arXiv subjects

Janghyeon Kim

Publications and source records attributed to Janghyeon Kim.

4 recordsLinked to original sources

arXiv2026

BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference

Large Reasoning Models (LRMs) achieve superior problem-solving through extended Chain-of-Thought (CoT) generation, but the resulting key-value (KV) cache grows linearly with sequence length and creates severe memory bottlenecks, often exceeding GPU capacity for long reasoning traces. Existing KV cache compression methods rely on recent queries to estimate future token importance, implicitly assuming these serve as reliable proxies for future attention patterns. We demonstrate that this assumption fails in long-horizon reasoning: certain decoding steps generate Thought Revisiting Tokens (TRT) that re-attend to distant previous context, such as task-solving plans formulated early in the trace. Through systematic analysis, we discover that queries corresponding to the TRT cluster into a small number of similarity groups in the embedding space. Based on this insight, we propose BeaconKV, a training-free KV cache compression method that maintains beacon queries, compact representatives for each global query cluster, to anticipate which KV pairs will be revisited without storing the entire query history. Across four open-source LRMs and diverse reasoning benchmarks, BeaconKV generally outperforms existing compression methods, achieving up to $5.8\times$ memory reduction while nearly preserving full cache accuracy and improving throughput by over $4.3\times$.

cs.LG↗

arXiv2025

Scalable Processing-Near-Memory for 1M-Token LLM Inference: CXL-Enabled KV-Cache Management Beyond GPU Limits

The expansion of context windows in large language models (LLMs) to multi-million tokens introduces severe memory and compute bottlenecks, particularly in managing the growing Key-Value (KV) cache. While Compute Express Link (CXL) enables non-eviction frameworks that offload the full KV-cache to scalable external memory, these frameworks still suffer from costly data transfers when recalling non-resident KV tokens to limited GPU memory as context lengths increase. This work proposes scalable Processing-Near-Memory (PNM) for 1M-Token LLM Inference, a CXL-enabled KV-cache management system that coordinates memory and computation beyond GPU limits. Our design offloads token page selection to a PNM accelerator within CXL memory, eliminating costly recalls and enabling larger GPU batch sizes. We further introduce a hybrid parallelization strategy and a steady-token selection mechanism to enhance compute efficiency and scalability. Implemented atop a state-of-the-art CXL-PNM system, our solution delivers consistent performance gains for LLMs with up to 405B parameters and 1M-token contexts. Our PNM-only offloading scheme (PNM-KV) and GPU-PNM hybrid with steady-token execution (PnG-KV) achieve up to 21.9x throughput improvement, up to 60x lower energy per token, and up to 7.3x better total cost efficiency than the baseline, demonstrating that CXL-enabled multi-PNM architectures can serve as a scalable backbone for future long-context LLM inference.

cs.AR↗

arXiv2024

PIMphony: Overcoming Bandwidth and Capacity Inefficiency in PIM-based Long-Context LLM Inference System

The expansion of long-context Large Language Models (LLMs) creates significant memory system challenges. While Processing-in-Memory (PIM) is a promising accelerator, we identify that it suffers from critical inefficiencies when scaled to long contexts: severe channel underutilization, performance-limiting I/O bottlenecks, and massive memory waste from static KV cache management. In this work, we propose PIMphony, a PIM orchestrator that systematically resolves these issues with three co-designed techniques. First, Token-Centric PIM Partitioning (TCP) ensures high channel utilization regardless of batch size. Second, Dynamic PIM Command Scheduling (DCS) mitigates the I/O bottleneck by overlapping data movement and computation. Finally, a Dynamic PIM Access (DPA) controller enables dynamic memory management to eliminate static memory waste. Implemented via an MLIR-based compiler and evaluated on a cycle-accurate simulator, PIMphony significantly improves throughput for long-context LLM inference (up to 72B parameters and 1M context length). Our evaluations show performance boosts of up to 11.3x on PIM-only systems and 8.4x on xPU+PIM systems, enabling more efficient deployment of LLMs in real-world long-context applications.

cs.AR↗

arXiv2022

Human-Robot Interface to Operate Robotic Systems via Muscle Synergy-Based Kinodynamic Information Transfer

When a human performs a given specific task, it has been known that the central nervous system controls modularized muscle group, which is called muscle synergy. For human-robot interface design problem, therefore, the muscle synergy can be utilized to reduce the dimensionality of control signal as well as the complexity of classifying human posture and motion. In this paper, we propose an approach to design a human-robot interface which enables a human operator to transfer a kinodynamic control command to robotic systems. A key feature of the proposed approach is that the muscle synergy and corresponding activation curve are employed to calculate a force generated by a tool at the robot end effector. A test bed for experiments consisted of two armband type surface electromyography sensors, an RGB-d camera, and a Kinova Gen2 robotic manipulator to verify the proposed approach. The result showed that both force and position commands could be successfully transferred to the robotic manipulator via our muscle synergy-based kinodynamic interface.

cs.RO↗