Vision Language Models excel in handling a wide range of complex tasks, including Optical Character Recognition (OCR), Visual Question Answering (VQA), and advanced geometric reasoning. However, these models fail to perform well on low-level basic visual tasks which are especially easy for humans. Our goal in this work was to determine if these models are truly "blind" to geometric reasoning or if there are ways to enhance their capabilities in this area. Our work presents a novel automatic pipeline designed to extract key information from images in response to specific questions. Instead of just relying on direct VQA, we use question-derived keywords to create a caption that highlights important details in the image related to the question. This caption is then used by a language model to provide a precise answer to the question without requiring external fine-tuning.
Generative models such as Large Language Models (LLM) and Multimodal Large Language models (MLLMs) trained on massive web corpora can memorize and disclose individuals' confidential and private data, raising legal and ethical concerns. While many previous works have addressed this issue in LLM via machine unlearning, it remains largely unexplored for MLLMs. To tackle this challenge, we introduce Multimodal Large Language Model Unlearning Benchmark (MLLMU-Bench), a novel benchmark aimed at advancing the understanding of multimodal machine unlearning. MLLMU-Bench consists of 500 fictitious profiles and 153 profiles for public celebrities, each profile feature over 14 customized question-answer pairs, evaluated from both multimodal (image+text) and unimodal (text) perspectives. The benchmark is divided into four sets to assess unlearning algorithms in terms of efficacy, generalizability, and model utility. Finally, we provide baseline results using existing generative model unlearning algorithms. Surprisingly, our experiments show that unimodal unlearning algorithms excel in generation and cloze tasks, while multimodal unlearning approaches perform better in classification tasks with multimodal inputs.
Multilingual large language models (LLMs) have gained prominence, but concerns arise regarding their reliability beyond English. This study addresses the gap in cross-lingual semantic evaluation by introducing a novel benchmark for cross-lingual sense disambiguation, StingrayBench. In this paper, we demonstrate using false friends -- words that are orthographically similar but have completely different meanings in two languages -- as a possible approach to pinpoint the limitation of cross-lingual sense disambiguation in LLMs. We collect false friends in four language pairs, namely Indonesian-Malay, Indonesian-Tagalog, Chinese-Japanese, and English-German; and challenge LLMs to distinguish the use of them in context. In our analysis of various models, we observe they tend to be biased toward higher-resource languages. We also propose new metrics for quantifying the cross-lingual sense bias and comprehension based on our benchmark. Our work contributes to developing more diverse and inclusive language modeling, promoting fairer access for the wider multilingual community.
Although large language models (LLMs) have demonstrated their strong intelligence ability, the high demand for computation and storage hinders their practical application. To this end, many model compression techniques are proposed to increase the efficiency of LLMs. However, current researches only validate their methods on limited models, datasets, metrics, etc, and still lack a comprehensive evaluation under more general scenarios. So it is still a question of which model compression approach we should use under a specific case. To mitigate this gap, we present the Large Language Model Compression Benchmark (LLMCBench), a rigorously designed benchmark with an in-depth analysis for LLM compression algorithms. We first analyze the actual model production requirements and carefully design evaluation tracks and metrics. Then, we conduct extensive experiments and comparison using multiple mainstream LLM compression approaches. Finally, we perform an in-depth analysis based on the evaluation and provide useful insight for LLM compression design. We hope our LLMCBench can contribute insightful suggestions for LLM compression algorithm design and serve as a foundation for future research. Our code is available at https://github.com/AboveParadise/LLMCBench.
Sparse Autoencoders (SAEs) have emerged as a powerful unsupervised method for extracting sparse representations from language models, yet scalable training remains a significant challenge. We introduce a suite of 256 SAEs, trained on each layer and sublayer of the Llama-3.1-8B-Base model, with 32K and 128K features. Modifications to a state-of-the-art SAE variant, Top-K SAEs, are evaluated across multiple dimensions. In particular, we assess the generalizability of SAEs trained on base models to longer contexts and fine-tuned models. Additionally, we analyze the geometry of learned SAE latents, confirming that \emph{feature splitting} enables the discovery of new features. The Llama Scope SAE checkpoints are publicly available at~\url{https://huggingface.co/fnlp/Llama-Scope}, alongside our scalable training, interpretation, and visualization tools at \url{https://github.com/OpenMOSS/Language-Model-SAEs}. These contributions aim to advance the open-source Sparse Autoencoder ecosystem and support mechanistic interpretability research by reducing the need for redundant SAE training.
