Authors: Connor Shorten, Charles Pierse, Thomas Benjamin Smith, Erika Cardenas, Akanksha Sharma, John Trengrove, Bob van Luijt
Abstract: The ability of Large Language Models (LLMs) to generate structured outputs, such as JSON, is crucial for their use in Compound AI Systems. However, evaluating and improving this capability remains challenging. In this work, we introduce StructuredRAG, a benchmark of six tasks designed to assess LLMs' proficiency in following response format instructions. We evaluate two state-of-the-art LLMs, Gemini 1.5 Pro and Llama 3 8B-instruct with 4-bit quantization using two distinct prompting strategies. We introduce these prompting strategies as f-String and Follow the Format (FF) prompting. Across 24 experiments, we find an average success rate of 82.55%. We further find a high variance in performance across tasks, models, and prompting strategies with success rates ranging from 0 to 100%. We find that Llama 3 8B-instruct often performs competitively with Gemini 1.5 Pro. We observe that task complexity significantly influences performance, with tasks involving lists or composite object outputs proving more challenging. Our findings highlight the need for further research into improving the reliability and consistency of structured output generation in LLMs. We have open-sourced our experimental code and results at github.com/weaviate/structured-rag.
Authors: Guanming Xiong, Junwei Bao, Hongfei Jiang, Yang Song, Wen Zhao
Abstract: This study explores text-to-SQL parsing by leveraging the powerful reasoning capabilities of large language models (LLMs). Despite recent advancements, existing LLM-based methods have not adequately addressed scalability, leading to inefficiencies when processing wide tables. Furthermore, current interaction-based approaches either lack a step-by-step, interpretable SQL generation process or fail to provide an efficient and universally applicable interaction design. To address these challenges, we introduce Interactive-T2S, a framework that generates SQL queries through direct interactions with databases. This framework includes four general tools that facilitate proactive and efficient information retrieval by the LLM. Additionally, we have developed detailed exemplars to demonstrate the step-wise reasoning processes within our framework. Our experiments on the BIRD-Dev dataset, employing a setting without oracle knowledge, reveal that our method achieves state-of-the-art results with only two exemplars, underscoring the effectiveness and robustness of our framework.
Authors: Jaehyun Nam, Woomin Song, Seong Hyeon Park, Jihoon Tack, Sukmin Yun, Jaehyung Kim, Kyu Hwan Oh, Jinwoo Shin
Abstract: Learning with a limited number of labeled data is a central problem in real-world applications of machine learning, as it is often expensive to obtain annotations. To deal with the scarcity of labeled data, transfer learning is a conventional approach; it suggests to learn a transferable knowledge by training a neural network from multiple other sources. In this paper, we investigate transfer learning of tabular tasks, which has been less studied and successful in the literature, compared to other domains, e.g., vision and language. This is because tables are inherently heterogeneous, i.e., they contain different columns and feature spaces, making transfer learning difficult. On the other hand, recent advances in natural language processing suggest that the label scarcity issue can be mitigated by utilizing in-context learning capability of large language models (LLMs). Inspired by this and the fact that LLMs can also process tables within a unified language space, we ask whether LLMs can be effective for tabular transfer learning, in particular, under the scenarios where the source and target datasets are of different format. As a positive answer, we propose a novel tabular transfer learning framework, coined Prompt to Transfer (P2T), that utilizes unlabeled (or heterogeneous) source data with LLMs. Specifically, P2T identifies a column feature in a source dataset that is strongly correlated with a target task feature to create examples relevant to the target task, thus creating pseudo-demonstrations for prompts. Experimental results demonstrate that P2T outperforms previous methods on various tabular learning benchmarks, showing good promise for the important, yet underexplored tabular transfer learning problem. Code is available at https://github.com/jaehyun513/P2T.
Authors: Evan Dufraisse, Adrian Popescu, Julien Tourille, Armelle Brun, Olivier Hamon
Abstract: Researchers and practitioners interested in computational politics rely on automatic content analysis tools to make sense of the large amount of political texts available on the Web. Such tools should provide objective and subjective aspects at different granularity levels to make the analyses useful in practice. Existing methods produce interesting insights for objective aspects, but are limited for subjective ones, are often limited to national contexts, and have limited explainability. We introduce a text analysis framework which integrates both perspectives and provides a fine-grained processing of subjective aspects. Information retrieval techniques and knowledge bases complement powerful natural language processing components to allow a flexible aggregation of results at different granularity levels. Importantly, the proposed bottom-up approach facilitates the explainability of the obtained results. We illustrate its functioning with insights on news outlets, political orientations, topics, individual entities, and demographic segments. The approach is instantiated on a large corpus of French news, but is designed to work seamlessly for other languages and countries.
Authors: Benjamin Reichman, Kartik Talamadupula, Toshish Jawale, Larry Heck
Abstract: Retrieval augmented generation (RAG) systems augment how knowledge language models are by integrating external information sources such as Wikipedia, internal documents, scientific papers, or the open internet. RAG systems that rely on the open internet as their knowledge source have to contend with the complexities of human-generated content. Human communication extends much deeper than just the words rendered as text. Intent, tonality, and connotation can all change the meaning of what is being conveyed. Recent real-world deployments of RAG systems have shown some difficulty in understanding these nuances of human communication. One significant challenge for these systems lies in processing sarcasm. Though the Large Language Models (LLMs) that make up the backbone of these RAG systems are able to detect sarcasm, they currently do not always use these detections for the subsequent processing of text. To address these issues, in this paper, we synthetically generate sarcastic passages from Natural Question's Wikipedia retrieval corpus. We then test the impact of these passages on the performance of both the retriever and reader portion of the RAG pipeline. We introduce a prompting system designed to enhance the model's ability to interpret and generate responses in the presence of sarcasm, thus improving overall system performance. Finally, we conduct ablation studies to validate the effectiveness of our approach, demonstrating improvements in handling sarcastic content within RAG systems.
Authors: Patrick Huber, Arash Einolghozati, Rylan Conway, Kanika Narang, Matt Smith, Waqar Nayyar, Adithya Sagar, Ahmed Aly, Akshat Shrivastava
Abstract: Distilling conversational skills into Small Language Models (SLMs) with approximately 1 billion parameters presents significant challenges. Firstly, SLMs have limited capacity in their model parameters to learn extensive knowledge compared to larger models. Secondly, high-quality conversational datasets are often scarce, small, and domain-specific. Addressing these challenges, we introduce a novel data distillation framework named CoDi (short for Conversational Distillation, pronounced "Cody"), allowing us to synthesize large-scale, assistant-style datasets in a steerable and diverse manner. Specifically, while our framework is task agnostic at its core, we explore and evaluate the potential of CoDi on the task of conversational grounded reasoning for question answering. This is a typical on-device scenario for specialist SLMs, allowing for open-domain model responses, without requiring the model to "memorize" world knowledge in its limited weights. Our evaluations show that SLMs trained with CoDi-synthesized data achieve performance comparable to models trained on human-annotated data in standard metrics. Additionally, when using our framework to generate larger datasets from web data, our models surpass larger, instruction-tuned models in zero-shot conversational grounded reasoning tasks.
Authors: Atmika Gorti, Manas Gaur, Aman Chadha
Abstract: Large Language Models (LLMs) are prone to inheriting and amplifying societal biases embedded within their training data, potentially reinforcing harmful stereotypes related to gender, occupation, and other sensitive categories. This issue becomes particularly problematic as biased LLMs can have far-reaching consequences, leading to unfair practices and exacerbating social inequalities across various domains, such as recruitment, online content moderation, or even the criminal justice system. Although prior research has focused on detecting bias in LLMs using specialized datasets designed to highlight intrinsic biases, there has been a notable lack of investigation into how these findings correlate with authoritative datasets, such as those from the U.S. National Bureau of Labor Statistics (NBLS). To address this gap, we conduct empirical research that evaluates LLMs in a ``bias-out-of-the-box" setting, analyzing how the generated outputs compare with the distributions found in NBLS data. Furthermore, we propose a straightforward yet effective debiasing mechanism that directly incorporates NBLS instances to mitigate bias within LLMs. Our study spans seven different LLMs, including instructable, base, and mixture-of-expert models, and reveals significant levels of bias that are often overlooked by existing bias detection techniques. Importantly, our debiasing method, which does not rely on external datasets, demonstrates a substantial reduction in bias scores, highlighting the efficacy of our approach in creating fairer and more reliable LLMs.
Authors: Sepehr Kamahi, Yadollah Yaghoobzadeh
Abstract: Despite the widespread adoption of autoregressive language models, explainability evaluation research has predominantly focused on span infilling and masked language models (MLMs). Evaluating the faithfulness of an explanation method -- how accurately the method explains the inner workings and decision-making of the model -- is very challenging because it is very hard to separate the model from its explanation. Most faithfulness evaluation techniques corrupt or remove some input tokens considered important according to a particular attribution (feature importance) method and observe the change in the model's output. This approach creates out-of-distribution inputs for causal language models (CLMs) due to their training objective of next token prediction. In this study, we propose a technique that leverages counterfactual generation to evaluate the faithfulness of attribution methods for autoregressive language modeling scenarios. Our technique creates fluent and in-distribution counterfactuals that makes evaluation protocol more reliable. Code is available at https://github.com/Sepehr-Kamahi/faith
Authors: Anh-Dung Vo, Minseong Jung, Wonbeen Lee, Daewoo Choi
Abstract: The field of Natural Language Processing (NLP) has seen significant advancements with the development of Large Language Models (LLMs). However, much of this research remains focused on English, often overlooking low-resource languages like Korean. This oversight presents challenges due to the unique non-alphabetic token structure of Korean and the substantial memory and computational demands required for LLM training, which frequently lead to memory constraints and out-of-memory errors. To address these issues, we present RedWhale, a model specifically tailored for Korean language processing. RedWhale is developed using an efficient continual pretraining approach that includes a comprehensive Korean corpus preprocessing pipeline, a specialized tokenizer, an optimized model initialization technique, and a multistage pretraining strategy. These innovations collectively reduce training time and computational costs while maintaining high levels of accuracy and comprehension. By leveraging cross-lingual transfer learning, which exploits shared linguistic similarities across languages, RedWhale builds on English models to enhance Korean language processing. Experimental results demonstrate that RedWhale outperforms other leading models on Korean NLP benchmarks, including the Korean Balanced Evaluation of Significant Tasks (KoBEST), showing superior understanding and generation of Korean text. Furthermore, RedWhale showed no signs of convergence even after pretraining on 9.7 billion tokens, indicating the potential for further improvements with additional training. This work represents a significant advancement in bridging the linguistic divide, particularly in enhancing NLP capabilities for the Korean language.
Authors: Yazhou Zhang, Chunwang Zou, Zheng Lian, Prayag Tiwari, Jing Qin
Abstract: In the era of large language models (LLMs), the task of ``System I''~-~the fast, unconscious, and intuitive tasks, e.g., sentiment analysis, text classification, etc., have been argued to be successfully solved. However, sarcasm, as a subtle linguistic phenomenon, often employs rhetorical devices like hyperbole and figuration to convey true sentiments and intentions, involving a higher level of abstraction than sentiment analysis. There is growing concern that the argument about LLMs' success may not be fully tenable when considering sarcasm understanding. To address this question, we select eleven SOTA LLMs and eight SOTA pre-trained language models (PLMs) and present comprehensive evaluations on six widely used benchmark datasets through different prompting approaches, i.e., zero-shot input/output (IO) prompting, few-shot IO prompting, chain of thought (CoT) prompting. Our results highlight three key findings: (1) current LLMs underperform supervised PLMs based sarcasm detection baselines across six sarcasm benchmarks. This suggests that significant efforts are still required to improve LLMs' understanding of human sarcasm. (2) GPT-4 consistently and significantly outperforms other LLMs across various prompting methods, with an average improvement of 14.0\%$\uparrow$. Claude 3 and ChatGPT demonstrate the next best performance after GPT-4. (3) Few-shot IO prompting method outperforms the other two methods: zero-shot IO and few-shot CoT. The reason is that sarcasm detection, being a holistic, intuitive, and non-rational cognitive process, is argued not to adhere to step-by-step logical reasoning, making CoT less effective in understanding sarcasm compared to its effectiveness in mathematical reasoning tasks.
Authors: Sai Koneru, Matthias Huck, Miriam Exel, Jan Niehues
Abstract: Recent advancements in NLP have resulted in models with specialized strengths, such as processing multimodal inputs or excelling in specific domains. However, real-world tasks, like multimodal translation, often require a combination of these strengths, such as handling both translation and image processing. While individual translation and vision models are powerful, they typically lack the ability to perform both tasks in a single system. Combining these models poses challenges, particularly due to differences in their vocabularies, which limit the effectiveness of traditional ensemble methods to post-generation techniques like N-best list re-ranking. In this work, we propose a novel zero-shot ensembling strategy that allows for the integration of different models during the decoding phase without the need for additional training. Our approach re-ranks beams during decoding by combining scores at the word level, using heuristics to predict when a word is completed. We demonstrate the effectiveness of this method in machine translation scenarios, showing that it enables the generation of translations that are both speech- and image-aware while also improving overall translation quality\footnote{We will release the code upon paper acceptance.}.
Authors: Yuxuan Chen, Haoyan Yang, Hengkai Pan, Fardeen Siddiqui, Antonio Verdone, Qingyang Zhang, Sumit Chopra, Chen Zhao, Yiqiu Shen
Abstract: Breast ultrasound is essential for detecting and diagnosing abnormalities, with radiology reports summarizing key findings like lesion characteristics and malignancy assessments. Extracting this critical information is challenging due to the unstructured nature of these reports, with varied linguistic styles and inconsistent formatting. While proprietary LLMs like GPT-4 are effective, they are costly and raise privacy concerns when handling protected health information. This study presents a pipeline for developing an in-house LLM to extract clinical information from radiology reports. We first use GPT-4 to create a small labeled dataset, then fine-tune a Llama3-8B model on it. Evaluated on clinician-annotated reports, our model achieves an average F1 score of 84.6%, which is on par with GPT-4. Our findings demonstrate the feasibility of developing an in-house LLM that not only matches GPT-4's performance but also offers cost reductions and enhanced data privacy.
Authors: Sonit Singh
Abstract: Recent developments in the field of Natural Language Processing, especially language models such as the transformer have brought state-of-the-art results in language understanding and language generation. In this work, we investigate the use of the transformer model for radiology report generation from chest X-rays. We also highlight limitations in evaluating radiology report generation using only the standard language generation metrics. We then applied a transformer based radiology report generation architecture, and also compare the performance of a transformer based decoder with the recurrence based decoder. Experiments were performed using the IU-CXR dataset, showing superior results to its LSTM counterpart and being significantly faster. Finally, we identify the need of evaluating radiology report generation system using both language generation metrics and classification metrics, which helps to provide robust measure of generated reports in terms of their coherence and diagnostic value.
Authors: Yibo Yan, Joey Lee
Abstract: In human reading and communication, individuals tend to engage in geospatial reasoning, which involves recognizing geographic entities and making informed inferences about their interrelationships. To mimic such cognitive process, current methods either utilize conventional natural language understanding toolkits, or directly apply models pretrained on geo-related natural language corpora. However, these methods face two significant challenges: i) they do not generalize well to unseen geospatial scenarios, and ii) they overlook the importance of integrating geospatial context from geographical databases with linguistic information from the Internet. To handle these challenges, we propose GeoReasoner, a language model capable of reasoning on geospatially grounded natural language. Specifically, it first leverages Large Language Models (LLMs) to generate a comprehensive location description based on linguistic and geospatial information. It also encodes direction and distance information into spatial embedding via treating them as pseudo-sentences. Consequently, the model is trained on both anchor-level and neighbor-level inputs to learn geo-entity representation. Extensive experimental results demonstrate GeoReasoner's superiority in three tasks: toponym recognition, toponym linking, and geo-entity typing, compared to the state-of-the-art baselines.
