PyTorchKR Papers

최신 논문

총 5,357편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2405.01016

Addressing Diverging Training Costs using BEVRestore for High-resolution Bird's Eye View Map Construction

Recent advancements in Bird's Eye View (BEV) fusion for map construction have demonstrated remarkable mapping of urban environments. However, their deep and bulky architecture incurs substantial amounts of backpropagation memory and computing latency. Consequently, the problem poses an unavoidable bottleneck in constructing high-resolution (HR) BEV maps, as their large-sized features cause significant increases in costs including GPU memory consumption and computing latency, named diverging training costs issue. Affected by the problem, most existing methods adopt low-resolution (LR) BEV and struggle to estimate the precise locations of urban scene components like road lanes, and sidewalks. As the imprecision leads to risky motion planning like collision avoidance, the diverging training costs issue has to be resolved. In this paper, we address the issue with our novel BEVRestore mechanism. Specifically, our proposed model encodes the features of each sensor to LR BEV space and restores them to HR space to establish a memory-efficient map constructor. To this end, we introduce the BEV restoration strategy, which restores aliasing, and blocky artifacts of the up-scaled BEV features, and narrows down the width of the labels. Our extensive experiments show that the proposed mechanism provides a plug-and-play, memory-efficient pipeline, enabling an HR map construction with a broad BEV scope.

arXiv:2405.01004

Deep Learning Models in Speech Recognition: Measuring GPU Energy Consumption, Impact of Noise and Model Quantization for Edge Deployment

Recent transformer-based ASR models have achieved word-error rates (WER) below 4%, surpassing human annotator accuracy, yet they demand extensive server resources, contributing to significant carbon footprints. The traditional server-based architecture of ASR also presents privacy concerns, alongside reliability and latency issues due to network dependencies. In contrast, on-device (edge) ASR enhances privacy, boosts performance, and promotes sustainability by effectively balancing energy use and accuracy for specific applications. This study examines the effects of quantization, memory demands, and energy consumption on the performance of various ASR model inference on the NVIDIA Jetson Orin Nano. By analyzing WER and transcription speed across models using FP32, FP16, and INT8 quantization on clean and noisy datasets, we highlight the crucial trade-offs between accuracy, speeds, quantization, energy efficiency, and memory needs. We found that changing precision from fp32 to fp16 halves the energy consumption for audio transcription across different models, with minimal performance degradation. A larger model size and number of parameters neither guarantees better resilience to noise, nor predicts the energy consumption for a given transcription load. These, along with several other findings offer novel insights for optimizing ASR systems within energy- and memory-limited environments, crucial for the development of efficient on-device ASR solutions. The code and input data needed to reproduce the results in this article are open sourced are available on [https://github.com/zzadiues3338/ASR-energy-jetson].

arXiv:2405.00732공식 코드 · Papers with Code

LoRA Land: 310 Fine-tuned LLMs that Rival GPT-4, A Technical Report

Low Rank Adaptation (LoRA) has emerged as one of the most widely adopted methods for Parameter Efficient Fine-Tuning (PEFT) of Large Language Models (LLMs). LoRA reduces the number of trainable parameters and memory usage while achieving comparable performance to full fine-tuning. We aim to assess the viability of training and serving LLMs fine-tuned with LoRA in real-world applications. First, we measure the quality of LLMs fine-tuned with quantized low rank adapters across 10 base models and 31 tasks for a total of 310 models. We find that 4-bit LoRA fine-tuned models outperform base models by 34 points and GPT-4 by 10 points on average. Second, we investigate the most effective base models for fine-tuning and assess the correlative and predictive capacities of task complexity heuristics in forecasting the outcomes of fine-tuning. Finally, we evaluate the latency and concurrency capabilities of LoRAX, an open-source Multi-LoRA inference server that facilitates the deployment of multiple LoRA fine-tuned models on a single GPU using shared base model weights and dynamic adapter loading. LoRAX powers LoRA Land, a web application that hosts 25 LoRA fine-tuned Mistral-7B LLMs on a single NVIDIA A100 GPU with 80GB memory. LoRA Land highlights the quality and cost-effectiveness of employing multiple specialized LLMs over a single, general-purpose LLM.

arXiv:2405.01413공식 코드 · Papers with Code

MiniGPT-3D: Efficiently Aligning 3D Point Clouds with Large Language Models using 2D Priors

