Showing 1–13 of 13 results for author: Kwon, W

R3eVision: A Survey on Robust Rendering, Restoration, and Enhancement for 3D Low-Level Vision

Authors: Weeyoung Kwon, Jeahun Sung, Minkyu Jeon, Chanho Eom, Jihyong Oh

Abstract: Neural rendering methods such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have achieved significant progress in photorealistic 3D scene reconstruction and novel view synthesis. However, most existing models assume clean and high-resolution (HR) multi-view inputs, which limits their robustness under real-world degradations such as noise, blur, low-resolution (LR), and weather-… ▽ More Neural rendering methods such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have achieved significant progress in photorealistic 3D scene reconstruction and novel view synthesis. However, most existing models assume clean and high-resolution (HR) multi-view inputs, which limits their robustness under real-world degradations such as noise, blur, low-resolution (LR), and weather-induced artifacts. To address these limitations, the emerging field of 3D Low-Level Vision (3D LLV) extends classical 2D Low-Level Vision tasks including super-resolution (SR), deblurring, weather degradation removal, restoration, and enhancement into the 3D spatial domain. This survey, referred to as R\textsuperscript{3}eVision, provides a comprehensive overview of robust rendering, restoration, and enhancement for 3D LLV by formalizing the degradation-aware rendering problem and identifying key challenges related to spatio-temporal consistency and ill-posed optimization. Recent methods that integrate LLV into neural rendering frameworks are categorized to illustrate how they enable high-fidelity 3D reconstruction under adverse conditions. Application domains such as autonomous driving, AR/VR, and robotics are also discussed, where reliable 3D perception from degraded inputs is critical. By reviewing representative methods, datasets, and evaluation protocols, this work positions 3D LLV as a fundamental direction for robust 3D content generation and scene-level reconstruction in real-world environments. △ Less

Submitted 23 June, 2025; v1 submitted 19 June, 2025; originally announced June 2025.

Comments: Please visit our project page at https://github.com/CMLab-Korea/Awesome-3D-Low-Level-Vision

Gemma 3 Technical Report

Authors: Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, Louis Rouillard, Thomas Mesnard, Geoffrey Cideron, Jean-bastien Grill, Sabela Ramos, Edouard Yvinec, Michelle Casbon, Etienne Pot, Ivo Penchev, Gaël Liu, Francesco Visin, Kathleen Kenealy, Lucas Beyer, Xiaohai Zhai, Anton Tsitsulin , et al. (191 additional authors not shown)

Abstract: We introduce Gemma 3, a multimodal addition to the Gemma family of lightweight open models, ranging in scale from 1 to 27 billion parameters. This version introduces vision understanding abilities, a wider coverage of languages and longer context - at least 128K tokens. We also change the architecture of the model to reduce the KV-cache memory that tends to explode with long context. This is achie… ▽ More We introduce Gemma 3, a multimodal addition to the Gemma family of lightweight open models, ranging in scale from 1 to 27 billion parameters. This version introduces vision understanding abilities, a wider coverage of languages and longer context - at least 128K tokens. We also change the architecture of the model to reduce the KV-cache memory that tends to explode with long context. This is achieved by increasing the ratio of local to global attention layers, and keeping the span on local attention short. The Gemma 3 models are trained with distillation and achieve superior performance to Gemma 2 for both pre-trained and instruction finetuned versions. In particular, our novel post-training recipe significantly improves the math, chat, instruction-following and multilingual abilities, making Gemma3-4B-IT competitive with Gemma2-27B-IT and Gemma3-27B-IT comparable to Gemini-1.5-Pro across benchmarks. We release all our models to the community. △ Less

Submitted 25 March, 2025; originally announced March 2025.

