PyTorchKR Papers

최신 논문

총 5,357편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2410.13842공식 코드 · Papers with Code

D-FINE: Redefine Regression Task in DETRs as Fine-grained Distribution Refinement

We introduce D-FINE, a powerful real-time object detector that achieves outstanding localization precision by redefining the bounding box regression task in DETR models. D-FINE comprises two key components: Fine-grained Distribution Refinement (FDR) and Global Optimal Localization Self-Distillation (GO-LSD). FDR transforms the regression process from predicting fixed coordinates to iteratively refining probability distributions, providing a fine-grained intermediate representation that significantly enhances localization accuracy. GO-LSD is a bidirectional optimization strategy that transfers localization knowledge from refined distributions to shallower layers through self-distillation, while also simplifying the residual prediction tasks for deeper layers. Additionally, D-FINE incorporates lightweight optimizations in computationally intensive modules and operations, achieving a better balance between speed and accuracy. Specifically, D-FINE-L / X achieves 54.0% / 55.8% AP on the COCO dataset at 124 / 78 FPS on an NVIDIA T4 GPU. When pretrained on Objects365, D-FINE-L / X attains 57.1% / 59.3% AP, surpassing all existing real-time detectors. Furthermore, our method significantly enhances the performance of a wide range of DETR models by up to 5.3% AP with negligible extra parameters and training costs. Our code and pretrained models: https://github.com/Peterande/D-FINE.

arXiv:2410.13415공식 코드 · Papers with Code

Shavette: Low Power Neural Network Acceleration via Algorithm-level Error Detection and Undervolting

Reduced voltage operation is an effective technique for substantial energy efficiency improvement in digital circuits. This brief introduces a simple approach for enabling reduced voltage operation of Deep Neural Network (DNN) accelerators by mere software modifications. Conventional approaches for enabling reduced voltage operation e.g., Timing Error Detection (TED) systems, incur significant development costs and overheads, while not being applicable to the off-the-shelf components. Contrary to those, the solution proposed in this paper relies on algorithm-based error detection, and hence, is implemented with low development costs, does not require any circuit modifications, and is even applicable to commodity devices. By showcasing the solution through experimenting on popular DNNs, i.e., LeNet and VGG16, on a GPU platform, we demonstrate 18% to 25% energy saving with no accuracy loss of the models and negligible throughput compromise (< 3.9%), considering the overheads from integration of the error detection schemes into the DNN. The integration of presented algorithmic solution into the design is simpler when compared conventional TED based techniques that require extensive circuit-level modifications, cell library characterizations or special support from the design tools.

arXiv:2410.11871

TinyClick: Single-Turn Agent for Empowering GUI Automation

We present an UI agent for user interface (UI) interaction tasks, using Vision-Language Model Florence-2-Base. The agent's primary task is identifying the screen coordinates of the UI element corresponding to the user's command. It demonstrates very strong performance on Screenspot and OmniAct annotations, while maintaining a very small size of 0.27B parameters and minimal latency. Moreover, training needs small compute budget of 56 GPU-hours (worth about 40 USD). Relevant improvement comes from vision-specific multi-task training and MLLM-based data augmentation. We hope that decreased needs for expensive compute resources and manually annotated data will allow to facilitate more inclusive and sustainable research of UI agents.

arXiv:2410.12642

Optimization and Application of Cloud-based Deep Learning Architecture for Multi-Source Data Prediction

This study develops a cloud-based deep learning system for early prediction of diabetes, leveraging the distributed computing capabilities of the AWS cloud platform and deep learning technologies to achieve efficient and accurate risk assessment. The system utilizes EC2 p3.8xlarge GPU instances to accelerate model training, reducing training time by 93.2% while maintaining a prediction accuracy of 94.2%. With an automated data processing and model training pipeline built using Apache Airflow, the system can complete end-to-end updates within 18.7 hours. In clinical applications, the system demonstrates a prediction accuracy of 89.8%, sensitivity of 92.3%, and specificity of 95.1%. Early interventions based on predictions lead to a 37.5% reduction in diabetes incidence among the target population. The system's high performance and scalability provide strong support for large-scale diabetes prevention and management, showcasing significant public health value.

arXiv:2410.12781

Long-LRM: Long-sequence Large Reconstruction Model for Wide-coverage Gaussian Splats