Large 2D vision-language models (2D-LLMs) have gained significant attention by bridging Large Language Models (LLMs) with images using a simple projector. Inspired by their success, large 3D point cloud-language models (3D-LLMs) also integrate point clouds into LLMs. However, directly aligning point clouds with LLM requires expensive training costs, typically in hundreds of GPU-hours on A100, which hinders the development of 3D-LLMs. In this paper, we introduce MiniGPT-3D, an efficient and powerful 3D-LLM that achieves multiple SOTA results while training for only 27 hours on one RTX 3090. Specifically, we propose to align 3D point clouds with LLMs using 2D priors from 2D-LLMs, which can leverage the similarity between 2D and 3D visual information. We introduce a novel four-stage training strategy for modality alignment in a cascaded way, and a mixture of query experts module to adaptively aggregate features with high efficiency. Moreover, we utilize parameter-efficient fine-tuning methods LoRA and Norm fine-tuning, resulting in only 47.8M learnable parameters, which is up to 260x fewer than existing methods. Extensive experiments show that MiniGPT-3D achieves SOTA on 3D object classification and captioning tasks, with significantly cheaper training costs. Notably, MiniGPT-3D gains an 8.12 increase on GPT-4 evaluation score for the challenging object captioning task compared to ShapeLLM-13B, while the latter costs 160 total GPU-hours on 8 A800. We are the first to explore the efficient 3D-LLM, offering new insights to the community. Code and weights are available at https://github.com/TangYuan96/MiniGPT-3D.

arXiv:2405.00476

A Comprehensive Survey of Dynamic Graph Neural Networks: Models, Frameworks, Benchmarks, Experiments and Challenges

Dynamic Graph Neural Networks (GNNs) combine temporal information with GNNs to capture structural, temporal, and contextual relationships in dynamic graphs simultaneously, leading to enhanced performance in various applications. As the demand for dynamic GNNs continues to grow, numerous models and frameworks have emerged to cater to different application needs. There is a pressing need for a comprehensive survey that evaluates the performance, strengths, and limitations of various approaches in this domain. This paper aims to fill this gap by offering a thorough comparative analysis and experimental evaluation of dynamic GNNs. It covers 81 dynamic GNN models with a novel taxonomy, 12 dynamic GNN training frameworks, and commonly used benchmarks. We also conduct experimental results from testing representative nine dynamic GNN models and three frameworks on six standard graph datasets. Evaluation metrics focus on convergence accuracy, training efficiency, and GPU memory usage, enabling a thorough comparison of performance across various models and frameworks. From the analysis and evaluation results, we identify key challenges and offer principles for future research to enhance the design of models and frameworks in the dynamic GNNs field.

arXiv:2405.00172

Bypassing Skip-Gram Negative Sampling: Dimension Regularization as a More Efficient Alternative for Graph Embeddings

A wide range of graph embedding objectives decompose into two components: one that enforces similarity, attracting the embeddings of nodes that are perceived as similar, and another that enforces dissimilarity, repelling the embeddings of nodes that are perceived as dissimilar. Without repulsion, the embeddings would collapse into trivial solutions. Skip-Gram Negative Sampling (SGNS) is a popular and efficient repulsion approach that prevents collapse by repelling each node from a sample of dissimilar nodes. In this work, we show that when repulsion is most needed and the embeddings approach collapse, SGNS node-wise repulsion is, in the aggregate, an approximate re-centering of the node embedding dimensions. Such dimension operations are more scalable than node operations and produce a simpler geometric interpretation of the repulsion. Our theoretical result establishes dimension regularization as an effective and more efficient, compared to skip-gram node contrast, approach to enforcing dissimilarity among embeddings of nodes. We use this result to propose a flexible algorithm augmentation framework that improves the scalability of any existing algorithm using SGNS. The framework prioritizes node attraction and replaces SGNS with dimension regularization. We instantiate this generic framework for LINE and node2vec and show that the augmented algorithms preserve downstream link-prediction performance while reducing GPU memory usage by up to 33.3% and training time by 23.4%. Moreover, we show that completely removing repulsion (a special case of our augmentation framework) in LINE reduces training time by 70.9% on average, while increasing link prediction performance, especially for graphs that are globally sparse but locally dense. In general, however, repulsion is needed, and dimension regularization provides an efficient alternative to SGNS.