Jenga: Effective Memory Management for Serving LLM with Heterogeneity

Authors: Chen Zhang, Kuntai Du, Shu Liu, Woosuk Kwon, Xiangxi Mo, Yufeng Wang, Xiaoxuan Liu, Kaichao You, Zhuohan Li, Mingsheng Long, Jidong Zhai, Joseph Gonzalez, Ion Stoica

Abstract: Large language models (LLMs) are widely used but expensive to run, especially as inference workloads grow. To lower costs, maximizing the request batch size by managing GPU memory efficiently is crucial. While PagedAttention has recently been proposed to improve the efficiency of memory management, we find that the growing heterogeneity in the embeddings dimensions, attention, and access patterns… ▽ More Large language models (LLMs) are widely used but expensive to run, especially as inference workloads grow. To lower costs, maximizing the request batch size by managing GPU memory efficiently is crucial. While PagedAttention has recently been proposed to improve the efficiency of memory management, we find that the growing heterogeneity in the embeddings dimensions, attention, and access patterns of modern LLM architectures introduces new challenges for memory allocation. In this paper, we present Jenga, a novel memory allocation framework for heterogeneous embeddings in LLMs. Jenga tackles two key challenges: (1) minimizing memory fragmentation when managing embeddings of different sizes, and (2) enabling flexible caching and eviction policies tailored to the specific token-dependency patterns of various layers. Jenga employs a two-level memory allocator, leveraging the least common multiple (LCM) of embedding sizes to optimize memory usage and providing APIs to express layer-specific caching logic to enhance memory reuse. We implemente Jenga on vLLM, a state-of-the-art LLM inference engine, and evaluate it with diverse LLMs, datasets, and GPU configurations. Evaluations show that Jenga improves GPU memory utilization by up to 79.6%, and increases serving throughput by up to 4.92x (1.80x on average). △ Less

Submitted 23 March, 2025; originally announced March 2025.

Comments: 16 pages, 19 figures

APEX: An Extensible and Dynamism-Aware Simulator for Automated Parallel Execution in LLM Serving

Authors: Yi-Chien Lin, Woosuk Kwon, Ronald Pineda, Fanny Nina Paravecino

Abstract: Efficiently serving Large Language Models (LLMs) requires selecting an optimal parallel execution plan, balancing computation, memory, and communication overhead. However, determining the best strategy is challenging due to varying parallelism techniques (data, pipeline, tensor) and workload characteristics (e.g., compute-intensive tasks with long prompts vs. memory-intensive tasks with long gener… ▽ More Efficiently serving Large Language Models (LLMs) requires selecting an optimal parallel execution plan, balancing computation, memory, and communication overhead. However, determining the best strategy is challenging due to varying parallelism techniques (data, pipeline, tensor) and workload characteristics (e.g., compute-intensive tasks with long prompts vs. memory-intensive tasks with long generation). We propose APEX, an LLM serving system simulator that efficiently identifies optimal parallel execution plans by considering key factors of LLM serving systems, such as memory usage, batching behavior, etc. APEX performs dynamism-aware simulation to model iteration-level batching, and leverages LLMs' repetitive structure to reduce design space, scaling efficiently to trillion-scale models. APEX abstracts the key components of LLM serving systems, including the model, batching module, quantization formats, and device clusters, enabling the simulator to be general and extensible. Simulating on a CPU, APEX evaluates execution plans for various device clusters, covering diverse LLMs and workloads. APEX finds plans up to 3.37x faster than heuristics, and also plans that reduce energy consumption by up to 45% compared to latency-optimal plans. APEX performs comprehensive evaluations, reporting key system metrics like time per output token and time to first token, which can help service providers meet SLOs. APEX identifies an optimal plan within 15 minutes on a CPU, making it 71x faster and 1234x more cost-effective than cloud-based GPU deployment. APEX can be accessed at https://github.com/microsoft/apex_plus △ Less

Submitted 29 April, 2025; v1 submitted 26 November, 2024; originally announced November 2024.

