Abstract:KV cache stores key and value states from previous tokens to avoid re-computation, yet it demands substantial storage space, especially for long sequences. Adaptive KV cache compression seeks to discern the saliency of tokens, preserving vital information while aggressively compressing those of less importance. However, previous methods of this approach exhibit significant performance degradation at high compression ratios due to inaccuracies in identifying salient tokens. In this paper, we present ZipCache, an accurate and efficient KV cache quantization method for LLMs. First, we construct a strong baseline for quantizing KV cache. Through the proposed channel-separable tokenwise quantization scheme, the memory overhead of quantization parameters are substantially reduced compared to fine-grained groupwise quantization. To enhance the compression ratio, we propose normalized attention score as an effective metric for identifying salient tokens by considering the lower triangle characteristics of the attention matrix. Moreover, we develop an efficient approximation method that decouples the saliency metric from full attention scores, enabling compatibility with fast attention implementations like FlashAttention. Extensive experiments demonstrate that ZipCache achieves superior compression ratios, fast generation speed and minimal performance losses compared with previous KV cache compression methods. For instance, when evaluating Mistral-7B model on GSM8k dataset, ZipCache is capable of compressing the KV cache by $4.98\times$, with only a $0.38\%$ drop in accuracy. In terms of efficiency, ZipCache also showcases a $37.3\%$ reduction in prefill-phase latency, a $56.9\%$ reduction in decoding-phase latency, and a $19.8\%$ reduction in GPU memory usage when evaluating LLaMA3-8B model with a input length of $4096$.
Abstract:Predicting future human pose is a fundamental application for machine intelligence, which drives robots to plan their behavior and paths ahead of time to seamlessly accomplish human-robot collaboration in real-world 3D scenarios. Despite encouraging results, existing approaches rarely consider the effects of the external scene on the motion sequence, leading to pronounced artifacts and physical implausibilities in the predictions. To address this limitation, this work introduces a novel multi-modal sense-informed motion prediction approach, which conditions high-fidelity generation on two modal information: external 3D scene, and internal human gaze, and is able to recognize their salience for future human activity. Furthermore, the gaze information is regarded as the human intention, and combined with both motion and scene features, we construct a ternary intention-aware attention to supervise the generation to match where the human wants to reach. Meanwhile, we introduce semantic coherence-aware attention to explicitly distinguish the salient point clouds and the underlying ones, to ensure a reasonable interaction of the generated sequence with the 3D scene. On two real-world benchmarks, the proposed method achieves state-of-the-art performance both in 3D human pose and trajectory prediction.
Abstract:Cloud computing (cloud computing) is a kind of distributed computing, referring to the network "cloud" will be a huge data calculation and processing program into countless small programs, and then, through the system composed of multiple servers to process and analyze these small programs to get the results and return to the user. This report explores the intersection of cloud computing and financial information processing, identifying risks and challenges faced by financial institutions in adopting cloud technology. It discusses the need for intelligent solutions to enhance data processing efficiency and accuracy while addressing security and privacy concerns. Drawing on regulatory frameworks, the report proposes policy recommendations to mitigate concentration risks associated with cloud computing in the financial industry. By combining intelligent forecasting and evaluation technologies with cloud computing models, the study aims to provide effective solutions for financial data processing and management, facilitating the industry's transition towards digital transformation.