arXiv:2405.00313

Streamlining Image Editing with Layered Diffusion Brushes

Denoising diffusion models have recently gained prominence as powerful tools for a variety of image generation and manipulation tasks. Building on this, we propose a novel tool for real-time editing of images that provides users with fine-grained region-targeted supervision in addition to existing prompt-based controls. Our novel editing technique, termed Layered Diffusion Brushes, leverages prompt-guided and region-targeted alteration of intermediate denoising steps, enabling precise modifications while maintaining the integrity and context of the input image. We provide an editor based on Layered Diffusion Brushes modifications, which incorporates well-known image editing concepts such as layer masks, visibility toggles, and independent manipulation of layers; regardless of their order. Our system renders a single edit on a 512x512 image within 140 ms using a high-end consumer GPU, enabling real-time feedback and rapid exploration of candidate edits. We validated our method and editing system through a user study involving both natural images (using inversion) and generated images, showcasing its usability and effectiveness compared to existing techniques such as InstructPix2Pix and Stable Diffusion Inpainting for refining images. Our approach demonstrates efficacy across a range of tasks, including object attribute adjustments, error correction, and sequential prompt-based object placement and manipulation, demonstrating its versatility and potential for enhancing creative workflows.

arXiv:2405.00263공식 코드 · Papers with Code

Clover: Regressive Lightweight Speculative Decoding with Sequential Knowledge

Large language models (LLMs) suffer from low efficiency as the mismatch between the requirement of auto-regressive decoding and the design of most contemporary GPUs. Specifically, billions to trillions of parameters must be loaded to the GPU cache through its limited memory bandwidth for computation, but only a small batch of tokens is actually computed. Consequently, the GPU spends most of its time on memory transfer instead of computation. Recently, parallel decoding, a type of speculative decoding algorithms, is becoming more popular and has demonstrated impressive efficiency improvement in generation. It introduces extra decoding heads to large models, enabling them to predict multiple subsequent tokens simultaneously and verify these candidate continuations in a single decoding step. However, this approach deviates from the training objective of next token prediction used during pre-training, resulting in a low hit rate for candidate tokens. In this paper, we propose a new speculative decoding algorithm, Clover, which integrates sequential knowledge into the parallel decoding process. This enhancement improves the hit rate of speculators and thus boosts the overall efficiency. Clover transmits the sequential knowledge from pre-speculated tokens via the Regressive Connection, then employs an Attention Decoder to integrate these speculated tokens. Additionally, Clover incorporates an Augmenting Block that modifies the hidden states to better align with the purpose of speculative generation rather than next token prediction. The experiment results demonstrate that Clover outperforms the baseline by up to 91% on Baichuan-Small and 146% on Baichuan-Large, respectively, and exceeds the performance of the previously top-performing method, Medusa, by up to 37% on Baichuan-Small and 57% on Baichuan-Large, respectively.

arXiv:2404.19553공식 코드 · Papers with Code

Extending Llama-3's Context Ten-Fold Overnight

We extend the context length of Llama-3-8B-Instruct from 8K to 80K via QLoRA fine-tuning. The entire training cycle is super efficient, which takes 8 hours on one 8xA800 (80G) GPU machine. The resulted model exhibits superior performances across a broad range of evaluation tasks, such as NIHS, topic retrieval, and long-context language understanding; meanwhile, it also well preserves the original capability over short contexts. The dramatic context extension is mainly attributed to merely 3.5K synthetic training samples generated by GPT-4 , which indicates the LLMs' inherent (yet largely underestimated) potential to extend its original context length. In fact, the context length could be extended far beyond 80K with more computation resources. Therefore, the team will publicly release the entire resources (including data, model, data generation pipeline, training code) so as to facilitate the future research from the community: \url{https://github.com/FlagOpen/FlagEmbedding}.

arXiv:2404.19024공식 코드 · Papers with Code

Multi-Page Document Visual Question Answering using Self-Attention Scoring Mechanism

