Turbo as a reformulation model. For example, the success rate of this simple attack on GPT-4o increases from 1% using direct requests to 88% using 20 past tense reformulation attempts on harmful requests from JailbreakBench with GPT-4 as a jailbreak judge. Interestingly, we also find that reformulations in the future tense are less effective, suggesting that refusal guardrails tend to consider past historical questions more benign than hypothetical future questions. Moreover, our experiments on fine-tuning GPT-3.5 Turbo show that defending against past reformulations is feasible when past tense examples are explicitly included in the fine-tuning data. Overall, our findings highlight that the widely used alignment techniques -- such as SFT, RLHF, and adversarial training -- employed to align the studied models can be brittle and do not always generalize as intended. We provide code and jailbreak artifacts at https://github.com/tml-epfl/llm-past-tense.
Authors: Arda Y\"uksel, Abdullatif K\"oksal, L\"utfi Kerem \c{S}enel, Anna Korhonen, Hinrich Sch\"utze
Abstract: Multiple choice question answering tasks evaluate the reasoning, comprehension, and mathematical abilities of Large Language Models (LLMs). While existing benchmarks employ automatic translation for multilingual evaluation, this approach is error-prone and potentially introduces culturally biased questions, especially in social sciences. We introduce the first multitask, multiple-choice Turkish QA benchmark, TurkishMMLU, to evaluate LLMs' understanding of the Turkish language. TurkishMMLU includes over 10,000 questions, covering 9 different subjects from Turkish high-school education curricula. These questions are written by curriculum experts, suitable for the high-school curricula in Turkey, covering subjects ranging from natural sciences and math questions to more culturally representative topics such as Turkish Literature and the history of the Turkish Republic. We evaluate over 20 LLMs, including multilingual open-source (e.g., Gemma, Llama, MT5), closed-source (GPT 4o, Claude, Gemini), and Turkish-adapted (e.g., Trendyol) models. We provide an extensive evaluation, including zero-shot and few-shot evaluation of LLMs, chain-of-thought reasoning, and question difficulty analysis along with model performance. We provide an in-depth analysis of the Turkish capabilities and limitations of current LLMs to provide insights for future LLMs for the Turkish language. We publicly release our code for the dataset and evaluation: https://github.com/ArdaYueksel/TurkishMMLU.
Authors: Rujun Han, Yuhao Zhang, Peng Qi, Yumo Xu, Jenyuan Wang, Lan Liu, William Yang Wang, Bonan Min, Vittorio Castelli
Abstract: Question answering based on retrieval augmented generation (RAG-QA) is an important research topic in NLP and has a wide range of real-world applications. However, most existing datasets for this task are either constructed using a single source corpus or consist of short extractive answers, which fall short of evaluating large language model (LLM) based RAG-QA systems on cross-domain generalization. To address these limitations, we create Long-form RobustQA (LFRQA), a new dataset comprising human-written long-form answers that integrate short extractive answers from multiple documents into a single, coherent narrative, covering 26K queries and large corpora across seven different domains. We further propose RAG-QA Arena by directly comparing model-generated answers against LFRQA's answers using LLMs as evaluators. We show via extensive experiments that RAG-QA Arena and human judgments on answer quality are highly correlated. Moreover, only 41.3% of the most competitive LLM's answers are preferred to LFRQA's answers, demonstrating RAG-QA Arena as a challenging evaluation platform for future research.
Authors: Xinyi Wang, Antonis Antoniades, Yanai Elazar, Alfonso Amayuelas, Alon Albalak, Kexun Zhang, William Yang Wang
Abstract: The impressive capabilities of large language models (LLMs) have sparked debate over whether these models genuinely generalize to unseen tasks or predominantly rely on memorizing vast amounts of pretraining data. To explore this issue, we introduce an extended concept of memorization, distributional memorization, which measures the correlation between the LLM output probabilities and the pretraining data frequency. To effectively capture task-specific pretraining data frequency, we propose a novel task-gram language model, which is built by counting the co-occurrence of semantically related $n$-gram pairs from task inputs and outputs in the pretraining corpus. Using the Pythia models trained on the Pile dataset, we evaluate three distinct tasks: machine translation, factual question answering, and reasoning. Our findings reveal varying levels of memorization, with the strongest effect observed in factual question answering. Furthermore, while model performance improves across all tasks as LLM size increases, only factual question answering shows an increase in memorization, whereas machine translation and reasoning tasks exhibit greater generalization, producing more novel outputs. This study demonstrates that memorization plays a larger role in simpler, knowledge-intensive tasks, while generalization is the key for harder, reasoning-based tasks, providing a scalable method for analyzing large pretraining corpora in greater depth.
Authors: He Bai, Tatiana Likhomanenko, Ruixiang Zhang, Zijin Gu, Zakaria Aldeneh, Navdeep Jaitly
Abstract: Large language models have revolutionized natural language processing by leveraging self-supervised pretraining on vast textual data. Inspired by this success, researchers have investigated complicated speech tokenization methods to discretize continuous speech signals so that language modeling techniques can be applied to speech data. However, existing approaches either model semantic (content) tokens, potentially losing acoustic information, or model acoustic tokens, risking the loss of semantic (content) information. Having multiple token types also complicates the architecture and requires additional pretraining. Here we show that discretizing mel-filterbank channels into discrete intensity bins produces a simple representation (dMel), that performs better than other existing speech tokenization methods. Using an LM-style transformer architecture for speech-text modeling, we comprehensively evaluate different speech tokenization methods on speech recognition (ASR) and speech synthesis (TTS). Our results demonstrate the effectiveness of dMel in achieving high performance on both tasks within a unified framework, paving the way for efficient and effective joint modeling of speech and text.
Authors: Yuhui Xu, Zhanming Jie, Hanze Dong, Lei Wang, Xudong Lu, Aojun Zhou, Amrita Saha, Caiming Xiong, Doyen Sahoo
Abstract: Large Language Models (LLMs) have revolutionized the field of natural language processing, achieving unprecedented performance across a variety of applications. However, their increased computational and memory demands present significant challenges, especially when handling long sequences. This paper focuses on the long-context scenario, addressing the inefficiencies in KV cache memory consumption during inference. Unlike existing approaches that optimize the memory based on the sequence length, we identify substantial redundancy in the channel dimension of the KV cache, as indicated by an uneven magnitude distribution and a low-rank structure in the attention weights. In response, we propose ThinK, a novel query-dependent KV cache pruning method designed to minimize attention weight loss while selectively pruning the least significant channels. Our approach not only maintains or enhances model accuracy but also achieves a reduction in KV cache memory costs by over 20% compared with vanilla KV cache eviction and quantization methods. For instance, ThinK integrated with KIVI can achieve a 2.8x reduction in peak memory usage while maintaining nearly the same quality, enabling up to a 5x increase in batch size when using a single GPU. Extensive evaluations on the LLaMA and Mistral models across various long-sequence datasets verified the efficiency of ThinK, establishing a new baseline algorithm for efficient LLM deployment without compromising performance.
Authors: Takashi Furuya, Maarten V. de Hoop, Gabriel Peyr\'e
Abstract: Transformers are deep architectures that define "in-context mappings" which enable predicting new tokens based on a given set of tokens (such as a prompt in NLP applications or a set of patches for a vision transformer). In this work, we study in particular the ability of these architectures to handle an arbitrarily large number of context tokens. To mathematically, uniformly address their expressivity, we consider the case that the mappings are conditioned on a context represented by a probability distribution of tokens which becomes discrete for a finite number of these. The relevant notion of smoothness then corresponds to continuity in terms of the Wasserstein distance between these contexts. We demonstrate that deep transformers are universal and can approximate continuous in-context mappings to arbitrary precision, uniformly over compact token domains. A key aspect of our results, compared to existing findings, is that for a fixed precision, a single transformer can operate on an arbitrary (even infinite) number of tokens. Additionally, it operates with a fixed embedding dimension of tokens (this dimension does not increase with precision) and a fixed number of heads (proportional to the dimension). The use of MLPs between multi-head attention layers is also explicitly controlled. We consider both unmasked attentions (as used for the vision transformer) and masked causal attentions (as used for NLP and time series applications). We tackle the causal setting leveraging a space-time lifting to analyze causal attention as a mapping over probability distributions of tokens.
