Myeongjae Jeon

Xinyue Ma, Suyeon Jeong, Minjia Zhang et al.

Continual learning (CL) trains NN models incrementally from a continuous stream of tasks. To remember previously learned knowledge, prior studies store old samples over a memory hierarchy and replay them when new tasks arrive. Edge devices that adopt CL to preserve data privacy are typically energy-sensitive and thus require high model accuracy while not compromising energy efficiency, i.e., cost-effectiveness. Our work is the first to explore the design space of hierarchical memory replay-based CL to gain insights into achieving cost-effectiveness on edge devices. We present Miro, a novel system runtime that carefully integrates our insights into the CL framework by enabling it to dynamically configure the CL system based on resource states for the best cost-effectiveness. To reach this goal, Miro also performs online profiling on parameters with clear accuracy-energy trade-offs and adapts to optimal values with low overhead. Extensive evaluations show that Miro significantly outperforms baseline systems we build for comparison, consistently achieving higher cost-effectiveness.

Xinyue Ma, Heelim Hong, Taegeon Um et al.

Serving long-context LLMs is challenging because request lengths and batch composition vary during token generation, causing the memory footprint to fluctuate significantly at runtime. Offloading KV caches to host memory limits effective memory usage, but existing static and predetermined offloading strategies cannot adapt to the rapidly shifting memory demands of long-context serving. This often leads to excessive CPU-to-GPU KV transfers that translate into latency spikes and frequent SLO violations. To address these challenges, we introduce OrbitFlow, a fine-grained and adaptive KV cache management system that meets latency SLOs in long-context LLM serving. OrbitFlow employs a lightweight ILP solver to decide which layers' KV caches to retain on the GPU for each request, within memory capacity constraints. It continuously refines KV placements based on runtime feedback when the active plan becomes suboptimal during token generation. Under heavy load, OrbitFlow invokes a fallback mechanism to temporarily defer in-flight requests with large memory footprints, preserving overall SLO attainment. Our experiments demonstrate that OrbitFlow improves SLO attainment for TPOT and TBT by up to 66% and 48%, respectively, while reducing the 95th percentile latency by 38% and achieving up to 3.3x higher throughput compared to existing offloading methods.

Jeffrey Willette, Heejun Lee, Youngwan Lee et al.

The transformer's context window is vital for tasks such as few-shot learning and conditional generation as it preserves previous tokens for active memory. However, as the context lengths increase, the computational costs grow quadratically, hindering the deployment of large language models (LLMs) in real-world, long sequence scenarios. Although some recent key-value caching (KV Cache) methods offer linear inference complexity, they naively manage the stored context, prematurely evicting tokens and losing valuable information. Moreover, they lack an optimized prefill/prompt stage strategy, resulting in higher latency than even quadratic attention for realistic context sizes. In response, we introduce a novel mechanism that leverages cascading sub-cache buffers to selectively retain the most relevant tokens, enabling the model to maintain longer context histories without increasing the cache size. Our approach outperforms linear caching baselines across key benchmarks, including streaming perplexity, question answering, book summarization, and passkey retrieval, where it retains better retrieval accuracy at 1M tokens after four doublings of the cache size of 65K. Additionally, our method reduces prefill stage latency by a factor of 6.8 when compared to flash attention on 1M tokens. These innovations not only enhance the computational efficiency of LLMs but also pave the way for their effective deployment in resource-constrained environments, enabling large-scale, real-time applications with significantly reduced latency.

Heejun Lee, Geon Park, Youngwan Lee et al.

In modern large language models (LLMs), increasing the context length is crucial for improving comprehension and coherence in long-context, multi-modal, and retrieval-augmented language generation. While many recent transformer models attempt to extend their context length over a million tokens, they remain impractical due to the quadratic time and space complexities. Although recent works on linear and sparse attention mechanisms can achieve this goal, their real-world applicability is often limited by the need to re-train from scratch and significantly worse performance. In response, we propose a novel approach, Hierarchically Pruned Attention (HiP), which reduces the time complexity of the attention mechanism to $O(T \log T)$ and the space complexity to $O(T)$, where $T$ is the sequence length. We notice a pattern in the attention scores of pretrained LLMs where tokens close together tend to have similar scores, which we call ``attention locality''. Based on this observation, we utilize a novel tree-search-like algorithm that estimates the top-$k$ key tokens for a given query on the fly, which is mathematically guaranteed to have better performance than random attention pruning. In addition to improving the time complexity of the attention mechanism, we further optimize GPU memory usage by implementing KV cache offloading, which stores only $O(\log T)$ tokens on the GPU while maintaining similar decoding throughput. Experiments on benchmarks show that HiP, with its training-free nature, significantly reduces both prefill and decoding latencies, as well as memory usage, while maintaining high-quality generation with minimal degradation. HiP enables pretrained LLMs to scale up to millions of tokens on commodity GPUs, potentially unlocking long-context LLM applications previously deemed infeasible.

Sungho Jeon, Xinyue Ma, Kwang In Kim et al.

Recent rehearsal-free continual learning (CL) methods guided by prompts achieve strong performance on vision tasks with non-stationary data but remain resource-intensive, hindering real-world edge deployment. We introduce resource-efficient prompting (REP), which improves the computational and memory efficiency of prompt-based rehearsal-free continual learning methods while minimizing accuracy trade-offs. Our approach employs swift prompt selection to refine input data using a carefully provisioned model and introduces adaptive token merging (AToM) and adaptive layer dropping (ALD) for efficient prompt updates. AToM and ALD selectively skip data and model layers while preserving task-specific features during the learning of new tasks. Extensive experiments on multiple image classification datasets demonstrate REP's superior resource efficiency over state-of-the-art rehearsal-free CL methods.

Soobee Lee, Minindu Weerakoon, Jonghyun Choi et al.

Continual Learning (CL) is an emerging machine learning paradigm that aims to learn from a continuous stream of tasks without forgetting knowledge learned from the previous tasks. To avoid performance decrease caused by forgetting, prior studies exploit episodic memory (EM), which stores a subset of the past observed samples while learning from new non-i.i.d. data. Despite the promising results, since CL is often assumed to execute on mobile or IoT devices, the EM size is bounded by the small hardware memory capacity and makes it infeasible to meet the accuracy requirements for real-world applications. Specifically, all prior CL methods discard samples overflowed from the EM and can never retrieve them back for subsequent training steps, incurring loss of information that would exacerbate catastrophic forgetting. We explore a novel hierarchical EM management strategy to address the forgetting issue. In particular, in mobile and IoT devices, real-time data can be stored not just in high-speed RAMs but in internal storage devices as well, which offer significantly larger capacity than the RAMs. Based on this insight, we propose to exploit the abundant storage to preserve past experiences and alleviate the forgetting by allowing CL to efficiently migrate samples between memory and storage without being interfered by the slow access speed of the storage. We call it Carousel Memory (CarM). As CarM is complementary to existing CL methods, we conduct extensive evaluations of our method with seven popular CL methods and show that CarM significantly improves the accuracy of the methods across different settings by large margins in final average accuracy (up to 28.4%) while retaining the same training efficiency.