| Total: 1908
Finetuning large language models (LLMs) has been empirically effective on a variety of downstream tasks. Existing approaches to finetuning an LLM either focus on parameter-efficient finetuning, which only updates a small number of trainable parameters, or attempt to reduce the memory footprint during the training phase of the finetuning. Typically, the memory footprint during finetuning stems from three contributors: model weights, optimizer states, and intermediate activations. However, existing works still require considerable memory, and none can simultaneously mitigate the memory footprint of all three sources. In this paper, we present quantized side tuing (QST), which enables memory-efficient and fast finetuning of LLMs by operating through a dual-stage process. First, QST quantizes an LLM’s model weights into 4-bit to reduce the memory footprint of the LLM’s original weights. Second, QST introduces a side network separated from the LLM, which utilizes the hidden states of the LLM to make task-specific predictions. Using a separate side network avoids performing back-propagation through the LLM, thus reducing the memory requirement of the intermediate activations. Finally, QST leverages several low-rank adaptors and gradient-free downsample modules to significantly reduce the trainable parameters, so as to save the memory footprint of the optimizer states. Experiments show that QST can reduce the total memory footprint by up to 2.3× and speed up the finetuning process by up to 3× while achieving competent performance compared with the state-of-the-art. When it comes to full finetuning, QST can reduce the total memory footprint up to 7×.
Discovering the semantics of multimodal utterances is essential for understanding human language and enhancing human-machine interactions. Existing methods manifest limitations in leveraging nonverbal information for discerning complex semantics in unsupervised scenarios. This paper introduces a novel unsupervised multimodal clustering method (UMC), making a pioneering contribution to this field. UMC introduces a unique approach to constructing augmentation views for multimodal data, which are then used to perform pre-training to establish well-initialized representations for subsequent clustering. An innovative strategy is proposed to dynamically select high-quality samples as guidance for representation learning, gauged by the density of each sample’s nearest neighbors. Besides, it is equipped to automatically determine the optimal value for the top-K parameter in each cluster to refine sample selection. Finally, both high- and low-quality samples are used to learn representations conducive to effective clustering. We build baselines on benchmark multimodal intent and dialogue act datasets. UMC shows remarkable improvements of 2-6% scores in clustering metrics over state-of-the-art methods, marking the first successful endeavor in this domain. The complete code and data are available at https://github.com/thuiar/UMC.
Large language models (LLMs) have achieved human-level text generation, emphasizing the need for effective deepfake text detection to mitigate risks like the spread of fake news and plagiarism. Existing research has been constrained by evaluating detection methods o specific domains or particular language models. In practical scenarios, however, the detector faces texts from various domains or LLMs without knowing their sources. To this end, we build a comprehensive testbed by gathering texts from diverse human writings and deepfake texts generated by different LLMs. Empirical results on mainstream detection methods demonstrate the difficulties associated with detecting deepfake text in a wide-ranging testbed, particularly in out-of-distribution scenarios. Such difficulties align with the diminishing linguistic differences between the two text sources. Despite challenges, the top-performing detector can identify 84.12% out-of-domain texts generated by a new LLM, indicating the feasibility for application scenarios.
The rapid development of language models (LMs) brings unprecedented accessibility and usage for both models and users. On the one hand, powerful LMs achieve state-of-the-art performance over numerous downstream NLP tasks. On the other hand, more and more attention is paid to unrestricted model accesses that may bring malicious privacy risks of data leakage. To address these issues, many recent works propose privacy-preserving language models (PPLMs) with differential privacy (DP). Unfortunately, different DP implementations make it challenging for a fair comparison among existing PPLMs. In this paper, we present PrivLM-Bench, a multi-perspective privacy evaluation benchmark to empirically and intuitively quantify the privacy leakage of LMs. Instead of only reporting DP parameters, PrivLM-Bench sheds light on the neglected inference data privacy during actual usage. PrivLM-Bench first clearly defines multi-faceted privacy objectives. Then, PrivLM-Bench constructs a unified pipeline to perform private fine-tuning. Lastly, PrivLM-Bench performs existing privacy attacks on LMs with pre-defined privacy objectives as the empirical evaluation results. The empirical attack results are used to fairly and intuitively evaluate the privacy leakage of various PPLMs. We conduct extensive experiments on three datasets of GLUE for mainstream LMs.
