arXiv subjects

Young Geun Kim

Publications and source records attributed to Young Geun Kim.

13 recordsLinked to original sources

arXiv2026

Text Capability Loss in Vision-Language Adaptation: An Attention-Sink Diagnosis

Fine-tuning a pretrained LLM into a vision-language model (VLM) can erode the backbone's text capability, with the damage concentrated on tasks that require following exact output rules, such as instruction following, chain-of-thought reasoning graded on a strictly parsed final answer, and similar evaluations with strict graders. We trace this gap to attention-sink corruption: VL fine-tuning perturbs the early sink position that anchors a large fraction of attention probability, and how well the base LLM preserves its sink tracks how much of the affected capability survives adaptation. Building on this view, we introduce Sink Strength, a single scalar computed on the base LLM in a few seconds on a single GPU that predicts post-VL degradation without any VL training. It consistently tracks relative degradation across the six VLM-LLM pairs and multiple format-sensitive tasks. Complementing this diagnostic, we find that post-pretraining QK-RMSNorm injection fails to reproduce the protection of native QK-RMSNorm, while several off-the-shelf weight-merging settings fail to recover the lost capability after VL training. These negative results underscore the value of screening backbones with Sink Strength before VL training and narrow the intervention space toward head-selective training-time protection.

cs.LG↗

arXiv2026

SwiftQK: Fast and Communication-Efficient Tensor Parallelism for Query-Key Normalization

Query-Key Normalization (QK-Norm) improves the training stability and quality of modern Large Language Models (LLMs). However, under Tensor Parallelism (TP), layerwise QK-Norm introduces additional cross-GPU communication because the normalization factor depends on the full hidden vector. We present SwiftQK, a multi-GPU RMSNorm kernel that exchanges only scalar normalization statistics and overlaps the remaining Peer-to-Peer reduction with independent element-wise computation in a deadlock-safe persistent kernel. Evaluations on recent LLMs show that SwiftQK reduces QK-Norm latency by 81.4--93.9% relative to the standard TP QK-Norm using full-vector All-Gather. In end-to-end serving, SwiftQK reduces TPOT on average by 29.5% over the All-Gather-based baseline and by 14.3% over an optimized scalar-aggregation implementation.

cs.LG↗

arXiv2026

Voltron: Enabling Elastic Multi-Device Execution of LLM Inference for Empowered Edge Intelligence

Large language models (LLMs) are widely used in intelligent services due to their remarkable capability in generative tasks. Typically, LLM-based services process the inference requests of the users in a centralized data center. Unfortunately, such centralized execution has limitations for end-users, such as increased response latency with communication overhead and privacy leakage risk. To alleviate the aforementioned limitations, there have been increasing pushes to execute LLM inference locally on user-end devices. However, the limited resources of a single edge device impose restrictions on achievable accuracy of LLMs. To overcome the issue, we first propose to leverage multiple user-end devices available at the edge for LLM inference, enabling the execution of larger models. Specifically, we propose Voltron, a novel on-device LLM inference framework that elastically utilizes multiple user-end devices for LLM inference execution while adapting to diverse real-world edge environments. In our evaluation, Voltron achieves up to 16.5% higher accuracy than state-of-the-art LLMs that can be executed on a single edge device, satisfying user QoS requirements.

cs.DC↗

arXiv2026

Entropy Meets Importance: A Unified Head Importance-Entropy Score for Stable and Efficient Transformer Pruning

Transformer-based models have achieved remarkable performance in NLP tasks. However, their structural characteristics-multiple layers and attention heads-introduce efficiency challenges in inference and deployment. To address these challenges, various pruning methods have recently been proposed. Notably, gradient-based methods using Head Importance Scores (HIS) have gained traction for interpretability, efficiency, and ability to identify redundant heads. However, HIS alone has limitations as it captures only the gradient-driven contribution, overlooking the diversity of attention patterns. To overcome these limitations, we introduce a novel pruning criterion, HIES (Head Importance-Entropy Score), which integrates head importance scores with attention entropy, providing complementary evidence on per-head contribution. Empirically, HIES-based pruning yields up to 15.2% improvement in model quality and 2.04x improvement in stability over HIS-only methods, enabling substantial model compression without sacrificing either accuracy or stability. Code will be released upon publication.

cs.CL↗

arXiv2025

CorGi: Contribution-Guided Block-Wise Interval Caching for Training-Free Acceleration of Diffusion Transformers