Authors: Xuanwang Zhang, Yunze Song, Yidong Wang, Shuyun Tang, Xinfeng Li, Zhengran Zeng, Zhen Wu, Wei Ye, Wenyuan Xu, Yue Zhang, Xinyu Dai, Shikun Zhang, Qingsong Wen
Abstract: Large Language Models (LLMs) demonstrate human-level capabilities in dialogue, reasoning, and knowledge retention. However, even the most advanced LLMs face challenges such as hallucinations and real-time updating of their knowledge. Current research addresses this bottleneck by equipping LLMs with external knowledge, a technique known as Retrieval Augmented Generation (RAG). However, two key issues constrained the development of RAG. First, there is a growing lack of comprehensive and fair comparisons between novel RAG algorithms. Second, open-source tools such as LlamaIndex and LangChain employ high-level abstractions, which results in a lack of transparency and limits the ability to develop novel algorithms and evaluation metrics. To close this gap, we introduce RAGLAB, a modular and research-oriented open-source library. RAGLAB reproduces 6 existing algorithms and provides a comprehensive ecosystem for investigating RAG algorithms. Leveraging RAGLAB, we conduct a fair comparison of 6 RAG algorithms across 10 benchmarks. With RAGLAB, researchers can efficiently compare the performance of various algorithms and develop novel algorithms.
Authors: Varun Gumma, Pranjal A. Chitale, Kalika Bali
Abstract: Standard Neural Machine Translation (NMT) models have traditionally been trained with Sinusoidal Positional Embeddings (PEs), which are inadequate for capturing long-range dependencies and are inefficient for long-context or document-level translation. In contrast, state-of-the-art large language models (LLMs) employ relative PEs, demonstrating superior length generalization. This work explores the potential for efficiently switching the Positional Embeddings of pre-trained NMT models from absolute sinusoidal PEs to relative approaches such as RoPE and ALiBi. Our findings reveal that sinusoidal PEs can be effectively replaced with RoPE and ALiBi with negligible or no performance loss, achieved by fine-tuning on a small fraction of high-quality data. Additionally, models trained without Positional Embeddings (NoPE) are not a viable solution for Encoder-Decoder architectures, as they consistently under-perform compared to models utilizing any form of Positional Embedding. Furthermore, even a model trained from scratch with these relative PEs slightly under-performs a fine-tuned model, underscoring the efficiency and validity of our hypothesis.
Authors: Chi Ma, Mincong Huang, Ying Zhang, Chao Wang, Yujie Wang, Lei Yu, Chuan Liu, Wei Lin
Abstract: Dynamic activation (DA) techniques, such as DejaVu and MoEfication, have demonstrated their potential to significantly enhance the inference efficiency of large language models (LLMs). However, these techniques often rely on ReLU activation functions or require additional parameters and training to maintain performance. This paper introduces a training-free Threshold-based Dynamic Activation(TDA) method that leverage sequence information to exploit the inherent sparsity of models across various architectures. This method is designed to accelerate generation speed by 18-25\% without significantly compromising task performance, thereby addressing the limitations of existing DA techniques. Moreover, we delve into the root causes of LLM sparsity and theoretically analyze two of its critical features: history-related activation uncertainty and semantic-irrelevant activation inertia. Our comprehensive analyses not only provide a robust theoretical foundation for DA methods but also offer valuable insights to guide future research in optimizing LLMs for greater efficiency and effectiveness.
Authors: Hao Zhou, Zhijun Wang, Shujian Huang, Xin Huang, Xue Han, Junlan Feng, Chao Deng, Weihua Luo, Jiajun Chen
Abstract: Large Language Models (LLMs) are often English-centric due to the disproportionate distribution of languages in their pre-training data. Enhancing non-English language capabilities through post-pretraining often results in catastrophic forgetting of the ability of original languages. Previous methods either achieve good expansion with severe forgetting or slight forgetting with poor expansion, indicating the challenge of balancing language expansion while preventing forgetting. In this paper, we propose a method called MoE-LPR (Mixture-of-Experts with Language Priors Routing) to alleviate this problem. MoE-LPR employs a two-stage training approach to enhance the multilingual capability. First, the model is post-pretrained into a Mixture-of-Experts (MoE) architecture by upcycling, where all the original parameters are frozen and new experts are added. In this stage, we focus improving the ability on expanded languages, without using any original language data. Then, the model reviews the knowledge of the original languages with replay data amounting to less than 1% of post-pretraining, where we incorporate language priors routing to better recover the abilities of the original languages. Evaluations on multiple benchmarks show that MoE-LPR outperforms other post-pretraining methods. Freezing original parameters preserves original language knowledge while adding new experts preserves the learning ability. Reviewing with LPR enables effective utilization of multilingual knowledge within the parameters. Additionally, the MoE architecture maintains the same inference overhead while increasing total model parameters. Extensive experiments demonstrate MoE-LPR's effectiveness in improving expanded languages and preserving original language proficiency with superior scalability. Code and scripts are freely available at https://github.com/zjwang21/MoE-LPR.git.
Authors: Simon M\"unker
Abstract: Contemporary research in social sciences is increasingly utilizing state-of-the-art statistical language models to annotate or generate content. While these models perform benchmark-leading on common language tasks and show exemplary task-independent emergent abilities, transferring them to novel out-of-domain tasks is only insufficiently explored. The implications of the statistical black-box approach - stochastic parrots - are prominently criticized in the language model research community; however, the significance for novel generative tasks is not. This work investigates the alignment between personalized language models and survey participants on a Moral Foundation Theory questionnaire. We adapt text-to-text models to different political personas and survey the questionnaire repetitively to generate a synthetic population of persona and model combinations. Analyzing the intra-group variance and cross-alignment shows significant differences across models and personas. Our findings indicate that adapted models struggle to represent the survey-captured assessment of political ideologies. Thus, using language models to mimic social interactions requires measurable improvements in in-context optimization or parameter manipulation to align with psychological and sociological stereotypes. Without quantifiable alignment, generating politically nuanced content remains unfeasible. To enhance these representations, we propose a testable framework to generate agents based on moral value statements for future research.
Authors: Kai Xiong, Xiao Ding, Li Du, Jiahao Ying, Ting Liu, Bing Qin, Yixin Cao
Abstract: Large Language Models (LLMs) are versatile and demonstrate impressive generalization ability by mining and learning information from extensive unlabeled text. However, they still exhibit reasoning mistakes, often stemming from knowledge deficiencies, which can affect their trustworthiness and reliability. Although users can provide diverse and comprehensive queries, obtaining sufficient and effective feedback is demanding. Furthermore, evaluating LLMs comprehensively with limited labeled samples is difficult. This makes it a challenge to diagnose and remedy the deficiencies of LLMs through rich label-free user queries. To tackle this challenge, we propose a label-free curricular meaningful learning framework (LaMer). LaMer first employs relative entropy to automatically diagnose and quantify the knowledge deficiencies of LLMs in a label-free setting. Next, to remedy the diagnosed knowledge deficiencies, we apply curricular meaningful learning: first, we adopt meaningful learning to adaptively synthesize augmentation data according to the severity of the deficiencies, and then design a curricular deficiency remedy strategy to remedy the knowledge deficiencies of LLMs progressively. Experiments show that LaMer efficiently and effectively diagnoses and remedies knowledge deficiencies in LLMs, improving various LLMs across seven out-of-distribution (OOD) reasoning and language understanding benchmarks, achieving comparable results to baselines with just 40\% training data. LaMer even surpasses methods that rely on labeled datasets for deficiency diagnosis. In application, our label-free method can offer an effective knowledge deficiency diagnostic tool for efficient LLM development.
Authors: Tahir Javed, Janki Nawale, Sakshi Joshi, Eldho George, Kaushal Bhogale, Deovrat Mehendale, Mitesh M. Khapra
Abstract: Hindi, one of the most spoken language of India, exhibits a diverse array of accents due to its usage among individuals from diverse linguistic origins. To enable a robust evaluation of Hindi ASR systems on multiple accents, we create a benchmark, LAHAJA, which contains read and extempore speech on a diverse set of topics and use cases, with a total of 12.5 hours of Hindi audio, sourced from 132 speakers spanning 83 districts of India. We evaluate existing open-source and commercial models on LAHAJA and find their performance to be poor. We then train models using different datasets and find that our model trained on multilingual data with good speaker diversity outperforms existing models by a significant margin. We also present a fine-grained analysis which shows that the performance declines for speakers from North-East and South India, especially with content heavy in named entities and specialized terminology.
Authors: Marco Cognetta, Vil\'em Zouhar, Naoaki Okazaki
Abstract: Subword regularization, used widely in NLP, improves model performance by reducing the dependency on exact tokenizations, augmenting the training corpus, and exposing the model to more unique contexts during training. BPE and MaxMatch, two popular subword tokenization schemes, have stochastic dropout regularization variants. However, there has not been an analysis of the distributions formed by them. We show that these stochastic variants are heavily biased towards a small set of tokenizations per word. If the benefits of subword regularization are as mentioned, we hypothesize that biasedness artificially limits the effectiveness of these schemes. Thus, we propose an algorithm to uniformly sample tokenizations that we use as a drop-in replacement for the stochastic aspects of existing tokenizers, and find that it improves machine translation quality.
Authors: Felermino D. M. Antonio Ali, Henrique Lopes Cardoso, Rui Sousa-Silva
Abstract: As part of the Open Language Data Initiative shared tasks, we have expanded the FLORES+ evaluation set to include Emakhuwa, a low-resource language widely spoken in Mozambique. We translated the dev and devtest sets from Portuguese into Emakhuwa, and we detail the translation process and quality assurance measures used. Our methodology involved various quality checks, including post-editing and adequacy assessments. The resulting datasets consist of multiple reference sentences for each source. We present baseline results from training a Neural Machine Translation system and fine-tuning existing multilingual translation models. Our findings suggest that spelling inconsistencies remain a challenge in Emakhuwa. Additionally, the baseline models underperformed on this evaluation set, underscoring the necessity for further research to enhance machine translation quality for Emakhuwa. The data is publicly available at https://huggingface.co/datasets/LIACC/Emakhuwa-FLORES.
URLs: https://huggingface.co/datasets/LIACC/Emakhuwa-FLORES.
Authors: David Kletz (Lattice, LLF - UMR7110, UPCit\'e), Pascal Amsili (Lattice), Marie Candito (LLF UMR7110, UPCit\'e)
Abstract: Several methodologies have recently been proposed to evaluate the ability of Pretrained Language Models (PLMs) to interpret negation. In this article, we build on Gubelmann and Handschuh (2022), which studies the modification of PLMs' predictions as a function of the polarity of inputs, in English. Crucially, this test uses ``self-contained'' inputs ending with a masked position: depending on the polarity of a verb in the input, a particular token is either semantically ruled out or allowed at the masked position. By replicating Gubelmann and Handschuh (2022) experiments, we have uncovered flaws that weaken the conclusions that can be drawn from this test. We thus propose an improved version, the Self-Contained Neg Test, which is more controlled, more systematic, and entirely based on examples forming minimal pairs varying only in the presence or absence of verbal negation in English. When applying our test to the roberta and bert base and large models, we show that only roberta-large shows trends that match the expectations, while bert-base is mostly insensitive to negation. For all the tested models though, in a significant number of test instances the top-1 prediction remains the token that is semantically forbidden by the context, which shows how much room for improvement remains for a proper treatment of the negation phenomenon.
Authors: Haochen Wang, Kai Hu, Haoyu Dong, Liangcai Gao
Abstract: We study a new problem setting of question answering (QA), referred to as DocTabQA. Within this setting, given a long document, the goal is to respond to questions by organizing the answers into structured tables derived directly from the document's content. Unlike traditional QA approaches which predominantly rely on unstructured text to formulate responses, DocTabQA aims to leverage structured tables as answers to convey information clearly and systematically, thereby enhancing user comprehension and highlighting relationships between data points. To the best of our knowledge, this problem has not been previously explored. In this paper, we introduce the QTabA dataset, encompassing 300 financial documents, accompanied by manually annotated 1.5k question-table pairs. Initially, we leverage Large Language Models (LLMs) such as GPT-4 to establish a baseline. However, it is widely acknowledged that LLMs encounter difficulties when tasked with generating intricate, structured outputs from long input sequences. To overcome these challenges, we present a two-stage framework, called DocTabTalk, which initially retrieves relevant sentences from extensive documents and subsequently generates hierarchical tables based on these identified sentences. DocTabTalk incorporates two key technological innovations: AlignLLaMA and TabTalk, which are specifically tailored to assist GPT-4 in tackling DocTabQA, enabling it to generate well-structured, hierarchical tables with improved organization and clarity. Comprehensive experimental evaluations conducted on both QTabA and RotoWire datasets demonstrate that our DocTabTalk significantly enhances the performances of the GPT-4 in our proposed DocTabQA task and the table generation task. The code and dataset are available at https://github.com/SmileWHC/DocTabQA for further research.
Authors: Baohao Liao, Christian Herold, Shahram Khadivi, Christof Monz
Abstract: This paper introduces two multilingual systems, IKUN and IKUN-C, developed for the general machine translation task in WMT24. IKUN and IKUN-C represent an open system and a constrained system, respectively, built on Llama-3-8b and Mistral-7B-v0.3. Both systems are designed to handle all 11 language directions using a single model. According to automatic evaluation metrics, IKUN-C achieved 6 first-place and 3 second-place finishes among all constrained systems, while IKUN secured 1 first-place and 2 second-place finishes across both open and constrained systems. These encouraging results suggest that large language models (LLMs) are nearing the level of proficiency required for effective multilingual machine translation. The systems are based on a two-stage approach: first, continuous pre-training on monolingual data in 10 languages, followed by fine-tuning on high-quality parallel data for 11 language directions. The primary difference between IKUN and IKUN-C lies in their monolingual pre-training strategy. IKUN-C is pre-trained using constrained monolingual data, whereas IKUN leverages monolingual data from the OSCAR dataset. In the second phase, both systems are fine-tuned on parallel data sourced from NTREX, Flores, and WMT16-23 for all 11 language pairs.
Authors: Edirlei Soares de Lima, Marco A. Casanova, Antonio L. Furtado
Abstract: A method for generating narratives by analyzing single images or image sequences is presented, inspired by the time immemorial tradition of Narrative Art. The proposed method explores the multimodal capabilities of GPT-4o to interpret visual content and create engaging stories, which are illustrated by a Stable Diffusion XL model. The method is supported by a fully implemented tool, called ImageTeller, which accepts images from diverse sources as input. Users can guide the narrative's development according to the conventions of fundamental genres - such as Comedy, Romance, Tragedy, Satire or Mystery -, opt to generate data-driven stories, or to leave the prototype free to decide how to handle the narrative structure. User interaction is provided along the generation process, allowing the user to request alternative chapters or illustrations, and even reject and restart the story generation based on the same input. Additionally, users can attach captions to the input images, influencing the system's interpretation of the visual content. Examples of generated stories are provided, along with details on how to access the prototype.
Authors: Shahriar Golchin, Mihai Surdeanu, Steven Bethard, Eduardo Blanco, Ellen Riloff
Abstract: In-context learning (ICL) has proven to be an effective strategy for improving the performance of large language models (LLMs) with no additional training. However, the exact mechanism behind these performance improvements remains unclear. This study is the first to show how ICL surfaces memorized training data and to explore the correlation between this memorization and performance across various ICL regimes: zero-shot, few-shot, and many-shot. Our most notable findings include: (1) ICL significantly surfaces memorization compared to zero-shot learning in most cases; (2) demonstrations, without their labels, are the most effective element in surfacing memorization; (3) ICL improves performance when the surfaced memorization in few-shot regimes reaches a high level (about 40%); and (4) there is a very strong correlation between performance and memorization in ICL when it outperforms zero-shot learning. Overall, our study uncovers a hidden phenomenon -- memorization -- at the core of ICL, raising an important question: to what extent do LLMs truly generalize from demonstrations in ICL, and how much of their success is due to memorization?