Watermarking is a technique that involves embedding nearly unnoticeable statistical signals within generated content to help trace its source. This work focuses on a scenario where an untrusted third-party user sends prompts to a trusted language model (LLM) provider, who then generates a text from their LLM with a watermark. This setup makes it possible for a detector to later identify the source of the text if the user publishes it. The user can modify the generated text by substitutions, insertions, or deletions. Our objective is to develop a statistical method to detect if a published text is LLM-generated from the perspective of a detector. We further propose a methodology to segment the published text into watermarked and non-watermarked sub-strings. The proposed approach is built upon randomization tests and change point detection techniques. We demonstrate that our method ensures Type I and Type II error control and can accurately identify watermarked sub-strings by finding the corresponding change point locations. To validate our technique, we apply it to texts generated by several language models with prompts extracted from Google's C4 dataset and obtain encouraging numerical results. We release all code publicly at https://github.com/doccstat/llm-watermark-cpd.
Establishing a unified theory of cognition has been a major goal of psychology. While there have been previous attempts to instantiate such theories by building computational models, we currently do not have one model that captures the human mind in its entirety. Here we introduce Centaur, a computational model that can predict and simulate human behavior in any experiment expressible in natural language. We derived Centaur by finetuning a state-of-the-art language model on a novel, large-scale data set called Psych-101. Psych-101 reaches an unprecedented scale, covering trial-by-trial data from over 60,000 participants performing over 10,000,000 choices in 160 experiments. Centaur not only captures the behavior of held-out participants better than existing cognitive models, but also generalizes to new cover stories, structural task modifications, and entirely new domains. Furthermore, we find that the model's internal representations become more aligned with human neural activity after finetuning. Taken together, Centaur is the first real candidate for a unified model of human cognition. We anticipate that it will have a disruptive impact on the cognitive sciences, challenging the existing paradigm for developing computational models.
The synthesis of inorganic crystalline materials is essential for modern technology, especially in quantum materials development. However, designing efficient synthesis workflows remains a significant challenge due to the precise experimental conditions and extensive trial and error. Here, we present a framework using large language models (LLMs) to predict synthesis pathways for inorganic materials, including quantum materials. Our framework contains three models: LHS2RHS, predicting products from reactants; RHS2LHS, predicting reactants from products; and TGT2CEQ, generating full chemical equations for target compounds. Fine-tuned on a text-mined synthesis database, our model raises accuracy from under 40% with pretrained models, to under 80% using conventional fine-tuning, and further to around 90% with our proposed generalized Tanimoto similarity, while maintaining robust to additional synthesis steps. Our model further demonstrates comparable performance across materials with varying degrees of quantumness quantified using quantum weight, indicating that LLMs offer a powerful tool to predict balanced chemical equations for quantum materials discovery.
This Paper presents an advanced approach for fine-tuning BiomedCLIP PubMedBERT, a multimodal model, to classify abnormalities in Video Capsule Endoscopy (VCE) frames, aiming to enhance diagnostic efficiency in gastrointestinal healthcare. By integrating the PubMedBERT language model with a Vision Transformer (ViT) to process endoscopic images, our method categorizes images into ten specific classes: angioectasia, bleeding, erosion, erythema, foreign body, lymphangiectasia, polyp, ulcer, worms, and normal. Our workflow incorporates image preprocessing and fine-tunes the BiomedCLIP model to generate high-quality embeddings for both visual and textual inputs, aligning them through similarity scoring for classification. Performance metrics, including classification, accuracy, recall, and F1 score, indicate the models strong ability to accurately identify abnormalities in endoscopic frames, showing promise for practical use in clinical diagnostics.