We propose Long-LRM, a generalizable 3D Gaussian reconstruction model that is capable of reconstructing a large scene from a long sequence of input images. Specifically, our model can process 32 source images at 960x540 resolution within only 1.3 seconds on a single A100 80G GPU. Our architecture features a mixture of the recent Mamba2 blocks and the classical transformer blocks which allowed many more tokens to be processed than prior work, enhanced by efficient token merging and Gaussian pruning steps that balance between quality and efficiency. Unlike previous feed-forward models that are limited to processing 1~4 input images and can only reconstruct a small portion of a large scene, Long-LRM reconstructs the entire scene in a single feed-forward step. On large-scale scene datasets such as DL3DV-140 and Tanks and Temples, our method achieves performance comparable to optimization-based approaches while being two orders of magnitude more efficient. Project page: https://arthurhero.github.io/projects/llrm

arXiv:2410.12074공식 코드 · Papers with Code

nvTorchCam: An Open-source Library for Camera-Agnostic Differentiable Geometric Vision

We introduce nvTorchCam, an open-source library under the Apache 2.0 license, designed to make deep learning algorithms camera model-independent. nvTorchCam abstracts critical camera operations such as projection and unprojection, allowing developers to implement algorithms once and apply them across diverse camera models--including pinhole, fisheye, and 360 equirectangular panoramas, which are commonly used in automotive and real estate capture applications. Built on PyTorch, nvTorchCam is fully differentiable and supports GPU acceleration and batching for efficient computation. Furthermore, deep learning models trained for one camera type can be directly transferred to other camera types without requiring additional modification. In this paper, we provide an overview of nvTorchCam, its functionality, and present various code examples and diagrams to demonstrate its usage. Source code and installation instructions can be found on the nvTorchCam GitHub page at https://github.com/NVlabs/nvTorchCam.

arXiv:2410.10847공식 코드 · Papers with Code

Lotus: learning-based online thermal and latency variation management for two-stage detectors on edge devices

Two-stage object detectors exhibit high accuracy and precise localization, especially for identifying small objects that are favorable for various edge applications. However, the high computation costs associated with two-stage detection methods cause more severe thermal issues on edge devices, incurring dynamic runtime frequency change and thus large inference latency variations. Furthermore, the dynamic number of proposals in different frames leads to various computations over time, resulting in further latency variations. The significant latency variations of detectors on edge devices can harm user experience and waste hardware resources. To avoid thermal throttling and provide stable inference speed, we propose Lotus, a novel framework that is tailored for two-stage detectors to dynamically scale CPU and GPU frequencies jointly in an online manner based on deep reinforcement learning (DRL). To demonstrate the effectiveness of Lotus, we implement it on NVIDIA Jetson Orin Nano and Mi 11 Lite mobile platforms. The results indicate that Lotus can consistently and significantly reduce latency variation, achieve faster inference, and maintain lower CPU and GPU temperatures under various settings.

arXiv:2410.10870

PortLLM: Personalizing Evolving Large Language Models with Training-Free and Portable Model Patches

As large language models (LLMs) increasingly shape the AI landscape, fine-tuning pretrained models has become more popular than in the pre-LLM era for achieving optimal performance in domain-specific tasks. However, pretrained LLMs such as ChatGPT are periodically evolved, i.e., model parameters are frequently updated), making it challenging for downstream users with limited resources to keep up with fine-tuning the newest LLMs for their domain application. Even though fine-tuning costs have nowadays been reduced thanks to the innovations of parameter-efficient fine-tuning such as LoRA, not all downstream users have adequate computing for frequent personalization. Moreover, access to fine-tuning datasets, particularly in sensitive domains such as healthcare, could be time-restrictive, making it crucial to retain the knowledge encoded in earlier fine-tuned rounds for future adaptation. In this paper, we present PortLLM, a training-free framework that (i) creates an initial lightweight model update patch to capture domain-specific knowledge, and (ii) allows a subsequent seamless plugging for the continual personalization of evolved LLM at minimal cost. Our extensive experiments cover seven representative datasets, from easier question-answering tasks {BoolQ, SST2} to harder reasoning tasks {WinoGrande, GSM8K}, and models including {Mistral-7B, Llama2, Llama3.1, and Gemma2}, validating the portability of our designed model patches and showcasing the effectiveness of our proposed framework. For instance, PortLLM achieves comparable performance to LoRA fine-tuning with reductions of up to 12.2x in GPU memory usage. Finally, we provide theoretical justifications to understand the portability of our model update patches, which offers new insights into the theoretical dimension of LLMs' personalization.

