Large language models (LLMs) are increasingly used in applications where LLM inputs may span many different tasks. Recent work has found that the choice of LLM is consequential, and different LLMs may be good for different input samples. Prior approaches have thus explored how engineers might select an LLM to use for each sample (i.e. routing). While existing routing methods mostly require training auxiliary models on human-annotated data, our work explores whether it is possible to perform unsupervised routing. We propose Smoothie, a weak supervision-inspired routing approach that requires no labeled data. Given a set of outputs from different LLMs, Smoothie constructs a latent variable graphical model over embedding representations of observable LLM outputs and unknown "true" outputs. Using this graphical model, we estimate sample-dependent quality scores for each LLM, and route each sample to the LLM with the highest corresponding score. We find that Smoothie's LLM quality-scores correlate with ground-truth model quality (correctly identifying the optimal model on 9/14 tasks), and that Smoothie outperforms baselines for routing by up to 10 points accuracy.
Text-to-SQL systems facilitate smooth interaction with databases by translating natural language queries into Structured Query Language (SQL), bridging the gap between non-technical users and complex database management systems. This survey provides a comprehensive overview of the evolution of AI-driven text-to-SQL systems, highlighting their foundational components, advancements in large language model (LLM) architectures, and the critical role of datasets such as Spider, WikiSQL, and CoSQL in driving progress. We examine the applications of text-to-SQL in domains like healthcare, education, and finance, emphasizing their transformative potential for improving data accessibility. Additionally, we analyze persistent challenges, including domain generalization, query optimization, support for multi-turn conversational interactions, and the limited availability of datasets tailored for NoSQL databases and dynamic real-world scenarios. To address these challenges, we outline future research directions, such as extending text-to-SQL capabilities to support NoSQL databases, designing datasets for dynamic multi-turn interactions, and optimizing systems for real-world scalability and robustness. By surveying current advancements and identifying key gaps, this paper aims to guide the next generation of research and applications in LLM-based text-to-SQL systems.
Recent studies show that large language models (LLMs) are powerful tools for working with natural language, bringing advances in many areas of computational linguistics. However, these models face challenges when applied to low-resource languages due to limited training data and difficulty in understanding cultural nuances. In this paper, we propose QueEn, a novel approach for Quechua-English translation that combines Retrieval-Augmented Generation (RAG) with parameter-efficient fine-tuning techniques. Our method leverages external linguistic resources through RAG and uses Low-Rank Adaptation (LoRA) for efficient model adaptation. Experimental results show that our approach substantially exceeds baseline models, with a BLEU score of 17.6 compared to 1.5 for standard GPT models. The integration of RAG with fine-tuning allows our system to address the challenges of low-resource language translation while maintaining computational efficiency. This work contributes to the broader goal of preserving endangered languages through advanced language technologies.
We introduce a novel method to enhance cross-language code translation from Fortran to C++ by integrating task-specific embedding alignment into a Retrieval-Augmented Generation (RAG) framework. Unlike conventional retrieval approaches that utilize generic embeddings agnostic to the downstream task, our strategy aligns the retrieval model directly with the objective of maximizing translation quality, as quantified by the CodeBLEU metric. This alignment ensures that the embeddings are semantically and syntactically meaningful for the specific code translation task. Our methodology involves constructing a dataset of 25,000 Fortran code snippets sourced from Stack-V2 dataset and generating their corresponding C++ translations using the LLaMA 3.1-8B language model. We compute pairwise CodeBLEU scores between the generated translations and ground truth examples to capture fine-grained similarities. These scores serve as supervision signals in a contrastive learning framework, where we optimize the embedding model to retrieve Fortran-C++ pairs that are most beneficial for improving the language model's translation performance. By integrating these CodeBLEU-optimized embeddings into the RAG framework, our approach significantly enhances both retrieval accuracy and code generation quality over methods employing generic embeddings. On the HPC Fortran2C++ dataset, our method elevates the average CodeBLEU score from 0.64 to 0.73, achieving a 14% relative improvement. On the Numerical Recipes dataset, we observe an increase from 0.52 to 0.60, marking a 15% relative improvement. Importantly, these gains are realized without any fine-tuning of the language model, underscoring the efficiency and practicality of our approach.