Authors: Wenqing Deng, Zhe Wang, Kewen Wang, Shirui Pan, Xiaowang Zhang, Zhiyong Feng
Abstract: Multiple-choice question answering (MCQA) becomes particularly challenging when all choices are relevant to the question and are semantically similar. Yet this setting of MCQA can potentially provide valuable clues for choosing the right answer. Existing models often rank each choice separately, overlooking the context provided by other choices. Specifically, they fail to leverage the semantic commonalities and nuances among the choices for reasoning. In this paper, we propose a novel MCQA model by differentiating choices through identifying and eliminating their commonality, called DCQA. Our model captures token-level attention of each choice to the question, and separates tokens of the question attended to by all the choices (i.e., commonalities) from those by individual choices (i.e., nuances). Using the nuances as refined contexts for the choices, our model can effectively differentiate choices with subtle differences and provide justifications for choosing the correct answer. We conduct comprehensive experiments across five commonly used MCQA benchmarks, demonstrating that DCQA consistently outperforms baseline models. Furthermore, our case study illustrates the effectiveness of the approach in directing the attention of the model to more differentiating features.
Authors: Ziqiang Li, Yueqi Zeng, Pengfei Xia, Lei Liu, Zhangjie Fu, Bin Li
Abstract: With the burgeoning advancements in the field of natural language processing (NLP), the demand for training data has increased significantly. To save costs, it has become common for users and businesses to outsource the labor-intensive task of data collection to third-party entities. Unfortunately, recent research has unveiled the inherent risk associated with this practice, particularly in exposing NLP systems to potential backdoor attacks. Specifically, these attacks enable malicious control over the behavior of a trained model by poisoning a small portion of the training data. Unlike backdoor attacks in computer vision, textual backdoor attacks impose stringent requirements for attack stealthiness. However, existing attack methods meet significant trade-off between effectiveness and stealthiness, largely due to the high information entropy inherent in textual data. In this paper, we introduce the Efficient and Stealthy Textual backdoor attack method, EST-Bad, leveraging Large Language Models (LLMs). Our EST-Bad encompasses three core strategies: optimizing the inherent flaw of models as the trigger, stealthily injecting triggers with LLMs, and meticulously selecting the most impactful samples for backdoor injection. Through the integration of these techniques, EST-Bad demonstrates an efficient achievement of competitive attack performance while maintaining superior stealthiness compared to prior methods across various text classifier datasets.
Authors: Xinhao Chen, Chong Yang, Man Lan, Li Cai, Yang Chen, Tu Hu, Xinlin Zhuang, Aimin Zhou
Abstract: Empathetic response generation endows agents with the capability to comprehend dialogue contexts and react to expressed emotions. Previous works predominantly focus on leveraging the speaker's emotional labels, but ignore the importance of emotion cause reasoning in empathetic response generation, which hinders the model's capacity for further affective understanding and cognitive inference. In this paper, we propose a cause-aware empathetic generation approach by integrating emotions and causes through a well-designed Chain-of-Thought (CoT) prompt on Large Language Models (LLMs). Our approach can greatly promote LLMs' performance of empathy by instruction tuning and enhancing the role awareness of an empathetic listener in the prompt. Additionally, we propose to incorporate cause-oriented external knowledge from COMET into the prompt, which improves the diversity of generation and alleviates conflicts between internal and external knowledge at the same time. Experimental results on the benchmark dataset demonstrate that our approach on LLaMA-7b achieves state-of-the-art performance in both automatic and human evaluations.
Authors: Yiquan Wu, Bo Tang, Chenyang Xi, Yu Yu, Pengyu Wang, Yifei Liu, Kun Kuang, Haiying Deng, Zhiyu Li, Feiyu Xiong, Jie Hu, Peng Cheng, Zhonghao Wang, Yi Wang, Yi Luo, Mingchuan Yang
Abstract: Commentary provides readers with a deep understanding of events by presenting diverse arguments and evidence. However, creating commentary is a time-consuming task, even for skilled commentators. Large language models (LLMs) have simplified the process of natural language generation, but their direct application in commentary creation still faces challenges due to unique task requirements. These requirements can be categorized into two levels: 1) fundamental requirements, which include creating well-structured and logically consistent narratives, and 2) advanced requirements, which involve generating quality arguments and providing convincing evidence. In this paper, we introduce Xinyu, an efficient LLM-based system designed to assist commentators in generating Chinese commentaries. To meet the fundamental requirements, we deconstruct the generation process into sequential steps, proposing targeted strategies and supervised fine-tuning (SFT) for each step. To address the advanced requirements, we present an argument ranking model for arguments and establish a comprehensive evidence database that includes up-to-date events and classic books, thereby strengthening the substantiation of the evidence with retrieval augmented generation (RAG) technology. To evaluate the generated commentaries more fairly, corresponding to the two-level requirements, we introduce a comprehensive evaluation metric that considers five distinct perspectives in commentary generation. Our experiments confirm the effectiveness of our proposed system. We also observe a significant increase in the efficiency of commentators in real-world scenarios, with the average time spent on creating a commentary dropping from 4 hours to 20 minutes. Importantly, such an increase in efficiency does not compromise the quality of the commentaries.
Authors: Zhenyu Li, Yike Zhang, Tengyu Pan, Yutao Sun, Zhichao Duan, Junjie Fang, Rong Han, Zixuan Wang, Jianyong Wang
Abstract: Empowering LLMs with the ability to utilize useful information from a long context is crucial for many downstream applications. However, achieving long context lengths with the conventional transformer architecture requires substantial training and inference resources. In this paper, we present FocusLLM, a framework designed to extend the context length of any decoder-only LLM, enabling the model to focus on relevant information from very long sequences. FocusLLM processes long text inputs by dividing them into chunks based on the model's original context length to alleviate the issue of attention distraction. Then, it appends the local context to each chunk as a prompt to extract essential information from each chunk based on a novel parallel decoding mechanism, and ultimately integrates the extracted information into the local context. FocusLLM stands out for great training efficiency and versatility: trained with an 8K input length with much less training cost than previous methods, FocusLLM exhibits superior performance across downstream long-context tasks and maintains strong language modeling ability when handling extensive long texts, even up to 400K tokens. Our code is available at https://github.com/leezythu/FocusLLM.
Authors: Yiyi Chen, Russa Biswas, Heather Lent, Johannes Bjerva
Abstract: Large Language Models (LLMs) are susceptible to malicious influence by cyber attackers through intrusions such as adversarial, backdoor, and embedding inversion attacks. In response, the burgeoning field of LLM Security aims to study and defend against such threats. Thus far, the majority of works in this area have focused on monolingual English models, however, emerging research suggests that multilingual LLMs may be more vulnerable to various attacks than their monolingual counterparts. While previous work has investigated embedding inversion over a small subset of European languages, it is challenging to extrapolate these findings to languages from different linguistic families and with differing scripts. To this end, we explore the security of multilingual LLMs in the context of embedding inversion attacks and investigate cross-lingual and cross-script inversion across 20 languages, spanning over 8 language families and 12 scripts. Our findings indicate that languages written in Arabic script and Cyrillic script are particularly vulnerable to embedding inversion, as are languages within the Indo-Aryan language family. We further observe that inversion models tend to suffer from language confusion, sometimes greatly reducing the efficacy of an attack. Accordingly, we systematically explore this bottleneck for inversion models, uncovering predictable patterns which could be leveraged by attackers. Ultimately, this study aims to further the field's understanding of the outstanding security vulnerabilities facing multilingual LLMs and raise awareness for the languages most at risk of negative impact from these attacks.
Authors: Omar Erak, Nouf Alabbasi, Omar Alhussein, Ismail Lotfi, Amr Hussein, Sami Muhaidat, Merouane Debbah
Abstract: Recent studies show that large language models (LLMs) struggle with technical standards in telecommunications. We propose a fine-tuned retrieval-augmented generation (RAG) system based on the Phi-2 small language model (SLM) to serve as an oracle for communication networks. Our developed system leverages forward-looking semantic chunking to adaptively determine parsing breakpoints based on embedding similarity, enabling effective processing of diverse document formats. To handle the challenge of multiple similar contexts in technical standards, we employ a re-ranking algorithm to prioritize the most relevant retrieved chunks. Recognizing the limitations of Phi-2's small context window, we implement a recent technique, namely SelfExtend, to expand the context window during inference, which not only boosts the performance but also can accommodate a wider range of user queries and design requirements from customers to specialized technicians. For fine-tuning, we utilize the low-rank adaptation (LoRA) technique to enhance computational efficiency during training and enable effective fine-tuning on small datasets. Our comprehensive experiments demonstrate substantial improvements over existing question-answering approaches in the telecom domain, achieving performance that exceeds larger language models such as GPT-4 (which is about 880 times larger in size). This work presents a novel approach to leveraging SLMs for communication networks, offering a balance of efficiency and performance. This work can serve as a foundation towards agentic language models for networks.
Authors: Minjun Zhu, Linyi Yang, Yue Zhang
Abstract: Current methods for aligning large language models (LLMs) typically aim to reflect general human values and behaviors, but they often fail to capture the unique characteristics and preferences of individual users. To address this gap, we introduce the concept of Personality Alignment. This approach tailors LLMs' responses and decisions to match the specific preferences of individual users or closely related groups. Inspired by psychometrics, we created the Personality Alignment with Personality Inventories (PAPI) dataset, which includes data from 300,000 real subjects, each providing behavioral preferences based on the Big Five Personality Factors. This dataset allows us to quantitatively evaluate the extent to which LLMs can align with each subject's behavioral patterns. Recognizing the challenges of personality alignments: such as limited personal data, diverse preferences, and scalability requirements: we developed an activation intervention optimization method. This method enhances LLMs' ability to efficiently align with individual behavioral preferences using minimal data and computational resources. Remarkably, our method, PAS, achieves superior performance while requiring only 1/5 of the optimization time compared to DPO, offering practical value for personality alignment. Our work paves the way for future AI systems to make decisions and reason in truly personality ways, enhancing the relevance and meaning of AI interactions for each user and advancing human-centered artificial intelligence.The code has released in \url{https://github.com/zhu-minjun/PAlign}.
Authors: Sharath Turuvekere Sreenivas, Saurav Muralidharan, Raviraj Joshi, Marcin Chochowski, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro, Jan Kautz, Pavlo Molchanov
Abstract: We present a comprehensive report on compressing the Llama 3.1 8B and Mistral NeMo 12B models to 4B and 8B parameters, respectively, using pruning and distillation. We explore two distinct pruning strategies: (1) depth pruning and (2) joint hidden/attention/MLP (width) pruning, and evaluate the results on common benchmarks from the LM Evaluation Harness. The models are then aligned with NeMo Aligner and tested in instruct-tuned versions. This approach produces a compelling 4B model from Llama 3.1 8B and a state-of-the-art Mistral-NeMo-Minitron-8B (MN-Minitron-8B for brevity) model from Mistral NeMo 12B. We found that with no access to the original data, it is beneficial to slightly fine-tune teacher models on the distillation dataset. We open-source our base model weights on Hugging Face with a permissive license.
Authors: Haode Qi, Cheng Qian, Jian Ni, Pratyush Singh, Reza Fazeli, Gengyu Wang, Zhongzheng Shu, Eric Wayne, Juergen Bross
Abstract: In an enterprise Virtual Assistant (VA) system, intent classification is the crucial component that determines how a user input is handled based on what the user wants. The VA system is expected to be a cost-efficient SaaS service with low training and inference time while achieving high accuracy even with a small number of training samples. We pretrain a transformer-based sentence embedding model with a contrastive learning objective and leverage the embedding of the model as features when training intent classification models. Our approach achieves the state-of-the-art results for few-shot scenarios and performs better than other commercial solutions on popular intent classification benchmarks. However, generating features via a transformer-based model increases the inference time, especially for longer user inputs, due to the quadratic runtime of the transformer's attention mechanism. On top of model distillation, we introduce a practical multi-task adaptation approach that configures dynamic token pruning without the need for task-specific training for intent classification. We demonstrate that this approach improves the inference speed of popular sentence transformer models without affecting model performance.
Authors: Rounak Meyur, Hung Phan, Sridevi Wagle, Jan Strube, Mahantesh Halappanavar, Sameera Horawalavithana, Anurag Acharya, Sai Munikoti
Abstract: In the rapidly evolving landscape of Natural Language Processing (NLP) and text generation, the emergence of Retrieval Augmented Generation (RAG) presents a promising avenue for improving the quality and reliability of generated text by leveraging information retrieved from user specified database. Benchmarking is essential to evaluate and compare the performance of the different RAG configurations in terms of retriever and generator, providing insights into their effectiveness, scalability, and suitability for the specific domain and applications. In this paper, we present a comprehensive framework to generate a domain relevant RAG benchmark. Our framework is based on automatic question-answer generation with Human (domain experts)-AI Large Language Model (LLM) teaming. As a case study, we demonstrate the framework by introducing PermitQA, a first-of-its-kind benchmark on the wind siting and permitting domain which comprises of multiple scientific documents/reports related to environmental impact of wind energy projects. Our framework systematically evaluates RAG performance using diverse metrics and multiple question types with varying complexity level. We also demonstrate the performance of different models on our benchmark.
Authors: Shangyi Geng, Wenting Zhao, Alexander M Rush
Abstract: $K$-nearest neighbor language models ($k$NN-LMs), which integrate retrieval with next-word prediction, have demonstrated strong performance in language modeling as well as downstream NLP benchmarks. These results have led researchers to argue that models trained on poor quality or outdated data could perform well by employing a $k$NN extension that has access to a higher-quality datastore. In this work, we ask whether this improved ability to recall information really translates into downstream abilities. We extensively evaluate $k$NN-LMs on a diverse set of tasks, ranging from sentiment classification and commonsense reasoning to multi-hop reasoning. Results show that $k$NN-LMs excel at memory-intensive tasks, where utilizing the patterns in the input is sufficient for determining the output, but struggle with reasoning tasks that require integrating multiple pieces of information to derive new knowledge. We further demonstrate through oracle experiments and qualitative analysis that even with perfect retrieval, $k$NN-LMs still fail to determine the correct answers, placing an upper bound on their reasoning performance. Code and datastores are released at https://github.com/GSYfate/knnlm-limits/.
Authors: Sarthak Jain (University of Illinois Urbana Champaign,Cisco), Aditya Dora (University of Illinois Urbana Champaign), Ka Seng Sam (University of Illinois Urbana Champaign), Prabhat Singh (Cisco)
Abstract: Code Search is a key task that many programmers often have to perform while developing solutions to problems. Current methodologies suffer from an inability to perform accurately on prompts that contain some ambiguity or ones that require additional context relative to a code-base. We introduce the approach of using Retrieval Augmented Generation (RAG) powered agents to inject information into user prompts allowing for better inputs into embedding models. By utilizing RAG, agents enhance user queries with relevant details from GitHub repositories, making them more informative and contextually aligned. Additionally, we introduce a multi-stream ensemble approach which when paired with agentic workflow can obtain improved retrieval accuracy, which we deploy on application called repo-rift.com. Experimental results on the CodeSearchNet dataset demonstrate that RepoRift significantly outperforms existing methods, achieving an 78.2% success rate at Success@10 and a 34.6% success rate at Success@1. This research presents a substantial advancement in semantic code search, highlighting the potential of agentic LLMs and RAG to enhance code retrieval systems.