Documents are 2-dimensional carriers of written communication, and as such their interpretation requires a multi-modal approach where textual and visual information are efficiently combined. Document Visual Question Answering (Document VQA), due to this multi-modal nature, has garnered significant interest from both the document understanding and natural language processing communities. The state-of-the-art single-page Document VQA methods show impressive performance, yet in multi-page scenarios, these methods struggle. They have to concatenate all pages into one large page for processing, demanding substantial GPU resources, even for evaluation. In this work, we propose a novel method and efficient training strategy for multi-page Document VQA tasks. In particular, we employ a visual-only document representation, leveraging the encoder from a document understanding model, Pix2Struct. Our approach utilizes a self-attention scoring mechanism to generate relevance scores for each document page, enabling the retrieval of pertinent pages. This adaptation allows us to extend single-page Document VQA models to multi-page scenarios without constraints on the number of pages during evaluation, all with minimal demand for GPU resources. Our extensive experiments demonstrate not only achieving state-of-the-art performance without the need for Optical Character Recognition (OCR), but also sustained performance in scenarios extending to documents of nearly 800 pages compared to a maximum of 20 pages in the MP-DocVQA dataset. Our code is publicly available at \url{https://github.com/leitro/SelfAttnScoring-MPDocVQA}.

arXiv:2404.19702

GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting

We propose GS-LRM, a scalable large reconstruction model that can predict high-quality 3D Gaussian primitives from 2-4 posed sparse images in 0.23 seconds on single A100 GPU. Our model features a very simple transformer-based architecture; we patchify input posed images, pass the concatenated multi-view image tokens through a sequence of transformer blocks, and decode final per-pixel Gaussian parameters directly from these tokens for differentiable rendering. In contrast to previous LRMs that can only reconstruct objects, by predicting per-pixel Gaussians, GS-LRM naturally handles scenes with large variations in scale and complexity. We show that our model can work on both object and scene captures by training it on Objaverse and RealEstate10K respectively. In both scenarios, the models outperform state-of-the-art baselines by a wide margin. We also demonstrate applications of our model in downstream 3D generation tasks. Our project webpage is available at: https://sai-bi.github.io/project/gs-lrm/ .

arXiv:2404.19525공식 코드 · Papers with Code

MicroDreamer: Efficient 3D Generation in $\sim$20 Seconds by Score-based Iterative Reconstruction

Optimization-based approaches, such as score distillation sampling (SDS), show promise in zero-shot 3D generation but suffer from low efficiency, primarily due to the high number of function evaluations (NFEs) required for each sample and the limitation of optimization confined to latent space. This paper introduces score-based iterative reconstruction (SIR), an efficient and general algorithm mimicking a differentiable 3D reconstruction process to reduce the NFEs and enable optimization in pixel space. Given a single set of images sampled from a multi-view score-based diffusion model, SIR repeatedly optimizes 3D parameters, unlike the single-step optimization in SDS. With other improvements in training, we present an efficient approach called MicroDreamer that generally applies to various 3D representations and 3D generation tasks. In particular, MicroDreamer is 5-20 times faster than SDS in generating neural radiance field while retaining a comparable performance and takes about 20 seconds to create meshes from 3D Gaussian splatting on a single A100 GPU, halving the time of the fastest optimization-based baseline DreamGaussian with significantly superior performance compared to the measurement standard deviation. Our code is available at https://github.com/ML-GSAI/MicroDreamer.

arXiv:2404.18617공식 코드 · Papers with Code

CoSense3D: an Agent-based Efficient Learning Framework for Collective Perception

Collective Perception has attracted significant attention in recent years due to its advantage for mitigating occlusion and expanding the field-of-view, thereby enhancing reliability, efficiency, and, most crucially, decision-making safety. However, developing collective perception models is highly resource demanding due to extensive requirements of processing input data for many agents, usually dozens of images and point clouds for a single frame. This not only slows down the model development process for collective perception but also impedes the utilization of larger models. In this paper, we propose an agent-based training framework that handles the deep learning modules and agent data separately to have a cleaner data flow structure. This framework not only provides an API for flexibly prototyping the data processing pipeline and defining the gradient calculation for each agent, but also provides the user interface for interactive training, testing and data visualization. Training experiment results of four collective object detection models on the prominent collective perception benchmark OPV2V show that the agent-based training can significantly reduce the GPU memory consumption and training time while retaining inference performance. The framework and model implementations are available at \url{https://github.com/YuanYunshuang/CoSense3D}