Authors: Zhihu Wang, Shiwan Zhao, Yu Wang, Heyuan Huang, Sitao Xie, Yubo Zhang, Jiaxin Shi, Zhixing Wang, Hongyan Li, Junchi Yan
Abstract: The Chain-of-Thought (CoT) paradigm has become a pivotal method for solving complex problems. However, its application to intricate, domain-specific tasks remains challenging, as large language models (LLMs) often struggle to accurately decompose these tasks and, even when decomposition is correct, fail to execute the subtasks effectively. This paper introduces the Re-TASK framework, a novel theoretical model that revisits LLM tasks from the perspectives of capability, skill, and knowledge, drawing on the principles of Bloom's Taxonomy and Knowledge Space Theory. While CoT offers a workflow perspective on tasks, the Re-TASK framework introduces a Chain-of-Learning view, illustrating how tasks and their corresponding subtasks depend on various capability items. Each capability item is further dissected into its constituent aspects of knowledge and skills. Our framework reveals that many CoT failures in domain-specific tasks stem from insufficient knowledge or inadequate skill adaptation. In response, we combine CoT with the Re-TASK framework and implement a carefully designed Re-TASK prompting strategy to improve task performance. Specifically, we identify core capability items linked to tasks and subtasks, then strengthen these capabilities through targeted knowledge injection and skill adaptation. We validate the Re-TASK framework on three datasets across the law, finance, and mathematics domains, achieving significant improvements over the baseline models. Notably, our approach yields a remarkable 44.42% improvement with the Yi-1.5-9B model and a 33.08% improvement with the Llama3-Chinese-8b on the legal dataset. These experimental results confirm the effectiveness of the Re-TASK framework, demonstrating substantial enhancements in both the performance and applicability of LLMs.
Authors: KaShun Shum, Minrui Xu, Jianshu Zhang, Zixin Chen, Shizhe Diao, Hanze Dong, Jipeng Zhang, Muhammad Omer Raza
Abstract: Large language models (LLMs) have become increasingly prevalent in our daily lives, leading to an expectation for LLMs to be trustworthy -- - both accurate and well-calibrated (the prediction confidence should align with its ground truth correctness likelihood). Nowadays, fine-tuning has become the most popular method for adapting a model to practical usage by significantly increasing accuracy on downstream tasks. Despite the great accuracy it achieves, we found fine-tuning is still far away from satisfactory trustworthiness due to "tuning-induced mis-calibration". In this paper, we delve deeply into why and how mis-calibration exists in fine-tuned models, and how distillation can alleviate the issue. Then we further propose a brand new method named Efficient Trustworthy Distillation (FIRST), which utilizes a small portion of teacher's knowledge to obtain a reliable language model in a cost-efficient way. Specifically, we identify the "concentrated knowledge" phenomenon during distillation, which can significantly reduce the computational burden. Then we apply a "trustworthy maximization" process to optimize the utilization of this small portion of concentrated knowledge before transferring it to the student. Experimental results demonstrate the effectiveness of our method, where better accuracy (+2.3%) and less mis-calibration (-10%) are achieved on average across both in-domain and out-of-domain scenarios, indicating better trustworthiness.
Authors: Xidong Wang, Dingjie Song, Shunian Chen, Chen Zhang, Benyou Wang
Abstract: Expanding the long-context capabilities of Multi-modal Large Language Models~(MLLMs) is crucial for video understanding, high-resolution image understanding, and multi-modal agents. This involves a series of systematic optimizations, including model architecture, data construction and training strategy, particularly addressing challenges such as \textit{degraded performance with more images} and \textit{high computational costs}. In this paper, we adapt the model architecture to a hybrid of Mamba and Transformer blocks, approach data construction with both temporal and spatial dependencies among multiple images and employ a progressive training strategy. The released model \textbf{LongLLaVA}~(\textbf{Long}-Context \textbf{L}arge \textbf{L}anguage \textbf{a}nd \textbf{V}ision \textbf{A}ssistant) is the first hybrid MLLM, which achieved a better balance between efficiency and effectiveness. LongLLaVA not only achieves competitive results across various benchmarks, but also maintains high throughput and low memory consumption. Especially, it could process nearly a thousand images on a single A100 80GB GPU, showing promising application prospects for a wide range of tasks.
Authors: Neeladri Bhuiya, Viktor Schlegel, Stefan Winkler
Abstract: State-of-the-art Large Language Models (LLMs) are accredited with an increasing number of different capabilities, ranging from reading comprehension, over advanced mathematical and reasoning skills to possessing scientific knowledge. In this paper we focus on their multi-hop reasoning capability: the ability to identify and integrate information from multiple textual sources. Given the concerns with the presence of simplifying cues in existing multi-hop reasoning benchmarks, which allow models to circumvent the reasoning requirement, we set out to investigate, whether LLMs are prone to exploiting such simplifying cues. We find evidence that they indeed circumvent the requirement to perform multi-hop reasoning, but they do so in more subtle ways than what was reported about their fine-tuned pre-trained language model (PLM) predecessors. Motivated by this finding, we propose a challenging multi-hop reasoning benchmark, by generating seemingly plausible multi-hop reasoning chains, which ultimately lead to incorrect answers. We evaluate multiple open and proprietary state-of-the-art LLMs, and find that their performance to perform multi-hop reasoning is affected, as indicated by up to 45% relative decrease in F1 score when presented with such seemingly plausible alternatives. We conduct a deeper analysis and find evidence that while LLMs tend to ignore misleading lexical cues, misleading reasoning paths indeed present a significant challenge.
Authors: Riona Matsuoka, Hiroki Matsumoto, Takahiro Yoshida, Tomohiro Watanabe, Ryoma Kondo, Ryohei Hisano
Abstract: Written texts reflect an author's perspective, making the thorough analysis of literature a key research method in fields such as the humanities and social sciences. However, conventional text mining techniques like sentiment analysis and topic modeling are limited in their ability to capture the hierarchical narrative structures that reveal deeper argumentative patterns. To address this gap, we propose a method that leverages large language models (LLMs) to extract and organize these structures into a hierarchical framework. We validate this approach by analyzing public opinions on generative AI collected by Japan's Agency for Cultural Affairs, comparing the narratives of supporters and critics. Our analysis provides clearer visualization of the factors influencing divergent opinions on generative AI, offering deeper insights into the structures of agreement and disagreement.
Authors: Kaige Xie, Ian Yang, John Gunerli, Mark Riedl
Abstract: World models, which encapsulate the dynamics of how actions affect environments, are foundational to the functioning of intelligent agents. In this work, we explore the potential of Large Language Models (LLMs) to operate as world models. Although LLMs are not inherently designed to model real-world dynamics, we show that they can be induced to perform two critical world model functions: determining the applicability of an action based on a given world state, and predicting the resulting world state upon action execution. This is achieved by fine-tuning two separate LLMs-one for precondition prediction and another for effect prediction-while leveraging synthetic data generation techniques. Through human-participant studies, we validate that the precondition and effect knowledge generated by our models aligns with human understanding of world dynamics. We also analyze the extent to which the world model trained on our synthetic data results in an inferred state space that supports the creation of action chains, a necessary property for planning.
Authors: Qinzhuo Wu, Weikai Xu, Wei Liu, Tao Tan, Jianfeng Liu, Ang Li, Jian Luan, Bin Wang, Shuo Shang
Abstract: Recently, mobile AI agents based on VLMs have been gaining increasing attention. These works typically utilize VLM as a foundation, fine-tuning it with instruction-based mobile datasets. However, these VLMs are typically pre-trained on general-domain data, which often results in a lack of fundamental capabilities specific to the mobile domain. Therefore, they may struggle to recognize specific UI elements and understand intra-UI fine-grained information. In addition, the current fine-tuning task focuses on interacting with the most relevant element for the given instruction. These fine-tuned VLMs may still ignore the relationships between UI pages, neglect the roles of elements in page transitions and lack inter-UI understanding. To address issues, we propose a VLM called MobileVLM, which includes two additional pre-training stages to enhance both intra- and inter-UI understanding. We defined four UI-based pre-training tasks, enabling the model to better perceive fine-grained elements and capture page transition actions. To address the lack of mobile pre-training data, we built a large Chinese mobile dataset Mobile3M from scratch, which contains 3 million UI pages, and real-world transition actions, forming a directed graph structure. Experimental results show MobileVLM excels on both our test set and public mobile benchmarks, outperforming existing VLMs.
