Vision-Language Models (VLMs) bring powerful understanding and reasoning capabilities to multimodal tasks. Meanwhile, the great need for capable aritificial intelligence on mobile devices also arises, such as the AI assistant software. Some efforts try to migrate VLMs to edge devices to expand their application scope. Simplifying the model structure is a common method, but as the model shrinks, the trade-off between performance and size becomes more and more difficult. Knowledge distillation (KD) can help models improve comprehensive capabilities without increasing size or data volume. However, most of the existing large model distillation techniques only consider applications on single-modal LLMs, or only use teachers to create new data environments for students. None of these methods take into account the distillation of the most important cross-modal alignment knowledge in VLMs. We propose a method called Align-KD to guide the student model to learn the cross-modal matching that occurs at the shallow layer. The teacher also helps student learn the projection of vision token into text embedding space based on the focus of text. Under the guidance of Align-KD, the 1.7B MobileVLM V2 model can learn rich knowledge from the 7B teacher model with light design of training loss, and achieve an average score improvement of 2.0 across 6 benchmarks under two training subsets respectively. Code is available at: https://github.com/fqhank/Align-KD.
Recommendation Systems have become integral to modern user experiences, but lack transparency in their decision-making processes. Existing explainable recommendation methods are hindered by reliance on a post-hoc paradigm, wherein explanation generators are trained independently of the underlying recommender models. This paradigm necessitates substantial human effort in data construction and raises concerns about explanation reliability. In this paper, we present ExpCTR, a novel framework that integrates large language model based explanation generation directly into the CTR prediction process. Inspired by recent advances in reinforcement learning, we employ two carefully designed reward mechanisms, LC alignment, which ensures explanations reflect user intentions, and IC alignment, which maintains consistency with traditional ID-based CTR models. Our approach incorporates an efficient training paradigm with LoRA and a three-stage iterative process. ExpCTR circumvents the need for extensive explanation datasets while fostering synergy between CTR prediction and explanation generation. Experimental results demonstrate that ExpCTR significantly enhances both recommendation accuracy and interpretability across three real-world datasets.
Retrieval Augmented Generation (RAG) has proven to be highly effective in boosting the generative performance of language model in knowledge-intensive tasks. However, existing RAG framework either indiscriminately perform retrieval or rely on rigid single-class classifiers to select retrieval methods, leading to inefficiencies and suboptimal performance across queries of varying complexity. To address these challenges, we propose a reinforcement learning-based framework that dynamically selects the most suitable retrieval strategy based on query complexity. % our solution Our approach leverages a multi-armed bandit algorithm, which treats each retrieval method as a distinct ``arm'' and adapts the selection process by balancing exploration and exploitation. Additionally, we introduce a dynamic reward function that balances accuracy and efficiency, penalizing methods that require more retrieval steps, even if they lead to a correct result. Our method achieves new state of the art results on multiple single-hop and multi-hop datasets while reducing retrieval costs. Our code are available at https://github.com/FUTUREEEEEE/MBA .
The GPT-4o represents a significant milestone in enabling real-time interaction with large language models (LLMs) through speech, its remarkable low latency and high fluency not only capture attention but also stimulate research interest in the field. This real-time speech interaction is particularly valuable in scenarios requiring rapid feedback and immediate responses, dramatically enhancing user experience. However, there is a notable lack of research focused on real-time large speech language models, particularly for Chinese. In this work, we present KE-Omni, a seamless large speech language model built upon Ke-SpeechChat, a large-scale high-quality synthetic speech interaction dataset consisting of 7 million Chinese and English conversations, featuring 42,002 speakers, and totaling over 60,000 hours, This contributes significantly to the advancement of research and development in this field. The demos can be accessed at \url{https://huggingface.co/spaces/KE-Team/KE-Omni}.