The proliferation of large language models has raised growing concerns about their misuse, particularly in cases where AI-generated text is falsely attributed to human authors. Machine-generated content detectors claim to effectively identify such text under various conditions and from any language model. This paper critically evaluates these claims by assessing several popular detectors (RADAR, Wild, T5Sentinel, Fast-DetectGPT, GPTID, LogRank, Binoculars) on a range of domains, datasets, and models that these detectors have not previously encountered. We employ various prompting strategies to simulate adversarial attacks, demonstrating that even moderate efforts can significantly evade detection. We emphasize the importance of the true positive rate at a specific false positive rate (TPR@FPR) metric and demonstrate that these detectors perform poorly in certain settings, with TPR@.01 as low as 0%. Our findings suggest that both trained and zero-shot detectors struggle to maintain high sensitivity while achieving a reasonable true positive rate.
We develop task scaling laws and model ladders to predict the individual task performance of pretrained language models (LMs) in the overtrained setting. Standard power laws for language modeling loss cannot accurately model task performance. Therefore, we leverage a two-step prediction approach: first use model and data size to predict a task-specific loss, and then use this task loss to predict task performance. We train a set of small-scale "ladder" models, collect data points to fit the parameterized functions of the two prediction steps, and make predictions for two target models: a 7B model trained to 4T tokens and a 13B model trained to 5T tokens. Training the ladder models only costs 1% of the compute used for the target models. On four multiple-choice tasks written in ranked classification format, we can predict the accuracy of both target models within 2 points of absolute error. We have higher prediction error on four other tasks (average absolute error 6.9) and find that these are often tasks with higher variance in task metrics. We also find that using less compute to train fewer ladder models tends to deteriorate predictions. Finally, we empirically show that our design choices and the two-step approach lead to superior performance in establishing scaling laws.
Procedural Knowledge is the know-how expressed in the form of sequences of steps needed to perform some tasks. Procedures are usually described by means of natural language texts, such as recipes or maintenance manuals, possibly spread across different documents and systems, and their interpretation and subsequent execution is often left to the reader. Representing such procedures in a Knowledge Graph (KG) can be the basis to build digital tools to support those users who need to apply or execute them. In this paper, we leverage Large Language Model (LLM) capabilities and propose a prompt engineering approach to extract steps, actions, objects, equipment and temporal information from a textual procedure, in order to populate a Procedural KG according to a pre-defined ontology. We evaluate the KG extraction results by means of a user study, in order to qualitatively and quantitatively assess the perceived quality and usefulness of the LLM-extracted procedural knowledge. We show that LLMs can produce outputs of acceptable quality and we assess the subjective perception of AI by human evaluators.
Recent approaches to large language model (LLM) alignment typically require millions of human annotations or rely on external aligned models for synthetic data generation. This paper introduces ALMA: Alignment with Minimal Annotation, demonstrating that effective alignment can be achieved using only 9,000 labeled examples -- less than 1% of conventional approaches. ALMA generates large amounts of high-quality synthetic alignment data through new techniques: diverse prompt synthesis via few-shot learning, diverse response generation with multiple model checkpoints, and judge (reward model) enhancement through score aggregation and self-distillation. Using only a pretrained Llama3 base model, 5,000 SFT examples, and 4,000 judge annotations, ALMA achieves performance close to Llama3-Instruct across diverse alignment benchmarks (e.g., 0.1% difference on AlpacaEval 2.0 score). These results are achieved with a multi-round, self-bootstrapped data synthesis and training recipe that continues to improve for 10 rounds, surpassing the typical 3-round ceiling of previous methods. These results suggest that base models already possess sufficient knowledge for effective alignment, and that synthetic data generation methods can expose it.