Authors: Qinzhuo Wu, Wei Liu, Jian Luan, Bin Wang
Abstract: Recently, tool-augmented LLMs have gained increasing attention. Given an instruction, tool-augmented LLMs can interact with various external tools in multiple rounds and provide a final answer. However, previous LLMs were trained on overly detailed instructions, which included API names or parameters, while real users would not explicitly mention these API details. This leads to a gap between trained LLMs and real-world scenarios. In addition, most works ignore whether the interaction process follows the instruction. To address these issues, we constructed a training dataset called MGToolBench, which contains statement and category-level instructions to better reflect real-world scenarios. In addition, we propose ToolPlanner, a two-stage reinforcement learning framework that utilizes path planning and two feedback mechanisms to enhance the LLM's task completion and instruction-following capabilities. Experimental results show that ToolPlanner significantly improves the Match Rate, Pass Rate and Win Rate by 26.8%, 20.2%, and 5.6% compared to the SOTA model. Human evaluation verifies that the multi-granularity instructions can better align with users' usage habits. Our data and code will be released upon acceptance.
Authors: Yixuan Tang, Yi Yang
Abstract: Embedding models play a crucial role in representing and retrieving information across various NLP applications. Recent advancements in Large Language Models (LLMs) have further enhanced the performance of embedding models, which are trained on massive amounts of text covering almost every domain. These models are often benchmarked on general-purpose datasets like Massive Text Embedding Benchmark (MTEB), where they demonstrate superior performance. However, a critical question arises: Is the development of domain-specific embedding models necessary when general-purpose models are trained on vast corpora that already include specialized domain texts? In this paper, we empirically investigate this question, choosing the finance domain as an example. We introduce the Finance Massive Text Embedding Benchmark (FinMTEB), a counterpart to MTEB that consists of financial domain-specific text datasets. We evaluate the performance of seven state-of-the-art embedding models on FinMTEB and observe a significant performance drop compared to their performance on MTEB. To account for the possibility that this drop is driven by FinMTEB's higher complexity, we propose four measures to quantify dataset complexity and control for this factor in our analysis. Our analysis provides compelling evidence that state-of-the-art embedding models struggle to capture domain-specific linguistic and semantic patterns. Moreover, we find that the performance of general-purpose embedding models on MTEB is not correlated with their performance on FinMTEB, indicating the need for domain-specific embedding benchmarks for domain-specific embedding models. This study sheds light on developing domain-specific embedding models in the LLM era. FinMTEB comes with open-source code at https://github.com/yixuantt/FinMTEB
Authors: Praneeth Vadlapati
Abstract: Classification tasks are typically handled using Machine Learning (ML) models, which lack a balance between accuracy and interpretability. This paper introduces a new approach to using Large Language Models (LLMs) for classification tasks in an explainable way. Unlike ML models that rely heavily on data cleaning and feature engineering, this method streamlines the process using LLMs. This paper proposes a new concept called "Language Model Learning (LML)" powered by a new method called "Data-Augmented Prediction (DAP)". The classification is performed by LLMs using a method similar to humans manually exploring and understanding the data and deciding classifications using data as a reference. In the LML process, a dataset is summarized and evaluated to determine the features that lead to the classification of each label the most. In the process of DAP, the system uses the data summary and a row of the testing dataset to automatically generate a query, which is used to retrieve relevant rows from the dataset. A classification is generated by the LLM using data summary and relevant rows, ensuring satisfactory accuracy even with complex data using context-aware decision-making. LML and DAP unlock the possibilities of new applications. The proposed method uses the words "Act as an Explainable Machine Learning Model" in the prompt to enhance the interpretability of the predictions by allowing users to review the logic behind each prediction. In some test cases, the system scored an accuracy above 90%, proving the effectiveness of the system and its potential to outperform conventional ML models in various scenarios. The code is available at https://github.com/Pro-GenAI/LML-DAP
Authors: Jia-Nan Li, Jian Guan, Wei Wu, Zhengtao Yu, Rui Yan
Abstract: Tables are ubiquitous across various domains for concisely representing structured information. Empowering large language models (LLMs) to reason over tabular data represents an actively explored direction. However, since typical LLMs only support one-dimensional~(1D) inputs, existing methods often flatten the two-dimensional~(2D) table structure into a sequence of tokens, which can severely disrupt the spatial relationships and result in an inevitable loss of vital contextual information. In this paper, we first empirically demonstrate the detrimental impact of such flattening operations on the performance of LLMs in capturing the spatial information of tables through two elaborate proxy tasks. Subsequently, we introduce a simple yet effective positional encoding method, termed ``2D-TPE'' (Two-Dimensional Table Positional Encoding), to address this challenge. 2D-TPE enables each attention head to dynamically select a permutation order of tokens within the context for attending to them, where each permutation represents a distinct traversal mode for the table, such as column-wise or row-wise traversal. 2D-TPE effectively mitigates the risk of losing essential spatial information while preserving computational efficiency, thus better preserving the table structure. Extensive experiments across five benchmarks demonstrate that 2D-TPE outperforms strong baselines, underscoring the importance of preserving the table structure for accurate table comprehension. Comprehensive analysis further reveals the substantially better scalability of 2D-TPE to large tables than baselines.
Authors: Linus Tze En Foo, Lynnette Hui Xian Ng
Abstract: Singlish, or formally Colloquial Singapore English, is an English-based creole language originating from the SouthEast Asian country Singapore. The language contains influences from Sinitic languages such as Chinese dialects, Malay, Tamil and so forth. A fundamental task to understanding Singlish is to first understand the pragmatic functions of its discourse particles, upon which Singlish relies heavily to convey meaning. This work offers a preliminary effort to disentangle the Singlish discourse particles (lah, meh and hor) with task-driven representation learning. After disentanglement, we cluster these discourse particles to differentiate their pragmatic functions, and perform Singlish-to-English machine translation. Our work provides a computational method to understanding Singlish discourse particles, and opens avenues towards a deeper comprehension of the language and its usage.
Authors: Xilong Wang, Hao Fu, Jindong Wang, Neil Zhenqiang Gong
Abstract: String processing, which mainly involves the analysis and manipulation of strings, is a fundamental component of modern computing. Despite the significant advancements of large language models (LLMs) in various natural language processing (NLP) tasks, their capability in string processing remains underexplored and underdeveloped. To bridge this gap, we present a comprehensive study of LLMs' string processing capability. In particular, we first propose StringLLM, a method to construct datasets for benchmarking string processing capability of LLMs. We use StringLLM to build a series of datasets, referred to as StringBench. It encompasses a wide range of string processing tasks, allowing us to systematically evaluate LLMs' performance in this area. Our evaluations indicate that LLMs struggle with accurately processing strings compared to humans. To uncover the underlying reasons for this limitation, we conduct an in-depth analysis and subsequently propose an effective approach that significantly enhances LLMs' string processing capability via fine-tuning. This work provides a foundation for future research to understand LLMs' string processing capability. Our code and data are available at https://github.com/wxl-lxw/StringLLM.
Authors: Ameen Ali, Lior Wolf, Ivan Titov
Abstract: Large language models (LLMs) have demonstrated impressive few-shot in-context learning (ICL) abilities. Still, we show that they are sometimes prone to a `copying bias', where they copy answers from provided examples instead of learning the underlying patterns. In this work, we propose a novel and simple method to mitigate such copying bias. First, we create a synthetic task and use the Integrated Gradients method to identify neurons that prioritize copying over generalization. We demonstrate that pruning these neurons consistently improves performance across a diverse set of ICL tasks. We also show that our method is applicable across various LLM architectures, including Transformers and State-Space Models, without requiring modifications. In our analysis, we adopt a task-recognition perspective on ICL and examine task vectors (Hendel et al., 2023) induced by the model. We find that pruning enhances the quality of these vectors, suggesting that the pruned neurons previously hindered effective task recognition.