Recent advances in large language models (LLMs) have stepped forward the development of multilingual speech and machine translation by its reduced representation errors and incorporated external knowledge. However, both translation tasks typically utilize beam search decoding and top-1 hypothesis selection for inference. These techniques struggle to fully exploit the rich information in the diverse N-best hypotheses, making them less optimal for translation tasks that require a single, high-quality output sequence. In this paper, we propose a new generative paradigm for translation tasks, namely GenTranslate, which builds upon LLMs to generate better results from the diverse translation versions in N-best list. Leveraging the rich linguistic knowledge and strong reasoning abilities of LLMs, our new paradigm can integrate the diverse N-best candidates to generate a higher-quality translation result. Furthermore, to support LLM finetuning, we build and release a HypoTranslate dataset that contains over 592K hypotheses-translation pairs in 11 languages. Experiments on various speech and machine translation benchmarks (e.g., FLEURS, CoVoST-2, WMT) demonstrate that our GenTranslate significantly outperforms the state-of-the-art model.
Metaphors are commonly found in advertising and internet memes. However, the free form of internet memes often leads to a lack of high-quality textual data. Metaphor detection demands a deep interpretation of both textual and visual elements, requiring extensive common-sense knowledge, which poses a challenge to language models. To address these challenges, we propose a compact framework called C4MMD, which utilizes a Chain-of-Thought(CoT) method for Multi-modal Metaphor Detection. Specifically, our approach designs a three-step process inspired by CoT that extracts and integrates knowledge from Multi-modal Large Language Models(MLLMs) into smaller ones. We also developed a modality fusion architecture to transform knowledge from large models into metaphor features, supplemented by auxiliary tasks to improve model performance. Experimental results on the MET-MEME dataset demonstrate that our method not only effectively enhances the metaphor detection capabilities of small models but also outperforms existing models. To our knowledge, this is the first systematic study leveraging MLLMs in metaphor detection tasks. The code for our method is publicly available at https://github.com/xyz189411yt/C4MMD.
The upscaling of Large Language Models (LLMs) has yielded impressive advances in natural language processing, yet it also poses significant deployment challenges. Weight quantization has emerged as a widely embraced solution to reduce memory and computational demands. This paper introduces BitDistiller, a framework that synergizes Quantization-Aware Training (QAT) with Knowledge Distillation (KD) to boost the performance of LLMs at ultra-low precisions (sub-4-bit). Specifically, BitDistiller first incorporates a tailored asymmetric quantization and clipping technique to maximally preserve the fidelity of quantized weights, and then proposes a novel Confidence-Aware Kullback-Leibler Divergence (CAKLD) objective, which is employed in a self-distillation manner to enable faster convergence and superior model performance. Empirical evaluations demonstrate that BitDistiller significantly surpasses existing methods in both 3-bit and 2-bit configurations on general language understanding and complex reasoning benchmarks. Notably, BitDistiller is shown to be more cost-effective, demanding fewer data and training resources. The code is available at https://github.com/DD-DuDa/BitDistiller.
Temporal knowledge graph (TKG) reasoning has two settings: interpolation reasoning and extrapolation reasoning. Both of them draw plenty of research interest and have great significance. Methods of the former de-emphasize the temporal correlations among facts sequences, while methods of the latter require strict chronological order of knowledge and ignore inferring clues provided by missing facts of the past. These limit the practicability of TKG applications as almost all of the existing TKG reasoning methods are designed specifically to address either one setting. To this end, this paper proposes an original Temporal PAth-based Reasoning (TPAR) model for both the interpolation and extrapolation reasoning settings. TPAR performs a neural-driven symbolic reasoning fashion that is robust to ambiguous and noisy temporal data, and with fine interpretability as well. Comprehensive experiments show that TPAR outperforms SOTA methods on the link prediction task for both the interpolation and the extrapolation settings. A novel pipeline experimental setting is designed to evaluate the performances of SOTA combinations and the proposed TPAR towards interpolation and extrapolation reasoning. And more diverse experiments are conducted to show the robustness and interpretability of TPAR.
Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating additional information from retrieval. However, studies have shown that LLMs still face challenges in effectively using the retrieved information, even ignore it or be misled by it. The key reason is that the training of LLMs does not clearly make LLMs learn how to utilize input retrieved texts with varied quality. In this paper, we propose a novel perspective that considers the role of LLMs in RAG as “Information Refiner”, which means that regardless of correctness, completeness, or usefulness of retrieved texts, LLMs can consistently integrate knowledge within the retrieved texts and model parameters to generate the texts that are more concise, accurate, and complete than the retrieved texts. To this end, we propose an information refinement training method named INFO-RAG that optimizes LLMs for RAG in an unsupervised manner. INFO-RAG is low-cost and general across various tasks. Extensive experiments on zero-shot prediction of 11 datasets in diverse tasks including Question Answering, Slot-Filling, Language Modeling, Dialogue, and Code Generation show that INFO-RAG improves the performance of LLaMA2 by an average of 9.39% relative points. INFO-RAG also shows advantages in in-context learning and robustness of RAG.
In this paper, we present CSCD-NS, the first Chinese spelling check (CSC) dataset designed for native speakers, containing 40,000 samples from a Chinese social platform. Compared with existing CSC datasets aimed at Chinese learners, CSCD-NS is ten times larger in scale and exhibits a distinct error distribution, with a significantly higher proportion of word-level errors. To further enhance the data resource, we propose a novel method that simulates the input process through an input method, generating large-scale and high-quality pseudo data that closely resembles the actual error distribution and outperforms existing methods. Moreover, we investigate the performance of various models in this scenario, including large language models (LLMs), such as ChatGPT. The result indicates that generative models underperform BERT-like classification models due to strict length and pronunciation constraints. The high prevalence of word-level errors also makes CSC for native speakers challenging enough, leaving substantial room for improvement.
There is a lack of quantitative measures to evaluate the progression of topics through time in dynamic topic models (DTMs). Filling this gap, we propose a novel evaluation measure for DTMs that analyzes the changes in the quality of each topic over time. Additionally, we propose an extension combining topic quality with the model’s temporal consistency. We demonstrate the utility of the proposed measure by applying it to synthetic data and data from existing DTMs, including DTMs from large language models (LLMs). We also show that the proposed measure correlates well with human judgment. Our findings may help in identifying changing topics, evaluating different DTMs and LLMs, and guiding future research in this area.
Large language models (LLMs) with enormous pre-training tokens and parameters emerge diverse abilities, including math reasoning, codegeneration, and instruction following. These abilities are further enhanced by supervised fine-tuning (SFT). While the open-source community has explored ad-hoc SFT for enhancing individual capabilities, proprietary LLMs exhibit versatility across various skills. Therefore, understanding the facilitation of multiple abilities via SFT is paramount. In this study, we specificially focuses on the interplay of data composition between mathematical reasoning, code generation, and general human-aligning abilities during SFT. We propose four intriguing research questions to explore the association between model performance and various factors including data amount, composition ratio, model size and SFT strategies. Our experiments reveal that distinct capabilities scale differently and larger models generally show superior performance with same amount of data. Mathematical reasoning and code generation consistently improve with increasing data amount, whereas general abilities plateau after roughly a thousand samples. Moreover, we observe data composition appears to enhance various abilities under limited data conditions, yet can lead to performance conflicts when data is plentiful. Our findings also suggest the amount of composition data influences performance more than the composition ratio. In analysis of SFT strategies, we find that sequentially learning multiple skills risks catastrophic forgetting. Our proposed Dual-stage Mixed Fine-tuning (DMT) strategy offers a promising solution to learn multiple abilities with different scaling patterns.