We introduce HackSynth, a novel Large Language Model (LLM)-based agent capable of autonomous penetration testing. HackSynth's dual-module architecture includes a Planner and a Summarizer, which enable it to generate commands and process feedback iteratively. To benchmark HackSynth, we propose two new Capture The Flag (CTF)-based benchmark sets utilizing the popular platforms PicoCTF and OverTheWire. These benchmarks include two hundred challenges across diverse domains and difficulties, providing a standardized framework for evaluating LLM-based penetration testing agents. Based on these benchmarks, extensive experiments are presented, analyzing the core parameters of HackSynth, including creativity (temperature and top-p) and token utilization. Multiple open source and proprietary LLMs were used to measure the agent's capabilities. The experiments show that the agent performed best with the GPT-4o model, better than what the GPT-4o's system card suggests. We also discuss the safety and predictability of HackSynth's actions. Our findings indicate the potential of LLM-based agents in advancing autonomous penetration testing and the importance of robust safeguards. HackSynth and the benchmarks are publicly available to foster research on autonomous cybersecurity solutions.
Multimodal LLMs (MLLMs) equip language models with visual capabilities by aligning vision encoders with language models. Existing methods to enhance the visual perception of MLLMs often involve designing more powerful vision encoders, which requires exploring a vast design space and re-aligning each potential encoder with the language model, resulting in prohibitively high training costs. In this paper, we introduce VisionFuse, a novel integration framework that efficiently utilizes multiple vision encoders from off-the-shelf MLLMs to enhance visual perception without requiring additional training. Our approach is motivated by the observation that different MLLMs tend to focus on distinct regions given the same query and image. Moreover, we find that the feature distributions of vision encoders within an MLLM family, a group of MLLMs sharing the same pretrained LLM, are highly aligned. Building on these insights, VisionFuse enriches the visual context by concatenating the tokens generated by the vision encoders of selected MLLMs within a family. By merging the parameters of language models from these MLLMs, VisionFuse allows a single language model to align with various vision encoders, significantly reducing deployment overhead. We conduct comprehensive evaluations across multiple multimodal benchmarks using various MLLM combinations, demonstrating substantial improvements in multimodal tasks. Notably, when integrating MiniGemini-8B and SLIME-8B, VisionFuse achieves an average performance increase of over 4%.
Large language models (LLMs) are effective at capturing complex, valuable conceptual representations from textual data for a wide range of real-world applications. However, in fields like Intelligent Fault Diagnosis (IFD), incorporating additional sensor data-such as vibration signals, temperature readings, and operational metrics-is essential but it is challenging to capture such sensor data information within traditional text corpora. This study introduces a novel IFD approach by effectively adapting LLMs to numerical data inputs for identifying various machine faults from time-series sensor data. We propose FD-LLM, an LLM framework specifically designed for fault diagnosis by formulating the training of the LLM as a multi-class classification problem. We explore two methods for encoding vibration signals: the first method uses a string-based tokenization technique to encode vibration signals into text representations, while the second extracts statistical features from both the time and frequency domains as statistical summaries of each signal. We assess the fault diagnosis capabilities of four open-sourced LLMs based on the FD-LLM framework, and evaluate the models' adaptability and generalizability under various operational conditions and machine components, namely for traditional fault diagnosis, cross-operational conditions, and cross-machine component settings. Our results show that LLMs such as Llama3 and Llama3-instruct demonstrate strong fault detection capabilities and significant adaptability across different operational conditions, outperforming state-of-the-art deep learning (DL) approaches in many cases.
Automated documentation of programming source code is a challenging task with significant practical and scientific implications for the developer community. We present a large language model (LLM)-based application that developers can use as a support tool to generate basic documentation for any publicly available repository. Over the last decade, several papers have been written on generating documentation for source code using neural network architectures. With the recent advancements in LLM technology, some open-source applications have been developed to address this problem. However, these applications typically rely on the OpenAI APIs, which incur substantial financial costs, particularly for large repositories. Moreover, none of these open-source applications offer a fine-tuned model or features to enable users to fine-tune. Additionally, finding suitable data for fine-tuning is often challenging. Our application addresses these issues which is available at https://pypi.org/project/readme-ready/.