Authors: Yingsheng Wu, Yuxuan Gu, Xiaocheng Feng, Weihong Zhong, Dongliang Xu, Qing Yang, Hongtao Liu, Bing Qin
Abstract: Scaling the rotary position embedding (RoPE) has become a common method for extending the context window of RoPE-based large language models (LLMs). However, existing scaling methods often rely on empirical approaches and lack a profound understanding of the internal distribution within RoPE, resulting in suboptimal performance in extending the context window length. In this paper, we propose to optimize the context window extending task from the view of rotary angle distribution. Specifically, we first estimate the distribution of the rotary angles within the model and analyze the extent to which length extension perturbs this distribution. Then, we present a novel extension strategy that minimizes the disturbance between rotary angle distributions to maintain consistency with the pre-training phase, enhancing the model's capability to generalize to longer sequences. Experimental results compared to the strong baseline methods demonstrate that our approach reduces by up to 72% of the distributional disturbance when extending LLaMA2's context window to 8k, and reduces by up to 32% when extending to 16k. On the LongBench-E benchmark, our method achieves an average improvement of up to 4.33% over existing state-of-the-art methods. Furthermore, Our method maintains the model's performance on the Hugging Face Open LLM benchmark after context window extension, with only an average performance fluctuation ranging from -0.12 to +0.22.
Authors: Yi Cheng, Xiao Liang, Yeyun Gong, Wen Xiao, Song Wang, Yuji Zhang, Wenjun Hou, Kaishuai Xu, Wenge Liu, Wenjie Li, Jian Jiao, Qi Chen, Peng Cheng, Wayne Xiong
Abstract: Self-consistency-based approaches, which involve repeatedly sampling multiple outputs and selecting the most consistent one as the final response, prove to be remarkably effective in improving the factual accuracy of large language models. Nonetheless, existing methods usually have strict constraints on the task format, largely limiting their applicability. In this paper, we present Integrative Decoding (ID), to unlock the potential of self-consistency in open-ended generation tasks. ID operates by constructing a set of inputs, each prepended with a previously sampled response, and then processes them concurrently, with the next token being selected by aggregating of all their corresponding predictions at each decoding step. In essence, this simple approach implicitly incorporates self-consistency in the decoding objective. Extensive evaluation shows that ID consistently enhances factuality over a wide range of language models, with substantial improvements on the TruthfulQA (+11.2%), Biographies (+15.4%) and LongFact (+8.5%) benchmarks. The performance gains amplify progressively as the number of sampled responses increases, indicating the potential of ID to scale up with repeated sampling.
Authors: Zhenting Qi, Hongyin Luo, Xuliang Huang, Zhuokai Zhao, Yibo Jiang, Xiangjun Fan, Himabindu Lakkaraju, James Glass
Abstract: While large language models (LLMs) have shown exceptional capabilities in understanding complex queries and performing sophisticated tasks, their generalization abilities are often deeply entangled with memorization, necessitating more precise evaluation. To address this challenge, we introduce Scylla, a dynamic evaluation framework that quantitatively measures the generalization abilities of LLMs. Scylla disentangles generalization from memorization via assessing model performance on both in-distribution (ID) and out-of-distribution (OOD) data through 20 tasks across 5 levels of complexity. Through extensive experiments, we uncover a non-monotonic relationship between task complexity and the performance gap between ID and OOD data, which we term the generalization valley. Specifically, this phenomenon reveals a critical threshold - referred to as critical complexity - where reliance on non-generalizable behavior peaks, indicating the upper bound of LLMs' generalization capabilities. As model size increases, the critical complexity shifts toward higher levels of task complexity, suggesting that larger models can handle more complex reasoning tasks before over-relying on memorization. Leveraging Scylla and the concept of critical complexity, we benchmark 28LLMs including both open-sourced models such as LLaMA and Qwen families, and close-sourced models like Claude and GPT, providing a more robust evaluation and establishing a clearer understanding of LLMs' generalization capabilities.
Authors: Keunwoo Peter Yu, Zheyuan Zhang, Fengyuan Hu, Shane Storks, Joyce Chai
Abstract: A major reason behind the recent success of large language models (LLMs) is their \textit{in-context learning} capability, which makes it possible to rapidly adapt them to downstream text-based tasks by prompting them with a small number of relevant demonstrations. While large vision-language models (VLMs) have recently been developed for tasks requiring both text and images, they largely lack in-context learning over visual information, especially in understanding and generating text about videos. In this work, we implement \textbf{E}mergent \textbf{I}n-context \textbf{Le}arning on \textbf{V}ideos (\eilev{}), a novel training paradigm that induces in-context learning over video and text by capturing key properties of pre-training data found by prior work to be essential for in-context learning in transformers. In our experiments, we show that \eilev-trained models outperform other off-the-shelf VLMs in few-shot video narration for novel, rare actions. Furthermore, we demonstrate that these key properties of bursty distributions, skewed marginal distributions, and dynamic meaning each contribute to varying degrees to VLMs' in-context learning capability in narrating procedural videos. Our results, analysis, and \eilev{}-trained models yield numerous insights about the emergence of in-context learning over video and text, creating a foundation for future work to optimize and scale VLMs for open-domain video understanding and reasoning. Our code and demo are available at \url{https://github.com/yukw777/EILEV}.
Authors: David de-Fitero-Dominguez, Eva Garcia-Lopez, Antonio Garcia-Cabot, Jose-Javier Martinez-Herraiz
Abstract: This research addresses the complex challenge of automated repair of code vulnerabilities, vital for enhancing digital security in an increasingly technology-driven world. The study introduces a novel and efficient format for the representation of code modification, using advanced Large Language Models (LLMs) such as Code Llama and Mistral. These models, fine-tuned on datasets featuring C code vulnerabilities, significantly improve the accuracy and adaptability of automated code repair techniques. A key finding is the enhanced repair accuracy of these models when compared to previous methods such as VulRepair, which underscores their practical utility and efficiency. The research also offers a critical assessment of current evaluation metrics, such as perfect predictions, and their limitations in reflecting the true capabilities of automated repair models in real-world scenarios. Following this, it underscores the importance of using test datasets devoid of train samples, emphasizing the need for dataset integrity to enhance the effectiveness of LLMs in code repair tasks. The significance of this work is its contribution to digital security, setting new standards for automated code vulnerability repair and paving the way for future advancements in the fields of cybersecurity and artificial intelligence. The study does not only highlight the potential of LLMs in enhancing code security but also fosters further exploration and research in these crucial areas.
Authors: Tianyang Han, Qing Lian, Rui Pan, Renjie Pi, Jipeng Zhang, Shizhe Diao, Yong Lin, Tong Zhang
Abstract: Large language models (LLMs) have recently experienced remarkable progress, where the advent of multi-modal large language models (MLLMs) has endowed LLMs with visual capabilities, leading to impressive performances in various multi-modal tasks. However, those powerful MLLMs such as GPT-4V still fail spectacularly when presented with certain image and text inputs. In this paper, we identify a typical class of inputs that baffles MLLMs, which consist of images that are highly relevant but inconsistent with answers, causing MLLMs to suffer from visual illusion. To quantify the effect, we propose CorrelationQA, the first benchmark that assesses the visual illusion level given spurious images. This benchmark contains 7,308 text-image pairs across 13 categories. Based on the proposed CorrelationQA, we conduct a thorough analysis on 9 mainstream MLLMs, illustrating that they universally suffer from this instinctive bias to varying degrees. We hope that our curated benchmark and evaluation results aid in better assessments of the MLLMs' robustness in the presence of misleading images. The code and datasets are available at https://github.com/MasaiahHan/CorrelationQA.