In legal decisions, split votes (SV) occur when judges cannot reach a unanimous decision, posing a difficulty for lawyers who must navigate diverse legal arguments and opinions. In high-stakes domains, %as human-AI interaction systems become increasingly important, understanding the alignment of perceived difficulty between humans and AI systems is crucial to build trust. However, existing NLP calibration methods focus on a classifier’s awareness of predictive performance, measured against the human majority class, overlooking inherent human label variation (HLV). This paper explores split votes as naturally observable human disagreement and value pluralism. We collect judges’ vote distributions from the European Court of Human Rights (ECHR), and present SV-ECHR, a case outcome classification (COC) dataset with SV information. We build a taxonomy of disagreement with SV-specific subcategories. We further assess the alignment of perceived difficulty between models and humans, as well as confidence- and human-calibration of COC models. We observe limited alignment with the judge vote distribution. To our knowledge, this is the first systematic exploration of calibration to human judgements in legal NLP. Our study underscores the necessity for further research on measuring and enhancing model calibration considering HLV in legal decision tasks.
While Large Language Models (LLMs) have found success in real-world applications, their underlying explanatory process is still poorly understood. This paper proposes IBE-Eval, a framework inspired by philosophical accounts on Inference to the Best Explanation (IBE) to advance the interpretation and evaluation of LLMs’ explanations. IBE-Eval estimates the plausibility of natural language explanations through a combination of explicit logical and linguistic features including: consistency, parsimony, coherence, and uncertainty. Extensive experiments are conducted on Causal Question Answering (CQA), where IBE-Eval is tasked to select the most plausible causal explanation amongst competing ones generated by LLMs (i.e., GPT 3.5 and Llama 2). The experiments reveal that IBE-Eval can successfully identify the best explanation with up to 77% accuracy (≈ 27% above random), improving upon a GPT 3.5-as-a-Judge baseline (≈+17%) while being intrinsically more efficient and interpretable. Additional analyses suggest that, despite model-specific variances, LLM-generated explanations tend to conform to IBE criteria and that IBE-Eval is significantly correlated with human judgment, opening up opportunities for future development of automated explanation verification tools.
Natural language inference (NLI), the task of recognizing the entailment relationship in sentence pairs, is an actively studied topic serving as a proxy for natural language understanding. Despite the relevance of the task in building conversational agents and improving text classification, machine translation and other NLP tasks, to the best of our knowledge, there is no publicly available NLI corpus for the Romanian language. To this end, we introduce the first Romanian NLI corpus (RoNLI) comprising 58K training sentence pairs, which are obtained via distant supervision, and 6K validation and test sentence pairs, which are manually annotated with the correct labels. We conduct experiments with multiple machine learning methods based on distant learning, ranging from shallow models based on word embeddings to transformer-based neural networks, to establish a set of competitive baselines. Furthermore, we improve on the best model by employing a new curriculum learning strategy based on data cartography. Our dataset and code to reproduce the baselines are available at https://github.com/Eduard6421/RONLI.
Recent advances in few-shot question answering (QA) mostly rely on the power of pre-trained large language models (LLMs) and fine-tuning in specific settings. Although the pre-training stage has already equipped LLMs with powerful reasoning capabilities, LLMs still need to be fine-tuned to adapt to specific domains to achieve the best results. In this paper, we propose to select the most informative data for fine-tuning, thereby improving the efficiency of the fine-tuning process with comparative or even better accuracy on the open-domain QA task. We present MinPrompt, a minimal data augmentation framework for open-domain QA based on an approximate graph algorithm and unsupervised question generation. We transform the raw text into a graph structure to build connections between different factual sentences, then apply graph algorithms to identify the minimal set of sentences needed to cover the most information in the raw text. We then generate QA pairs based on the identified sentence subset and train the model on the selected sentences to obtain the final model. Empirical results on several benchmark datasets and theoretical analysis show that MinPrompt is able to achieve comparable or better results than baselines with a high degree of efficiency, bringing consistent improvements in F-1 scores.
Large language models hold significant potential for integrating various data types, such as text documents and database records, for advanced analytics. However, blending text and numerical data presents substantial challenges. LLMs need to process and cross-reference entities and numbers, handle data inconsistencies and redundancies, and develop planning capabilities such as building a working memory for managing complex data queries. In this paper, we introduce four novel tasks centered around sports data analytics to evaluate the numerical reasoning and information fusion capabilities of LLMs. These tasks involve providing LLMs with detailed, play-by-play sports game descriptions, then challenging them with adversarial scenarios such as new game rules, longer durations, scrambled narratives, and analyzing key statistics in game summaries. We conduct extensive experiments on NBA and NFL games to assess the performance of LLMs on these tasks. Our benchmark, SportsMetrics, introduces a new mechanism for assessing LLMs’ numerical reasoning and fusion skills.