Cooking meals can be difficult, causing many to resort to cookbooks and online recipes. However, relying on these traditional methods of cooking often results in missing ingredients, nutritional hazards, and unsatisfactory meals. Using Augmented Reality (AR) can address these issues; however, current AR cooking applications have poor user interfaces and limited accessibility. This paper proposes a prototype of an iOS application that integrates AR and Computer Vision (CV) into the cooking process. We leverage Google's Gemini Large Language Model (LLM) to identify ingredients in the camera's field of vision and generate recipe choices with detailed nutritional information. Additionally, this application uses Apple's ARKit to create an AR user interface compatible with iOS devices. Users can personalize their meal suggestions by inputting their dietary preferences and rating each meal. The application's effectiveness is evaluated through three rounds of user experience surveys. This application advances the field of accessible cooking assistance technologies, aiming to reduce food wastage and improve the meal planning experience.
The current large language models are mainly based on decode-only structure transformers, which have great in-context learning (ICL) capabilities. It is generally believed that the important foundation of its ICL capability is the induction heads mechanism, which requires at least two layers attention. In order to more efficiently implement the ability of the model's induction, we revisit the induction heads mechanism and proposed a KV shifting attention. We theoretically prove that the KV shifting attention reducing the model's requirements for the depth and width of the induction heads mechanism. Our experimental results demonstrate that KV shifting attention is beneficial to learning induction heads and language modeling, which lead to better performance or faster convergence from toy models to the pre-training models with more than 10 B parameters.
Bioinformatics workflows are essential for complex biological data analyses and are often described in scientific articles with source code in public repositories. Extracting detailed workflow information from articles can improve accessibility and reusability but is hindered by limited annotated corpora. To address this, we framed the problem as a low-resource extraction task and tested four strategies: 1) creating a tailored annotated corpus, 2) few-shot named-entity recognition (NER) with an autoregressive language model, 3) NER using masked language models with existing and new corpora, and 4) integrating workflow knowledge into NER models. Using BioToFlow, a new corpus of 52 articles annotated with 16 entities, a SciBERT-based NER model achieved a 70.4 F-measure, comparable to inter-annotator agreement. While knowledge integration improved performance for specific entities, it was less effective across the entire information schema. Our results demonstrate that high-performance information extraction for bioinformatics workflows is achievable.
Large language models (LLMs) have demonstrated remarkable capabilities, but their adoption is limited by high computational costs during inference. While increasing parameter counts enhances accuracy, it also widens the gap between state-of-the-art capabilities and practical deployability. We present Puzzle, a framework to accelerate LLM inference on specific hardware while preserving their capabilities. Through an innovative application of neural architecture search (NAS) at an unprecedented scale, Puzzle systematically optimizes models with tens of billions of parameters under hardware constraints. Our approach utilizes blockwise local knowledge distillation (BLD) for parallel architecture exploration and employs mixed-integer programming for precise constraint optimization. We demonstrate the real-world impact of our framework through Llama-3.1-Nemotron-51B-Instruct (Nemotron-51B), a publicly available model derived from Llama-3.1-70B-Instruct. Nemotron-51B achieves a 2.17x inference throughput speedup, fitting on a single NVIDIA H100 GPU while preserving 98.4% of the original model's capabilities. Nemotron-51B currently stands as the most accurate language model capable of inference on a single GPU with large batch sizes. Remarkably, this transformation required just 45B training tokens, compared to over 15T tokens used for the 70B model it was derived from. This establishes a new paradigm where powerful models can be optimized for efficient deployment with only negligible compromise of their capabilities, demonstrating that inference performance, not parameter count alone, should guide model selection. With the release of Nemotron-51B and the presentation of the Puzzle framework, we provide practitioners immediate access to state-of-the-art language modeling capabilities at significantly reduced computational costs.