The development of large language models and multi-modal models has enabled the appealing idea of generating novel molecules from text descriptions. Generative modeling would shift the paradigm from relying on large-scale chemical screening to find molecules with desired properties to directly generating those molecules. However, multi-modal models combining text and molecules are often trained from scratch, without leveraging existing high-quality pretrained models. Training from scratch consumes more computational resources and prohibits model scaling. In contrast, we propose a lightweight adapter-based strategy named Chemical Language Model Linker (ChemLML). ChemLML blends the two single domain models and obtains conditional molecular generation from text descriptions while still operating in the specialized embedding spaces of the molecular domain. ChemLML can tailor diverse pretrained text models for molecule generation by training relatively few adapter parameters. We find that the choice of molecular representation used within ChemLML, SMILES versus SELFIES, has a strong influence on conditional molecular generation performance. SMILES is often preferable despite not guaranteeing valid molecules. We raise issues in using the entire PubChem dataset of molecules and their associated descriptions for evaluating molecule generation and provide a filtered version of the dataset as a generation test set. To demonstrate how ChemLML could be used in practice, we generate candidate protein inhibitors and use docking to assess their quality and also generate candidate membrane permeable molecules.
Recent advancements in Large Language Models (LLMs) have significantly improved text generation capabilities, but these systems are still known to hallucinate, and granular uncertainty estimation for long-form LLM generations remains challenging. In this work, we propose Graph Uncertainty -- which represents the relationship between LLM generations and claims within them as a bipartite graph and estimates the claim-level uncertainty with a family of graph centrality metrics. Under this view, existing uncertainty estimation methods based on the concept of self-consistency can be viewed as using degree centrality as an uncertainty measure, and we show that more sophisticated alternatives such as closeness centrality provide consistent gains at claim-level uncertainty estimation. Moreover, we present uncertainty-aware decoding techniques that leverage both the graph structure and uncertainty estimates to improve the factuality of LLM generations by preserving only the most reliable claims. Compared to existing methods, our graph-based uncertainty metrics lead to an average of 6.8% relative gains on AUPRC across various long-form generation settings, and our end-to-end system provides consistent 2-4% gains in factuality over existing decoding techniques while significantly improving the informativeness of generated responses.
Trajectory modeling, which includes research on trajectory data pattern mining and future prediction, has widespread applications in areas such as life services, urban transportation, and public administration. Numerous methods have been proposed to address specific problems within trajectory modelling. However, due to the heterogeneity of data and the diversity of trajectory tasks, achieving unified trajectory modelling remains an important yet challenging task. In this paper, we propose TrajAgent, a large language model-based agentic framework, to unify various trajectory modelling tasks. In TrajAgent, we first develop UniEnv, an execution environment with a unified data and model interface, to support the execution and training of various models. Building on UniEnv, we introduce TAgent, an agentic workflow designed for automatic trajectory modelling across various trajectory tasks. Specifically, we design AutOpt, a systematic optimization module within TAgent, to further improve the performance of the integrated model. With diverse trajectory tasks input in natural language, TrajAgent automatically generates competitive results via training and executing appropriate models. Extensive experiments on four tasks using four real-world datasets demonstrate the effectiveness of TrajAgent in unified trajectory modelling, achieving an average performance improvement of 15.43% over baseline methods.
This study introduces SLLMBO, an innovative framework leveraging large language models (LLMs) for hyperparameter optimization (HPO), incorporating dynamic search space adaptability, enhanced parameter space exploitation, and a novel LLM-tree-structured parzen estimator (LLM-TPE) sampler. By addressing limitations in recent fully LLM-based methods and traditional bayesian optimization (BO), SLLMBO achieves more robust optimization. This comprehensive benchmarking evaluates multiple LLMs, including GPT-3.5-Turbo, GPT-4o, Claude-Sonnet-3.5, and Gemini-1.5-Flash, extending prior work and establishing SLLMBO as the first framework to benchmark a diverse set of LLMs for HPO. By integrating LLMs' established strengths in parameter initialization with the exploitation abilities demonstrated in this study, alongside TPE's exploration capabilities, the LLM-TPE sampler achieves a balanced exploration-exploitation trade-off, reduces API costs, and mitigates premature early stoppings for more effective parameter searches. Across 14 tabular tasks in classification and regression, the LLM-TPE sampler outperformed fully LLM-based methods and achieved superior results over BO methods in 9 tasks. Testing early stopping in budget-constrained scenarios demonstrated competitive performance, indicating that LLM-based methods generally benefit from extended iterations for optimal results. This work lays the foundation for future research exploring open-source LLMs, reproducibility of LLM results in HPO, and benchmarking SLLMBO on complex datasets, such as image classification, segmentation, and machine translation.