arXiv:2410.10887

ActNAS : Generating Efficient YOLO Models using Activation NAS

Activation functions introduce non-linearity into Neural Networks, enabling them to learn complex patterns. Different activation functions vary in speed and accuracy, ranging from faster but less accurate options like ReLU to slower but more accurate functions like SiLU or SELU. Typically, same activation function is used throughout an entire model architecture. In this paper, we conduct a comprehensive study on the effects of using mixed activation functions in YOLO-based models, evaluating their impact on latency, memory usage, and accuracy across CPU, NPU, and GPU edge devices. We also propose a novel approach that leverages Neural Architecture Search (NAS) to design YOLO models with optimized mixed activation functions.The best model generated through this method demonstrates a slight improvement in mean Average Precision (mAP) compared to baseline model (SiLU), while it is 22.28% faster and consumes 64.15% less memory on the reference NPU device.

arXiv:2410.11019

ET-Former: Efficient Triplane Deformable Attention for 3D Semantic Scene Completion From Monocular Camera

We introduce ET-Former, a novel end-to-end algorithm for semantic scene completion using a single monocular camera. Our approach generates a semantic occupancy map from single RGB observation while simultaneously providing uncertainty estimates for semantic predictions. By designing a triplane-based deformable attention mechanism, our approach improves geometric understanding of the scene than other SOTA approaches and reduces noise in semantic predictions. Additionally, through the use of a Conditional Variational AutoEncoder (CVAE), we estimate the uncertainties of these predictions. The generated semantic and uncertainty maps will aid in the formulation of navigation strategies that facilitate safe and permissible decision-making in the future. Evaluated on the Semantic-KITTI dataset, ET-Former achieves the highest IoU and mIoU, surpassing other methods by 15.16% in IoU and 24.24% in mIoU, while reducing GPU memory usage of existing methods by 25%-50.5%.

arXiv:2410.11419공식 코드 · Papers with Code

GS^3: Efficient Relighting with Triple Gaussian Splatting

We present a spatial and angular Gaussian based representation and a triple splatting process, for real-time, high-quality novel lighting-and-view synthesis from multi-view point-lit input images. To describe complex appearance, we employ a Lambertian plus a mixture of angular Gaussians as an effective reflectance function for each spatial Gaussian. To generate self-shadow, we splat all spatial Gaussians towards the light source to obtain shadow values, which are further refined by a small multi-layer perceptron. To compensate for other effects like global illumination, another network is trained to compute and add a per-spatial-Gaussian RGB tuple. The effectiveness of our representation is demonstrated on 30 samples with a wide variation in geometry (from solid to fluffy) and appearance (from translucent to anisotropic), as well as using different forms of input data, including rendered images of synthetic/reconstructed objects, photographs captured with a handheld camera and a flash, or from a professional lightstage. We achieve a training time of 40-70 minutes and a rendering speed of 90 fps on a single commodity GPU. Our results compare favorably with state-of-the-art techniques in terms of quality/performance. Our code and data are publicly available at https://GSrelight.github.io/.

arXiv:2410.10989공식 코드 · Papers with Code

Liger Kernel: Efficient Triton Kernels for LLM Training

Training Large Language Models (LLMs) efficiently at scale presents a formidable challenge, driven by their ever-increasing computational demands and the need for enhanced performance. In this work, we introduce Liger-Kernel, an open-sourced set of Triton kernels developed specifically for LLM training. With kernel optimization techniques like kernel operation fusing and input chunking, our kernels achieve on average a 20% increase in training throughput and a 60% reduction in GPU memory usage for popular LLMs compared to HuggingFace implementations. In addition, Liger-Kernel is designed with modularity, accessibility, and adaptability in mind, catering to both casual and expert users. Comprehensive benchmarks and integration tests are built in to ensure compatibility, performance, correctness, and convergence across diverse computing environments and model architectures. The source code is available under a permissive license at: github.com/linkedin/Liger-Kernel.

arXiv:2410.11457공식 코드 · Papers with Code

LR-SQL: A Supervised Fine-Tuning Method for Text2SQL Tasks under Low-Resource Scenarios