Authors: Andrei Constantin, Deaglan Bartlett, Harry Desmond, Pedro G. Ferreira
Abstract: Physics, as a fundamental science, aims to understand the laws of Nature and describe them in mathematical equations. While the physical reality manifests itself in a wide range of phenomena with varying levels of complexity, the equations that describe them display certain statistical regularities and patterns, which we begin to explore here. By drawing inspiration from linguistics, where Zipf's law states that the frequency of any word in a large corpus of text is roughly inversely proportional to its rank in the frequency table, we investigate whether similar patterns for the distribution of operators emerge in the equations of physics. We analyse three corpora of formulae and find, using sophisticated implicit-likelihood methods, that the frequency of operators as a function of their rank in the frequency table is best described by an exponential law with a stable exponent, in contrast with Zipf's inverse power-law. Understanding the underlying reasons behind this statistical pattern may shed light on Nature's modus operandi or reveal recurrent patterns in physicists' attempts to formalise the laws of Nature. It may also provide crucial input for symbolic regression, potentially augmenting language models to generate symbolic models for physical phenomena. By pioneering the study of statistical regularities in the equations of physics, our results open the door for a meta-law of Nature, a (probabilistic) law that all physical laws obey.
Authors: Nickil Maveli, Antonio Vergari, Shay B. Cohen
Abstract: Code-LLMs, LLMs pre-trained on large code corpora, have shown great progress in learning rich representations of the structure and syntax of code, successfully using it to generate or classify code fragments. At the same time, understanding if they are able to do so because they capture code semantics, and how well, is still an open question. In this paper, we tackle this problem by introducing SeqCoBench, a benchmark for systematically assessing how Code-LLMs can capture code functional equivalence. SeqCoBench contains over 20 code transformations that either preserve or alter the semantics of Python programs. We conduct extensive evaluations in different settings, including zero-shot and parameter-efficient finetuning methods on state-of-the-art (Code-)LLMs to see if they can discern semantically equivalent or different pairs of programs in SeqCoBench. We find that the performance gap between these LLMs and classical match-based retrieval scores is minimal, with both approaches showing a concerning lack of depth in understanding code semantics.
Authors: Meet Doshi, Vishwajeet Kumar, Rudra Murthy, Vignesh P, Jaydeep Sen
Abstract: Learned Sparse Retrievers (LSR) have evolved into an effective retrieval strategy that can bridge the gap between traditional keyword-based sparse retrievers and embedding-based dense retrievers. At its core, learned sparse retrievers try to learn the most important semantic keyword expansions from a query and/or document which can facilitate better retrieval with overlapping keyword expansions. LSR like SPLADE has typically been using encoder only models with MLM (masked language modeling) style objective in conjunction with known ways of retrieval performance improvement such as hard negative mining, distillation, etc. In this work, we propose to use decoder-only model for learning semantic keyword expansion. We posit, decoder only models that have seen much higher magnitudes of data are better equipped to learn keyword expansions needed for improved retrieval. We use Mistral as the backbone to develop our Learned Sparse Retriever similar to SPLADE and train it on a subset of sentence-transformer data which is often used for training text embedding models. Our experiments support the hypothesis that a sparse retrieval model based on decoder only large language model (LLM) surpasses the performance of existing LSR systems, including SPLADE and all its variants. The LLM based model (Echo-Mistral-SPLADE) now stands as a state-of-the-art learned sparse retrieval model on the BEIR text retrieval benchmark.
Authors: Tom Segal, Asaf Shabtai, Yuval Elovici
Abstract: The utility of large language models (LLMs) depends heavily on the quality and quantity of their training data. Many organizations possess large data corpora that could be leveraged to train or fine-tune LLMs tailored to their specific needs. However, these datasets often come with access restrictions that are based on user privileges and enforced by access control mechanisms. Training LLMs on such datasets could result in exposure of sensitive information to unauthorized users. A straightforward approach for preventing such exposure is to train a separate model for each access level. This, however, may result in low utility models due to the limited amount of training data per model compared to the amount in the entire organizational corpus. Another approach is to train a single LLM on all the data while limiting the exposure of unauthorized information. However, current exposure-limiting methods for LLMs are ineffective for access-controlled data, where sensitive information appears frequently across many training examples. We propose DOMBA - double model balancing - a simple approach for training and deploying LLMs that provides high utility and access-control functionality with security guarantees. DOMBA aggregates the probability distributions of two models, each trained on documents with (potentially many) different access levels, using a "min-bounded" average function (a function that is bounded by the smaller value, e.g., harmonic mean). A detailed mathematical analysis and extensive evaluation show that DOMBA safeguards restricted information while offering utility comparable to non-secure models.
Authors: Thomas Hoang, Quynh Anh Nguyen, Long Nguyen
Abstract: Countless disasters have resulted from climate change, causing severe damage to infrastructure and the economy. These disasters have significant societal impacts, necessitating mental health services for the millions affected. To prepare for and respond effectively to such events, it is important to understand people's emotions and the life incidents they experience before and after a disaster strikes. In this case study, we collected a dataset of approximately 400,000 public tweets related to the storm. Using a BERT-based model, we predicted the emotions associated with each tweet. To efficiently identify these topics, we utilized the Latent Dirichlet Allocation (LDA) technique for topic modeling, which allowed us to bypass manual content analysis and extract meaningful patterns from the data. However, rather than stopping at topic identification like previous methods \cite{math11244910}, we further refined our analysis by integrating Graph Neural Networks (GNN) and Large Language Models (LLM). The GNN was employed to generate embeddings and construct a similarity graph of the tweets, which was then used to optimize clustering. Subsequently, we used an LLM to automatically generate descriptive names for each event cluster, offering critical insights for disaster preparedness and response strategies.
Authors: Xueliang Zhao, Lin Zheng, Haige Bo, Changran Hu, Urmish Thakker, Lingpeng Kong
Abstract: Formal theorem proving, a field at the intersection of mathematics and computer science, has seen renewed interest with advancements in large language models (LLMs). This paper introduces SubgoalXL, a novel approach that synergizes subgoal-based proofs with expert learning to enhance LLMs' capabilities in formal theorem proving within the Isabelle environment. SubgoalXL addresses two critical challenges: the scarcity of specialized mathematics and theorem-proving data, and the need for improved multi-step reasoning abilities in LLMs. By optimizing data efficiency and employing subgoal-level supervision, SubgoalXL extracts richer information from limited human-generated proofs. The framework integrates subgoal-oriented proof strategies with an expert learning system, iteratively refining formal statement, proof, and subgoal generators. Leveraging the Isabelle environment's advantages in subgoal-based proofs, SubgoalXL achieves a new state-of-the-art performance of 56.1\% in Isabelle on the standard miniF2F dataset, marking an absolute improvement of 4.9\%. Notably, SubgoalXL successfully solves 41 AMC12, 9 AIME, and 3 IMO problems from miniF2F. These results underscore the effectiveness of maximizing limited data utility and employing targeted guidance for complex reasoning in formal theorem proving, contributing to the ongoing advancement of AI reasoning capabilities. The implementation is available at \url{https://github.com/zhaoxlpku/SubgoalXL}.
Authors: Abhinav Kumar, Atharva Khedkar, Aviral Shrivastava
Abstract: Traditional Digital Signal Processing ( DSP ) compilers work at low level ( C-level / assembly level ) and hence lose much of the optimization opportunities present at high-level ( domain-level ). The emerging multi-level compiler infrastructure MLIR ( Multi-level Intermediate Representation ) allows to specify optimizations at higher level. In this paper, we utilize MLIR framework to introduce a DSP Dialect and perform domain-specific optimizations at dialect -level ( high-level ) and show the usefulness of these optimizations on sample DSP apps. In particular, we develop a compiler for DSP and a DSL (Domain Specific Language) to ease the development of apps. We show the performance improvement in execution time for these sample apps by upto 10x which would have been difficult if the IR were at C/ affine level.
Authors: Christos Constantinou, Georgios Ioannides, Aman Chadha, Aaron Elkins, Edwin Simpson
Abstract: Detecting out-of-distribution (OOD) data is crucial in machine learning applications to mitigate the risk of model overconfidence, thereby enhancing the reliability and safety of deployed systems. The majority of existing OOD detection methods predominantly address uni-modal inputs, such as images or texts. In the context of multi-modal documents, there is a notable lack of extensive research on the performance of these methods, which have primarily been developed with a focus on computer vision tasks. We propose a novel methodology termed as attention head masking (AHM) for multi-modal OOD tasks in document classification systems. Our empirical results demonstrate that the proposed AHM method outperforms all state-of-the-art approaches and significantly decreases the false positive rate (FPR) compared to existing solutions up to 7.5\%. This methodology generalizes well to multi-modal data, such as documents, where visual and textual information are modeled under the same Transformer architecture. To address the scarcity of high-quality publicly available document datasets and encourage further research on OOD detection for documents, we introduce FinanceDocs, a new document AI dataset. Our code and dataset are publicly available.
Authors: John Scoville, Shang Gao, Devanshu Agrawal, Javed Qadrud-Din
Abstract: We introduce a group of related methods for binary classification tasks using probes of the hidden state activations in large language models (LLMs). Performance is on par with the largest and most advanced LLMs currently available, but requiring orders of magnitude fewer computational resources and not requiring labeled data. This approach involves translating class labels into a semantically rich description, spontaneous symmetry breaking of multilayer perceptron probes for unsupervised learning and inference, training probes to generate confidence scores (prior probabilities) from hidden state activations subject to known constraints via entropy maximization, and selecting the most confident probe model from an ensemble for prediction. These techniques are evaluated on four datasets using five base LLMs.
Authors: Prashant Serai, Peidong Wang, Eric Fosler-Lussier
Abstract: Modeling the errors of a speech recognizer can help simulate errorful recognized speech data from plain text, which has proven useful for tasks like discriminative language modeling, improving robustness of NLP systems, where limited or even no audio data is available at train time. Previous work typically considered replicating behavior of GMM-HMM based systems, but the behavior of more modern posterior-based neural network acoustic models is not the same and requires adjustments to the error prediction model. In this work, we extend a prior phonetic confusion based model for predicting speech recognition errors in two ways: first, we introduce a sampling-based paradigm that better simulates the behavior of a posterior-based acoustic model. Second, we investigate replacing the confusion matrix with a sequence-to-sequence model in order to introduce context dependency into the prediction. We evaluate the error predictors in two ways: first by predicting the errors made by a Switchboard ASR system on unseen data (Fisher), and then using that same predictor to estimate the behavior of an unrelated cloud-based ASR system on a novel task. Sampling greatly improves predictive accuracy within a 100-guess paradigm, while the sequence model performs similarly to the confusion matrix.
Authors: Yunpu Zhao, Rui Zhang, Junbin Xiao, Changxin Ke, Ruibo Hou, Yifan Hao, Qi Guo, Yunji Chen
Abstract: Large Vision-Language Models (LVLMs) have shown significant capability in vision-language understanding. However, one critical issue that persists in these models is sycophancy, which means models are unduly influenced by leading or deceptive prompts, resulting in biased outputs and hallucinations. Despite the progress in LVLMs, evaluating and mitigating sycophancy is yet much under-explored. In this work, we fill this gap by systematically analyzing sycophancy on various VL benchmarks with curated leading queries and further proposing a text contrastive decoding method for mitigation. While the specific sycophantic behavior varies significantly among models, our analysis reveals the severe deficiency of all LVLMs in resilience of sycophancy across various tasks. For improvement, we propose Leading Query Contrastive Decoding (LQCD), a model-agnostic method focusing on calibrating the LVLMs' over-reliance on leading cues by identifying and suppressing the probabilities of sycophancy tokens at the decoding stage. Extensive experiments show that LQCD effectively mitigate sycophancy, outperforming both prompt engineering methods and common methods for hallucination mitigation. We further demonstrate that LQCD does not hurt but even slightly improves LVLMs' responses to neutral queries, suggesting it being a more effective strategy for general-purpose decoding but not limited to sycophancy.
Authors: Yuze Zhao, Zhenya Huang, Yixiao Ma, Rui Li, Kai Zhang, Hao Jiang, Qi Liu, Linbo Zhu, Yu Su
Abstract: The gap between the trepidation of program reliability and the expense of repairs underscores the indispensability of Automated Program Repair (APR). APR is instrumental in transforming vulnerable programs into more robust ones, bolstering program reliability while simultaneously diminishing the financial burden of manual repairs. Commercial-scale language models (LM) have taken APR to unprecedented levels. However, the emergence reveals that for models fewer than 100B parameters, making single-step modifications may be difficult to achieve the desired effect. Moreover, humans interact with the LM through explicit prompts, which hinders the LM from receiving feedback from compiler and test cases to automatically optimize its repair policies. In this literature, we explore how small-scale LM (less than 20B) achieve excellent performance through process supervision and feedback. We start by constructing a dataset named CodeNet4Repair, replete with multiple repair records, which supervises the fine-tuning of a foundational model. Building upon the encouraging outcomes of reinforcement learning, we develop a reward model that serves as a critic, providing feedback for the fine-tuned LM's action, progressively optimizing its policy. During inference, we require the LM to generate solutions iteratively until the repair effect no longer improves or hits the maximum step limit. The results show that process-based not only outperforms larger outcome-based generation methods, but also nearly matches the performance of closed-source commercial large-scale LMs.
Authors: Chongwen Zhao, Zhihao Dou, Kaizhu Huang
Abstract: Large Language Models (LLMs) are increasingly attracting attention in various applications. Nonetheless, there is a growing concern as some users attempt to exploit these models for malicious purposes, including the synthesis of controlled substances and the propagation of disinformation. In an effort to mitigate such risks, the concept of "Alignment" technology has been developed. However, recent studies indicate that this alignment can be undermined using sophisticated prompt engineering or adversarial suffixes, a technique known as "Jailbreak." Our research takes cues from the human-like generate process of LLMs. We identify that while jailbreaking prompts may yield output logits similar to benign prompts, their initial embeddings within the model's latent space tend to be more analogous to those of malicious prompts. Leveraging this finding, we propose utilizing the early transformer outputs of LLMs as a means to detect malicious inputs, and terminate the generation immediately. Built upon this idea, we introduce a simple yet significant defense approach called EEG-Defender for LLMs. We conduct comprehensive experiments on ten jailbreak methods across three models. Our results demonstrate that EEG-Defender is capable of reducing the Attack Success Rate (ASR) by a significant margin, roughly 85\% in comparison with 50\% for the present SOTAs, with minimal impact on the utility and effectiveness of LLMs.
Authors: Xun Zhou, Liang Feng, Xingyu Wu, Zhichao Lu, Kay Chen Tan
Abstract: Transferable neural architecture search (TNAS) has been introduced to design efficient neural architectures for multiple tasks, to enhance the practical applicability of NAS in real-world scenarios. In TNAS, architectural knowledge accumulated in previous search processes is reused to warm up the architecture search for new tasks. However, existing TNAS methods still search in an extensive search space, necessitating the evaluation of numerous architectures. To overcome this challenge, this work proposes a novel transfer paradigm, i.e., design principle transfer. In this work, the linguistic description of various structural components' effects on architectural performance is termed design principles. They are learned from established architectures and then can be reused to reduce the search space by discarding unpromising architectures. Searching in the refined search space can boost both the search performance and efficiency for new NAS tasks. To this end, a large language model (LLM)-assisted design principle transfer (LAPT) framework is devised. In LAPT, LLM is applied to automatically reason the design principles from a set of given architectures, and then a principle adaptation method is applied to refine these principles progressively based on the new search results. Experimental results show that LAPT can beat the state-of-the-art TNAS methods on most tasks and achieve comparable performance on others.
Authors: Martin Pichlmair, Riddhi Raj, Charlene Putney
Abstract: This technical report presents the Drama Engine, a novel framework for agentic interaction with large language models designed for narrative purposes. The framework adapts multi-agent system principles to create dynamic, context-aware companions that can develop over time and interact with users and each other. Key features include multi-agent workflows with delegation, dynamic prompt assembly, and model-agnostic design. The Drama Engine introduces unique elements such as companion development, mood systems, and automatic context summarising. It is implemented in TypeScript. The framework's applications include multi-agent chats and virtual co-workers for creative writing. The paper discusses the system's architecture, prompt assembly process, delegation mechanisms, and moderation techniques, as well as potential ethical considerations and future extensions.