Enzyme engineering enables the modification of wild-type proteins to meet industrial and research demands by enhancing catalytic activity, stability, binding affinities, and other properties. The emergence of deep learning methods for protein modeling has demonstrated superior results at lower costs compared to traditional approaches such as directed evolution and rational design. In mutation effect prediction, the key to pre-training deep learning models lies in accurately interpreting the complex relationships among protein sequence, structure, and function. This study introduces a retrieval-enhanced protein language model for comprehensive analysis of native properties from sequence and local structural interactions, as well as evolutionary properties from retrieved homologous sequences. The state-of-the-art performance of the proposed ProtREM is validated on over 2 million mutants across 217 assays from an open benchmark (ProteinGym). We also conducted post-hoc analyses of the model's ability to improve the stability and binding affinity of a VHH antibody. Additionally, we designed 10 new mutants on a DNA polymerase and conducted wet-lab experiments to evaluate their enhanced activity at higher temperatures. Both in silico and experimental evaluations confirmed that our method provides reliable predictions of mutation effects, offering an auxiliary tool for biologists aiming to evolve existing enzymes. The implementation is publicly available at https://github.com/tyang816/ProtREM.
In recent years, natural language processing (NLP) models have demonstrated remarkable capabilities in various domains beyond traditional text generation. In this work, we introduce PeptideGPT, a protein language model tailored to generate protein sequences with distinct properties: hemolytic activity, solubility, and non-fouling characteristics. To facilitate a rigorous evaluation of these generated sequences, we established a comprehensive evaluation pipeline consisting of ideas from bioinformatics to retain valid proteins with ordered structures. First, we rank the generated sequences based on their perplexity scores, then we filter out those lying outside the permissible convex hull of proteins. Finally, we predict the structure using ESMFold and select the proteins with pLDDT values greater than 70 to ensure ordered structure. The properties of generated sequences are evaluated using task-specific classifiers - PeptideBERT and HAPPENN. We achieved an accuracy of 76.26% in hemolytic, 72.46% in non-hemolytic, 78.84% in non-fouling, and 68.06% in solubility protein generation. Our experimental results demonstrate the effectiveness of PeptideGPT in de novo protein design and underscore the potential of leveraging NLP-based approaches for paving the way for future innovations and breakthroughs in synthetic biology and bioinformatics. Codes, models, and data used in this study are freely available at: https://github.com/aayush-shah14/PeptideGPT.
Large Language Models (LLMs) have demonstrated strong reasoning abilities, making them suitable for complex tasks such as graph computation. Traditional reasoning steps paradigm for graph problems is hindered by unverifiable steps, limited long-term reasoning, and poor generalization to graph variations. To overcome these limitations, we introduce GCoder, a code-based LLM designed to enhance problem-solving in generalized graph computation problems. Our method involves constructing an extensive training dataset, GraphWild, featuring diverse graph formats and algorithms. We employ a multi-stage training process, including Supervised Fine-Tuning (SFT) and Reinforcement Learning from Compiler Feedback (RLCF), to refine model capabilities. For unseen tasks, a hybrid retrieval technique is used to augment performance. Experiments demonstrate that GCoder outperforms GPT-4o, with an average accuracy improvement of 16.42% across various graph computational problems. Furthermore, GCoder efficiently manages large-scale graphs with millions of nodes and diverse input formats, overcoming the limitations of previous models focused on the reasoning steps paradigm. This advancement paves the way for more intuitive and effective graph problem-solving using LLMs. Code and data are available at here: https://github.com/Bklight999/WWW25-GCoder/tree/master.