Large language models revolutionize Text2SQL through supervised fine-tuning, yet a crucial limitation is overlooked: the complexity of databases leads to an increased context length, consequently resulting in higher GPU memory demands for model fine-tuning. To address this issue, we propose LR-SQL. LR-SQL comprises two supervised fine-tuning models: the schema\_link model and the SQL\_generation model, with the schema\_link model serving as the focal point for streamlining the overall process. During the fine-tuning of the schema\_link model, LR-SQL breaks down the complete database into flexible combinations of tables with adjustable quantities, enabling the model to learn the relationships within the entire database from these dispersed slices. Furthermore, to enhance the model's ability to perceive the relationships among various discrete slices during inference, LR-SQL trains the model's Chain-of-Thought capability for this task. Experimental results demonstrate that LR-SQL can reduce the total GPU memory usage by 40\% compared to existing fine-tuning methods, while only losing 2\% of table prediction accuracy in schema\_link task. For the overall Text2SQL task, the Execution Accuracy decrease by 0.6\%.Our project is now available on https://github.com/hongWin/LR-SQL

arXiv:2410.10819공식 코드 · Papers with Code

DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads

Deploying long-context large language models (LLMs) is essential but poses significant computational and memory challenges. Caching all Key and Value (KV) states across all attention heads consumes substantial memory. Existing KV cache pruning methods either damage the long-context capabilities of LLMs or offer only limited efficiency improvements. In this paper, we identify that only a fraction of attention heads, a.k.a, Retrieval Heads, are critical for processing long contexts and require full attention across all tokens. In contrast, all other heads, which primarily focus on recent tokens and attention sinks--referred to as Streaming Heads--do not require full attention. Based on this insight, we introduce DuoAttention, a framework that only applies a full KV cache to retrieval heads while using a light-weight, constant-length KV cache for streaming heads, which reduces both LLM's decoding and pre-filling memory and latency without compromising its long-context abilities. DuoAttention uses a lightweight, optimization-based algorithm with synthetic data to identify retrieval heads accurately. Our method significantly reduces long-context inference memory by up to 2.55x for MHA and 1.67x for GQA models while speeding up decoding by up to 2.18x and 1.50x and accelerating pre-filling by up to 1.73x and 1.63x for MHA and GQA models, respectively, with minimal accuracy loss compared to full attention. Notably, combined with quantization, DuoAttention enables Llama-3-8B decoding with 3.3 million context length on a single A100 GPU. Code is provided in https://github.com/mit-han-lab/duo-attention.

arXiv:2410.10122공식 코드 · Papers with Code

MuseTalk: Real-Time High-Fidelity Video Dubbing via Spatio-Temporal Sampling

Real-time video dubbing that preserves identity consistency while achieving accurate lip synchronization remains a critical challenge. Existing approaches face a trilemma: diffusion-based methods achieve high visual fidelity but suffer from prohibitive computational costs, while GAN-based solutions sacrifice lip-sync accuracy or dental details for real-time performance. We present MuseTalk, a novel two-stage training framework that resolves this trade-off through latent space optimization and spatio-temporal data sampling strategy. Our key innovations include: (1) During the Facial Abstract Pretraining stage, we propose Informative Frame Sampling to temporally align reference-source pose pairs, eliminating redundant feature interference while preserving identity cues. (2) In the Lip-Sync Adversarial Finetuning stage, we employ Dynamic Margin Sampling to spatially select the most suitable lip-movement-promoting regions, balancing audio-visual synchronization and dental clarity. (3) MuseTalk establishes an effective audio-visual feature fusion framework in the latent space, delivering 30 FPS output at 256*256 resolution on an NVIDIA V100 GPU. Extensive experiments demonstrate that MuseTalk outperforms state-of-the-art methods in visual fidelity while achieving comparable lip-sync accuracy. %The codes and models will be made publicly available upon acceptance. The code is made available at \href{https://github.com/TMElyralab/MuseTalk}{https://github.com/TMElyralab/MuseTalk}

arXiv:2410.10629

SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers

We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096$\times$4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8$\times$, we trained an AE that can compress images 32$\times$, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolutions without sacrificing quality. (3) Decoder-only text encoder: we replaced T5 with modern decoder-only small LLM as the text encoder and designed complex human instruction with in-context learning to enhance the image-text alignment. (4) Efficient training and sampling: we propose Flow-DPM-Solver to reduce sampling steps, with efficient caption labeling and selection to accelerate convergence. As a result, Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput. Moreover, Sana-0.6B can be deployed on a 16GB laptop GPU, taking less than 1 second to generate a 1024$\times$1024 resolution image. Sana enables content creation at low cost. Code and model will be publicly released.

arXiv:2410.10510공식 코드 · Papers with Code

Exploiting Local Features and Range Images for Small Data Real-Time Point Cloud Semantic Segmentation