Clinical language models are important for many applications in healthcare, but their development depends on access to extensive clinical text for pretraining. However, obtaining clinical notes from electronic health records (EHRs) at scale is challenging due to patient privacy concerns. In this study, we rephrase existing clinical notes using LLMs to generate synthetic pretraining corpora, drawing inspiration from previous work on rephrasing web data. We examine four popular small-sized LLMs (<10B) to create synthetic clinical text to pretrain both decoder-based and encoder-based language models. The method yields better results in language modeling and downstream tasks than previous synthesis approaches without referencing real clinical text. We find that augmenting original clinical notes with synthetic corpora from different LLMs improves performances even at a small token budget, showing the potential of this method to support pretraining at the institutional level or be scaled to synthesize large-scale clinical corpora.
Sentiment analysis (SA) is a process of identifying the emotional tone or polarity within a given text and aims to uncover the user's complex emotions and inner feelings. While sentiment analysis has been extensively studied for languages like English, research in Bengali, remains limited, particularly for fine-grained sentiment categorization. This work aims to connect this gap by developing a novel approach that integrates rule-based algorithms with pre-trained language models. We developed a dataset from scratch, comprising over 15,000 manually labeled reviews. Next, we constructed a Lexicon Data Dictionary, assigning polarity scores to the reviews. We developed a novel rule based algorithm Bangla Sentiment Polarity Score (BSPS), an approach capable of generating sentiment scores and classifying reviews into nine distinct sentiment categories. To assess the performance of this method, we evaluated the classified sentiments using BanglaBERT, a pre-trained transformer-based language model. We also performed sentiment classification directly with BanglaBERT on the original data and evaluated this model's results. Our analysis revealed that the BSPS + BanglaBERT hybrid approach outperformed the standalone BanglaBERT model, achieving higher accuracy, precision, and nuanced classification across the nine sentiment categories. The results of our study emphasize the value and effectiveness of combining rule-based and pre-trained language model approaches for enhanced sentiment analysis in Bengali and suggest pathways for future research and application in languages with similar linguistic complexities.
Medical vision-language model models often struggle with generating accurate quantitative measurements in radiology reports, leading to hallucinations that undermine clinical reliability. We introduce FactCheXcker, a modular framework that de-hallucinates radiology report measurements by leveraging an improved query-code-update paradigm. Specifically, FactCheXcker employs specialized modules and the code generation capabilities of large language models to solve measurement queries generated based on the original report. After extracting measurable findings, the results are incorporated into an updated report. We evaluate FactCheXcker on endotracheal tube placement, which accounts for an average of 78% of report measurements, using the MIMIC-CXR dataset and 11 medical report-generation models. Our results show that FactCheXcker significantly reduces hallucinations, improves measurement precision, and maintains the quality of the original reports. Specifically, FactCheXcker improves the performance of all 11 models and achieves an average improvement of 94.0% in reducing measurement hallucinations measured by mean absolute error.
Humans recognize objects after observing only a few examples, a remarkable capability enabled by their inherent language understanding of the real-world environment. Developing verbalized and interpretable representation can significantly improve model generalization in low-data settings. In this work, we propose Verbalized Representation Learning (VRL), a novel approach for automatically extracting human-interpretable features for object recognition using few-shot data. Our method uniquely captures inter-class differences and intra-class commonalities in the form of natural language by employing a Vision-Language Model (VLM) to identify key discriminative features between different classes and shared characteristics within the same class. These verbalized features are then mapped to numeric vectors through the VLM. The resulting feature vectors can be further utilized to train and infer with downstream classifiers. Experimental results show that, at the same model scale, VRL achieves a 24% absolute improvement over prior state-of-the-art methods while using 95% less data and a smaller mode. Furthermore, compared to human-labeled attributes, the features learned by VRL exhibit a 20% absolute gain when used for downstream classification tasks. Code is available at: https://github.com/joeyy5588/VRL/tree/main.