Dialectal Arabic (DA) varieties are under-served by language technologies, particularly large language models (LLMs). This trend threatens to exacerbate existing social inequalities and limits LLM applications, yet the research community lacks operationalized performance measurements in DA. We present a framework that comprehensively assesses LLMs' DA modeling capabilities across four dimensions: fidelity, understanding, quality, and diglossia. We evaluate nine LLMs in eight DA varieties and provide practical recommendations. Our evaluation suggests that LLMs do not produce DA as well as they understand it, not because their DA fluency is poor, but because they are reluctant to generate DA. Further analysis suggests that current post-training can contribute to bias against DA, that few-shot examples can overcome this deficiency, and that otherwise no measurable features of input text correlate well with LLM DA performance.
The rapid growth of academic publications has exacerbated the issue of author name ambiguity in online digital libraries. Despite advances in name disambiguation algorithms, cumulative errors continue to undermine the reliability of academic systems. It is estimated that over 10% paper-author assignments are rectified when constructing the million-scale WhoIsWho benchmark. Existing endeavors to detect incorrect assignments are either semantic-based or graph-based approaches, which fall short of making full use of the rich text attributes of papers and implicit structural features defined via the co-occurrence of paper attributes. To this end, this paper introduces a structure-enhanced language model that combines key structural features from graph-based methods with fine-grained semantic features from rich paper attributes to detect incorrect assignments. The proposed model is trained with a highly effective multi-modal multi-turn instruction tuning framework, which incorporates task-guided instruction tuning, text-attribute modality, and structural modality. Experimental results demonstrate that our model outperforms previous approaches, achieving top performance on the leaderboard of KDD Cup 2024. Our code has been publicly available.
Game-theoretic scenarios have become pivotal in evaluating the social intelligence of Large Language Model (LLM)-based social agents. While numerous studies have explored these agents in such settings, there is a lack of a comprehensive survey summarizing the current progress. To address this gap, we systematically review existing research on LLM-based social agents within game-theoretic scenarios. Our survey organizes the findings into three core components: Game Framework, Social Agent, and Evaluation Protocol. The game framework encompasses diverse game scenarios, ranging from choice-focusing to communication-focusing games. The social agent part explores agents' preferences, beliefs, and reasoning abilities. The evaluation protocol covers both game-agnostic and game-specific metrics for assessing agent performance. By reflecting on the current research and identifying future research directions, this survey provides insights to advance the development and evaluation of social agents in game-theoretic scenarios.
We propose a suite of tasks to evaluate the instrumental self-reasoning ability of large language model (LLM) agents. Instrumental self-reasoning ability could improve adaptability and enable self-modification, but it could also pose significant risks, such as enabling deceptive alignment. Prior work has only evaluated self-reasoning in non-agentic settings or in limited domains. In this paper, we propose evaluations for instrumental self-reasoning ability in agentic tasks in a wide range of scenarios, including self-modification, knowledge seeking, and opaque self-reasoning. We evaluate agents built using state-of-the-art LLMs, including commercial and open source systems. We find that instrumental self-reasoning ability emerges only in the most capable frontier models and that it is highly context-dependent. No model passes the the most difficult versions of our evaluations, hence our evaluation can be used to measure increases in instrumental self-reasoning ability in future models. We open-source our evaluations at https://github.com/kaifronsdal/Self-Reasoning-Evals.
Pretrained transformer-based Language Models (LMs) are well-known for their ability to achieve significant improvement on NLP tasks, but their black-box nature, which leads to a lack of interpretability, has been a major concern. My dissertation focuses on developing intrinsically interpretable models when using LMs as encoders while maintaining their superior performance via prototypical networks. I initiated my research by investigating enhancements in performance for interpretable models of sarcasm detection. My proposed approach focuses on capturing sentiment incongruity to enhance accuracy while offering instance-based explanations for the classification decisions. Later, I developed a novel white-box multi-head graph attention-based prototype network designed to explain the decisions of text classification models without sacrificing the accuracy of the original black-box LMs. In addition, I am working on extending the attention-based prototype network with contrastive learning to redesign an interpretable graph neural network, aiming to enhance both the interpretability and performance of the model in document classification.