Authors: Yuxuan Wang, Yueqian Wang, Pengfei Wu, Jianxin Liang, Dongyan Zhao, Yang Liu, Zilong Zheng
Abstract: Despite progress in multimodal large language models (MLLMs), the challenge of interpreting long-form videos in response to linguistic queries persists, largely due to the inefficiency in temporal grounding and limited pre-trained context window size. In this work, we introduce Temporal Grounding Bridge (TGB), a novel framework that bootstraps MLLMs with advanced temporal grounding capabilities and broadens their contextual scope. Our framework significantly enhances the temporal capabilities of current MLLMs through three key innovations: an efficient multi-span temporal grounding algorithm applied to low-dimension temporal features projected from flow; a multimodal length extrapolation training paradigm that utilizes low-dimension temporal features to extend the training context window size; and a bootstrapping framework that bridges our model with pluggable MLLMs without requiring annotation. We validate TGB across seven video benchmarks and demonstrate substantial performance improvements compared with prior MLLMs. Notably, our model, initially trained on sequences of four frames, effectively handles sequences up to 16 longer without sacrificing performance, highlighting its scalability and effectiveness in real-world applications. Our code is publicly available at https://github.com/bigai-nlco/VideoTGB
Authors: Max Lamparth, Anthony Corso, Jacob Ganz, Oriana Skylar Mastro, Jacquelyn Schneider, Harold Trinkunas
Abstract: To some, the advent of artificial intelligence (AI) promises better decision-making and increased military effectiveness while reducing the influence of human error and emotions. However, there is still debate about how AI systems, especially large language models (LLMs) that can be applied to many tasks, behave compared to humans in high-stakes military decision-making scenarios with the potential for increased risks towards escalation. To test this potential and scrutinize the use of LLMs for such purposes, we use a new wargame experiment with 214 national security experts designed to examine crisis escalation in a fictional U.S.-China scenario and compare the behavior of human player teams to LLM-simulated team responses in separate simulations. Here, we find that the LLM-simulated responses can be more aggressive and significantly affected by changes in the scenario. We show a considerable high-level agreement in the LLM and human responses and significant quantitative and qualitative differences in individual actions and strategic tendencies. These differences depend on intrinsic biases in LLMs regarding the appropriate level of violence following strategic instructions, the choice of LLM, and whether the LLMs are tasked to decide for a team of players directly or first to simulate dialog between a team of players. When simulating the dialog, the discussions lack quality and maintain a farcical harmony. The LLM simulations cannot account for human player characteristics, showing no significant difference even for extreme traits, such as "pacifist" or "aggressive sociopath." When probing behavioral consistency across individual moves of the simulation, the tested LLMs deviated from each other but generally showed somewhat consistent behavior. Our results motivate policymakers to be cautious before granting autonomy or following AI-based strategy recommendations.
Authors: Huaisheng Zhu, Teng Xiao, Vasant G Honavar
Abstract: Generating molecular structures with desired properties is a critical task with broad applications in drug discovery and materials design. We propose 3M-Diffusion, a novel multi-modal molecular graph generation method, to generate diverse, ideally novel molecular structures with desired properties. 3M-Diffusion encodes molecular graphs into a graph latent space which it then aligns with the text space learned by encoder-based LLMs from textual descriptions. It then reconstructs the molecular structure and atomic attributes based on the given text descriptions using the molecule decoder. It then learns a probabilistic mapping from the text space to the latent molecular graph space using a diffusion model. The results of our extensive experiments on several datasets demonstrate that 3M-Diffusion can generate high-quality, novel and diverse molecular graphs that semantically match the textual description provided.
Authors: Azmine Toushik Wasi, Raima Islam, Mst Rafia Islam
Abstract: Individuality and personalization comprise the distinctive characteristics that make each writer unique and influence their words in order to effectively engage readers while conveying authenticity. However, our growing reliance on LLM-based writing assistants risks compromising our creativity and individuality over time. We often overlook the negative impacts of this trend on our creativity and uniqueness, despite the possible consequences. This study investigates these concerns by performing a brief survey to explore different perspectives and concepts, as well as trying to understand people's viewpoints, in conjunction with past studies in the area. Addressing these issues is essential for improving human-computer interaction systems and enhancing writing assistants for personalization and individuality.
Authors: Azmine Toushik Wasi, Mst Rafia Islam, Raima Islam
Abstract: Sense of ownership in writing confines our investment of thoughts, time, and contribution, leading to attachment to the output. However, using writing assistants introduces a mental dilemma, as some content isn't directly our creation. For instance, we tend to credit Large Language Models (LLMs) more in creative tasks, even though all tasks are equal for them. Additionally, while we may not claim complete ownership of LLM-generated content, we freely claim authorship. We conduct a short survey to examine these issues and understand underlying cognitive processes in order to gain a better knowledge of human-computer interaction in writing and improve writing aid systems.
Authors: Yiqing Xie, Alex Xie, Divyanshu Sheth, Pengfei Liu, Daniel Fried, Carolyn Rose
Abstract: To adequately test modern code generation systems, evaluation benchmarks must execute and test the code generated by the system. However, these execution and testing requirements have largely limited benchmarks to settings where code is easily executable or has human-written tests. To facilitate evaluation of code generation systems across diverse scenarios, we present CodeBenchGen, a framework to create scalable execution-based benchmarks from naturally occurring code sources. Specifically, we leverage a large language model (LLM) to sandbox arbitrary pieces of code into evaluation examples, including test cases for execution-based evaluation. We illustrate the usefulness of our framework by creating a dataset, Exec-CSN, which includes 1,931 examples involving 293 libraries converted from code in 367 GitHub repositories taken from the Code- SearchNet dataset. To demonstrate the solvability of examples in Exec-CSN, we present a human study demonstrating that 81.3% of the examples can be solved by humans and 61% are rated as "requires effort to solve". We conduct code generation experiments on open-source and proprietary models and analyze the performance of both humans and models. We provide code and data at: https://github.com/yiqingxyq/CodeBenchGen.
Authors: Pranab Sahoo, Prabhash Meharia, Akash Ghosh, Sriparna Saha, Vinija Jain, Aman Chadha
Abstract: The rapid advancement of foundation models (FMs) across language, image, audio, and video domains has shown remarkable capabilities in diverse tasks. However, the proliferation of FMs brings forth a critical challenge: the potential to generate hallucinated outputs, particularly in high-stakes applications. The tendency of foundation models to produce hallucinated content arguably represents the biggest hindrance to their widespread adoption in real-world scenarios, especially in domains where reliability and accuracy are paramount. This survey paper presents a comprehensive overview of recent developments that aim to identify and mitigate the problem of hallucination in FMs, spanning text, image, video, and audio modalities. By synthesizing recent advancements in detecting and mitigating hallucination across various modalities, the paper aims to provide valuable insights for researchers, developers, and practitioners. Essentially, it establishes a clear framework encompassing definition, taxonomy, and detection strategies for addressing hallucination in multimodal foundation models, laying the foundation for future research in this pivotal area.
Authors: Dominik Wagner, Seanie Lee, Ilja Baumann, Philipp Seeberger, Korbinian Riedhammer, Tobias Bocklet
Abstract: In this work, we optimize speculative sampling for parallel hardware accelerators to improve sampling speed. We notice that substantial portions of the intermediate matrices necessary for speculative sampling can be computed concurrently. This allows us to distribute the workload across multiple GPU threads, enabling simultaneous operations on matrix segments within thread blocks. This results in profiling time improvements ranging from 6% to 13% relative to the baseline implementation, without compromising accuracy. To further accelerate speculative sampling, probability distributions parameterized by softmax are approximated by sigmoid. This approximation approach results in significantly greater relative improvements in profiling time, ranging from 37% to 94%, with a minor decline in accuracy. We conduct extensive experiments on both automatic speech recognition and summarization tasks to validate the effectiveness of our optimization methods.
Authors: Shwai He, Guoheng Sun, Zheyu Shen, Ang Li
Abstract: While scaling Transformer-based large language models (LLMs) has demonstrated promising performance across various tasks, it also introduces redundant architectures, posing efficiency challenges for real-world deployment. Despite some recognition of redundancy in LLMs, the variability of redundancy across different architectures in transformers, such as MLP and Attention layers, is under-explored. In this work, we investigate redundancy across different modules within Transformers, including Blocks, MLP, and Attention layers, using a similarity-based metric. Surprisingly, despite the critical role of attention layers in distinguishing transformers from other architectures, we found that a large portion of these layers exhibit excessively high similarity and can be pruned without degrading performance. For instance, Llama-2-70B achieved a 48.4\% speedup with only a 2.4\% performance drop by pruning half of the attention layers. Furthermore, by tracing model checkpoints throughout the training process, we observed that attention layer redundancy is inherent and consistent across training stages. Additionally, we further propose a method that jointly drops Attention and MLP layers, allowing us to more aggressively drop additional layers. For instance, when dropping 31 layers (Attention + MLP), Llama-2-13B still retains 90\% of the performance on the MMLU task. Our work provides valuable insights for future network architecture design. The code is released at: \url{https://github.com/Shwai-He/LLM-Drop}.