Gemma 2: Improving Open Language Models at a Practical Size

Authors: Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Johan Ferret, Peter Liu, Pouya Tafti, Abe Friesen, Michelle Casbon, Sabela Ramos, Ravin Kumar, Charline Le Lan, Sammy Jerome, Anton Tsitsulin, Nino Vieillard, Piotr Stanczyk, Sertan Girgin, Nikola Momchev, Matt Hoffman , et al. (173 additional authors not shown)

Abstract: In this work, we introduce Gemma 2, a new addition to the Gemma family of lightweight, state-of-the-art open models, ranging in scale from 2 billion to 27 billion parameters. In this new version, we apply several known technical modifications to the Transformer architecture, such as interleaving local-global attentions (Beltagy et al., 2020a) and group-query attention (Ainslie et al., 2023). We al… ▽ More In this work, we introduce Gemma 2, a new addition to the Gemma family of lightweight, state-of-the-art open models, ranging in scale from 2 billion to 27 billion parameters. In this new version, we apply several known technical modifications to the Transformer architecture, such as interleaving local-global attentions (Beltagy et al., 2020a) and group-query attention (Ainslie et al., 2023). We also train the 2B and 9B models with knowledge distillation (Hinton et al., 2015) instead of next token prediction. The resulting models deliver the best performance for their size, and even offer competitive alternatives to models that are 2-3 times bigger. We release all our models to the community. △ Less

Submitted 2 October, 2024; v1 submitted 31 July, 2024; originally announced August 2024.

Optimizing Speculative Decoding for Serving Large Language Models Using Goodput

Authors: Xiaoxuan Liu, Cade Daniel, Langxiang Hu, Woosuk Kwon, Zhuohan Li, Xiangxi Mo, Alvin Cheung, Zhijie Deng, Ion Stoica, Hao Zhang

Abstract: Reducing the inference latency of large language models (LLMs) is crucial, and speculative decoding (SD) stands out as one of the most effective techniques. Rather than letting the LLM generate all tokens directly, speculative decoding employs effective proxies to predict potential outputs, which are then verified by the LLM without compromising the generation quality. Yet, deploying SD in real on… ▽ More Reducing the inference latency of large language models (LLMs) is crucial, and speculative decoding (SD) stands out as one of the most effective techniques. Rather than letting the LLM generate all tokens directly, speculative decoding employs effective proxies to predict potential outputs, which are then verified by the LLM without compromising the generation quality. Yet, deploying SD in real online LLM serving systems (with continuous batching) does not always yield improvement -- under higher request rates or low speculation accuracy, it paradoxically increases latency. Furthermore, there is no best speculation length work for all workloads under different system loads. Based on the observations, we develop a dynamic framework SmartSpec. SmartSpec dynamically determines the best speculation length for each request (from 0, i.e., no speculation, to many tokens) -- hence the associated speculative execution costs -- based on a new metric called goodput, which characterizes the current observed load of the entire system and the speculation accuracy. We show that SmartSpec consistently reduces average request latency by up to 3.2x compared to non-speculative decoding baselines across different sizes of target models, draft models, request rates, and datasets. Moreover, SmartSpec can be applied to different styles of speculative decoding, including traditional, model-based approaches as well as model-free methods like prompt lookup and tree-style decoding. △ Less

Submitted 25 June, 2024; v1 submitted 20 June, 2024; originally announced June 2024.

Efficient Memory Management for Large Language Model Serving with PagedAttention

Authors: Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, Ion Stoica

Abstract: High throughput serving of large language models (LLMs) requires batching sufficiently many requests at a time. However, existing systems struggle because the key-value cache (KV cache) memory for each request is huge and grows and shrinks dynamically. When managed inefficiently, this memory can be significantly wasted by fragmentation and redundant duplication, limiting the batch size. To address… ▽ More High throughput serving of large language models (LLMs) requires batching sufficiently many requests at a time. However, existing systems struggle because the key-value cache (KV cache) memory for each request is huge and grows and shrinks dynamically. When managed inefficiently, this memory can be significantly wasted by fragmentation and redundant duplication, limiting the batch size. To address this problem, we propose PagedAttention, an attention algorithm inspired by the classical virtual memory and paging techniques in operating systems. On top of it, we build vLLM, an LLM serving system that achieves (1) near-zero waste in KV cache memory and (2) flexible sharing of KV cache within and across requests to further reduce memory usage. Our evaluations show that vLLM improves the throughput of popular LLMs by 2-4$\times$ with the same level of latency compared to the state-of-the-art systems, such as FasterTransformer and Orca. The improvement is more pronounced with longer sequences, larger models, and more complex decoding algorithms. vLLM's source code is publicly available at https://github.com/vllm-project/vllm △ Less

Submitted 12 September, 2023; originally announced September 2023.