arXiv:2404.18174공식 코드 · Papers with Code

Mamba-FETrack: Frame-Event Tracking via State Space Model

RGB-Event based tracking is an emerging research topic, focusing on how to effectively integrate heterogeneous multi-modal data (synchronized exposure video frames and asynchronous pulse Event stream). Existing works typically employ Transformer based networks to handle these modalities and achieve decent accuracy through input-level or feature-level fusion on multiple datasets. However, these trackers require significant memory consumption and computational complexity due to the use of self-attention mechanism. This paper proposes a novel RGB-Event tracking framework, Mamba-FETrack, based on the State Space Model (SSM) to achieve high-performance tracking while effectively reducing computational costs and realizing more efficient tracking. Specifically, we adopt two modality-specific Mamba backbone networks to extract the features of RGB frames and Event streams. Then, we also propose to boost the interactive learning between the RGB and Event features using the Mamba network. The fused features will be fed into the tracking head for target object localization. Extensive experiments on FELT and FE108 datasets fully validated the efficiency and effectiveness of our proposed tracker. Specifically, our Mamba-based tracker achieves 43.5/55.6 on the SR/PR metric, while the ViT-S based tracker (OSTrack) obtains 40.0/50.9. The GPU memory cost of ours and ViT-S based tracker is 13.98GB and 15.44GB, which decreased about $9.5\%$. The FLOPs and parameters of ours/ViT-S based OSTrack are 59GB/1076GB and 7MB/60MB, which decreased about $94.5\%$ and $88.3\%$, respectively. We hope this work can bring some new insights to the tracking field and greatly promote the application of the Mamba architecture in tracking. The source code of this work will be released on \url{https://github.com/Event-AHU/Mamba_FETrack}.

arXiv:2404.17962

Deep Learning for Low-Latency, Quantum-Ready RF Sensing

Recent work has shown the promise of applying deep learning to enhance software processing of radio frequency (RF) signals. In parallel, hardware developments with quantum RF sensors based on Rydberg atoms are breaking longstanding barriers in frequency range, resolution, and sensitivity. In this paper, we describe our implementations of quantum-ready machine learning approaches for RF signal classification. Our primary objective is latency: while deep learning offers a more powerful computational paradigm, it also traditionally incurs latency overheads that hinder wider scale deployment. Our work spans three axes. (1) A novel continuous wavelet transform (CWT) based recurrent neural network (RNN) architecture that enables flexible online classification of RF signals on-the-fly with reduced sampling time. (2) Low-latency inference techniques for both GPU and CPU that span over 100x reductions in inference time, enabling real-time operation with sub-millisecond inference. (3) Quantum-readiness validated through application of our models to physics-based simulation of Rydberg atom QRF sensors. Altogether, our work bridges towards next-generation RF sensors that use quantum technology to surpass previous physical limits, paired with latency-optimized AI/ML software that is suitable for real-time deployment.

arXiv:2404.17107공식 코드 · Papers with Code

Exploring Pre-trained General-purpose Audio Representations for Heart Murmur Detection

To reduce the need for skilled clinicians in heart sound interpretation, recent studies on automating cardiac auscultation have explored deep learning approaches. However, despite the demands for large data for deep learning, the size of the heart sound datasets is limited, and no pre-trained model is available. On the contrary, many pre-trained models for general audio tasks are available as general-purpose audio representations. This study explores the potential of general-purpose audio representations pre-trained on large-scale datasets for transfer learning in heart murmur detection. Experiments on the CirCor DigiScope heart sound dataset show that the recent self-supervised learning Masked Modeling Duo (M2D) outperforms previous methods with the results of a weighted accuracy of 0.832 and an unweighted average recall of 0.713. Experiments further confirm improved performance by ensembling M2D with other models. These results demonstrate the effectiveness of general-purpose audio representation in processing heart sounds and open the way for further applications. Our code is available online which runs on a 24 GB consumer GPU at https://github.com/nttcslab/m2d/tree/master/app/circor

arXiv:2404.17394

Child Speech Recognition in Human-Robot Interaction: Problem Solved?

