arXiv subjects

Gyeongsik Yang

Publications and source records attributed to Gyeongsik Yang.

10 recordsLinked to original sources

arXiv2026

Xronos: Heterogeneity-Aware Tensor Parallelism for Collaborative LLM Fine-Tuning on Edge CPUs

Collaborative fine-tuning on edge devices adapts large language models to domain-specific data while keeping each device's data local. State-of-the-art (SOTA) collaborative fine-tuning techniques are largely designed for GPU-based edge devices and rely on pipeline parallelism (PP). However, many edge platforms, including IoT gateways, smart-home hubs, and in-vehicle computers, are primarily CPU-based. This paper reports that PP is ineffective on CPU-based edge devices because the same CPU handles both model computation and communication, which causes severe CPU contention. Our analysis shows that this leads to 5.75$\times$ higher computation stall ratios than on GPU devices on average. Tensor parallelism (TP) can alleviate this contention by separating computation and communication, but existing TP techniques assume homogeneous devices. On heterogeneous CPU edge devices, we find that this assumption causes faster workers to remain idle for up to 34% while waiting for slower devices at synchronization points. To address the limitations, we propose Xronos, a collaborative fine-tuning framework for heterogeneous CPU edge devices. Xronos uses TP as its execution backbone and combines lightweight profiling with heterogeneity-aware tensor partitioning to reduce the straggler bottleneck. Across diverse devices, models, and benchmark tasks, Xronos reduces fine-tuning time by 18% (TP) to 56% (PP) and the ratio of device idle time by $\sim$5.9$\times$ over SOTA techniques, while maintaining the accuracy.

cs.DC↗

arXiv2026

Not All AI Agents Are Equal: Characterizing Resource and Performance Dynamics

LLM-based AI agents process user requests through iterative reasoning and tool execution, often involving the invocation of remote LLM APIs with local tool containers. This execution model can make the optimization of agent serving difficult because latency, local resource demand, and container bottlenecks inter-mix across requests. However, the current agent ecosystem runs without much consideration of resource dynamics, which results in significant waste of the precious resources. This paper analyzes the resource inter-mix of AI agents for three representative tasks: retrieval-augmented question answering, web search, and software coding. To this end, we characterize the latency with respect to the resource dynamics of processing multiple requests and tasks concurrently. Our measurements show that agents have a wide range of behaviors depending on tasks, so that even the same tool can differ substantially in resource dynamics. We also find that running multiple requests concurrently exposes task-dependent bottlenecks in resource dynamics such as CPU, disk I/O, and memory. Furthermore, we uncover that faster LLM responses or more CPU cores do not always accelerate agents. Based on these observations, we demonstrate new optimization opportunities that exploit the resource dynamics of tasks: CPU-aware tool admission and task-aware CPU allocation. Our results show that the latency of CPU-sensitive agent tasks improves $\sim$5.4$\times$, and the average latency across multiple tasks is reduced $\sim$32% compared to native agents.

cs.AI↗

arXiv2026

GeoMesh: Workload-Balanced and Sign-Compressed Geo-Distributed LLM Training

Large language models are increasingly trained on GPUs distributed across multiple regions, but geo-distributed training is challenging in practice. Real clusters often contain GPUs with different speeds and memory capacities, and they communicate over slow wide-area networks. Our analysis shows that this creates serious problems: existing synchronous methods preserve stable updates, but fast GPUs wait up to 20.9% of their runtime for slower ones, and all workers spend, on average, 65.8% of their runtime on synchronization. Recent asynchronous methods reduce waiting time but worsen the model accuracy due to stale updates. To address the problems, we present GeoMesh, a synchronous geo-distributed training framework for heterogeneous GPUs. GeoMesh balances per-worker workloads by assigning each GPU a suitable batch size and number of inner steps, so faster GPUs do more useful work instead of waiting. It also reduces communication volume by nearly 32x by exchanging compressed sign-based pseudo-gradients with lightweight magnitude and token count. Across heterogeneous GPUs and Azure-derived WAN, GeoMesh reduces time-to-target perplexity by up to 70.2% over representative baselines and lowers straggler- and WAN-induced GPU idle by up to 8.0x and 5.6x, respectively, while preserving comparable zero-shot accuracy.

cs.DC↗

arXiv2026

Parallelism Strategy Chaining for Fast Training Convergence

Selecting a parallelism strategy - the configuration of data, tensor, and pipeline parallelism degrees together with micro- and global-batch sizes - largely determines the training efficiency of large language models. State-of-the-art methods search for a parallelism strategy offline and select the single strategy that minimizes per-iteration time. But we find that they neglect the target validation perplexity and time-to-perplexity (TTP). In particular, our analysis reveals that the best strategy yielding the fastest perplexity improvement changes multiple times during training. As a result, state-of-the-art methods are 1.8-11.4x slower in TTP than the strategy sequence that selects the best strategy at each iteration. This paper proposes CONA, a new training method that introduces online strategy chaining. Instead of a single strategy selected offline, CONA ranks candidate strategies during training using a surrogate metric built from compute throughput and gradient statistics, and switches the current strategy to a new strategy with a higher metric. In our evaluation with GPT-3 1.3B, BERT-Large, and Llama-3.2-1B, CONA reaches the target validation perplexity 1.4-9.6x faster than state-of-the-art methods. Moreover, CONA closely tracks the perplexity achieved by the sequence that selects the best strategy at each iteration, within 2.6%.