Authors: Yunlong Feng, Dechuan Teng, Yang Xu, Honglin Mu, Xiao Xu, Libo Qin, Qingfu Zhu, Wanxiang Che
Abstract: Decompilation transforms compiled code back into a high-level programming language for analysis when source code is unavailable. Previous work has primarily focused on enhancing decompilation performance by increasing the scale of model parameters or training data for pre-training. Based on the characteristics of the decompilation task, we propose two methods: (1) Without fine-tuning, the Self-Constructed Context Decompilation (sc$^2$dec) method recompiles the LLM's decompilation results to construct pairs for in-context learning, helping the model improve decompilation performance. (2) Fine-grained Alignment Enhancement (FAE), which meticulously aligns assembly code with source code at the statement level by leveraging debugging information, is employed during the fine-tuning phase to achieve further improvements in decompilation. By integrating these two methods, we achieved a Re-Executability performance improvement of approximately 3.90% on the Decompile-Eval benchmark, establishing a new state-of-the-art performance of 52.41%. The code, data, and models are available at https://github.com/AlongWY/sccdec.
Authors: Mansour Al Ghanim, Saleh Almohaimeed, Mengxin Zheng, Yan Solihin, Qian Lou
Abstract: This study identifies the potential vulnerabilities of Large Language Models (LLMs) to 'jailbreak' attacks, specifically focusing on the Arabic language and its various forms. While most research has concentrated on English-based prompt manipulation, our investigation broadens the scope to investigate the Arabic language. We initially tested the AdvBench benchmark in Standardized Arabic, finding that even with prompt manipulation techniques like prefix injection, it was insufficient to provoke LLMs into generating unsafe content. However, when using Arabic transliteration and chatspeak (or arabizi), we found that unsafe content could be produced on platforms like OpenAI GPT-4 and Anthropic Claude 3 Sonnet. Our findings suggest that using Arabic and its various forms could expose information that might remain hidden, potentially increasing the risk of jailbreak attacks. We hypothesize that this exposure could be due to the model's learned connection to specific words, highlighting the need for more comprehensive safety training across all language forms.
Authors: Yuheng Zhang, Dian Yu, Baolin Peng, Linfeng Song, Ye Tian, Mingyue Huo, Nan Jiang, Haitao Mi, Dong Yu
Abstract: Reinforcement Learning with Human Feedback (RLHF) has achieved great success in aligning large language models (LLMs) with human preferences. Prevalent RLHF approaches are reward-based, following the Bradley-Terry (BT) model assumption, which may not fully capture the complexity of human preferences. In this paper, we explore RLHF under a general preference framework and approach it from a game-theoretic perspective. Specifically, we formulate the problem as a two-player game and propose a novel online algorithm, iterative Nash policy optimization (INPO). The key idea is to let the policy play against itself via no-regret learning, thereby approximating the Nash policy. Unlike previous methods, INPO bypasses the need for estimating the expected win rate for individual responses, which typically incurs high computational or annotation costs. Instead, we introduce a new loss objective that is directly minimized over a preference dataset. We provide theoretical analysis for our approach and demonstrate its effectiveness through experiments on various representative benchmarks. With an LLaMA-3-8B-based SFT model, INPO achieves a 42.6% length-controlled win rate on AlpacaEval 2.0 and a 37.8% win rate on Arena-Hard, showing substantial improvement over the state-of-the-art online RLHF algorithms.
Authors: Han Guo, William Brandon, Radostin Cholakov, Jonathan Ragan-Kelley, Eric P. Xing, Yoon Kim
Abstract: The deployment of large language models (LLMs) is often constrained by memory bandwidth, where the primary bottleneck is the cost of transferring model parameters from the GPU's global memory to its registers. When coupled with custom kernels that fuse the dequantization and matmul operations, weight-only quantization can thus enable faster inference by reducing the amount of memory movement. However, developing high-performance kernels for weight-quantized LLMs presents substantial challenges, especially when the weights are compressed to non-evenly-divisible bit widths (e.g., 3 bits) with non-uniform, lookup table (LUT) quantization. This paper describes FLUTE, a flexible lookup table engine for LUT-quantized LLMs, which uses offline restructuring of the quantized weight matrix to minimize bit manipulations associated with unpacking, and vectorization and duplication of the lookup table to mitigate shared memory bandwidth constraints. At batch sizes < 32 and quantization group size of 128 (typical in LLM inference), the FLUTE kernel can be 2-4x faster than existing GEMM kernels. As an application of FLUTE, we explore a simple extension to lookup table-based NormalFloat quantization and apply it to quantize LLaMA3 to various configurations, obtaining competitive quantization performance against strong baselines while obtaining an end-to-end throughput increase of 1.5 to 2 times.
Authors: Jiewen Hu, Thomas Zhu, Sean Welleck
Abstract: Real-world formal theorem proving often depends on a wealth of context, including definitions, lemmas, comments, file structure, and other information. We introduce miniCTX, which tests a model's ability to prove formal mathematical theorems that depend on new context that is not seen during training. miniCTX contains theorems sourced from real Lean projects and textbooks, each associated with a context that can span tens of thousands of tokens. Models are tasked with proving a theorem given access to code from the theorem's repository, which contains context that is needed for the proof. As a baseline for miniCTX, we tested fine-tuning and prompting methods that condition theorem proving on preceding context. Both approaches substantially outperform traditional methods that rely solely on state information. We found that this ability to use context is not captured by previous benchmarks such as miniF2F. Alongside miniCTX, we offer ntp-toolkit for automatically extracting and annotating theorem proving data, making it easy to add new projects into miniCTX to ensure that contexts are not seen during training. miniCTX offers a challenging and realistic evaluation of neural theorem provers.
Authors: Mohammadreza Pourreza, Ruoxi Sun, Hailong Li, Lesly Miculicich, Tomas Pfister, Sercan O. Arik
Abstract: Recent advances in Text-to-SQL have largely focused on the SQLite dialect, neglecting the diverse landscape of SQL dialects like BigQuery and PostgreSQL. This limitation is due to the diversity in SQL syntaxes and functions, along with the high cost of collecting and curating SQL-specific training data. To address this, we introduce SQL-GEN, a framework for generating high-quality synthetic training data for any SQL dialect, guided by readily available dialect-specific tutorials. SQL-GEN significantly improves cross-dialect Text-to-SQL performance, boosting execution accuracy by up to 20\% over existing methods. This performance gain narrows the gap with models trained on large-scale human-annotated data. Furthermore, combining synthetic data from SQL-GEN with human-annotated data yields additional improvements of up to 5.6\%. To unify multi-dialect capabilities within a single model, we propose a novel Mixture-of-Experts (MoE) initialization that leverages the shared knowledge across dialects. Our approach merges self-attention layers from dialect-specific models and initializes expert gates using dialect-specific keywords. This leads to a versatile model optimized for multiple SQL dialects, outperforming single-dialect models and significantly enhancing overall performance.
Authors: Yang Cao
Abstract: The rapid advancement in large language models (LLMs) comes with a significant increase in their parameter size, presenting challenges for adaptation and fine-tuning. Parameter-efficient fine-tuning (PEFT) methods are widely used to adapt LLMs for downstream tasks efficiently. In this paper, we propose Singular Values and Orthonormal Regularized Singular Vectors Adaptation, or SORSA, a novel PEFT method. We introduce a method to analyze the variation of the parameters by performing singular value decomposition (SVD) and discuss and analyze SORSA's superiority in minimizing the alteration in the SVD aspect. Each SORSA adapter consists of two main parts: trainable principal singular weights $W_p = U_p \text{diag}(S_p) V^\top_p$, and frozen residual weights $W_r = U_r \text{diag}(S_r) V^\top_r$. These parts are initialized by performing SVD on pre-trained weights. Moreover, we implement and analyze an orthonormal regularizer, which we prove could decrease the condition number of $W_p$ and allows the optimization to be more efficient. SORSA adapters could be merged during inference, thus eliminating any inference latency. After all, SORSA shows a faster convergence than PiSSA and LoRA in our experiments. On the GSM-8K benchmark, Llama 2 7B adapted using SORSA achieved 56.03% accuracy, surpassing LoRA (42.30%), Full FT (49.05%), and PiSSA (53.07%). On the MATH benchmark, SORSA achieved 10.36% accuracy, outperforming LoRA (5.50%), Full FT (7.22%), and PiSSA (7.44%). We conclude that SORSA offers a new perspective on parameter-efficient fine-tuning, demonstrating remarkable performance. The code is available at https://github.com/Gunale0926/SORSA.