Automated Speech Recognition shows superhuman performance for adult English speech on a range of benchmarks, but disappoints when fed children's speech. This has long sat in the way of child-robot interaction. Recent evolutions in data-driven speech recognition, including the availability of Transformer architectures and unprecedented volumes of training data, might mean a breakthrough for child speech recognition and social robot applications aimed at children. We revisit a study on child speech recognition from 2017 and show that indeed performance has increased, with newcomer OpenAI Whisper doing markedly better than leading commercial cloud services. Performance improves even more in highly structured interactions when priming models with specific phrases. While transcription is not perfect yet, the best model recognises 60.3% of sentences correctly barring small grammatical differences, with sub-second transcription time running on a local GPU, showing potential for usable autonomous child-robot speech interactions.

arXiv:2404.16221

NeRF-XL: Scaling NeRFs with Multiple GPUs

We present NeRF-XL, a principled method for distributing Neural Radiance Fields (NeRFs) across multiple GPUs, thus enabling the training and rendering of NeRFs with an arbitrarily large capacity. We begin by revisiting existing multi-GPU approaches, which decompose large scenes into multiple independently trained NeRFs, and identify several fundamental issues with these methods that hinder improvements in reconstruction quality as additional computational resources (GPUs) are used in training. NeRF-XL remedies these issues and enables the training and rendering of NeRFs with an arbitrary number of parameters by simply using more hardware. At the core of our method lies a novel distributed training and rendering formulation, which is mathematically equivalent to the classic single-GPU case and minimizes communication between GPUs. By unlocking NeRFs with arbitrarily large parameter counts, our approach is the first to reveal multi-GPU scaling laws for NeRFs, showing improvements in reconstruction quality with larger parameter counts and speed improvements with more GPUs. We demonstrate the effectiveness of NeRF-XL on a wide variety of datasets, including the largest open-source dataset to date, MatrixCity, containing 258K images covering a 25km^2 city area.

arXiv:2404.10777

Divide-Conquer-and-Merge: Memory- and Time-Efficient Holographic Displays

Recently, deep learning-based computer-generated holography (CGH) has demonstrated tremendous potential in three-dimensional (3D) displays and yielded impressive display quality. However, most existing deep learning-based CGH techniques can only generate holograms of 1080p resolution, which is far from the ultra-high resolution (16K+) required for practical virtual reality (VR) and augmented reality (AR) applications to support a wide field of view and large eye box. One of the major obstacles in current CGH frameworks lies in the limited memory available on consumer-grade GPUs which could not facilitate the generation of higher-definition holograms. To overcome the aforementioned challenge, we proposed a divide-conquer-and-merge strategy to address the memory and computational capacity scarcity in ultra-high-definition CGH generation. This algorithm empowers existing CGH frameworks to synthesize higher-definition holograms at a faster speed while maintaining high-fidelity image display quality. Both simulations and experiments were conducted to demonstrate the capabilities of the proposed framework. By integrating our strategy into HoloNet and CCNNs, we achieved significant reductions in GPU memory usage during the training period by 64.3\% and 12.9\%, respectively. Furthermore, we observed substantial speed improvements in hologram generation, with an acceleration of up to 3$\times$ and 2 $\times$, respectively. Particularly, we successfully trained and inferred 8K definition holograms on an NVIDIA GeForce RTX 3090 GPU for the first time in simulations. Furthermore, we conducted full-color optical experiments to verify the effectiveness of our method. We believe our strategy can provide a novel approach for memory- and time-efficient holographic displays.

arXiv:2404.16283

Andes: Defining and Enhancing Quality-of-Experience in LLM-Based Text Streaming Services

Large language models (LLMs) are now at the core of conversational AI services such as real-time translation and chatbots, which provide live user interaction by incrementally streaming text to the user. However, existing LLM serving systems fail to provide good user experience because their optimization metrics are not always aligned with user experience. In this paper, we first introduce and define the notion of Quality-of-Experience (QoE) for text streaming services by considering each user's end-to-end interaction timeline. Based on this, we propose Andes, a QoE-aware LLM serving system that enhances user experience by ensuring that users receive the first token promptly and subsequent tokens at a smooth, digestible pace, even during surge periods. This is enabled by Andes's preemptive request scheduler that dynamically prioritizes requests at the token granularity based on each request's expected QoE gain and GPU resource usage. Our evaluations demonstrate that, compared to state-of-the-art LLM serving systems, Andes improves the average QoE by up to $4.7\times$ given the same GPU resource, or saves up to 61% GPU resources while maintaining the same high QoE.