Comments: SOSP 2023

Real Robot Challenge 2022: Learning Dexterous Manipulation from Offline Data in the Real World

Authors: Nico Gürtler, Felix Widmaier, Cansu Sancaktar, Sebastian Blaes, Pavel Kolev, Stefan Bauer, Manuel Wüthrich, Markus Wulfmeier, Martin Riedmiller, Arthur Allshire, Qiang Wang, Robert McCarthy, Hangyeol Kim, Jongchan Baek, Wookyong Kwon, Shanliang Qian, Yasunori Toshimitsu, Mike Yan Michelis, Amirhossein Kazemipour, Arman Raayatsanati, Hehui Zheng, Barnabas Gavin Cangan, Bernhard Schölkopf, Georg Martius

Abstract: Experimentation on real robots is demanding in terms of time and costs. For this reason, a large part of the reinforcement learning (RL) community uses simulators to develop and benchmark algorithms. However, insights gained in simulation do not necessarily translate to real robots, in particular for tasks involving complex interactions with the environment. The Real Robot Challenge 2022 therefore… ▽ More Experimentation on real robots is demanding in terms of time and costs. For this reason, a large part of the reinforcement learning (RL) community uses simulators to develop and benchmark algorithms. However, insights gained in simulation do not necessarily translate to real robots, in particular for tasks involving complex interactions with the environment. The Real Robot Challenge 2022 therefore served as a bridge between the RL and robotics communities by allowing participants to experiment remotely with a real robot - as easily as in simulation. In the last years, offline reinforcement learning has matured into a promising paradigm for learning from pre-collected datasets, alleviating the reliance on expensive online interactions. We therefore asked the participants to learn two dexterous manipulation tasks involving pushing, grasping, and in-hand orientation from provided real-robot datasets. An extensive software documentation and an initial stage based on a simulation of the real set-up made the competition particularly accessible. By giving each team plenty of access budget to evaluate their offline-learned policies on a cluster of seven identical real TriFinger platforms, we organized an exciting competition for machine learners and roboticists alike. In this work we state the rules of the competition, present the methods used by the winning teams and compare their results with a benchmark of state-of-the-art offline RL algorithms on the challenge datasets. △ Less

Submitted 24 November, 2023; v1 submitted 15 August, 2023; originally announced August 2023.

Comments: Typo in author list fixed

Deep Learning Generates Synthetic Cancer Histology for Explainability and Education

Authors: James M. Dolezal, Rachelle Wolk, Hanna M. Hieromnimon, Frederick M. Howard, Andrew Srisuwananukorn, Dmitry Karpeyev, Siddhi Ramesh, Sara Kochanny, Jung Woo Kwon, Meghana Agni, Richard C. Simon, Chandni Desai, Raghad Kherallah, Tung D. Nguyen, Jefree J. Schulte, Kimberly Cole, Galina Khramtsova, Marina Chiara Garassino, Aliya N. Husain, Huihua Li, Robert Grossman, Nicole A. Cipriani, Alexander T. Pearson