Authors: Yi Liu, Junzhe Yu, Huijia Sun, Ling Shi, Gelei Deng, Yuqi Chen, Yang Liu
Abstract: Large language models (LLMs) like ChatGPT and Gemini have significantly advanced natural language processing, enabling various applications such as chatbots and automated content generation. However, these models can be exploited by malicious individuals who craft toxic prompts to elicit harmful or unethical responses. These individuals often employ jailbreaking techniques to bypass safety mechanisms, highlighting the need for robust toxic prompt detection methods. Existing detection techniques, both blackbox and whitebox, face challenges related to the diversity of toxic prompts, scalability, and computational efficiency. In response, we propose ToxicDetector, a lightweight greybox method designed to efficiently detect toxic prompts in LLMs. ToxicDetector leverages LLMs to create toxic concept prompts, uses embedding vectors to form feature vectors, and employs a Multi-Layer Perceptron (MLP) classifier for prompt classification. Our evaluation on various versions of the LLama models, Gemma-2, and multiple datasets demonstrates that ToxicDetector achieves a high accuracy of 96.39\% and a low false positive rate of 2.00\%, outperforming state-of-the-art methods. Additionally, ToxicDetector's processing time of 0.0780 seconds per prompt makes it highly suitable for real-time applications. ToxicDetector achieves high accuracy, efficiency, and scalability, making it a practical method for toxic prompt detection in LLMs.
Authors: Zhifei Xie, Daniel Tang, Dingwei Tan, Jacques Klein, Tegawend F. Bissyand, Saad Ezzini
Abstract: Current video generation models excel at creating short, realistic clips, but struggle with longer, multi-scene videos. We introduce \texttt{DreamFactory}, an LLM-based framework that tackles this challenge. \texttt{DreamFactory} leverages multi-agent collaboration principles and a Key Frames Iteration Design Method to ensure consistency and style across long videos. It utilizes Chain of Thought (COT) to address uncertainties inherent in large language models. \texttt{DreamFactory} generates long, stylistically coherent, and complex videos. Evaluating these long-form videos presents a challenge. We propose novel metrics such as Cross-Scene Face Distance Score and Cross-Scene Style Consistency Score. To further research in this area, we contribute the Multi-Scene Videos Dataset containing over 150 human-rated videos.
Authors: Mehrdad Ranjbar Khadivi, Shahin Akbarpour, Mohammad-Reza Feizi-Derakhshi, Babak Anari
Abstract: With the widespread use of social networks, detecting the topics discussed on these platforms has become a significant challenge. Current approaches primarily rely on frequent pattern mining or semantic relations, often neglecting the structure of the language. Language structural methods aim to discover the relationships between words and how humans understand them. Therefore, this paper introduces a topic detection framework for social networks based on the concept of imitating the mental ability of word association. This framework employs the Human Word Association method and includes a specially designed extraction algorithm. The performance of this method is evaluated using the FA-CUP dataset, a benchmark in the field of topic detection. The results indicate that the proposed method significantly improves topic detection compared to other methods, as evidenced by Topic-recall and the keyword F1 measure. Additionally, to assess the applicability and generalizability of the proposed method, a dataset of Telegram posts in the Persian language is used. The results demonstrate that this method outperforms other topic detection methods.
Authors: Adam Davies, Jize Jiang, ChengXiang Zhai
Abstract: Despite the recent successes of large, pretrained neural language models (LLMs), comparatively little is known about the representations of linguistic structure they learn during pretraining, which can lead to unexpected behaviors in response to prompt variation or distribution shift. To better understand these models and behaviors, we introduce a general model analysis framework to study LLMs with respect to their representation and use of human-interpretable linguistic properties. Our framework, CALM (Competence-based Analysis of Language Models), is designed to investigate LLM competence in the context of specific tasks by intervening on models' internal representations of different linguistic properties using causal probing, and measuring models' alignment under these interventions with a given ground-truth causal model of the task. We also develop a new approach for performing causal probing interventions using gradient-based adversarial attacks, which can target a broader range of properties and representations than prior techniques. Finally, we carry out a case study of CALM using these interventions to analyze and compare LLM competence across a variety of lexical inference tasks, showing that CALM can be used to explain and predict behaviors across these tasks.
Authors: Ronja Stern, Vishvaksenan Rasiah, Veton Matoshi, Srinanda Br\"ugger Bose, Matthias St\"urmer, Ilias Chalkidis, Daniel E. Ho, Joel Niklaus
Abstract: Recent strides in Large Language Models (LLMs) have saturated many Natural Language Processing (NLP) benchmarks, emphasizing the need for more challenging ones to properly assess LLM capabilities. However, domain-specific and multilingual benchmarks are rare because they require in-depth expertise to develop. Still, most public models are trained predominantly on English corpora, while other languages remain understudied, particularly for practical domain-specific NLP tasks. In this work, we introduce a novel NLP benchmark for the legal domain that challenges LLMs in five key dimensions: processing \emph{long documents} (up to 50K tokens), using \emph{domain-specific knowledge} (embodied in legal texts), \emph{multilingual} understanding (covering five languages), \emph{multitasking} (comprising legal document-to-document Information Retrieval, Court View Generation, Leading Decision Summarization, Citation Extraction, and eight challenging Text Classification tasks) and \emph{reasoning} (comprising especially Court View Generation, but also the Text Classification tasks). Our benchmark contains diverse datasets from the Swiss legal system, allowing for a comprehensive study of the underlying non-English, inherently multilingual legal system. Despite the large size of our datasets (some with hundreds of thousands of examples), existing publicly available multilingual models struggle with most tasks, even after extensive in-domain pre-training and fine-tuning. We publish all resources (benchmark suite, pre-trained models, code) under permissive open CC BY-SA licenses.
Authors: Tengchao Lv, Yupan Huang, Jingye Chen, Yuzhong Zhao, Yilin Jia, Lei Cui, Shuming Ma, Yaoyao Chang, Shaohan Huang, Wenhui Wang, Li Dong, Weiyao Luo, Shaoxiang Wu, Guoxin Wang, Cha Zhang, Furu Wei
Abstract: The automatic reading of text-intensive images represents a significant advancement toward achieving Artificial General Intelligence (AGI). In this paper we present KOSMOS-2.5, a multimodal literate model for machine reading of text-intensive images. Pre-trained on a large-scale corpus of text-intensive images, KOSMOS-2.5 excels in two distinct yet complementary transcription tasks: (1) generating spatially-aware text blocks, where each block of text is assigned spatial coordinates within the image, and (2) producing structured text output that captures both style and structure in markdown format. This unified multimodal literate capability is achieved through a shared decoder-only autoregressive Transformer architecture and task-specific prompts. Building on this foundation, we fine-tune KOSMOS-2.5 for document understanding tasks, resulting in a document understanding generalist named KOSMOS-2.5-CHAT. Additionally, a large corpus of 357.4 million document pages spanning diverse domains was curated for pre-training. We evaluate KOSMOS-2.5 on two newly proposed benchmarks, OCREval and MarkdownEval, for document-level text recognition and image-to-markdown generation, demonstrating impressive literate capabilities comparable to GPT-4o. KOSMOS-2.5-CHAT achieves performance comparable to other state-of-the-art generalists that are five times larger (1.3B vs. 7B) across nine text-rich visual question answering benchmarks. Models and code have been available at \url{https://aka.ms/kosmos25}.
URLs: https://aka.ms/kosmos25
Authors: Shanglin Lei, Guanting Dong, Xiaoping Wang, Keheng Wang, Sirui Wang
Abstract: The field of emotion recognition of conversation (ERC) has been focusing on separating sentence feature encoding and context modeling, lacking exploration in generative paradigms based on unified designs. In this study, we propose a novel approach, InstructERC, to reformulate the ERC task from a discriminative framework to a generative framework based on Large Language Models (LLMs). InstructERC makes three significant contributions: (1) it introduces a simple yet effective retrieval template module, which helps the model explicitly integrate multi-granularity dialogue supervision information. (2) We introduce two additional emotion alignment tasks, namely speaker identification and emotion prediction tasks, to implicitly model the dialogue role relationships and future emotional tendencies in conversations. (3) Pioneeringly, we unify emotion labels across benchmarks through the feeling wheel to fit real application scenarios. InstructERC still perform impressively on this unified dataset. Our LLM-based plugin framework significantly outperforms all previous models and achieves comprehensive SOTA on three commonly used ERC datasets. Extensive analysis of parameter-efficient and data-scaling experiments provides empirical guidance for applying it in practical scenarios.
Authors: Xiangru Tang, Yuliang Liu, Zefan Cai, Yanjun Shao, Junjie Lu, Yichi Zhang, Zexuan Deng, Helan Hu, Kaikai An, Ruijun Huang, Shuzheng Si, Sheng Chen, Haozhe Zhao, Liang Chen, Yan Wang, Tianyu Liu, Zhiwei Jiang, Baobao Chang, Yin Fang, Yujia Qin, Wangchunshu Zhou, Yilun Zhao, Arman Cohan, Mark Gerstein
Abstract: Despite Large Language Models (LLMs) like GPT-4 achieving impressive results in function-level code generation, they struggle with repository-scale code understanding (e.g., coming up with the right arguments for calling routines), requiring a deeper comprehension of complex file interactions. Also, recently, people have developed LLM agents that attempt to interact with repository code (e.g., compiling and evaluating its execution), prompting the need to evaluate their performance. These gaps have motivated our development of ML-Bench, a benchmark rooted in real-world programming applications that leverage existing code repositories to perform tasks. Addressing the need for LLMs to interpret long code contexts and translate instructions into precise, executable scripts, ML-Bench encompasses annotated 9,641 examples across 18 GitHub repositories, challenging LLMs to accommodate user-specified arguments and documentation intricacies effectively. To evaluate both LLMs and AI agents, two setups are employed: ML-LLM-Bench for assessing LLMs' text-to-code conversion within a predefined deployment environment, and ML-Agent-Bench for testing autonomous agents in an end-to-end task execution within a Linux sandbox environment. Our findings indicate that while GPT-4o leads with a Pass@5 rate surpassing 50%, there remains significant scope for improvement, highlighted by issues such as hallucinated outputs and difficulties with bash script generation. Notably, in the more demanding ML-Agent-Bench, GPT-4o achieves a 76.47% success rate, reflecting the efficacy of iterative action and feedback in complex task resolution. Our code, dataset, and models are available at https://github.com/gersteinlab/ML-bench.
Authors: Yining Hua, Fenglin Liu, Kailai Yang, Zehan Li, Hongbin Na, Yi-han Sheu, Peilin Zhou, Lauren V. Moran, Sophia Ananiadou, Andrew Beam, John Torous
Abstract: The integration of large language models (LLMs) in mental health care is an emerging field. There is a need to systematically review the application outcomes and delineate the advantages and limitations in clinical settings. This review aims to provide a comprehensive overview of the use of LLMs in mental health care, assessing their efficacy, challenges, and potential for future applications. A systematic search was conducted across multiple databases including PubMed, Web of Science, Google Scholar, arXiv, medRxiv, and PsyArXiv in November 2023. All forms of original research, peer-reviewed or not, published or disseminated between October 1, 2019, and December 2, 2023, are included without language restrictions if they used LLMs developed after T5 and directly addressed research questions in mental health care settings. From an initial pool of 313 articles, 34 met the inclusion criteria based on their relevance to LLM application in mental health care and the robustness of reported outcomes. Diverse applications of LLMs in mental health care are identified, including diagnosis, therapy, patient engagement enhancement, etc. Key challenges include data availability and reliability, nuanced handling of mental states, and effective evaluation methods. Despite successes in accuracy and accessibility improvement, gaps in clinical applicability and ethical considerations were evident, pointing to the need for robust data, standardized evaluations, and interdisciplinary collaboration. LLMs hold substantial promise for enhancing mental health care. For their full potential to be realized, emphasis must be placed on developing robust datasets, development and evaluation frameworks, ethical guidelines, and interdisciplinary collaborations to address current limitations.
Authors: Chenkai Sun, Ke Yang, Revanth Gangi Reddy, Yi R. Fung, Hou Pong Chan, Kevin Small, ChengXiang Zhai, Heng Ji
Abstract: The increasing demand for personalized interactions with large language models (LLMs) calls for methodologies capable of accurately and efficiently identifying user opinions and preferences. Retrieval augmentation emerges as an effective strategy, as it can accommodate a vast number of users without the costs from fine-tuning. Existing research, however, has largely focused on enhancing the retrieval stage and devoted limited exploration toward optimizing the representation of the database, a crucial aspect for tasks such as personalization. In this work, we examine the problem from a novel angle, focusing on how data can be better represented for more data-efficient retrieval in the context of LLM customization. To tackle this challenge, we introduce Persona-DB, a simple yet effective framework consisting of a hierarchical construction process to improve generalization across task contexts and collaborative refinement to effectively bridge knowledge gaps among users. In the evaluation of response prediction, Persona-DB demonstrates superior context efficiency in maintaining accuracy with a significantly reduced retrieval size, a critical advantage in scenarios with extensive histories or limited context windows. Our experiments also indicate a marked improvement of over 10% under cold-start scenarios, when users have extremely sparse data. Furthermore, our analysis reveals the increasing importance of collaborative knowledge as the retrieval capacity expands.
Authors: Preetam Prabhu Srikar Dammu, Himanshu Naidu, Mouly Dewan, YoungMin Kim, Tanya Roosta, Aman Chadha, Chirag Shah
Abstract: In the midst of widespread misinformation and disinformation through social media and the proliferation of AI-generated texts, it has become increasingly difficult for people to validate and trust information they encounter. Many fact-checking approaches and tools have been developed, but they often lack appropriate explainability or granularity to be useful in various contexts. A text validation method that is easy to use, accessible, and can perform fine-grained evidence attribution has become crucial. More importantly, building user trust in such a method requires presenting the rationale behind each prediction, as research shows this significantly influences people's belief in automated systems. Localizing and bringing users' attention to the specific problematic content is also paramount, instead of providing simple blanket labels. In this paper, we present ClaimVer, a human-centric framework tailored to meet users' informational and verification needs by generating rich annotations and thereby reducing cognitive load. Designed to deliver comprehensive evaluations of texts, it highlights each claim, verifies it against a trusted knowledge graph (KG), presents the evidence, and provides succinct, clear explanations for each claim prediction. Finally, our framework introduces an attribution score, enhancing applicability across a wide range of downstream tasks.
Authors: Mathieu Ravaut, Bosheng Ding, Fangkai Jiao, Hailin Chen, Xingxuan Li, Ruochen Zhao, Chengwei Qin, Caiming Xiong, Shafiq Joty
Abstract: With the rise of Large Language Models (LLMs) in recent years, abundant new opportunities are emerging, but also new challenges, among which contamination is quickly becoming critical. Business applications and fundraising in AI have reached a scale at which a few percentage points gained on popular question-answering benchmarks could translate into dozens of millions of dollars, placing high pressure on model integrity. At the same time, it is becoming harder and harder to keep track of the data that LLMs have seen; if not impossible with closed-source models like GPT-4 and Claude-3 not divulging any information on the training set. As a result, contamination becomes a major issue: LLMs' performance may not be reliable anymore, as the high performance may be at least partly due to their previous exposure to the data. This limitation jeopardizes the entire progress in the field of NLP, yet, there remains a lack of methods on how to efficiently detect contamination.In this paper, we survey all recent work on contamination detection with LLMs, and help the community track contamination levels of LLMs by releasing an open-source Python library named LLMSanitize implementing major contamination detection algorithms.