Authors: Sean I. Young
Abstract: In recent years, compression of large language models (LLMs) has emerged as an important problem to enable language model deployment on resource-constrained devices, reduce computational costs, and mitigate the environmental footprint of large-scale AI infrastructure. In this paper, we lay down the foundation for LLM quantization from a convex optimization perspective and propose a quantization technique that builds on this foundation for optimum quantization outcomes. Our quantization framework, CVXQ, scales to models containing hundreds of billions of weight parameters and provides users with the flexibility to compress models to any specified model size, post-training. A reference implementation of CVXQ can be obtained from github.com/seannz/cvxq.
Authors: Yong Lin, Skyler Seto, Maartje ter Hoeve, Katherine Metcalf, Barry-John Theobald, Xuan Wang, Yizhe Zhang, Chen Huang, Tong Zhang
Abstract: Reinforcement Learning from Human Feedback (RLHF) is an effective approach for aligning language models to human preferences. Central to RLHF is learning a reward function for scoring human preferences. Two main approaches for learning a reward model are 1) training an EXplicit Reward Model (EXRM) as in RLHF, and 2) using an implicit reward learned from preference data through methods such as Direct Preference Optimization (DPO). Prior work has shown that the implicit reward model of DPO (denoted as DPORM) can approximate an EXRM in the limit. DPORM's effectiveness directly implies the optimality of the learned policy, and also has practical implication for LLM alignment methods including iterative DPO. However, it is unclear how well DPORM empirically matches the performance of EXRM. This work studies the accuracy at distinguishing preferred and rejected answers for both DPORM and EXRM. Our findings indicate that even though DPORM fits the training dataset comparably, it generalizes less effectively than EXRM, especially when the validation datasets contain distribution shifts. Across five out-of-distribution settings, DPORM has a mean drop in accuracy of 3% and a maximum drop of 7%. These findings highlight that DPORM has limited generalization ability and substantiates the integration of an explicit reward model in iterative DPO approaches.
Authors: Zitong Yang, Neil Band, Shuangping Li, Emmanuel Cand\`es, Tatsunori Hashimoto
Abstract: Pretraining on large-scale, unstructured internet text enables language models to acquire a significant amount of world knowledge. However, this knowledge acquisition is data-inefficient--to learn a given fact, models must be trained on hundreds to thousands of diverse representations of it. This poses a challenge when adapting a pretrained model to a small corpus of domain-specific documents, where each fact may appear rarely or only once. We propose to bridge this gap with synthetic continued pretraining: using the small domain-specific corpus to synthesize a large corpus more amenable to learning, and then performing continued pretraining on the synthesized corpus. We instantiate this proposal with EntiGraph, a synthetic data augmentation algorithm that extracts salient entities from the source documents and then generates diverse text by drawing connections between the sampled entities. Synthetic continued pretraining with EntiGraph enables a language model to answer questions and follow generic instructions related to the source documents without access to them. If, instead, the source documents are available at inference time, we show that the knowledge acquired through our approach compounds with retrieval-augmented generation. To better understand these results, we build a simple mathematical model of EntiGraph, and show how synthetic data augmentation can "rearrange" knowledge to enable more data-efficient learning.
Authors: Zeyi Liao, Lingbo Mo, Chejian Xu, Mintong Kang, Jiawei Zhang, Chaowei Xiao, Yuan Tian, Bo Li, Huan Sun
Abstract: Generalist web agents have demonstrated remarkable potential in autonomously completing a wide range of tasks on real websites, significantly boosting human productivity. However, web tasks, such as booking flights, usually involve users' PII, which may be exposed to potential privacy risks if web agents accidentally interact with compromised websites, a scenario that remains largely unexplored in the literature. In this work, we narrow this gap by conducting the first study on the privacy risks of generalist web agents in adversarial environments. First, we present a realistic threat model for attacks on the website, where we consider two adversarial targets: stealing users' specific PII or the entire user request. Then, we propose a novel attack method, termed Environmental Injection Attack (EIA). EIA injects malicious content designed to adapt well to environments where the agents operate and our work instantiates EIA specifically for privacy scenarios in web environments. We collect 177 action steps that involve diverse PII categories on realistic websites from the Mind2Web, and conduct experiments using one of the most capable generalist web agent frameworks to date. The results demonstrate that EIA achieves up to 70% ASR in stealing specific PII and 16% ASR for full user request. Additionally, by accessing the stealthiness and experimenting with a defensive system prompt, we indicate that EIA is hard to detect and mitigate. Notably, attacks that are not well adapted for a webpage can be detected via human inspection, leading to our discussion about the trade-off between security and autonomy. However, extra attackers' efforts can make EIA seamlessly adapted, rendering such supervision ineffective. Thus, we further discuss the defenses at the pre- and post-deployment stages of the websites without relying on human supervision and call for more advanced defense strategies.
Authors: Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, Junyang Lin
Abstract: We present the Qwen2-VL Series, an advanced upgrade of the previous Qwen-VL models that redefines the conventional predetermined-resolution approach in visual processing. Qwen2-VL introduces the Naive Dynamic Resolution mechanism, which enables the model to dynamically process images of varying resolutions into different numbers of visual tokens. This approach allows the model to generate more efficient and accurate visual representations, closely aligning with human perceptual processes. The model also integrates Multimodal Rotary Position Embedding (M-RoPE), facilitating the effective fusion of positional information across text, images, and videos. We employ a unified paradigm for processing both images and videos, enhancing the model's visual perception capabilities. To explore the potential of large multimodal models, Qwen2-VL investigates the scaling laws for large vision-language models (LVLMs). By scaling both the model size-with versions at 2B, 8B, and 72B parameters-and the amount of training data, the Qwen2-VL Series achieves highly competitive performance. Notably, the Qwen2-VL-72B model achieves results comparable to leading models such as GPT-4o and Claude3.5-Sonnet across various multimodal benchmarks, outperforming other generalist models. Code is available at https://github.com/QwenLM/Qwen2-VL .
Authors: Prateek Verma
Abstract: Large Language Models (LLMs) have ushered in a new wave of artificial intelligence advancements impacting every scientific field and discipline. They are trained on a simple objective: to predict the next token given the previous context. We live in a world where most of the data around us, e.g., text, audio, and music, has a multi-scale structure associated with it. This paper infuses LLMs with traditional signal processing ideas, namely wavelets, during pre-training to take advantage of the structure. Without adding \textbf{any extra parameters} to a GPT-style LLM architecture, we achieve the same pre-training performance almost twice as fast in text, raw audio, and symbolic music. This is achieved by imposing a structure on intermediate embeddings. When trained for the same number of training steps, we achieve significant gains in performance, which is comparable to pre-training a larger neural architecture. Our architecture allows every next token prediction access to intermediate embeddings at different temporal resolutions in every Transformer decoder block. This work will hopefully pave the way for incorporating multi-rate signal processing ideas into traditional LLM pre-training. Further, we showcase pushing model performance by improving internal structure instead of just going after scale.
Authors: Jon Saad-Falcon, Adrian Gamarra Lafuente, Shlok Natarajan, Nahum Maru, Hristo Todorov, Etash Guha, E. Kelly Buchanan, Mayee Chen, Neel Guha, Christopher R\'e, Azalia Mirhoseini
Abstract: Inference-time techniques are emerging as highly effective tools to enhance large language model (LLM) capabilities. However, best practices for developing systems that combine these techniques remain underdeveloped due to our limited understanding of the utility of individual inference-time techniques and the interactions between them. Additionally, efficiently and automatically searching the space of model choices, inference-time techniques, and their compositions is challenging due to the large design space. To address these challenges, we introduce Archon, a modular framework for selecting, combining, and stacking layers of inference-time techniques to construct optimized LLM systems for target benchmarks. Rather than relying on a single LLM called once, we leverage a diverse set of LLMs and inference-time techniques, creating LLM systems greater than the sum of their parts. Archon defines an extensible design space, encompassing techniques such as generation ensembling, repeated sampling, ranking, fusion, critiquing, verification, and unit testing. It transforms the problem of building LLM systems into a hyperparameter optimization objective. Given the available LLMs, inference-time techniques, and compute budget, Archon utilizes hyperparameter search techniques to discover optimized architectures for target benchmark(s). We evaluate Archon architectures across a range of instruction-following, reasoning, and coding benchmarks, including MT-Bench, Arena-Hard-Auto, AlpacaEval 2.0, MixEval, MixEval Hard, MATH, and CodeContests. Archon architectures outperform frontier models, such as GPT-4o and Claude 3.5 Sonnet, on these benchmarks, achieving an average accuracy increase of 15.1 percentage points by using all available LLMs. We make our code and datasets available publicly on Github: https://github.com/ScalingIntelligence/Archon.