Recently, human motion analysis has experienced great improvement due to inspiring generative models such as the denoising diffusion model and large language model. While the existing approaches mainly focus on generating motions with textual descriptions and overlook the reciprocal task. In this paper, we present~\textbf{MoTe}, a unified multi-modal model that could handle diverse tasks by learning the marginal, conditional, and joint distributions of motion and text simultaneously. MoTe enables us to handle the paired text-motion generation, motion captioning, and text-driven motion generation by simply modifying the input context. Specifically, MoTe is composed of three components: Motion Encoder-Decoder (MED), Text Encoder-Decoder (TED), and Moti-on-Text Diffusion Model (MTDM). In particular, MED and TED are trained for extracting latent embeddings, and subsequently reconstructing the motion sequences and textual descriptions from the extracted embeddings, respectively. MTDM, on the other hand, performs an iterative denoising process on the input context to handle diverse tasks. Experimental results on the benchmark datasets demonstrate the superior performance of our proposed method on text-to-motion generation and competitive performance on motion captioning.
Identifying defects and anomalies in industrial products is a critical quality control task. Traditional manual inspection methods are slow, subjective, and error-prone. In this work, we propose a novel zero-shot training-free approach for automated industrial image anomaly detection using a multimodal machine learning pipeline, consisting of three foundation models. Our method first uses a large language model, i.e., GPT-3. generate text prompts describing the expected appearances of normal and abnormal products. We then use a grounding object detection model, called Grounding DINO, to locate the product in the image. Finally, we compare the cropped product image patches to the generated prompts using a zero-shot image-text matching model, called CLIP, to identify any anomalies. Our experiments on two datasets of industrial product images, namely MVTec-AD and VisA, demonstrate the effectiveness of this method, achieving high accuracy in detecting various types of defects and anomalies without the need for model training. Our proposed model enables efficient, scalable, and objective quality control in industrial manufacturing settings.
In this paper, we introduce an open-source Korean-English vision-language model (VLM), VARCO-VISION. We incorporate a step-by-step training strategy that allows a model learn both linguistic and visual information while preserving the backbone model's knowledge. Our model demonstrates outstanding performance in diverse settings requiring bilingual image-text understanding and generation abilities compared to models of similar size. VARCO-VISION is also capable of grounding, referring, and OCR, expanding its usage and potential applications for real-world scenarios. In addition to the model, we release five Korean evaluation datasets, including four closed-set and one openset benchmarks. We anticipate that our milestone will broaden the opportunities for AI researchers aiming to train VLMs. VARCO-VISION is available at https://huggingface.co/NCSOFT/VARCO-VISION-14B.
Hybrid models that combine the language modeling capabilities of Attention layers with the efficiency of Recurrent layers (e.g., State Space Models) have gained traction in practically supporting long contexts in Large Language Model serving. Yet, the unique properties of these models complicate the usage of complementary efficiency optimizations such as prefix caching that skip redundant computations across requests. Most notably, their use of in-place state updates for recurrent layers precludes rolling back cache entries for partial sequence overlaps, and instead mandates only exact-match cache hits; the effect is a deluge of (large) cache entries per sequence, most of which yield minimal reuse opportunities. We present Marconi, the first system that supports efficient prefix caching with Hybrid LLMs. Key to Marconi are its novel admission and eviction policies that more judiciously assess potential cache entries based not only on recency, but also on (1) forecasts of their reuse likelihood across a taxonomy of different hit scenarios, and (2) the compute savings that hits deliver relative to memory footprints. Across diverse workloads and Hybrid models, Marconi achieves up to 34.4$\times$ higher token hit rates (71.1% or 617 ms lower TTFT) compared to state-of-the-art prefix caching systems.