cs.LG↗

arXiv2026

Enabling KV Caching of Shared Prefix for Diffusion Language Models

Key-value (KV) caching for shared prefixes is essential for high-throughput large language model (LLM) serving, but it faces critical challenges in emerging diffusion language models (DLMs). In DLMs, bidirectional attention means that updating any token dynamically alters the entire context and its corresponding KVs. Thus, existing caching techniques developed for LLMs, which assume that KVs remain invariant once computed, corrupt the shared prefix KVs. Our experiments show that applying these techniques to DLMs causes model accuracy to collapse to near zero. To unlock high-throughput DLM serving, we propose bidirectional prefix caching, BiCache, the first KV caching technique for shared prefixes in DLMs. BiCache is designed based on key observations from our comprehensive analysis: shared prefix KVs remain stable and reusable in shallow layers, while the depth of shallow layers depends on the fraction of shared prefix tokens in each request. Thus, BiCache dynamically identifies a safe layer depth for reusing shared prefix KVs and eliminates redundant computation. Evaluations demonstrate that BiCache significantly improves serving throughput by 36.3%-98.3% compared to existing techniques without accuracy collapse (only 0-1.8% difference).

cs.LG↗

arXiv2026

Residual Sparsification via Output Importance for Compressing Mixture-of-Experts LLMs

Mixture-of-experts (MoE) architectures scale large language models efficiently, but they demand massive GPU memory. To cope with such demand, models are commonly compressed to reduce their memory footprint. Residual sparsification is a representative compression technique that decomposes each projection matrix of an expert into a shared base matrix and per-expert residual matrix, and then compresses the residuals. Existing sparsification methods compress each residual matrix independently by minimizing its compression error, thereby minimizing the error of each projection matrix. However, our analysis shows that this objective is misaligned with preserving model accuracy after compression. In an expert, the final output is produced through computations coupled across multiple projections and hidden representations. Therefore, even small errors in individual matrices can propagate through hidden representations and projection interactions, leading to large expert output errors and accuracy degradation. To address this misalignment, we propose PARSER, a new residual sparsification method that shifts the compression objective from minimizing isolated matrix errors to preserving the expert output error. PARSER achieves this by introducing output importance, which measures the actual contribution to the expert output error. Our experiments show that, compared with existing methods, PARSER narrows the accuracy gap to the uncompressed model by 1.41$\times$ on Qwen and 1.44$\times$ on DeepSeek, while achieving the same peak memory reduction. Our code is available at https://github.com/OSSS-KU/PARSER.

cs.AI↗

arXiv2026

Unified KV Pooling to Accelerate Long-Context LLM Serving

Long-context LLM serving requires offloading KV caches to host-memory and SSDs, but existing mechanisms are not designed for such long contexts. We observe significant inefficiencies in current KV caching in long contexts: high serving latency ~30.7 s, exceeding the typical TTFT requirement of 10 s by more than 3x. Our in-depth analysis explains two major reasons: (1) retrieval is serialized through host-memory and SSD, leaving other host-memory modules and SSDs underutilized, and (2) SSD-based KV retrieval spends 84% of its time in the kernel filesystem rather than actual device access. To address the problems, we propose unified KV pooling, which aggregates multiple host-memory modules and SSDs into a single logical pool and distributes KV caches across devices based on their bandwidth. To eliminate the filesystem overhead, we design KV-passthrough, which bypasses the kernel filesystem and directly accesses SSD-resident KV caches from user space via SPDK. Across evaluations on LLaMA 3.1-8B, GPT-OSS-20B, and Qwen3-30B-A3B, unified KV pooling reduces TTFT in long-contexts ~4.1x over state-of-the-art techniques, all making under 10 s. It also reduces blocked I/O time by up to 23.2x by eliminating filesystem overhead.

cs.AR↗

arXiv2026

Training Time Prediction for Mixed Precision-based Distributed Training

Accurate prediction of training time in distributed deep learning is crucial for resource allocation, cost estimation, and job scheduling. We observe that the floating-point precision setting is a key determinant of training time, leading to training time variations of ~2.4x over its minimum. However, existing studies on distributed training time prediction rely on static model computation graphs that do not capture precision variations, including mixed precision. According to our experiments, training time prediction without considering precision results in significant prediction errors - reaching up to 147.85% in mean absolute percentage error (MAPE). To address this issue, we propose a precision-aware distributed training time predictor that achieves robust accuracy across diverse precision settings, including mixed precision, with 9.8% MAPE.

cs.LG↗

arXiv2025

GPU Memory Prediction for Multimodal Model Training

As deep learning models in agentic AI systems grow in scale and complexity, GPU memory requirements increase and often exceed the available GPU memory capacity, so that out-of-memory (OoM) errors occur. It is well known that OoM interrupts the whole training itself and wastes substantial computational resources. Therefore, to prevent OoM, accurate prediction of GPU memory usage is essential. However, previous studies focus only on unimodal architectures and fail to generalize to multimodal models, even though the multimodal models are a common choice in agentic AI systems. To address this limitation, we propose a framework that predicts the peak GPU memory usage by analyzing the model architecture and training behavior of multimodal models. Specifically, the framework decomposes the multimodal model into its constituent layers and applies factorization to estimate the memory usage of each layer. Our evaluation shows that our framework achieves high prediction accuracy of ~8.7% average MAPE.

cs.LG↗

arXiv2025

Prediction of Permissioned Blockchain Performance for Resource Scaling Configurations

Blockchain is increasingly offered as blockchain-as-a-service (BaaS) by cloud service providers. However, configuring BaaS appropriately for optimal performance and reliability resorts to try-and-error. A key challenge is that BaaS is often perceived as a ``black-box,'' leading to uncertainties in performance and resource provisioning. Previous studies attempted to address this challenge; however, the impacts of both vertical and horizontal scaling remain elusive. To this end, we present machine learning-based models to predict network reliability and throughput based on scaling configurations. In our evaluation, the models exhibit prediction errors of ~1.9%, which is highly accurate and can be applied in the real-world.

cs.DC↗