Authors: Yu Zhang, Ziyue Jiang, Ruiqi Li, Changhao Pan, Jinzheng He, Rongjie Huang, Chuxin Wang, Zhou Zhao
Abstract: Zero-shot singing voice synthesis (SVS) with style transfer and style control aims to generate high-quality singing voices with unseen timbres and styles (including singing method, emotion, rhythm, technique, and pronunciation) from audio and text prompts. However, the multifaceted nature of singing styles poses a significant challenge for effective modeling, transfer, and control. Furthermore, current SVS models often fail to generate singing voices rich in stylistic nuances for unseen singers. To address these challenges, we introduce TCSinger, the first zero-shot SVS model for style transfer across cross-lingual speech and singing styles, along with multi-level style control. Specifically, TCSinger proposes three primary modules: 1) the clustering style encoder employs a clustering vector quantization model to stably condense style information into a compact latent space; 2) the Style and Duration Language Model (S\&D-LM) concurrently predicts style information and phoneme duration, which benefits both; 3) the style adaptive decoder uses a novel mel-style adaptive normalization method to generate singing voices with enhanced details. Experimental results show that TCSinger outperforms all baseline models in synthesis quality, singer similarity, and style controllability across various tasks, including zero-shot style transfer, multi-level style control, cross-lingual style transfer, and speech-to-singing style transfer. Singing voice samples can be accessed at https://tcsinger.github.io/.
Authors: Yotam Wolf, Binyamin Rothberg, Dorin Shteyman, Amnon Shashua
Abstract: A common practice in large language model (LLM) usage for complex analytical tasks such as code generation, is to sample a solution for the entire task within the model's context window. Previous works have shown that subtask decomposition within the model's context (chain of thought), is beneficial for solving such tasks. In this work, we point a limitation of LLMs' ability to perform several sub-tasks within the same context window - an in-context hardness of composition, pointing to an advantage for distributing a decomposed problem in a multi-agent system of LLMs. The hardness of composition is quantified by a generation complexity metric, i.e., the number of LLM generations required to sample at least one correct solution. We find a gap between the generation complexity of solving a compositional problem within the same context relative to distributing it among multiple agents, that increases exponentially with the solution's length. We prove our results theoretically and demonstrate them empirically.
Authors: Diogo Gl\'oria-Silva, David Semedo, Jo\~ao Magalh\~aes
Abstract: Guiding users through complex procedural plans is an inherently multimodal task in which having visually illustrated plan steps is crucial to deliver an effective plan guidance. However, existing works on plan-following language models (LMs) often are not capable of multimodal input and output. In this work, we present MM-PlanLLM, the first multimodal LLM designed to assist users in executing instructional tasks by leveraging both textual plans and visual information. Specifically, we bring cross-modality through two key tasks: Conversational Video Moment Retrieval, where the model retrieves relevant step-video segments based on user queries, and Visually-Informed Step Generation, where the model generates the next step in a plan, conditioned on an image of the user's current progress. MM-PlanLLM is trained using a novel multitask-multistage approach, designed to gradually expose the model to multimodal instructional-plans semantic layers, achieving strong performance on both multimodal and textual dialogue in a plan-grounded setting. Furthermore, we show that the model delivers cross-modal temporal and plan-structure representations aligned between textual plan steps and instructional video moments.
Authors: Ming Zhong, Aston Zhang, Xuewei Wang, Rui Hou, Wenhan Xiong, Chenguang Zhu, Zhengxing Chen, Liang Tan, Chloe Bi, Mike Lewis, Sravya Popuri, Sharan Narang, Melanie Kambadur, Dhruv Mahajan, Sergey Edunov, Jiawei Han, Laurens van der Maaten
Abstract: The development and evaluation of Large Language Models (LLMs) have largely focused on individual capabilities. However, this overlooks the intersection of multiple abilities across different types of expertise that are often required for real-world tasks, which we term cross capabilities. To systematically explore this concept, we first define seven core individual capabilities and then pair them to form seven common cross capabilities, each supported by a manually constructed taxonomy. Building on these definitions, we introduce CrossEval, a benchmark comprising 1,400 human-annotated prompts, with 100 prompts for each individual and cross capability. To ensure reliable evaluation, we involve expert annotators to assess 4,200 model responses, gathering 8,400 human ratings with detailed explanations to serve as reference examples. Our findings reveal that, in both static evaluations and attempts to enhance specific abilities, current LLMs consistently exhibit the "Law of the Weakest Link," where cross-capability performance is significantly constrained by the weakest component. Specifically, across 58 cross-capability scores from 17 models, 38 scores are lower than all individual capabilities, while 20 fall between strong and weak, but closer to the weaker ability. These results highlight the under-performance of LLMs in cross-capability tasks, making the identification and improvement of the weakest capabilities a critical priority for future research to optimize performance in complex, multi-dimensional scenarios.
Authors: Haolin Jin, Zechao Sun, Huaming Chen
Abstract: Large Language Models (LLMs) have shown incredible potential in code generation tasks, and recent research in prompt engineering have enhanced LLMs' understanding of textual information. However, ensuring the accuracy of generated code often requires extensive testing and validation by programmers. While LLMs can typically generate code based on task descriptions, their accuracy remains limited, especially for complex tasks that require a deeper understanding of both the problem statement and the code generation process. This limitation is primarily due to the LLMs' need to simultaneously comprehend text and generate syntactically and semantically correct code, without having the capability to automatically refine the code. In real-world software development, programmers rarely produce flawless code in a single attempt based on the task description alone, they rely on iterative feedback and debugging to refine their programs. Inspired by this process, we introduce a novel architecture of LLM-based agents for code generation and automatic debugging: Refinement and Guidance Debugging (RGD). The RGD framework is a multi-LLM-based agent debugger that leverages three distinct LLM agents-Guide Agent, Debug Agent, and Feedback Agent. RGD decomposes the code generation task into multiple steps, ensuring a clearer workflow and enabling iterative code refinement based on self-reflection and feedback. Experimental results demonstrate that RGD exhibits remarkable code generation capabilities, achieving state-of-the-art performance with a 9.8% improvement on the HumanEval dataset and a 16.2% improvement on the MBPP dataset compared to the state-of-the-art approaches and traditional direct prompting approaches. We highlight the effectiveness of the RGD framework in enhancing LLMs' ability to generate and refine code autonomously.
Authors: Mengzhao Jia, Wenhao Yu, Kaixin Ma, Tianqing Fang, Zhihan Zhang, Siru Ouyang, Hongming Zhang, Meng Jiang, Dong Yu
Abstract: Text-rich images, where text serves as the central visual element guiding the overall understanding, are prevalent in real-world applications, such as presentation slides, scanned documents, and webpage snapshots. Tasks involving multiple text-rich images are especially challenging, as they require not only understanding the content of individual images but reasoning about inter-relationships and logical flows across multiple visual inputs. Despite the importance of these scenarios, current multimodal large language models (MLLMs) struggle to handle such tasks due to two key challenges: (1) the scarcity of high-quality instruction tuning datasets for text-rich multi-image scenarios, and (2) the difficulty in balancing image resolution with visual feature sequence length. To address these challenges, we propose Leopard, a MLLM designed specifically for handling vision-language tasks involving multiple text-rich images. First, we curated about one million high-quality multimodal instruction-tuning data, tailored to text-rich, multi-image scenarios. Second, we developed an adaptive high-resolution multi-image encoding module to dynamically optimize the allocation of visual sequence length based on the original aspect ratios and resolutions of the input images. Experiments across a wide range of benchmarks demonstrate our model's superior capabilities in text-rich, multi-image evaluations and competitive performance in general domain evaluations.