Post-training quantization reduces the computational demand of Large Language Models (LLMs) but can weaken some of their capabilities. Since LLM abilities emerge with scale, smaller LLMs are more sensitive to quantization. In this paper, we explore how quantization affects smaller LLMs' ability to perform retrieval-augmented generation (RAG), specifically in longer contexts. We chose personalization for evaluation because it is a challenging domain to perform using RAG as it requires long-context reasoning over multiple documents. We compare the original FP16 and the quantized INT4 performance of multiple 7B and 8B LLMs on two tasks while progressively increasing the number of retrieved documents to test how quantized models fare against longer contexts. To better understand the effect of retrieval, we evaluate three retrieval models in our experiments. Our findings reveal that if a 7B LLM performs the task well, quantization does not impair its performance and long-context reasoning capabilities. We conclude that it is possible to utilize RAG with quantized smaller LLMs.
相關內容
Star-shaped Tilted Hexarotors are rapidly emerging for applications highly demanding in terms of robustness and maneuverability. To ensure improvement in such features, a careful selection of the tilt angles is mandatory. In this work, we present a rigorous analysis of how the force subspace varies with the tilt cant angles, namely the tilt angles along the vehicle arms, taking into account gravity compensation and torque decoupling to abide by the hovering condition. Novel metrics are introduced to assess the performance of existing tilted platforms, as well as to provide some guidelines for the selection of the tilt cant angle in the design phase.
The emergence of Large Language Models (LLMs) as chat assistants capable of generating human-like conversations has amplified the need for robust evaluation methods, particularly for open-ended tasks. Conventional metrics like BLEU and ROUGE, while useful, are increasingly inadequate for capturing the subtle semantics and contextual richness of such generative outputs. We propose a reference-guided verdict method that automates the evaluation process by leveraging multiple LLMs-as-judges. Through experiments on three open-ended question-answering tasks, we demonstrate that combining multiple LLMs-as-judges significantly improves the reliability and accuracy of evaluations, particularly in complex tasks where a single model might struggle. Our findings reveal a strong correlation with human evaluations, establishing our method as a viable and effective alternative to traditional metrics and human judgments, particularly in the context of LLM-based chat assistants where the complexity and diversity of responses challenge existing benchmarks.
Schema linking is a crucial step in Text-to-SQL pipelines. Its goal is to retrieve the relevant tables and columns of a target database for a user's query while disregarding irrelevant ones. However, imperfect schema linking can often exclude required columns needed for accurate query generation. In this work, we revisit schema linking when using the latest generation of large language models (LLMs). We find empirically that newer models are adept at utilizing relevant schema elements during generation even in the presence of large numbers of irrelevant ones. As such, our Text-to-SQL pipeline entirely forgoes schema linking in cases where the schema fits within the model's context window in order to minimize issues due to filtering required schema elements. Furthermore, instead of filtering contextual information, we highlight techniques such as augmentation, selection, and correction, and adopt them to improve the accuracy of our Text-to-SQL pipeline. Our approach ranks first on the BIRD benchmark achieving an accuracy of 71.83%.
Routine computed tomography (CT) scans often detect a wide range of renal cysts, some of which may be malignant. Early and precise localization of these cysts can significantly aid quantitative image analysis. Current segmentation methods, however, do not offer sufficient interpretability at the feature and pixel levels, emphasizing the necessity for an explainable framework that can detect and rectify model inaccuracies. We developed an interpretable segmentation framework and validated it on a multi-centric dataset. A Variational Autoencoder Generative Adversarial Network (VAE-GAN) was employed to learn the latent representation of 3D input patches and reconstruct input images. Modifications in the latent representation using the gradient of the segmentation model generated counterfactual explanations for varying dice similarity coefficients (DSC). Radiomics features extracted from these counterfactual images, using a ground truth cyst mask, were analyzed to determine their correlation with segmentation performance. The DSCs for the original and VAE-GAN reconstructed images for counterfactual image generation showed no significant differences. Counterfactual explanations highlighted how variations in cyst image features influence segmentation outcomes and showed model discrepancies. Radiomics features correlating positively and negatively with dice scores were identified. The uncertainty of the predicted segmentation masks was estimated using posterior sampling of the weight space. The combination of counterfactual explanations and uncertainty maps provided a deeper understanding of the image features within the segmented renal cysts that lead to high uncertainty. The proposed segmentation framework not only achieved high segmentation accuracy but also increased interpretability regarding how image features impact segmentation performance.
Regulators in the US and EU are using thresholds based on training compute--the number of computational operations used in training--to identify general-purpose artificial intelligence (GPAI) models that may pose risks of large-scale societal harm. We argue that training compute currently is the most suitable metric to identify GPAI models that deserve regulatory oversight and further scrutiny. Training compute correlates with model capabilities and risks, is quantifiable, can be measured early in the AI lifecycle, and can be verified by external actors, among other advantageous features. These features make compute thresholds considerably more suitable than other proposed metrics to serve as an initial filter to trigger additional regulatory requirements and scrutiny. However, training compute is an imperfect proxy for risk. As such, compute thresholds should not be used in isolation to determine appropriate mitigation measures. Instead, they should be used to detect potentially risky GPAI models that warrant regulatory oversight, such as through notification requirements, and further scrutiny, such as via model evaluations and risk assessments, the results of which may inform which mitigation measures are appropriate. In fact, this appears largely consistent with how compute thresholds are used today. As GPAI technology and market structures evolve, regulators should update compute thresholds and complement them with other metrics into regulatory review processes.