FP8 training has emerged as a promising method for improving training efficiency. Existing frameworks accelerate training by applying FP8 computation to linear layers while leaving optimizer states and activations in higher precision, which fails to fully optimize memory usage. This paper introduces COAT (Compressing Optimizer States and Activations for FP8 Training), a novel FP8 training framework designed to significantly reduce memory footprint when training large models. COAT addresses current limitations through two key innovations: (1) Dynamic Range Expansion, which aligns optimizer state distributions more closely with the FP8 representation range, thereby reducing quantization error, and (2) Mixed-Granularity Activation Quantization, which optimizes activation memory using a combination of per-tensor and per-group quantization strategies. Experiments demonstrate that COAT effectively reduces end-to-end training memory footprint by 1.54x compared to BF16 while achieving nearly lossless performance across various tasks, such as Large Language Model pretraining and fine-tuning and Vision Language Model training. COAT also achieves a 1.43x end-to-end training speedup compared to BF16, performing on par with or surpassing TransformerEngine's speedup. COAT enables efficient full-parameter training of large models on fewer GPUs, and facilitates doubling the batch size in distributed training settings, providing a practical solution for scaling large-scale model training. The code is available at https://github.com/NVlabs/COAT.
Humans develop their grammars by making structural generalizations from finite input. We ask how filler-gap dependencies, which share a structural generalization despite diverse surface forms, might arise from the input. We explicitly control the input to a neural language model (NLM) to uncover whether the model posits a shared representation for filler-gap dependencies. We show that while NLMs do have success differentiating grammatical from ungrammatical filler-gap dependencies, they rely on superficial properties of the input, rather than on a shared generalization. Our work highlights the need for specific linguistic inductive biases to model language acquisition.
Long-context models(LCMs) have shown great potential in processing long input sequences(even more than 100M tokens) conveniently and effectively. With significant progress, recent research has pointed out that LCMs can accurately locate token-level salient information within the context. Yet, the generation performance of these LCMs is far from satisfactory and might result in misaligned responses, such as hallucinations. To enhance the generation capability of LCMs, existing works have investigated the effects of data size and quality for both pre-training and instruction tuning. Though achieving meaningful improvement, previous methods fall short in either effectiveness or efficiency. In this paper, we introduce LOGO(Long cOntext aliGnment via efficient preference Optimization), a training strategy that first introduces preference optimization for long-context alignment. To overcome the GPU memory-bound issue caused by the long sequence, LOGO employs a reference-free preference optimization strategy and adopts a position synthesis method to construct the training data. By training with only 0.3B data on a single 8$\times$A800 GPU machine for 16 hours, LOGO allows the Llama-3-8B-Instruct-80K model to achieve comparable performance with GPT-4 in real-world long-context tasks while preserving the model's original capabilities on other tasks, e.g., language modeling and MMLU. Moreover, LOGO can extend the model's context window size while enhancing its generation performance.
Masked diffusion models (MDMs) have shown promise in language modeling, yet their scalability and effectiveness in core language tasks, such as text generation and language understanding, remain underexplored. This paper establishes the first scaling law for MDMs, demonstrating a scaling rate comparable to autoregressive models (ARMs) and a relatively small compute gap. Motivated by their scalability, we train a family of MDMs with up to 1.1 billion (B) parameters to systematically evaluate their performance against ARMs of comparable or larger sizes. Fully leveraging the probabilistic formulation of MDMs, we propose a simple yet effective unsupervised classifier-free guidance that effectively exploits large-scale unpaired data, boosting performance for conditional inference. In language understanding, the 1.1B MDM outperforms the 1.1B TinyLlama model trained on the same data across four of eight zero-shot benchmarks. Notably, it achieves competitive math reasoning ability with the 7B Llama-2 model on the GSM8K dataset. In text generation, MDMs provide a flexible trade-off compared to ARMs utilizing KV-cache: MDMs match the performance of ARMs while being 1.4 times faster or achieving higher quality than ARMs at a higher computational cost. Moreover, MDMs address challenging tasks for ARMs by effectively handling bidirectional reasoning and adapting to temporal shifts in data. Notably, a 1.1B MDM breaks the reverse curse encountered by much larger ARMs with significantly more data and computation, such as 13B Llama-2 and 175B GPT-3. Our code is available at https://github.com/ML-GSAI/SMDM.