Diffusion transformer (DiT) achieves remarkable performance in visual generation, but its iterative denoising process combined with larger capacity leads to a high inference cost. Recent works have demonstrated that the iterative denoising process of DiT models involves substantial redundant computation across steps. To effectively reduce the redundant computation in DiT, we propose CorGi (Contribution-Guided Block-Wise Interval Caching), training-free DiT inference acceleration framework that selectively reuses the outputs of transformer blocks in DiT across denoising steps. CorGi caches low-contribution blocks and reuses them in later steps within each interval to reduce redundant computation while preserving generation quality. For text-to-image tasks, we further propose CorGi+, which leverages per-block cross-attention maps to identify salient tokens and applies partial attention updates to protect important object details. Evaluation on the state-of-the-art DiT models demonstrates that CorGi and CorGi+ achieve up to 2.0x speedup on average, while preserving high generation quality.

cs.CV↗

arXiv2025

First Attentions Last: Better Exploiting First Attentions for Efficient Transformer Training

As training billion-scale transformers becomes increasingly common, employing multiple distributed GPUs along with parallel training methods has become a standard practice. However, existing transformer designs suffer from significant communication overhead, especially in Tensor Parallelism (TP), where each block's MHA-MLP connection requires an all-reduce communication. Through our investigation, we show that the MHA-MLP connections can be bypassed for efficiency, while the attention output of the first layer can serve as an alternative signal for the bypassed connection. Motivated by the observations, we propose FAL (First Attentions Last), an efficient transformer architecture that redirects the first MHA output to the MLP inputs of the following layers, eliminating the per-block MHA-MLP connections. This removes the all-reduce communication and enables parallel execution of MHA and MLP on a single GPU. We also introduce FAL+, which adds the normalized first attention output to the MHA outputs of the following layers to augment the MLP input for the model quality. Our evaluation shows that FAL reduces multi-GPU training time by up to 44%, improves single-GPU throughput by up to 1.18x, and achieves better perplexity compared to the baseline GPT. FAL+ achieves even lower perplexity without increasing the training time than the baseline. Codes are available at: https://github.com/CASL-KU/FAL

cs.LG↗

arXiv2025

Not All Adapters Matter: Selective Adapter Freezing for Memory-Efficient Fine-Tuning of Language Models

Transformer-based large-scale pre-trained models achieve great success. Fine-tuning is the standard practice for leveraging these models in downstream tasks. Among the fine-tuning methods, adapter-tuning provides a parameter-efficient fine-tuning by introducing lightweight trainable modules while keeping most pre-trained parameters frozen. However, existing adapter-tuning methods still impose substantial resource usage. Through our investigation, we show that each adapter unequally contributes to both task performance and resource usage. Motivated by this insight, we propose Selective Adapter FrEezing (SAFE), which gradually freezes less important adapters early to reduce unnecessary resource usage while maintaining performance. In our experiments, SAFE reduces memory usage, computation amount, and training time by 42.85\%, 34.59\%, and 11.82\%, respectively, while achieving comparable or better task performance compared to the baseline. We also demonstrate that SAFE induces regularization effect, thereby smoothing the loss landscape, which enables the model to generalize better by avoiding sharp minima.

cs.CL↗

arXiv2024

HeteroSwitch: Characterizing and Taming System-Induced Data Heterogeneity in Federated Learning

Federated Learning (FL) is a practical approach to train deep learning models collaboratively across user-end devices, protecting user privacy by retaining raw data on-device. In FL, participating user-end devices are highly fragmented in terms of hardware and software configurations. Such fragmentation introduces a new type of data heterogeneity in FL, namely \textit{system-induced data heterogeneity}, as each device generates distinct data depending on its hardware and software configurations. In this paper, we first characterize the impact of system-induced data heterogeneity on FL model performance. We collect a dataset using heterogeneous devices with variations across vendors and performance tiers. By using this dataset, we demonstrate that \textit{system-induced data heterogeneity} negatively impacts accuracy, and deteriorates fairness and domain generalization problems in FL. To address these challenges, we propose HeteroSwitch, which adaptively adopts generalization techniques (i.e., ISP transformation and SWAD) depending on the level of bias caused by varying HW and SW configurations. In our evaluation with a realistic FL dataset (FLAIR), HeteroSwitch reduces the variance of averaged precision by 6.3\% across device types.

cs.LG↗

arXiv2023

GreenScale: Carbon-Aware Systems for Edge Computing

To improve the environmental implications of the growing demand of computing, future applications need to improve the carbon-efficiency of computing infrastructures. State-of-the-art approaches, however, do not consider the intermittent nature of renewable energy. The time and location-based carbon intensity of energy fueling computing has been ignored when determining how computation is carried out. This poses a new challenge -- deciding when and where to run applications across consumer devices at the edge and servers in the cloud. Such scheduling decisions become more complicated with the stochastic runtime variance and the amortization of the rising embodied emissions. This work proposes GreenScale, a framework to understand the design and optimization space of carbon-aware scheduling for green applications across the edge-cloud infrastructure. Based on the quantified carbon output of the infrastructure components, we demonstrate that optimizing for carbon, compared to performance and energy efficiency, yields unique scheduling solutions. Our evaluation with three representative categories of applications (i.e., AI, Game, and AR/VR) demonstrate that the carbon emissions of the applications can be reduced by up to 29.1% with the GreenScale. The analysis in this work further provides a detailed road map for edge-cloud application developers to build green applications.