Abstract: Artificial intelligence methods including deep neural networks (DNN) can provide rapid molecular classification of tumors from routine histology with accuracy that matches or exceeds human pathologists. Discerning how neural networks make their predictions remains a significant challenge, but explainability tools help provide insights into what models have learned when corresponding histologic fea… ▽ More Artificial intelligence methods including deep neural networks (DNN) can provide rapid molecular classification of tumors from routine histology with accuracy that matches or exceeds human pathologists. Discerning how neural networks make their predictions remains a significant challenge, but explainability tools help provide insights into what models have learned when corresponding histologic features are poorly defined. Here, we present a method for improving explainability of DNN models using synthetic histology generated by a conditional generative adversarial network (cGAN). We show that cGANs generate high-quality synthetic histology images that can be leveraged for explaining DNN models trained to classify molecularly-subtyped tumors, exposing histologic features associated with molecular state. Fine-tuning synthetic histology through class and layer blending illustrates nuanced morphologic differences between tumor subtypes. Finally, we demonstrate the use of synthetic histology for augmenting pathologist-in-training education, showing that these intuitive visualizations can reinforce and improve understanding of histologic manifestations of tumor biology. △ Less

Submitted 9 December, 2022; v1 submitted 11 November, 2022; originally announced November 2022.

A Fast Post-Training Pruning Framework for Transformers

Authors: Woosuk Kwon, Sehoon Kim, Michael W. Mahoney, Joseph Hassoun, Kurt Keutzer, Amir Gholami

Abstract: Pruning is an effective way to reduce the huge inference cost of Transformer models. However, prior work on pruning Transformers requires retraining the models. This can add high training cost and high complexity to model deployment, making it difficult to use in many practical situations. To address this, we propose a fast post-training pruning framework for Transformers that does not require any… ▽ More Pruning is an effective way to reduce the huge inference cost of Transformer models. However, prior work on pruning Transformers requires retraining the models. This can add high training cost and high complexity to model deployment, making it difficult to use in many practical situations. To address this, we propose a fast post-training pruning framework for Transformers that does not require any retraining. Given a resource constraint and a sample dataset, our framework automatically prunes the Transformer model using structured sparsity methods. To retain high accuracy without retraining, we introduce three novel techniques: (i) a lightweight mask search algorithm that finds which heads and filters to prune based on the Fisher information; (ii) mask rearrangement that complements the search algorithm; and (iii) mask tuning that reconstructs the output activations for each layer. We apply our method to BERT-base and DistilBERT, and we evaluate its effectiveness on GLUE and SQuAD benchmarks. Our framework achieves up to 2.0x reduction in FLOPs and 1.56x speedup in inference latency, while maintaining < 1% loss in accuracy. Importantly, our framework prunes Transformers in less than 3 minutes on a single GPU, which is over two orders of magnitude faster than existing pruning approaches that retrain the models. △ Less

Submitted 17 October, 2022; v1 submitted 29 March, 2022; originally announced April 2022.

Comments: NeurIPS 2022

Learned Token Pruning for Transformers

Authors: Sehoon Kim, Sheng Shen, David Thorsley, Amir Gholami, Woosuk Kwon, Joseph Hassoun, Kurt Keutzer

Abstract: Deploying transformer models in practice is challenging due to their inference cost, which scales quadratically with input sequence length. To address this, we present a novel Learned Token Pruning (LTP) method which adaptively removes unimportant tokens as an input sequence passes through transformer layers. In particular, LTP prunes tokens with an attention score below a threshold value which is… ▽ More Deploying transformer models in practice is challenging due to their inference cost, which scales quadratically with input sequence length. To address this, we present a novel Learned Token Pruning (LTP) method which adaptively removes unimportant tokens as an input sequence passes through transformer layers. In particular, LTP prunes tokens with an attention score below a threshold value which is learned for each layer during training. Our threshold-based method allows the length of the pruned sequence to vary adaptively based on the input sequence, and avoids algorithmically expensive operations such as top-k token selection. We extensively test the performance of LTP on GLUE tasks and show that our method outperforms the prior state-of-the-art token pruning methods by up to ~2.5% higher accuracy with the same amount of FLOPs. In particular, LTP achieves up to 2.1x FLOPs reduction with less than 1% accuracy drop, which results in up to 1.9x and 2.0x throughput improvement on Intel Haswell CPUs and NVIDIA V100 GPUs, respectively. Furthermore, we demonstrate that LTP is more robust than prior methods to variations on input sentence lengths. Our code has been developed in PyTorch and has been open-sourced. △ Less

Submitted 2 June, 2022; v1 submitted 2 July, 2021; originally announced July 2021.