Modern language models are internally -- and mathematically -- distributions over token strings rather than \emph{character} strings, posing numerous challenges for programmers building user applications on top of them. For example, if a prompt is specified as a character string, it must be tokenized before passing it to the token-level language model. Thus, the tokenizer and consequent analyses are very sensitive to the specification of the prompt (e.g., if the prompt ends with a space or not). This paper presents algorithms for converting token-level language models to character-level ones. We present both exact and approximate algorithms. In the empirical portion of the paper, we benchmark the practical runtime and approximation quality. We find that -- even with a small computation budget -- our method is able to accurately approximate the character-level distribution (less than 0.00021 excess bits / character) at reasonably fast speeds (46.3 characters / second) on the Llama 3.1 8B language model.
Given the increasing use of synthetic data in language model (LM) post-training, an LM's ability to generate high-quality data has become nearly as crucial as its ability to solve problems directly. While prior works have focused on developing effective data generation methods, they lack systematic comparison of different LMs as data generators in a unified setting. To address this gap, we propose AgoraBench, a benchmark that provides standardized settings and metrics to evaluate LMs' data generation abilities. Through synthesizing 1.26 million training instances using 6 LMs and training 99 student models, we uncover key insights about LMs' data generation capabilities. First, we observe that LMs exhibit distinct strengths. For instance, GPT-4o excels at generating new problems, while Claude-3.5-Sonnet performs better at enhancing existing ones. Furthermore, our analysis reveals that an LM's data generation ability doesn't necessarily correlate with its problem-solving ability. Instead, multiple intrinsic features of data quality-including response quality, perplexity, and instruction difficulty-collectively serve as better indicators. Finally, we demonstrate that strategic choices in output format and cost-conscious model selection significantly impact data generation effectiveness.
Large language models have transformed the comprehension and generation of natural language tasks, but they come with substantial memory and computational requirements. Quantization techniques have emerged as a promising avenue for addressing these challenges while preserving accuracy and making energy efficient. We propose CPTQuant, a comprehensive strategy that introduces correlation-based (CMPQ), pruning-based (PMPQ), and Taylor decomposition-based (TDMPQ) mixed precision techniques. CMPQ adapts the precision level based on canonical correlation analysis of different layers. PMPQ optimizes precision layer-wise based on their sensitivity to sparsity. TDMPQ modifies precision using Taylor decomposition to assess each layer's sensitivity to input perturbation. These strategies allocate higher precision to more sensitive layers while diminishing precision to robust layers. CPTQuant assesses the performance across BERT, OPT-125M, OPT-350M, OPT-1.3B, and OPT-2.7B. We demonstrate up to 4x compression and a 2x-fold increase in efficiency with minimal accuracy drop compared to Hugging Face FP16. PMPQ stands out for achieving a considerably higher model compression. Sensitivity analyses across various LLMs show that the initial and final 30% of layers exhibit higher sensitivities than the remaining layers. PMPQ demonstrates an 11% higher compression ratio than other methods for classification tasks, while TDMPQ achieves a 30% greater compression ratio for language modeling tasks.
Coronavirus Disease 2019 (COVID-19), which emerged in 2019, has caused millions of deaths worldwide. Although effective vaccines have been developed to mitigate severe symptoms, certain populations, particularly the elderly and those with comorbidities, remain at high risk for severe outcomes and increased mortality. Consequently, early identification of the severity and clinical outcomes of the disease in these patients is vital to prevent adverse prognoses. Although traditional machine learning and deep learning models have been widely employed in this area, the potential of large language models (LLMs) remains largely unexplored. Our research focuses primarily on constructing specialized prompts and adopting multi-objective learning strategies. We started by selecting serological indicators that significantly correlate with clinical outcomes and disease severity to serve as input data for the model. Blood test samples often contain numerous missing values, and traditional models generally rely on imputation to handle these gaps in the data. In contrast, LLMs offer the advantage of robust semantic understanding. By setting prompts, we can explicitly inform the model when a feature's value is missing, without the need for imputation. For the multi-objective learning strategy, the model is designed to first predict disease severity and then predict clinical outcomes. Given that LLMs utilize both the input text and the generated tokens as input for generating the next token, the predicted severity is used as a basis for generating the clinical outcome. During the fine-tuning of the LLM, the two objectives influence and improve each other. Our experiments were implemented based on the ChatGLM model. The results demonstrate the effectiveness of LLMs in this task, suggesting promising potential for further development.