Authors: Jingyu Zhang, Marc Marone, Tianjian Li, Benjamin Van Durme, Daniel Khashabi
Abstract: To trust the fluent generations of large language models (LLMs), humans must be able to verify their correctness against trusted, external sources. Recent efforts, such as providing citations via retrieved documents or post-hoc provenance, enhance verifiability but still provide no guarantees on their correctness. To address these limitations, we tackle the verifiability goal with a different philosophy: trivializing the verification process by developing models that quote verbatim statements from trusted sources in pre-training data. We propose Quote-Tuning, and demonstrate it is feasible to align LLMs to provide quoted statements from data memorized during pre-training. The core of Quote-Tuning is a fast membership inference function (Marone and Van Durme, 2023) that efficiently verifies text against a trusted corpus. We leverage this tool to design a reward function to quantify quotes in model responses, which is then used to create a dataset for preference learning. Experimental results show that Quote-Tuning significantly increases verbatim quotes from high-quality pre-training documents by 55% to 130% relative to un-tuned models while maintaining response quality. Quote-Tuning also generalizes quoting to out-of-domain data, is applicable in different tasks, and provides additional benefits to truthfulness. Our method not only serves as a hassle-free method to increase quoting but also opens up avenues for improving LLM trustworthiness through better verifiability.
Authors: Jiaming He, Wenbo Jiang, Guanyu Hou, Wenshu Fan, Rui Zhang, Hongwei Li
Abstract: Mainstream backdoor attacks on large language models (LLMs) typically set a fixed trigger in the input instance and specific responses for triggered queries. However, the fixed trigger setting (e.g., unusual words) may be easily detected by human detection, limiting the effectiveness and practicality in real-world scenarios. To enhance the stealthiness of backdoor activation, we present a new poisoning paradigm against LLMs triggered by specifying generation conditions, which are commonly adopted strategies by users during model inference. The poisoned model performs normally for output under normal/other generation conditions, while becomes harmful for output under target generation conditions. To achieve this objective, we introduce BrieFool, an efficient attack framework. It leverages the characteristics of generation conditions by efficient instruction sampling and poisoning data generation, thereby influencing the behavior of LLMs under target conditions. Our attack can be generally divided into two types with different targets: Safety unalignment attack and Ability degradation attack. Our extensive experiments demonstrate that BrieFool is effective across safety domains and ability domains, achieving higher success rates than baseline methods, with 94.3 % on GPT-3.5-turbo
Authors: Dipankar Srirag, Nihar Ranjan Sahoo, Aditya Joshi
Abstract: With an evergrowing number of LLMs reporting superlative performance for English, their ability to perform equitably for different dialects of English ($\textit{i.e.}$, dialect robustness) needs to be ascertained. Specifically, we use English language (US English or Indian English) conversations between humans who play the word-guessing game of 'taboo'. We formulate two evaluative tasks: target word prediction (TWP) ($\textit{i.e.}$, predict the masked target word in a conversation) and target word selection (TWS) ($\textit{i.e.}$, select the most likely masked target word in a conversation, from among a set of candidate words). Extending MD3, an existing dialectic dataset of taboo-playing conversations, we introduce M-MD3, a target-word-masked version of MD3 with the en-US and en-IN subsets. We create two subsets: en-MV (where en-US is transformed to include dialectal information) and en-TR (where dialectal information is removed from en-IN). We evaluate one open-source (Llama3) and two closed-source (GPT-4/3.5) LLMs. LLMs perform significantly better for US English than Indian English for both TWP and TWS tasks, for all settings, exhibiting marginalisation against the Indian dialect of English. While GPT-based models perform the best, the comparatively smaller models work more equitably after fine-tuning. Our error analysis shows that the LLMs can understand the dialect better after fine-tuning using dialectal data. Our evaluation methodology exhibits a novel way to examine attributes of language models using pre-existing dialogue datasets.
Authors: Trinh Pham, Khoi M. Le, Luu Anh Tuan
Abstract: In this paper, we introduce UniBridge (Cross-Lingual Transfer Learning with Optimized Embeddings and Vocabulary), a comprehensive approach developed to improve the effectiveness of Cross-Lingual Transfer Learning, particularly in languages with limited resources. Our approach tackles two essential elements of a language model: the initialization of embeddings and the optimal vocabulary size. Specifically, we propose a novel embedding initialization method that leverages both lexical and semantic alignment for a language. In addition, we present a method for systematically searching for the optimal vocabulary size, ensuring a balance between model complexity and linguistic coverage. Our experiments across multilingual datasets show that our approach greatly improves the F1-Score in several languages. UniBridge is a robust and adaptable solution for cross-lingual systems in various languages, highlighting the significance of initializing embeddings and choosing the right vocabulary size in cross-lingual environments.
Authors: Xiaoze Liu, Ting Sun, Tianyang Xu, Feijie Wu, Cunxiang Wang, Xiaoqian Wang, Jing Gao
Abstract: Large Language Models (LLMs) have transformed machine learning but raised significant legal concerns due to their potential to produce text that infringes on copyrights, resulting in several high-profile lawsuits. The legal landscape is struggling to keep pace with these rapid advancements, with ongoing debates about whether generated text might plagiarize copyrighted materials. Current LLMs may infringe on copyrights or overly restrict non-copyrighted texts, leading to these challenges: (i) the need for a comprehensive evaluation benchmark to assess copyright compliance from multiple aspects; (ii) evaluating robustness against safeguard bypassing attacks; and (iii) developing effective defense targeted against the generation of copyrighted text. To tackle these challenges, we introduce a curated dataset to evaluate methods, test attack strategies, and propose lightweight, real-time defense to prevent the generation of copyrighted text, ensuring the safe and lawful use of LLMs. Our experiments demonstrate that current LLMs frequently output copyrighted text, and that jailbreaking attacks can significantly increase the volume of copyrighted output. Our proposed defense mechanism significantly reduces the volume of copyrighted text generated by LLMs by effectively refusing malicious requests. Code is publicly available at https://github.com/xz-liu/SHIELD
Authors: Oluyemi Enoch Amujo, Shanchieh Jay Yang
Abstract: Recently, large language models (LLMs) have expanded into various domains. However, there remains a need to evaluate how these models perform when prompted with commonplace queries compared to domain-specific queries, which may be useful for benchmarking prior to fine-tuning for domain-specific downstream tasks. This study evaluates LLMs, specifically Gemma-2B and Gemma-7B, across diverse domains, including cybersecurity, medicine, and finance, compared to common knowledge queries. This study utilizes a comprehensive methodology to assess foundational models, which includes problem formulation, data analysis, and the development of ThroughCut, a novel outlier detection technique that automatically identifies response throughput outliers based on their conciseness. This methodological rigor enhances the credibility of the presented evaluation frameworks. This study focused on assessing inference time, response length, throughput, quality, and resource utilization and investigated the correlations between these factors. The results indicate that model size and types of prompts used for inference significantly influenced response length and quality. In addition, common prompts, which include various types of queries, generate diverse and inconsistent responses at irregular intervals. In contrast, domain-specific prompts consistently generate concise responses within a reasonable time. Overall, this study underscores the need for comprehensive evaluation frameworks to enhance the reliability of benchmarking procedures in multidomain AI research.
Authors: Gleb Kuzmin, Neemesh Yadav, Ivan Smirnov, Timothy Baldwin, Artem Shelmanov
Abstract: We propose selective debiasing -- an inference-time safety mechanism that aims to increase the overall quality of models in terms of prediction performance and fairness in the situation when re-training a model is prohibitive. The method is inspired by selective prediction, where some predictions that are considered low quality are discarded at inference time. In our approach, we identify the potentially biased model predictions and, instead of discarding them, we debias them using LEACE -- a post-processing debiasing method. To select problematic predictions, we propose a bias quantification approach based on KL divergence, which achieves better results than standard UQ methods. Experiments with text classification datasets demonstrate that selective debiasing helps to close the performance gap between post-processing methods and at-training and pre-processing debiasing techniques.
Authors: Nunzio Lore, Alireza Sepehr Ilami, Babak Heydari
Abstract: As the performance of larger, newer Large Language Models continues to improve for strategic Theory of Mind (ToM) tasks, the demand for these state-of-the-art models increases commensurately. However, their deployment is costly both in terms of processing power and time. In this paper, we investigate the feasibility of creating smaller, highly-performing specialized algorithms by way of fine-tuning. To do this, we first present a large pre-trained model with 20 unique scenarios that combine different social contexts with games of varying social dilemmas, record its answers, and use them for Q&A fine-tuning on a smaller model of the same family. Our focus is on in-context game-theoretic decision-making, the same domain within which human interaction occurs and that requires both a theory of mind (or a semblance thereof) and an understanding of social dynamics. The smaller model is therefore trained not just on the answers provided, but also on the motivations provided by the larger model, which should contain advice and guidelines to navigate both strategic dilemmas and social cues. We find that the fine-tuned smaller language model consistently bridged the gap in performance between the smaller pre-trained version of the model and its larger relative and that its improvements extended in areas and contexts beyond the ones provided in the training examples, including on out-of-sample scenarios that include completely different game structures. On average for all games, through fine-tuning, the smaller model showed a 46% improvement measured as alignment towards the behavior of the larger model, with 100% representing indistinguishable behavior. When presented with out-of-sample social contexts and games, the fine-tuned model still displays remarkable levels of alignment, reaching an improvement of 18% and 28% respectively.
Authors: Mara Finkelstein, David Vilar, Markus Freitag
Abstract: Recent research in neural machine translation (NMT) has shown that training on high-quality machine-generated data can outperform training on human-generated data. This work accompanies the first-ever release of a LLM-generated, MBR-decoded and QE-reranked dataset with both sentence-level and multi-sentence examples. We perform extensive experiments to demonstrate the quality of our dataset in terms of its downstream impact on NMT model performance. We find that training from scratch on our (machine-generated) dataset outperforms training on the (web-crawled) WMT'23 training dataset (which is 300 times larger), and also outperforms training on the top-quality subset of the WMT'23 training dataset. We also find that performing self-distillation by finetuning the LLM which generated this dataset outperforms the LLM's strong few-shot baseline. These findings corroborate the quality of our dataset, and demonstrate the value of high-quality machine-generated data in improving performance of NMT models.
Authors: Shantanu Ghosh, Chenyu Wang, Kayhan Batmanghelich
Abstract: Error slice discovery associates structured patterns with model errors. Existing methods discover error slices by clustering the error-prone samples with similar patterns or assigning discrete attributes to each sample for post-hoc analysis. While these methods aim for interpretability and easier mitigation through reweighting or rebalancing, they may not capture the full complexity of error patterns due to incomplete or missing attributes. Contrary to the existing approach, this paper utilizes the reasoning capabilities of the Large Language Model (LLM) to analyze complex error patterns and generate testable hypotheses. This paper proposes LADDER: Language Driven slice Discovery and Error Rectification. It first projects the model's representation into a language-aligned feature space (\eg CLIP) to preserve semantics in the original model feature space. This ensures the accurate retrieval of sentences that highlight the model's errors. Next, the LLM utilizes the sentences and generates hypotheses to discover error slices. Finally, we mitigate the error by fine-tuning the classification head by creating a group-balanced dataset using the hypotheses. Our entire method does not require any attribute annotation, either explicitly or through external tagging models. We validate our method with \textbf{five} image classification datasets. The code is available\footnote{\url{https://github.com/batmanlab/Ladder}}
Authors: Xiaochuang Han, Marjan Ghazvininejad, Pang Wei Koh, Yulia Tsvetkov
Abstract: Recent work in image and video generation has been adopting the autoregressive LLM architecture due to its generality and potentially easy integration into multi-modal systems. The crux of applying autoregressive training in language generation to visual generation is discretization -- representing continuous data like images and videos as discrete tokens. Common methods of discretizing images and videos include modeling raw pixel values, which are prohibitively lengthy, or vector quantization, which requires convoluted pre-hoc training. In this work, we propose to directly model images and videos as compressed files saved on computers via canonical codecs (e.g., JPEG, AVC/H.264). Using the default Llama architecture without any vision-specific modifications, we pretrain JPEG-LM from scratch to generate images (and AVC-LM to generate videos as a proof of concept), by directly outputting compressed file bytes in JPEG and AVC formats. Evaluation of image generation shows that this simple and straightforward approach is more effective than pixel-based modeling and sophisticated vector quantization baselines (on which our method yields a 31% reduction in FID). Our analysis shows that JPEG-LM has an especial advantage over vector quantization models in generating long-tail visual elements. Overall, we show that using canonical codec representations can help lower the barriers between language generation and visual generation, facilitating future research on multi-modal language/image/video LLMs.
Authors: Hongyin Zhu
Abstract: The development of a large language model (LLM) infrastructure is a pivotal undertaking in artificial intelligence. This paper explores the intricate landscape of LLM infrastructure, software, and data management. By analyzing these core components, we emphasize the pivotal considerations and safeguards crucial for successful LLM development. This work presents a concise synthesis of the challenges and strategies inherent in constructing a robust and effective LLM infrastructure, offering valuable insights for researchers and practitioners alike.
Authors: Hongyin Zhu
Abstract: This paper carefully summarizes extensive and profound questions from all walks of life, focusing on the current high-profile AI field, covering multiple dimensions such as industry trends, academic research, technological innovation and business applications. This paper meticulously curates questions that are both thought-provoking and practically relevant, providing nuanced and insightful answers to each. To facilitate readers' understanding and reference, this paper specifically classifies and organizes these questions systematically and meticulously from the five core dimensions of computing power infrastructure, software architecture, data resources, application scenarios, and brain science. This work aims to provide readers with a comprehensive, in-depth and cutting-edge AI knowledge framework to help people from all walks of life grasp the pulse of AI development, stimulate innovative thinking, and promote industrial progress.
Authors: Emmanuel Chemla, Ryan M. Nefdt
Abstract: What role can the otherwise successful Large Language Models (LLMs) play in the understanding of human cognition, and in particular in terms of informing language acquisition debates? To contribute to this question, we first argue that neither humans nor LLMs are general learners, in a variety of senses. We make a novel case for how in particular LLMs follow a dual-optimization process: they are optimized during their training (which is typically compared to language acquisition), and modern LLMs have also been selected, through a process akin to natural selection in a species. From this perspective, we argue that the performance of LLMs, whether similar or dissimilar to that of humans, does not weigh easily on important debates about the importance of human cognitive biases for language.
Authors: Ken Gu, Ruoxi Shang, Ruien Jiang, Keying Kuang, Richard-John Lin, Donghe Lyu, Yue Mao, Youran Pan, Teng Wu, Jiaqian Yu, Yikun Zhang, Tianmai M. Zhang, Lanyi Zhu, Mike A. Merrill, Jeffrey Heer, Tim Althoff
Abstract: Data-driven scientific discovery requires the iterative integration of scientific domain knowledge, statistical expertise, and an understanding of data semantics to make nuanced analytical decisions, e.g., about which variables, transformations, and statistical models to consider. LM-based agents equipped with planning, memory, and code execution capabilities have the potential to support data-driven science. However, evaluating agents on such open-ended tasks is challenging due to multiple valid approaches, partially correct steps, and different ways to express the same decisions. To address these challenges, we present BLADE, a benchmark to automatically evaluate agents' multifaceted approaches to open-ended research questions. BLADE consists of 12 datasets and research questions drawn from existing scientific literature, with ground truth collected from independent analyses by expert data scientists and researchers. To automatically evaluate agent responses, we developed corresponding computational methods to match different representations of analyses to this ground truth. Though language models possess considerable world knowledge, our evaluation shows that they are often limited to basic analyses. However, agents capable of interacting with the underlying data demonstrate improved, but still non-optimal, diversity in their analytical decision making. Our work enables the evaluation of agents for data-driven science and provides researchers deeper insights into agents' analysis approaches.