We explore and enhance the ability of neural language models to generate novel scientific directions grounded in literature. Work on literature-based hypothesis generation has traditionally focused on binary link prediction—severely limiting the expressivity of hypotheses. This line of work also does not focus on optimizing novelty. We take a dramatic departure with a novel setting in which models use as input background contexts (e.g., problems, experimental settings, goals), and output natural language ideas grounded in literature. We present SciMON, a modeling framework that uses retrieval of “inspirations” from past scientific papers, and explicitly optimizes for novelty by iteratively comparing to prior papers and updating idea suggestions until sufficient novelty is achieved. Comprehensive evaluations reveal that GPT-4 tends to generate ideas with overall low technical depth and novelty, while our methods partially mitigate this issue. Our work represents a first step toward evaluating and developing language models that generate new ideas derived from the scientific literature. Code, data, and resources are publicly available for research purposes: https://github.com/eaglew/clbd.
We introduce EVLGen, a streamlined framework designed for the pre-training of visually conditioned language generation models with high computational demands, utilizing frozen pre-trained large language models (LLMs). The conventional approach in vision-language pre-training (VLP) typically involves a two-stage optimization process: an initial resource-intensive phase dedicated to general-purpose vision-language representation learning, focused on extracting and consolidating relevant visual features. This is followed by a subsequent phase that emphasizes end-to-end alignment between visual and linguistic modalities. Our novel one-stage, single-loss framework bypasses the computationally demanding first training stage by gradually merging similar visual tokens during training, while avoiding model collapse caused by single-stage training of BLIP-2 type models. The gradual merging process effectively condenses visual information while preserving semantic richness, resulting in rapid convergence without compromising performance. Our experimental findings demonstrate that our approach accelerates the training of vision-language models by a factor of 5 without a noticeable impact on overall performance. Furthermore, we illustrate that our models significantly narrow the performance gap to current vision-language models using only 1/10 of the data. Finally, we showcase how our image-text models can seamlessly adapt to video-conditioned language generation tasks through novel soft attentive temporal token contextualizing modules. Code: https://github.com/yiren-jian/EVLGen
As the use of Large Language Models (LLMs) becomes more widespread, understanding their self-evaluation of confidence in generated responses becomes increasingly important as it is integral to the reliability of the output of these models. We introduce the concept of Confidence-Probability Alignment, that connects an LLM’s internal confidence, quantified by token probabilities, to the confidence conveyed in the model’s response when explicitly asked about its certainty. Using various datasets and prompting techniques that encourage model introspection, we probe the alignment between models’ internal and expressed confidence. These techniques encompass using structured evaluation scales to rate confidence, including answer options when prompting, and eliciting the model’s confidence level for outputs it does not recognize as its own. Notably, among the models analyzed, OpenAI’s GPT-4 showed the strongest confidence-probability alignment, with an average Spearman’s ̂𝜌 of 0.42, across a wide range of tasks. Our work contributes to the ongoing efforts to facilitate risk assessment in the application of LLMs and to further our understanding of model trustworthiness.
Knowledge represented in Large Language Models (LLMs) is quite often incorrect and can also become obsolete over time. Updating knowledge via fine-tuning is computationally resource-hungry and not reliable, and so knowledge editing (KE) has developed as an effective and economical alternative to inject new knowledge or to fix factual errors in LLMs. Although there has been considerable interest in this area, current KE research exclusively focuses on monolingual settings, typically in English. However, what happens if the new knowledge is supplied in one language, but we would like to query an LLM in a different language? To address the problem of multilingual knowledge editing, we propose Retrieval-Augmented Multilingual Knowledge Editor (ReMaKE) to update knowledge in LLMs. ReMaKE can be used to perform model-agnostic knowledge editing in a multilingual setting. ReMaKE concatenates the new knowledge retrieved from a multilingual knowledge base with users’ prompts before querying an LLM. Our experimental results show that ReMaKE outperforms baseline knowledge editing methods by a significant margin and is scalable to real-word application scenarios. Our multilingual knowledge editing dataset (MzsRE) in 12 languages, the code, and additional project information are available at https://github.com/weixuan-wang123/ReMaKE.