The advent of Multimodal Large Language Models, leveraging the power of Large Language Models, has recently demonstrated superior multimodal understanding and reasoning abilities, heralding a new era for artificial general intelligence. However, achieving AGI necessitates more than just comprehension and reasoning. A crucial capability required is effective planning in diverse scenarios, which involves making reasonable decisions based on complex environments to solve real-world problems. Despite its importance, the planning abilities of current MLLMs in varied scenarios remain underexplored. In this paper, we introduce EgoPlan-Bench2, a rigorous and comprehensive benchmark designed to assess the planning capabilities of MLLMs across a wide range of real-world scenarios. EgoPlan-Bench2 encompasses everyday tasks spanning 4 major domains and 24 detailed scenarios, closely aligned with human daily life. EgoPlan-Bench2 is constructed through a semi-automatic process utilizing egocentric videos, complemented by manual verification. Grounded in a first-person perspective, it mirrors the way humans approach problem-solving in everyday life. We evaluate 21 competitive MLLMs and provide an in-depth analysis of their limitations, revealing that they face significant challenges in real-world planning. To further improve the planning proficiency of current MLLMs, we propose a training-free approach using multimodal Chain-of-Thought (CoT) prompting through investigating the effectiveness of various multimodal prompts in complex planning. Our approach enhances the performance of GPT-4V by 10.24 on EgoPlan-Bench2 without additional training. Our work not only sheds light on the current limitations of MLLMs in planning, but also provides insights for future enhancements in this critical area. We have made data and code available at https://qiulu66.github.io/egoplanbench2/.
We present Liquid, an auto-regressive generation paradigm that seamlessly integrates visual comprehension and generation by tokenizing images into discrete codes and learning these code embeddings alongside text tokens within a shared feature space for both vision and language. Unlike previous multimodal large language model (MLLM), Liquid achieves this integration using a single large language model (LLM), eliminating the need for external pretrained visual embeddings such as CLIP. For the first time, Liquid uncovers a scaling law that performance drop unavoidably brought by the unified training of visual and language tasks diminishes as the model size increases. Furthermore, the unified token space enables visual generation and comprehension tasks to mutually enhance each other, effectively removing the typical interference seen in earlier models. We show that existing LLMs can serve as strong foundations for Liquid, saving 100x in training costs while outperforming Chameleon in multimodal capabilities and maintaining language performance comparable to mainstream LLMs like LLAMA2. Liquid also outperforms models like SD v2.1 and SD-XL (FID of 5.47 on MJHQ-30K), excelling in both vision-language and text-only tasks. This work demonstrates that LLMs such as LLAMA3.2 and GEMMA2 are powerful multimodal generators, offering a scalable solution for enhancing both vision-language understanding and generation. The code and models will be released at https://github.com/FoundationVision/Liquid.
The digitization of vocal music scores presents unique challenges that go beyond traditional Optical Music Recognition (OMR) and Optical Character Recognition (OCR), as it necessitates preserving the critical alignment between music notation and lyrics. This alignment is essential for proper interpretation and processing in practical applications. This paper introduces and formalizes, for the first time, the Aligned Music Notation and Lyrics Transcription (AMNLT) challenge, which addresses the complete transcription of vocal scores by jointly considering music symbols, lyrics, and their synchronization. We analyze different approaches to address this challenge, ranging from traditional divide-and-conquer methods that handle music and lyrics separately, to novel end-to-end solutions including direct transcription, unfolding mechanisms, and language modeling. To evaluate these methods, we introduce four datasets of Gregorian chants, comprising both real and synthetic sources, along with custom metrics specifically designed to assess both transcription and alignment accuracy. Our experimental results demonstrate that end-to-end approaches generally outperform heuristic methods in the alignment challenge, with language models showing particular promise in scenarios where sufficient training data is available. This work establishes the first comprehensive framework for AMNLT, providing both theoretical foundations and practical solutions for preserving and digitizing vocal music heritage.