Authors: Yeyong Yu, Rusheng Yu, Haojie Wei, Zhanqiu Zhang, Quan Qian
Abstract: The rapid advancement of large language models (LLMs) has revolutionized role-playing, enabling the development of general role-playing models. However, current role-playing training has two significant issues: (I) Using a predefined role profile to prompt dialogue training for specific scenarios usually leads to inconsistencies and even conflicts between the dialogue and the profile, resulting in training biases. (II) The model learns to imitate the role based solely on the profile, neglecting profile-dialogue alignment at the sentence level. In this work, we propose a simple yet effective framework called BEYOND DIALOGUE, designed to overcome these hurdles. This framework innovatively introduces "beyond dialogue" tasks to align dialogue with profile traits based on each specific scenario, thereby eliminating biases during training. Furthermore, by adopting an innovative prompting mechanism that generates reasoning outcomes for training, the framework allows the model to achieve fine-grained alignment between profile and dialogue at the sentence level. The aforementioned methods are fully automated and low-cost. Additionally, the integration of automated dialogue and objective evaluation methods forms a comprehensive framework, paving the way for general role-playing. Experimental results demonstrate that our model excels in adhering to and reflecting various dimensions of role profiles, outperforming most proprietary general and specialized role-playing baselines. All code and datasets are available at https://github.com/yuyouyu32/BeyondDialogue.
Authors: Kangjun Noh, Baekryun Seong, Hoyoon Byun, Youngjun Choi, Sungjin Song, Kyungwoo Song
Abstract: Large Language Models (LLMs) have great success in natural language processing tasks such as response generation. However, their use in tabular data has been limited due to their inferior performance compared to traditional machine learning models (TMLs) such as XGBoost. We find that the pre-trained knowledge of LLMs enables them to interpret new variables that appear in a test without additional training, a capability central to the concept of Out-of-Variable (OOV). From the findings, we propose a Language-Based-Classifier (LBC), a classifier that maximizes the benefits of LLMs to outperform TMLs on OOV tasks. LBC employs three key methodological strategies: 1) Categorical changes to adjust data to better fit the model's understanding, 2) Advanced order and indicator to enhance data representation to the model, and 3) Using verbalizer to map logit scores to classes during inference to generate model predictions. These strategies, combined with the pre-trained knowledge of LBC, emphasize the model's ability to effectively handle OOV tasks. We empirically and theoretically validate the superiority of LBC. LBC is the first study to apply an LLM-based model to OOV tasks. The source code is at https://github.com/ASDASDanonymous/Language-Based-Classifier-forOOVtasks.
URLs: https://github.com/ASDASDanonymous/Language-Based-Classifier-forOOVtasks.
Authors: Jian Chen, Vashisth Tiwari, Ranajoy Sadhukhan, Zhuoming Chen, Jinyuan Shi, Ian En-Hsu Yen, Beidi Chen
Abstract: Large Language Models (LLMs) have become more prevalent in long-context applications such as interactive chatbots, document analysis, and agent workflows, but it is challenging to serve long-context requests with low latency and high throughput. Speculative decoding (SD) is a widely used technique to reduce latency without sacrificing performance but the conventional wisdom suggests that its efficacy is limited to small batch sizes. In MagicDec, we show that surprisingly SD can achieve speedup even for a high throughput inference regime for moderate to long sequences. More interestingly, an intelligent drafting strategy can achieve better speedup with increasing batch size based on our rigorous analysis. MagicDec first identifies the bottleneck shifts with increasing batch size and sequence length, and uses these insights to deploy speculative decoding more effectively for high throughput inference. Then, it leverages draft models with sparse KV cache to address the KV bottleneck that scales with both sequence length and batch size. This finding underscores the broad applicability of speculative decoding in long-context serving, as it can enhance throughput and reduce latency without compromising accuracy. For moderate to long sequences, we demonstrate up to 2x speedup for LLaMA-2-7B-32K and 1.84x speedup for LLaMA-3.1-8B when serving batch sizes ranging from 32 to 256 on 8 NVIDIA A100 GPUs. The code is available at https://github.com/Infini-AI-Lab/MagicDec/.
Authors: Qiushi Sun, Zhangyue Yin, Xiang Li, Zhiyong Wu, Xipeng Qiu, Lingpeng Kong
Abstract: Large Language Models (LLMs) are evolving at an unprecedented pace and have exhibited considerable capability in the realm of natural language processing (NLP) with world knowledge. Benefiting from ultra-large-scale training corpora, a single LLM can manage typical NLP tasks competently. However, its performance in executing reasoning tasks is still confined by the limitations of its internal representations. To push this boundary further, we introduce Corex in this paper, a suite of novel general-purpose strategies that transform LLMs into autonomous agents pioneering multi-model collaborations for complex task-solving. Inspired by human behaviors, Corex is constituted by diverse collaboration paradigms including Debate, Review, and Retrieve modes, which collectively work towards enhancing the factuality, faithfulness, and reliability of the reasoning process. These paradigms foster task-agnostic approaches that enable LLMs to ''think outside the box,'' thereby overcoming hallucinations and providing better solutions. Through extensive experiments across four different types of reasoning tasks, we demonstrate that orchestrating multiple LLMs to work in concert yields substantially better performance compared to existing methods. Further results and in-depth analysis demonstrate the cost-effectiveness of our method, facilitating collaboration among different LLMs and promoting annotation efficiency.
Authors: Yoonho Lee, Michelle S. Lam, Helena Vasconcelos, Michael S. Bernstein, Chelsea Finn
Abstract: The standard way to teach models is by feeding them lots of data. However, this approach often teaches models incorrect ideas because they pick up on misleading signals in the data. To prevent such misconceptions, we must necessarily provide additional information beyond the training data. Prior methods incorporate additional instance-level supervision, such as labels for misleading features or additional labels for debiased data. However, such strategies require a large amount of labeler effort. We hypothesize that people are good at providing textual feedback at the concept level, a capability that existing teaching frameworks do not leverage. We propose Clarify, a novel interface and method for interactively correcting model misconceptions. Through Clarify, users need only provide a short text description of a model's consistent failure patterns. Then, in an entirely automated way, we use such descriptions to improve the training process. Clarify is the first end-to-end system for user model correction. Our user studies show that non-expert users can successfully describe model misconceptions via Clarify, leading to increased worst-case performance in two datasets. We additionally conduct a case study on a large-scale image dataset, ImageNet, using Clarify to find and rectify 31 novel hard subpopulations.
Authors: Taylor Sorensen, Jared Moore, Jillian Fisher, Mitchell Gordon, Niloofar Mireshghallah, Christopher Michael Rytting, Andre Ye, Liwei Jiang, Ximing Lu, Nouha Dziri, Tim Althoff, Yejin Choi
Abstract: With increased power and prevalence of AI systems, it is ever more critical that AI systems are designed to serve all, i.e., people with diverse values and perspectives. However, aligning models to serve pluralistic human values remains an open research question. In this piece, we propose a roadmap to pluralistic alignment, specifically using language models as a test bed. We identify and formalize three possible ways to define and operationalize pluralism in AI systems: 1) Overton pluralistic models that present a spectrum of reasonable responses; 2) Steerably pluralistic models that can steer to reflect certain perspectives; and 3) Distributionally pluralistic models that are well-calibrated to a given population in distribution. We also formalize and discuss three possible classes of pluralistic benchmarks: 1) Multi-objective benchmarks, 2) Trade-off steerable benchmarks, which incentivize models to steer to arbitrary trade-offs, and 3) Jury-pluralistic benchmarks which explicitly model diverse human ratings. We use this framework to argue that current alignment techniques may be fundamentally limited for pluralistic AI; indeed, we highlight empirical evidence, both from our own experiments and from other work, that standard alignment procedures might reduce distributional pluralism in models, motivating the need for further research on pluralistic alignment.
Authors: Ke Zhu, Zheng Ge, Liang Zhao, Xiangyu Zhang
Abstract: This paper makes the first attempt towards unsupervised preference alignment in Vision-Language Models (VLMs). We generate chosen and rejected responses with regard to the original and augmented image pairs, and conduct preference alignment with direct preference optimization. It is based on a core idea: properly designed augmentation to the image input will induce VLM to generate false but hard negative responses, which helps the model to learn from and produce more robust and powerful answers. The whole pipeline no longer hinges on supervision from GPT-4 or human involvement during alignment, and is highly efficient with few lines of code. With only 8k randomly sampled unsupervised data, it achieves 90\% relative score to GPT-4 on complex reasoning in LLaVA-Bench, and improves LLaVA-7B/13B by 6.7\%/5.6\% score on complex multi-modal benchmark MM-Vet. Visualizations shows its improved ability to align with user-intentions. A series of ablations are firmly conducted to reveal the latent mechanism of the approach, which also indicates its potential towards further scaling. Code are available in https://github.com/Kevinz-code/SeVa.
Authors: Sunhao Dai, Chen Xu, Shicheng Xu, Liang Pang, Zhenhua Dong, Jun Xu
Abstract: With the rapid advancements of large language models (LLMs), information retrieval (IR) systems, such as search engines and recommender systems, have undergone a significant paradigm shift. This evolution, while heralding new opportunities, introduces emerging challenges, particularly in terms of biases and unfairness, which may threaten the information ecosystem. In this paper, we present a comprehensive survey of existing works on emerging and pressing bias and unfairness issues in IR systems when the integration of LLMs. We first unify bias and unfairness issues as distribution mismatch problems, providing a groundwork for categorizing various mitigation strategies through distribution alignment. Subsequently, we systematically delve into the specific bias and unfairness issues arising from three critical stages of LLMs integration into IR systems: data collection, model development, and result evaluation. In doing so, we meticulously review and analyze recent literature, focusing on the definitions, characteristics, and corresponding mitigation strategies associated with these issues. Finally, we identify and highlight some open problems and challenges for future work, aiming to inspire researchers and stakeholders in the IR field and beyond to better understand and mitigate bias and unfairness issues of IR in this LLM era. We also consistently maintain a GitHub repository for the relevant papers and resources in this rising direction at https://github.com/KID-22/LLM-IR-Bias-Fairness-Survey.
URLs: https://github.com/KID-22/LLM-IR-Bias-Fairness-Survey.
Authors: Xijie Huang, Xinyuan Wang, Hantao Zhang, Yinghao Zhu, Jiawen Xi, Jingkun An, Hao Wang, Hao Liang, Chengwei Pan
Abstract: Security concerns related to Large Language Models (LLMs) have been extensively explored, yet the safety implications for Multimodal Large Language Models (MLLMs), particularly in medical contexts (MedMLLMs), remain insufficiently studied. This paper delves into the underexplored security vulnerabilities of MedMLLMs, especially when deployed in clinical environments where the accuracy and relevance of question-and-answer interactions are critically tested against complex medical challenges. By combining existing clinical medical data with atypical natural phenomena, we define the mismatched malicious attack (2M-attack) and introduce its optimized version, known as the optimized mismatched malicious attack (O2M-attack or 2M-optimization). Using the voluminous 3MAD dataset that we construct, which covers a wide range of medical image modalities and harmful medical scenarios, we conduct a comprehensive analysis and propose the MCM optimization method, which significantly enhances the attack success rate on MedMLLMs. Evaluations with this dataset and attack methods, including white-box attacks on LLaVA-Med and transfer attacks (black-box) on four other SOTA models, indicate that even MedMLLMs designed with enhanced security features remain vulnerable to security breaches. Our work underscores the urgent need for a concerted effort to implement robust security measures and enhance the safety and efficacy of open-source MedMLLMs, particularly given the potential severity of jailbreak attacks and other malicious or clinically significant exploits in medical settings. Our code is available at https://github.com/dirtycomputer/O2M_attack.
Authors: Xuannan Liu, Zekun Li, Peipei Li, Shuhan Xia, Xing Cui, Linzhi Huang, Huaibo Huang, Weihong Deng, Zhaofeng He
Abstract: Current multimodal misinformation detection (MMD) methods often assume a single source and type of forgery for each sample, which is insufficient for real-world scenarios where multiple forgery sources coexist. The lack of a benchmark for mixed-source misinformation has hindered progress in this field. To address this, we introduce MMFakeBench, the first comprehensive benchmark for mixed-source MMD. MMFakeBench includes 3 critical sources: textual veracity distortion, visual veracity distortion, and cross-modal consistency distortion, along with 12 sub-categories of misinformation forgery types. We further conduct an extensive evaluation of 6 prevalent detection methods and 15 large vision-language models (LVLMs) on MMFakeBench under a zero-shot setting. The results indicate that current methods struggle under this challenging and realistic mixed-source MMD setting. Additionally, we propose an innovative unified framework, which integrates rationales, actions, and tool-use capabilities of LVLM agents, significantly enhancing accuracy and generalization. We believe this study will catalyze future research into more realistic mixed-source multimodal misinformation and provide a fair evaluation of misinformation detection methods.
Authors: Cheol Jun Cho, Peter Wu, Tejas S. Prabhune, Dhruv Agarwal, Gopala K. Anumanchipalli
Abstract: Vocal tract articulation is a natural, grounded control space of speech production. The spatiotemporal coordination of articulators combined with the vocal source shapes intelligible speech sounds to enable effective spoken communication. Based on this physiological grounding of speech, we propose a new framework of neural encoding-decoding of speech -- Articulatory Encodec. Articulatory Encodec comprises an articulatory analysis model that infers articulatory features from speech audio, and an articulatory synthesis model that synthesizes speech audio from articulatory features. The articulatory features are kinematic traces of vocal tract articulators and source features, which are intuitively interpretable and controllable, being the actual physical interface of speech production. An additional speaker identity encoder is jointly trained with the articulatory synthesizer to inform the voice texture of individual speakers. By training on large-scale speech data, we achieve a fully intelligible, high-quality articulatory synthesizer that generalizes to unseen speakers. Furthermore, the speaker embedding is effectively disentangled from articulations, which enables accent-perserving zero-shot voice conversion. To the best of our knowledge, this is the first demonstration of universal, high-performance articulatory inference and synthesis, suggesting the proposed framework as a powerful coding system of speech.
Authors: Yanshu Wang, Wang Li, Tong Yang
Abstract: Matrix quantization compresses matrix elements into a more compact form to reduce storage requirements, with dequantization enabling reconstruction for use. We define the Quantization Error Minimization (QEM) problem as minimizing the difference between the original and quantized matrices while ensuring the quantized matrix remains within fixed memory constraints. This technique is crucial in applications like Large Language Model (LLM) weight compression and KV cache compression, where large matrix sizes demand efficient storage solutions. As modern LLMs like GPT-4 and BERT continue to grow, effective matrix compression is increasingly important. These models contain billions of parameters in matrix form, making efficient weight quantization essential for both storage and computational efficiency. Similarly, KV caches, storing intermediate inference results, are matrix-based and benefit significantly from optimized compression techniques. To address the QEM problem in the context of LLM weight and KV cache compression, we propose Quantum Entanglement Trees (QET). QET leverages the local structure of matrix elements by iteratively swapping elements to create a locally ordered matrix, which is then grouped and quantized column by column. To enhance QET, we introduce two optimizations: residual quantization to further reduce Mean Squared Error (MSE) and masking with batch processing to accelerate the algorithm. Our experiments demonstrate that QET can reduce MSE to 12.3% of its original value at the same compression ratio, outperforming leading baseline methods. Our contributions include framing the QEM problem specifically for LLM and KV cache compression, developing the QET algorithm, and implementing optimizations that improve accuracy and processing speed.