Abstract:Model reparameterization is a widely accepted technique for improving inference speed without compromising performance. However, current Post-training Quantization (PTQ) methods often lead to significant accuracy degradation when applied to reparameterized models. This is primarily caused by channel-specific and sample-specific outliers, which appear only at specific samples and channels and impact on the selection of quantization parameters. To address this issue, we propose RepAPQ, a novel framework that preserves the accuracy of quantized reparameterization models. Different from previous frameworks using Mean Squared Error (MSE) as a measurement, we utilize Mean Absolute Error (MAE) to mitigate the influence of outliers on quantization parameters. Our framework comprises two main components: Quantization Protecting Reparameterization and Across-block Calibration. For effective calibration, Quantization Protecting Reparameterization combines multiple branches into a single convolution with an affine layer. During training, the affine layer accelerates convergence and amplifies the output of the convolution to better accommodate samples with outliers. Additionally, Across-block Calibration leverages the measurement of stage output as supervision to address the gradient problem introduced by MAE and enhance the interlayer correlation with quantization parameters. Comprehensive experiments demonstrate the effectiveness of RepAPQ across various models and tasks. Our framework outperforms previous methods by approximately 1\% for 8-bit PTQ and 2\% for 6-bit PTQ, showcasing its superior performance. The code is available at \url{https://github.com/ilur98/DLMC-QUANT}.
Abstract:Nowadays, Vision Transformer (ViT) is widely utilized in various computer vision tasks, owing to its unique self-attention mechanism. However, the model architecture of ViT is complex and often challenging to comprehend, leading to a steep learning curve. ViT developers and users frequently encounter difficulties in interpreting its inner workings. Therefore, a visualization system is needed to assist ViT users in understanding its functionality. This paper introduces EL-VIT, an interactive visual analytics system designed to probe the Vision Transformer and facilitate a better understanding of its operations. The system consists of four layers of visualization views. The first three layers include model overview, knowledge background graph, and model detail view. These three layers elucidate the operation process of ViT from three perspectives: the overall model architecture, detailed explanation, and mathematical operations, enabling users to understand the underlying principles and the transition process between layers. The fourth interpretation view helps ViT users and experts gain a deeper understanding by calculating the cosine similarity between patches. Our two usage scenarios demonstrate the effectiveness and usability of EL-VIT in helping ViT users understand the working mechanism of ViT.
Abstract:Recently, video text detection, tracking, and recognition in natural scenes are becoming very popular in the computer vision community. However, most existing algorithms and benchmarks focus on common text cases (e.g., normal size, density) and single scenario, while ignoring extreme video text challenges, i.e., dense and small text in various scenarios. In this paper, we establish a video text reading benchmark, named DSText V2, which focuses on Dense and Small text reading challenges in the video with various scenarios. Compared with the previous datasets, the proposed dataset mainly include three new challenges: 1) Dense video texts, a new challenge for video text spotters to track and read. 2) High-proportioned small texts, coupled with the blurriness and distortion in the video, will bring further challenges. 3) Various new scenarios, e.g., Game, Sports, etc. The proposed DSText V2 includes 140 video clips from 7 open scenarios, supporting three tasks, i.e., video text detection (Task 1), video text tracking (Task 2), and end-to-end video text spotting (Task 3). In this article, we describe detailed statistical information of the dataset, tasks, evaluation protocols, and the results summaries. Most importantly, a thorough investigation and analysis targeting three unique challenges derived from our dataset are provided, aiming to provide new insights. Moreover, we hope the benchmark will promise video text research in the community. DSText v2 is built upon DSText v1, which was previously introduced to organize the ICDAR 2023 competition for dense and small video text.
Abstract:Image segmentation based on continual learning exhibits a critical drop of performance, mainly due to catastrophic forgetting and background shift, as they are required to incorporate new classes continually. In this paper, we propose a simple, yet effective Continual Image Segmentation method with incremental Dynamic Query (CISDQ), which decouples the representation learning of both old and new knowledge with lightweight query embedding. CISDQ mainly includes three contributions: 1) We define dynamic queries with adaptive background class to exploit past knowledge and learn future classes naturally. 2) CISDQ proposes a class/instance-aware Query Guided Knowledge Distillation strategy to overcome catastrophic forgetting by capturing the inter-class diversity and intra-class identity. 3) Apart from semantic segmentation, CISDQ introduce the continual learning for instance segmentation in which instance-wise labeling and supervision are considered. Extensive experiments on three datasets for two tasks (i.e., continual semantic and instance segmentation are conducted to demonstrate that CISDQ achieves the state-of-the-art performance, specifically, obtaining 4.4% and 2.9% mIoU improvements for the ADE 100-10 (6 steps) setting and ADE 100-5 (11 steps) setting.