With breakthroughs in large-scale modeling, the Segment Anything Model (SAM) and its extensions have been attempted for applications in various underwater visualization tasks in marine sciences, and have had a significant impact on the academic community. Recently, Meta has further developed the Segment Anything Model 2 (SAM2), which significantly improves running speed and segmentation accuracy compared to its predecessor. This report aims to explore the potential of SAM2 in marine science by evaluating it on the underwater instance segmentation benchmark datasets UIIS and USIS10K. The experiments show that the performance of SAM2 is extremely dependent on the type of user-provided prompts. When using the ground truth bounding box as prompt, SAM2 performed excellently in the underwater instance segmentation domain. However, when running in automatic mode, SAM2's ability with point prompts to sense and segment underwater instances is significantly degraded. It is hoped that this paper will inspire researchers to further explore the SAM model family in the underwater domain. The results and evaluation codes in this paper are available at //github.com/LiamLian0727/UnderwaterSAM2Eval.
Auditing the use of data in training machine-learning (ML) models is an increasingly pressing challenge, as myriad ML practitioners routinely leverage the effort of content creators to train models without their permission. In this paper, we propose a general method to audit an ML model for the use of a data-owner's data in training, without prior knowledge of the ML task for which the data might be used. Our method leverages any existing black-box membership inference method, together with a sequential hypothesis test of our own design, to detect data use with a quantifiable, tunable false-detection rate. We show the effectiveness of our proposed framework by applying it to audit data use in two types of ML models, namely image classifiers and foundation models.
Despite the availability of commercial QRNG devices, distinguishing between PRNG and QRNG outputs computationally remains challenging. This paper presents two significant contributions from the perspectives of QRNG manufacturers and users. For manufacturers, the conventional method of assessing the quantumness of single-photon-based QRNGs through mean and variance comparisons of photon counts is statistically unreliable due to finite sample sizes. Given the sub-Poissonian statistics of single photons, confirming the underlying distribution is crucial for validating a QRNG's quantumness. We propose a more efficient two-fold statistical approach to ensure the quantumness of optical sources with the desired confidence level. Additionally, we demonstrate that the output of QRNGs from exponential and uniform distributions exhibit similarity under device noise, deriving corresponding photon statistics and conditions for $\epsilon$-randomness. From the user's perspective, the fundamental parameters of a QRNG are quantumness (security), efficiency (randomness and random number generation rate), and cost. Our analysis reveals that these parameters depend on three factors, expected photon count per unit time, external reference cycle duration, and detection efficiency. A lower expected photon count enhances security but increases cost and decreases the generation rate. A shorter external reference cycle boosts security but must exceed a minimum threshold to minimize timing errors, with minor impacts on cost and rate. Lower detection efficiency enhances security and lowers cost but reduces the generation rate. Finally, to validate our results, we perform statistical tests like NIST, Dieharder, AIS-31, ENT etc. over the data simulated with different values of the above parameters. Our findings can empower manufacturers to customize QRNGs to meet user needs effectively.
The concept of distributed consensus gained widespread attention following the publication of "Byzantine Generals Problem" by Leslie Lamport in the 1980s. This research topic has been active and extensively studied over the last four decades, particularly since the advent of blockchain technology in 2009. Blockchain technology employs Proof-of-X (PoX) or Byzantine-fault-tolerant (BFT) systems, where all participants follow a protocol to achieve a common state (i.e., consistency) eventually. However, because PoX consensus such as Proof-of-Work is is resource-intensive with high power consumption, most permissioned blockchains employ BFT to achieve consistency. In this article, we provide an introduction to the fundamental principles and history of distributed consensus. We then explore the well-known fault-tolerant state machine replication (SMR) in partially synchronous networks, as well as consensus protocols in asynchronous models and recently proposed DAG-based consensus. Additionally, we examine the relationship between BFT consensus and blockchain technology and discuss the following questions: What is the history and evolution of BFT? Why are BFT protocols designed in the way they are and what core components do they use? What is the connection between BFT and blockchain technology, and what are the driving needs for future BFT research?
The dominating NLP paradigm of training a strong neural predictor to perform one task on a specific dataset has led to state-of-the-art performance in a variety of applications (eg. sentiment classification, span-prediction based question answering or machine translation). However, it builds upon the assumption that the data distribution is stationary, ie. that the data is sampled from a fixed distribution both at training and test time. This way of training is inconsistent with how we as humans are able to learn from and operate within a constantly changing stream of information. Moreover, it is ill-adapted to real-world use cases where the data distribution is expected to shift over the course of a model's lifetime. The first goal of this thesis is to characterize the different forms this shift can take in the context of natural language processing, and propose benchmarks and evaluation metrics to measure its effect on current deep learning architectures. We then proceed to take steps to mitigate the effect of distributional shift on NLP models. To this end, we develop methods based on parametric reformulations of the distributionally robust optimization framework. Empirically, we demonstrate that these approaches yield more robust models as demonstrated on a selection of realistic problems. In the third and final part of this thesis, we explore ways of efficiently adapting existing models to new domains or tasks. Our contribution to this topic takes inspiration from information geometry to derive a new gradient update rule which alleviate catastrophic forgetting issues during adaptation.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191