Authors: Samyak Jain, Ekdeep Singh Lubana, Kemal Oksuz, Tom Joy, Philip H. S. Torr, Amartya Sanyal, Puneet K. Dokania
Abstract: Safety fine-tuning helps align Large Language Models (LLMs) with human preferences for their safe deployment. To better understand the underlying factors that make models safe via safety fine-tuning, we design a synthetic data generation framework that captures salient aspects of an unsafe input by modeling the interaction between the task the model is asked to perform (e.g., "design") versus the specific concepts the task is asked to be performed upon (e.g., a "cycle" vs. a "bomb"). Using this, we investigate three well-known safety fine-tuning methods -- supervised safety fine-tuning, direct preference optimization, and unlearning -- and provide significant evidence demonstrating that these methods minimally transform MLP weights to specifically align unsafe inputs into its weights' null space. This yields a clustering of inputs based on whether the model deems them safe or not. Correspondingly, when an adversarial input (e.g., a jailbreak) is provided, its activations are closer to safer samples, leading to the model processing such an input as if it were safe. We validate our findings, wherever possible, on real-world models -- specifically, Llama-2 7B and Llama-3 8B.
Authors: Liu Qi, He Yongyi, Lian Defu, Zheng Zhi, Xu Tong, Liu Che, Chen Enhong
Abstract: Multimodal Entity Linking (MEL) is a crucial task that aims at linking ambiguous mentions within multimodal contexts to the referent entities in a multimodal knowledge base, such as Wikipedia. Existing methods focus heavily on using complex mechanisms and extensive model tuning methods to model the multimodal interaction on specific datasets. However, these methods overcomplicate the MEL task and overlook the visual semantic information, which makes them costly and hard to scale. Moreover, these methods can not solve the issues like textual ambiguity, redundancy, and noisy images, which severely degrade their performance. Fortunately, the advent of Large Language Models (LLMs) with robust capabilities in text understanding and reasoning, particularly Multimodal Large Language Models (MLLMs) that can process multimodal inputs, provides new insights into addressing this challenge. However, how to design a universally applicable LLMs-based MEL approach remains a pressing challenge. To this end, we propose UniMEL, a unified framework which establishes a new paradigm to process multimodal entity linking tasks using LLMs. In this framework, we employ LLMs to augment the representation of mentions and entities individually by integrating textual and visual information and refining textual information. Subsequently, we employ the embedding-based method for retrieving and re-ranking candidate entities. Then, with only ~0.26% of the model parameters fine-tuned, LLMs can make the final selection from the candidate entities. Extensive experiments on three public benchmark datasets demonstrate that our solution achieves state-of-the-art performance, and ablation studies verify the effectiveness of all modules. Our code is available at https://github.com/Javkonline/UniMEL.
Authors: Aochuan Chen, Jiashun Cheng, Zijing Liu, Ziqi Gao, Fugee Tsung, Yu Li, Jia Li
Abstract: Low-Rank Adaptation (LoRA) has gained popularity for fine-tuning large foundation models, leveraging low-rank matrices $\mathbf{A}$ and $\mathbf{B}$ to represent weight changes (i.e., $\Delta \mathbf{W} = \mathbf{B} \mathbf{A}$). This method reduces trainable parameters and mitigates heavy memory consumption associated with full delta matrices by sequentially multiplying $\mathbf{A}$ and $\mathbf{B}$ with the activation. Despite its success, the intrinsic low-rank characteristic may limit its performance. Although several variants have been proposed to address this issue, they often overlook the crucial computational and memory efficiency brought by LoRA. In this paper, we propose Circular Convolution Adaptation (C$^3$A), which not only achieves high-rank adaptation with enhanced performance but also excels in both computational power and memory utilization. Extensive experiments demonstrate that C$^3$A consistently outperforms LoRA and its variants across various fine-tuning tasks.
Authors: Wenjun Huang, Jiakai Pan, Jiahao Tang, Yanyu Ding, Yifei Xing, Yuhe Wang, Zhengzhuo Wang, Jianguo Hu
Abstract: Multimodal Large Language Models (MLLMs) have attracted much attention for their multifunctionality. However, traditional Transformer architectures incur significant overhead due to their secondary computational complexity. To address this issue, we introduce ML-Mamba, a multimodal language model, which utilizes the latest and efficient Mamba-2 model for inference. Mamba-2 is known for its linear scalability and fast processing of long sequences. We replace the Transformer-based backbone with a pre-trained Mamba-2 model and explore methods for integrating 2D visual selective scanning mechanisms into multimodal learning while also trying various visual encoders and Mamba-2 model variants. Our extensive experiments in various multimodal benchmark tests demonstrate the competitive performance of ML-Mamba and highlight the potential of state space models in multimodal tasks. The experimental results show that: (1) we empirically explore how to effectively apply the 2D vision selective scan mechanism for multimodal learning. We propose a novel multimodal connector called the Mamba-2 Scan Connector (MSC), which enhances representational capabilities. (2) ML-Mamba achieves performance comparable to state-of-the-art methods such as TinyLaVA and MobileVLM v2 through its linear sequential modeling while faster inference speed; (3) Compared to multimodal models utilizing Mamba-1, the Mamba-2-based ML-Mamba exhibits superior inference performance and effectiveness.
Authors: Kyra Wilson, Aylin Caliskan
Abstract: Artificial intelligence (AI) hiring tools have revolutionized resume screening, and large language models (LLMs) have the potential to do the same. However, given the biases which are embedded within LLMs, it is unclear whether they can be used in this scenario without disadvantaging groups based on their protected attributes. In this work, we investigate the possibilities of using LLMs in a resume screening setting via a document retrieval framework that simulates job candidate selection. Using that framework, we then perform a resume audit study to determine whether a selection of Massive Text Embedding (MTE) models are biased in resume screening scenarios. We simulate this for nine occupations, using a collection of over 500 publicly available resumes and 500 job descriptions. We find that the MTEs are biased, significantly favoring White-associated names in 85.1\% of cases and female-associated names in only 11.1\% of cases, with a minority of cases showing no statistically significant differences. Further analyses show that Black males are disadvantaged in up to 100\% of cases, replicating real-world patterns of bias in employment settings, and validate three hypotheses of intersectionality. We also find an impact of document length as well as the corpus frequency of names in the selection of resumes. These findings have implications for widely used AI tools that are automating employment, fairness, and tech policy.
Authors: Jielong Tang, Zhenxing Wang, Ziyang Gong, Jianxing Yu, Xiangwei Zhu, Jian Yin
Abstract: Grounded Multimodal Named Entity Recognition (GMNER) is an emerging information extraction (IE) task, aiming to simultaneously extract entity spans, types, and corresponding visual regions of entities from given sentence-image pairs data. Recent unified methods employing machine reading comprehension or sequence generation-based frameworks show limitations in this difficult task. The former, utilizing human-designed queries, struggles to differentiate ambiguous entities, such as Jordan (Person) and off-White x Jordan (Shoes). The latter, following the one-by-one decoding order, suffers from exposure bias issues. We maintain that these works misunderstand the relationships of multimodal entities. To tackle these, we propose a novel unified framework named Multi-grained Query-guided Set Prediction Network (MQSPN) to learn appropriate relationships at intra-entity and inter-entity levels. Specifically, MQSPN consists of a Multi-grained Query Set (MQS) and a Multimodal Set Prediction Network (MSP). MQS explicitly aligns entity regions with entity spans by employing a set of learnable queries to strengthen intra-entity connections. Based on distinct intra-entity modeling, MSP reformulates GMNER as a set prediction, guiding models to establish appropriate inter-entity relationships from a global matching perspective. Additionally, we incorporate a query-guided Fusion Net (QFNet) to work as a glue network between MQS and MSP. Extensive experiments demonstrate that our approach achieves state-of-the-art performances in widely used benchmarks.
Authors: Wenbin Hu, Huihao Jing, Qi Hu, Haoran Li, Yangqiu Song
Abstract: Textual graphs are ubiquitous in real-world applications, featuring rich text information with complex relationships, which enables advanced research across various fields. Textual graph representation learning aims to generate low-dimensional feature embeddings from textual graphs that can improve the performance of downstream tasks. A high-quality feature embedding should effectively capture both the structural and the textual information in a textual graph. However, most textual graph dataset benchmarks rely on word2vec techniques to generate feature embeddings, which inherently limits their capabilities. Recent works on textual graph representation learning can be categorized into two folds: supervised and unsupervised methods. Supervised methods finetune a language model on labeled nodes, which have limited capabilities when labeled data is scarce. Unsupervised methods, on the other hand, extract feature embeddings by developing complex training pipelines. To address these limitations, we propose a novel unified unsupervised learning autoencoder framework, named Node Level Graph AutoEncoder (NodeGAE). We employ language models as the backbone of the autoencoder, with pretraining on text reconstruction. Additionally, we add an auxiliary loss term to make the feature embeddings aware of the local graph structure. Our method maintains simplicity in the training process and demonstrates generalizability across diverse textual graphs and downstream tasks. We evaluate our method on two core graph representation learning downstream tasks: node classification and link prediction. Comprehensive experiments demonstrate that our approach substantially enhances the performance of diverse graph neural networks (GNNs) across multiple textual graph datasets.
Authors: Enneng Yang, Li Shen, Guibing Guo, Xingwei Wang, Xiaochun Cao, Jie Zhang, Dacheng Tao
Abstract: Model merging is an efficient empowerment technique in the machine learning community that does not require the collection of raw training data and does not require expensive computation. As model merging becomes increasingly prevalent across various fields, it is crucial to understand the available model merging techniques comprehensively. However, there is a significant gap in the literature regarding a systematic and thorough review of these techniques. This survey provides a comprehensive overview of model merging methods and theories, their applications in various domains and settings, and future research directions. Specifically, we first propose a new taxonomic approach that exhaustively discusses existing model merging methods. Secondly, we discuss the application of model merging techniques in large language models, multimodal large language models, and 10+ machine learning subfields, including continual learning, multi-task learning, few-shot learning, etc. Finally, we highlight the remaining challenges of model merging and discuss future research directions. A comprehensive list of papers about model merging is available at \url{https://github.com/EnnengYang/Awesome-Model-Merging-Methods-Theories-Applications}.
URLs: https://github.com/EnnengYang/Awesome-Model-Merging-Methods-Theories-Applications
Authors: Yuan Tian, Tianyi Zhang
Abstract: Recent advances in large language models (LLMs) such as Copilot and ChatGPT have transformed software development by automating coding tasks. Despite these advancements, challenges remain in reducing error rates and fully meeting user expectations. Our empirical study reveals LLMs tend to dilute their self-attention on the initial prompt as more code tokens are generated. We hypothesize this self-attention dilution issue is one of the root causes of inaccuracies in LLM-generated code. To mitigate this issue, we propose Selective Prompt Anchoring (SPA). SPA amplifies the influence of the selected parts in the initial prompt, which we refer to as ``anchored text'', during code generation. Specifically, SPA calculates the logit distribution difference with and without the anchored text. We prove this difference approximates the anchored text's contextual contribution to the output logits. SPA creates an augmented logit distribution by linearly combining the original logit distribution and the logit difference. We evaluate SPA with five LLMs on four benchmarks. Our results demonstrate that using SPA can consistently improve Pass@1 rates by up to 9.7% in all settings. Notably, with selective text anchoring, a small version of DeepSeek-Coder (6.7B) can achieve better performance than an original much larger version (33B). Our code is available at https://github.com/magic-YuanTian/Selective-Prompt-Anchoring.
URLs: https://github.com/magic-YuanTian/Selective-Prompt-Anchoring.
Authors: Zitian Gao, Yihao Xiao
Abstract: In the Venture Capital(VC) industry, predicting the success of startups is challenging due to limited financial data and the need for subjective revenue forecasts. Previous methods based on time series analysis or deep learning often fall short as they fail to incorporate crucial inter-company relationships such as competition and collaboration. Regarding the issues, we propose a novel approach using GrahphRAG augmented time series model. With GraphRAG, time series predictive methods are enhanced by integrating these vital relationships into the analysis framework, allowing for a more dynamic understanding of the startup ecosystem in venture capital. Our experimental results demonstrate that our model significantly outperforms previous models in startup success predictions. To the best of our knowledge, our work is the first application work of GraphRAG.
Authors: Fuzhao Xue, Yukang Chen, Dacheng Li, Qinghao Hu, Ligeng Zhu, Xiuyu Li, Yunhao Fang, Haotian Tang, Shang Yang, Zhijian Liu, Ethan He, Hongxu Yin, Pavlo Molchanov, Jan Kautz, Linxi Fan, Yuke Zhu, Yao Lu, Song Han
Abstract: Long-context capability is critical for multi-modal foundation models, especially for long video understanding. We introduce LongVILA, a full-stack solution for long-context visual-language models by co-designing the algorithm and system. For model training, we upgrade existing VLMs to support long video understanding by incorporating two additional stages, i.e., long context extension and long supervised fine-tuning. However, training on long video is computationally and memory intensive. We introduce the long-context Multi-Modal Sequence Parallelism (MM-SP) system that efficiently parallelizes long video training and inference, enabling 2M context length training on 256 GPUs without any gradient checkpointing. LongVILA efficiently extends the number of video frames of VILA from 8 to 1024, improving the long video captioning score from 2.00 to 3.26 (out of 5), achieving 99.5% accuracy in 1400-frame (274k context length) video needle-in-a-haystack. LongVILA-8B demonstrates consistent accuracy improvements on long videos in the VideoMME benchmark as the number of frames increases. Besides, MM-SP is 2.1x - 5.7x faster than ring sequence parallelism and 1.1x - 1.4x faster than Megatron with context parallelism + tensor parallelism. Moreover, it seamlessly integrates with Hugging Face Transformers.
Authors: Jinxin Liu, Zao Yang
Abstract: The responses generated by Large Language Models (LLMs) can include sensitive information from individuals and organizations, leading to potential privacy leakage. This work implements Influence Functions (IFs) to trace privacy leakage back to the training data, thereby mitigating privacy concerns of Language Models (LMs). However, we notice that current IFs struggle to accurately estimate the influence of tokens with large gradient norms, potentially overestimating their influence. When tracing the most influential samples, this leads to frequently tracing back to samples with large gradient norm tokens, overshadowing the actual most influential samples even if their influences are well estimated. To address this issue, we propose Heuristically Adjusted IF (HAIF), which reduces the weight of tokens with large gradient norms, thereby significantly improving the accuracy of tracing the most influential samples. To establish easily obtained groundtruth for tracing privacy leakage, we construct two datasets, PII-E and PII-CR, representing two distinct scenarios: one with identical text in the model outputs and pre-training data, and the other where models leverage their reasoning abilities to generate text divergent from pre-training data. HAIF significantly improves tracing accuracy, enhancing it by 20.96\% to 73.71\% on the PII-E dataset and 3.21\% to 45.93\% on the PII-CR dataset, compared to the best SOTA IFs against various GPT-2 and QWen-1.5 models. HAIF also outperforms SOTA IFs on real-world pretraining data CLUECorpus2020, demonstrating strong robustness regardless prompt and response lengths.
Authors: Chenxing Wei, Yao Shu, Ying Tiffany He, Fei Richard Yu
Abstract: Large Language Models (LLMs) are driving advancements in artificial intelligence by increasing the scale of model parameters, which has significantly enhanced generalization ability and unlocked new capabilities in practice. However, their performance in specific downstream tasks is usually hindered by their knowledge boundaries on these tasks. Thus, fine-tuning techniques, especially the widely used Low-Rank Adaptation (LoRA) method, have been introduced to expand the boundaries on these tasks, whereas LoRA would underperform on certain tasks owing to its potential overfitting on these tasks. To overcome this overfitting and improve the performance of LoRA, we propose the flexible low rank adaptation (Flexora) method to automatically and flexibly select the most important layers needing to be fine-tuned to achieve the best performance on different downstream tasks. Specifically, Flexora firstly frames this layer selection problem as a well-defined hyperparameter optimization (HPO) problem, then addresses it using the unrolled differentiation (UD) method, and finally selects the most useful layers based on the optimized hyperparameters. Our extensive experiments on many pretrained models and natural language tasks show that Flexora is able to consistently improve over the existing baselines, indicating the effectiveness of our Flexora in practice. We additionally provide insightful theoretical results and many ablation studies to deliver a comprehensive understanding of our Flexora.