cs.DC↗

arXiv2022

FedGPO: Heterogeneity-Aware Global Parameter Optimization for Efficient Federated Learning

Federated learning (FL) has emerged as a solution to deal with the risk of privacy leaks in machine learning training. This approach allows a variety of mobile devices to collaboratively train a machine learning model without sharing the raw on-device training data with the cloud. However, efficient edge deployment of FL is challenging because of the system/data heterogeneity and runtime variance. This paper optimizes the energy-efficiency of FL use cases while guaranteeing model convergence, by accounting for the aforementioned challenges. We propose FedGPO based on a reinforcement learning, which learns how to identify optimal global parameters (B, E, K) for each FL aggregation round adapting to the system/data heterogeneity and stochastic runtime variance. In our experiments, FedGPO improves the model convergence time by 2.4 times, and achieves 3.6 times higher energy efficiency over the baseline settings, respectively.

cs.LG↗

arXiv2021

AutoFL: Enabling Heterogeneity-Aware Energy Efficient Federated Learning

Federated learning enables a cluster of decentralized mobile devices at the edge to collaboratively train a shared machine learning model, while keeping all the raw training samples on device. This decentralized training approach is demonstrated as a practical solution to mitigate the risk of privacy leakage. However, enabling efficient FL deployment at the edge is challenging because of non-IID training data distribution, wide system heterogeneity and stochastic-varying runtime effects in the field. This paper jointly optimizes time-to-convergence and energy efficiency of state-of-the-art FL use cases by taking into account the stochastic nature of edge execution. We propose AutoFL by tailor-designing a reinforcement learning algorithm that learns and determines which K participant devices and per-device execution targets for each FL model aggregation round in the presence of stochastic runtime variance, system and data heterogeneity. By considering the unique characteristics of FL edge deployment judiciously, AutoFL achieves 3.6 times faster model convergence time and 4.7 and 5.2 times higher energy efficiency for local clients and globally over the cluster of K participants, respectively.

cs.LG↗

arXiv2020

Chasing Carbon: The Elusive Environmental Footprint of Computing

Given recent algorithm, software, and hardware innovation, computing has enabled a plethora of new applications. As computing becomes increasingly ubiquitous, however, so does its environmental impact. This paper brings the issue to the attention of computer-systems researchers. Our analysis, built on industry-reported characterization, quantifies the environmental effects of computing in terms of carbon emissions. Broadly, carbon emissions have two sources: operational energy consumption, and hardware manufacturing and infrastructure. Although carbon emissions from the former are decreasing thanks to algorithmic, software, and hardware innovations that boost performance and power efficiency, the overall carbon footprint of computer systems continues to grow. This work quantifies the carbon output of computer systems to show that most emissions related to modern mobile and data-center equipment come from hardware manufacturing and infrastructure. We therefore outline future directions for minimizing the environmental impact of computing systems.

cs.AR↗

arXiv2020

AutoScale: Optimizing Energy Efficiency of End-to-End Edge Inference under Stochastic Variance

Deep learning inference is increasingly run at the edge. As the programming and system stack support becomes mature, it enables acceleration opportunities within a mobile system, where the system performance envelope is scaled up with a plethora of programmable co-processors. Thus, intelligent services designed for mobile users can choose between running inference on the CPU or any of the co-processors on the mobile system, or exploiting connected systems, such as the cloud or a nearby, locally connected system. By doing so, the services can scale out the performance and increase the energy efficiency of edge mobile systems. This gives rise to a new challenge - deciding when inference should run where. Such execution scaling decision becomes more complicated with the stochastic nature of mobile-cloud execution, where signal strength variations of the wireless networks and resource interference can significantly affect real-time inference performance and system energy efficiency. To enable accurate, energy-efficient deep learning inference at the edge, this paper proposes AutoScale. AutoScale is an adaptive and light-weight execution scaling engine built upon the custom-designed reinforcement learning algorithm. It continuously learns and selects the most energy-efficient inference execution target by taking into account characteristics of neural networks and available systems in the collaborative cloud-edge execution environment while adapting to the stochastic runtime variance. Real system implementation and evaluation, considering realistic execution scenarios, demonstrate an average of 9.8 and 1.6 times energy efficiency improvement for DNN edge inference over the baseline mobile CPU and cloud offloading, while meeting the real-time performance and accuracy requirement.

cs.LG↗