Large Language Models (LLMs) have demonstrated remarkable success in tasks like the Winograd Schema Challenge (WSC), showcasing advanced textual common-sense reasoning. However, applying this reasoning to multimodal domains, where understanding text and images together is essential, remains a substantial challenge. To address this, we introduce WinoVis, a novel dataset specifically designed to probe text-to-image models on pronoun disambiguation within multimodal contexts. Utilizing GPT-4 for prompt generation and Diffusion Attentive Attribution Maps (DAAM) for heatmap analysis, we propose a novel evaluation framework that isolates the models’ ability in pronoun disambiguation from other visual processing challenges. Evaluation of successive model versions reveals that, despite incremental advancements, Stable Diffusion 2.0 achieves a precision of 56.7% on WinoVis, only marginally surpassing random guessing. Further error analysis identifies important areas for future research aimed at advancing text-to-image models in their ability to interpret and interact with the complex visual world.
Research on Large Language Models (LLMs) has often neglected subtle biases that, although less apparent, can significantly influence the models’ outputs toward particular social narratives. This study addresses two such biases within LLMs: representative bias, which denotes a tendency of LLMs to generate outputs that mirror the experiences of certain identity groups, and affinity bias, reflecting the models’ evaluative preferences for specific narratives or viewpoints. We introduce two novel metrics to measure these biases: the Representative Bias Score (RBS) and the Affinity Bias Score (ABS), and present the Creativity-Oriented Generation Suite (CoGS), a collection of open-ended tasks such as short story writing and poetry composition, designed with customized rubrics to detect these subtle biases. Our analysis uncovers marked representative biases in prominent LLMs, with a preference for identities associated with being white, straight, and men. Furthermore, our investigation of affinity bias reveals distinctive evaluative patterns within each model, akin to ‘bias fingerprints’. This trend is also seen in human evaluators, highlighting a complex interplay between human and machine bias perceptions.
The mainstream media has much leeway in what it chooses to cover and how it covers it. These choices have real-world consequences on what people know and their subsequent behaviors. However, the lack of objective measures to evaluate editorial choices makes research in this area particularly difficult. In this paper, we argue that there are newsworthy topics where objective measures exist in the form of supporting data and propose a computational framework to analyze editorial choices in this setup. We focus on the economy because the reporting of economic indicators presents us with a relatively easy way to determine both the selection and framing of various publications. Their values provide a ground truth of how the economy is doing relative to how the publications choose to cover it. To do this, we define frame prediction as a set of interdependent tasks. At the article level, we learn to identify the reported stance towards the general state of the economy. Then, for every numerical quantity reported in the article, we learn to identify whether it corresponds to an economic indicator and whether it is being reported in a positive or negative way. To perform our analysis, we track six American publishers and each article that appeared in the top 10 slots of their landing page between 2015 and 2023.
Multimodal Large Language Models (MLLMs) have demonstrated proficiency in handling a variety of visual-language tasks. However, current MLLM benchmarks are predominantly designed to evaluate reasoning based on static information about a single image, and the ability of modern MLLMs to extrapolate from image sequences, which is essential for understanding our ever-changing world, has been less investigated. To address this challenge, this paper introduces Mementos, a new benchmark designed to assess MLLMs’ sequential image reasoning abilities. Mementos features 4,761 diverse image sequences with varying lengths. We also employ a GPT-4 assisted method to evaluate MLLM reasoning performance. Through a careful evaluation of nine recent MLLMs on Mementos, including GPT-4V and Gemini, we find that they struggle to accurately describe dynamic information about given image sequences, often leading to hallucinations/misrepresentations of objects and their corresponding behaviors. Our quantitative analysis and case studies identify three key factors impacting MLLMs’ sequential image reasoning: the correlation between object and behavioral hallucinations, the influence of co-occurring behaviors, and the compounding impact of behavioral hallucinations.