Comments: KDD 2022 (Research Track)

Nimble: Lightweight and Parallel GPU Task Scheduling for Deep Learning

Authors: Woosuk Kwon, Gyeong-In Yu, Eunji Jeong, Byung-Gon Chun

Abstract: Deep learning (DL) frameworks take advantage of GPUs to improve the speed of DL inference and training. Ideally, DL frameworks should be able to fully utilize the computation power of GPUs such that the running time depends on the amount of computation assigned to GPUs. Yet, we observe that in scheduling GPU tasks, existing DL frameworks suffer from inefficiencies such as large scheduling overhead… ▽ More Deep learning (DL) frameworks take advantage of GPUs to improve the speed of DL inference and training. Ideally, DL frameworks should be able to fully utilize the computation power of GPUs such that the running time depends on the amount of computation assigned to GPUs. Yet, we observe that in scheduling GPU tasks, existing DL frameworks suffer from inefficiencies such as large scheduling overhead and unnecessary serial execution. To this end, we propose Nimble, a DL execution engine that runs GPU tasks in parallel with minimal scheduling overhead. Nimble introduces a novel technique called ahead-of-time (AoT) scheduling. Here, the scheduling procedure finishes before executing the GPU kernel, thereby removing most of the scheduling overhead during run time. Furthermore, Nimble automatically parallelizes the execution of GPU tasks by exploiting multiple GPU streams in a single GPU. Evaluation on a variety of neural networks shows that compared to PyTorch, Nimble speeds up inference and training by up to 22.34$\times$ and 3.61$\times$, respectively. Moreover, Nimble outperforms state-of-the-art inference systems, TensorRT and TVM, by up to 2.81$\times$ and 1.70$\times$, respectively. △ Less

Submitted 4 December, 2020; originally announced December 2020.

Comments: In NeurIPS 2020

Selective Distillation of Weakly Annotated GTD for Vision-based Slab Identification System

Authors: Sang Jun Lee, Sang Woo Kim, Wookyong Kwon, Gyogwon Koo, Jong Pil Yun

Abstract: This paper proposes an algorithm for recognizing slab identification numbers in factory scenes. In the development of a deep-learning based system, manual labeling to make ground truth data (GTD) is an important but expensive task. Furthermore, the quality of GTD is closely related to the performance of a supervised learning algorithm. To reduce manual work in the labeling process, we generated we… ▽ More This paper proposes an algorithm for recognizing slab identification numbers in factory scenes. In the development of a deep-learning based system, manual labeling to make ground truth data (GTD) is an important but expensive task. Furthermore, the quality of GTD is closely related to the performance of a supervised learning algorithm. To reduce manual work in the labeling process, we generated weakly annotated GTD by marking only character centroids. Whereas bounding-boxes for characters require at least a drag-and-drop operation or two clicks to annotate a character location, the weakly annotated GTD requires a single click to record a character location. The main contribution of this paper is on selective distillation to improve the quality of the weakly annotated GTD. Because manual GTD are usually generated by many people, it may contain personal bias or human error. To address this problem, the information in manual GTD is integrated and refined by selective distillation. In the process of selective distillation, a fully convolutional network is trained using the weakly annotated GTD, and its prediction maps are selectively used to revise locations and boundaries of semantic regions of characters in the initial GTD. The modified GTD are used in the main training stage, and a post-processing is conducted to retrieve text information. Experiments were thoroughly conducted on actual industry data collected at a steelmaking factory to demonstrate the effectiveness of the proposed method. △ Less

Submitted 13 December, 2018; v1 submitted 9 October, 2018; originally announced October 2018.

Comments: 10 pages, 12 figures, submitted to a journal

Journal ref: IEEE Access 7 (2019) 23177-23186