Semantic segmentation of point clouds is an essential task for understanding the environment in autonomous driving and robotics. Recent range-based works achieve real-time efficiency, while point- and voxel-based methods produce better results but are affected by high computational complexity. Moreover, highly complex deep learning models are often not suited to efficiently learn from small datasets. Their generalization capabilities can easily be driven by the abundance of data rather than the architecture design. In this paper, we harness the information from the three-dimensional representation to proficiently capture local features, while introducing the range image representation to incorporate additional information and facilitate fast computation. A GPU-based KDTree allows for rapid building, querying, and enhancing projection with straightforward operations. Extensive experiments on SemanticKITTI and nuScenes datasets demonstrate the benefits of our modification in a ``small data'' setup, in which only one sequence of the dataset is used to train the models, but also in the conventional setup, where all sequences except one are used for training. We show that a reduced version of our model not only demonstrates strong competitiveness against full-scale state-of-the-art models but also operates in real-time, making it a viable choice for real-world case applications. The code of our method is available at https://github.com/Bender97/WaffleAndRange.

arXiv:2410.10166공식 코드 · Papers with Code

Automated Filtering of Human Feedback Data for Aligning Text-to-Image Diffusion Models

Fine-tuning text-to-image diffusion models with human feedback is an effective method for aligning model behavior with human intentions. However, this alignment process often suffers from slow convergence due to the large size and noise present in human feedback datasets. In this work, we propose FiFA, a novel automated data filtering algorithm designed to enhance the fine-tuning of diffusion models using human feedback datasets with direct preference optimization (DPO). Specifically, our approach selects data by solving an optimization problem to maximize three components: preference margin, text quality, and text diversity. The concept of preference margin is used to identify samples that contain high informational value to address the noisy nature of feedback dataset, which is calculated using a proxy reward model. Additionally, we incorporate text quality, assessed by large language models to prevent harmful contents, and consider text diversity through a k-nearest neighbor entropy estimator to improve generalization. Finally, we integrate all these components into an optimization process, with approximating the solution by assigning importance score to each data pair and selecting the most important ones. As a result, our method efficiently filters data automatically, without the need for manual intervention, and can be applied to any large-scale dataset. Experimental results show that FiFA significantly enhances training stability and achieves better performance, being preferred by humans 17% more, while using less than 0.5% of the full data and thus 1% of the GPU hours compared to utilizing full human feedback datasets.

arXiv:2410.10089

PromptGCN: Bridging Subgraph Gaps in Lightweight GCNs

Graph Convolutional Networks (GCNs) are widely used in graph-based applications, such as social networks and recommendation systems. Nevertheless, large-scale graphs or deep aggregation layers in full-batch GCNs consume significant GPU memory, causing out of memory (OOM) errors on mainstream GPUs (e.g., 29GB memory consumption on the Ogbnproducts graph with 5 layers). The subgraph sampling methods reduce memory consumption to achieve lightweight GCNs by partitioning the graph into multiple subgraphs and sequentially training GCNs on each subgraph. However, these methods yield gaps among subgraphs, i.e., GCNs can only be trained based on subgraphs instead of global graph information, which reduces the accuracy of GCNs. In this paper, we propose PromptGCN, a novel prompt-based lightweight GCN model to bridge the gaps among subgraphs. First, the learnable prompt embeddings are designed to obtain global information. Then, the prompts are attached into each subgraph to transfer the global information among subgraphs. Extensive experimental results on seven largescale graphs demonstrate that PromptGCN exhibits superior performance compared to baselines. Notably, PromptGCN improves the accuracy of subgraph sampling methods by up to 5.48% on the Flickr dataset. Overall, PromptGCN can be easily combined with any subgraph sampling method to obtain a lightweight GCN model with higher accuracy.

arXiv:2410.09687

MoIN: Mixture of Introvert Experts to Upcycle an LLM

The goal of this paper is to improve (upcycle) an existing large language model without the prohibitive requirements of continued pre-training of the full-model. The idea is to split the pre-training data into semantically relevant groups and train an expert on each subset. An expert takes the form of a lightweight adapter added on the top of a frozen base model. During inference, an incoming query is first routed to the most relevant expert which is then loaded onto the base model for the forward pass. Unlike typical Mixture of Experts (MoE) models, the experts in our method do not work with other experts for a single query. Hence, we dub them "introvert" experts. Freezing the base model and keeping the experts as lightweight adapters allows extreme parallelism during training and inference. Training of all experts can be done in parallel without any communication channels between them. Similarly, the inference can also be heavily parallelized by distributing experts on different GPUs and routing each request to the GPU containing its relevant expert. We implement a proof-of-concept version of this method and show the validity of our approach.