Abstract:Visual commonsense reasoning (VCR) is a challenging multi-modal task, which requires high-level cognition and commonsense reasoning ability about the real world. In recent years, large-scale pre-training approaches have been developed and promoted the state-of-the-art performance of VCR. However, the existing approaches almost employ the BERT-like objectives to learn multi-modal representations. These objectives motivated from the text-domain are insufficient for the excavation on the complex scenario of visual modality. Most importantly, the spatial distribution of the visual objects is basically neglected. To address the above issue, we propose to construct the spatial relation graph based on the given visual scenario. Further, we design two pre-training tasks named object position regression (OPR) and spatial relation classification (SRC) to learn to reconstruct the spatial relation graph respectively. Quantitative analysis suggests that the proposed method can guide the representations to maintain more spatial context and facilitate the attention on the essential visual regions for reasoning. We achieve the state-of-the-art results on VCR and two other vision-and-language reasoning tasks VQA, and NLVR.
Abstract:Diffusion models have demonstrated remarkable capabilities in image synthesis and related generative tasks. Nevertheless, their practicality for low-latency real-world applications is constrained by substantial computational costs and latency issues. Quantization is a dominant way to compress and accelerate diffusion models, where post-training quantization (PTQ) and quantization-aware training (QAT) are two main approaches, each bearing its own properties. While PTQ exhibits efficiency in terms of both time and data usage, it may lead to diminished performance in low bit-width. On the other hand, QAT can alleviate performance degradation but comes with substantial demands on computational and data resources. To capitalize on the advantages while avoiding their respective drawbacks, we introduce a data-free and parameter-efficient fine-tuning framework for low-bit diffusion models, dubbed EfficientDM, to achieve QAT-level performance with PTQ-like efficiency. Specifically, we propose a quantization-aware variant of the low-rank adapter (QALoRA) that can be merged with model weights and jointly quantized to low bit-width. The fine-tuning process distills the denoising capabilities of the full-precision model into its quantized counterpart, eliminating the requirement for training data. We also introduce scale-aware optimization and employ temporal learned step-size quantization to further enhance performance. Extensive experimental results demonstrate that our method significantly outperforms previous PTQ-based diffusion models while maintaining similar time and data efficiency. Specifically, there is only a marginal 0.05 sFID increase when quantizing both weights and activations of LDM-4 to 4-bit on ImageNet 256x256. Compared to QAT-based methods, our EfficientDM also boasts a 16.2x faster quantization speed with comparable generation quality.
Abstract:Large Language Models (LLMs) pose significant hardware challenges related to memory requirements and computational ability. There are two mainstream quantization schemes for LLMs: coarse-grained ($\textit{e.g.,}$ channel-wise) quantization and fine-grained ($\textit{e.g.,}$ group-wise) quantization. Fine-grained quantization has smaller quantization loss, consequently achieving superior performance. However, when applied to weight-activation quantization, it disrupts continuous integer matrix multiplication, leading to inefficient inference. In this paper, we introduce Dual Grained Quantization (DGQ), a novel A8W4 quantization for LLM that maintains superior performance while ensuring fast inference speed. DSQ dequantizes the fine-grained INT4 weight into coarse-grained INT8 representation and preform matrix multiplication using INT8 kernels. Besides, we develop a two-phase grid search algorithm to simplify the determination of fine-grained and coarse-grained quantization scales. We also devise a percentile clipping schema for smoothing the activation outliers without the need for complex optimization techniques. Experimental results demonstrate that DGQ consistently outperforms prior methods across various LLM architectures and a wide range of tasks. Remarkably, by our implemented efficient CUTLASS kernel, we achieve $\textbf{1.12}$ $\times$ memory reduction and $\textbf{3.24}$ $\times$ speed gains comparing A16W4 implementation. These advancements enable efficient deployment of A8W4 LLMs for real-world applications.