Training molecular machine-learning models on ephemeral or memory-constrained accelerator instances can require repeatedly retrieving preprocessed molecular graphs from remote storage. FoldPipe is a lightweight Python orchestration layer for already-sharded PyTorch and PyTorch Geometric data. It retrieves one shard ahead in a background thread while the consumer trains on the current shard, keeping the number of live shard payloads bounded with respect to total dataset size. Asynchronous prefetch and bounded buffering are established systems techniques rather than novel scheduling algorithms. FoldPipe's contribution is a small integration targeted at native .pt molecular shards together with a source-pinned empirical characterization of its operating regime. We evaluate a SchNet energy-and-force workload on MD17 aspirin using 20 paired, order-alternating benchmark passes on a Tesla T4. Each pass processes five pinned shards containing 25,000 structures. FoldPipe records 16.33 s mean I/O-compute overlap, compared with zero by construction for the sequential bounded baseline. Mean pass time is 76.78 s for FoldPipe and 83.37 s for the baseline. However, the geometric mean paired speedup is $1.059\times$ with a 95% bootstrap interval from $0.878\times$ to $1.288\times$. The experiment therefore verifies the overlap mechanism but is inconclusive about a reliable wall-clock speed advantage under the observed public-network variability.
Nobel Dhar, Md Romyull Islam, Xuechen Zhang +4cs.DC cs.LG
Deploying large language models on edge devices is increasingly limited by a widening gap between model size and available memory. Existing approaches such as quantization, smaller models, and offloading can raise the effective memory limit, but they still assume that the model can be compressed or partitioned to fit within some budget. We target the harder model-exceeds-memory setting, in which the model remains larger than resident memory throughout execution and storage becomes an active source of weights on the critical path. We observe that MLP activity during autoregressive decoding has strong temporal locality: approximately 82-85% of active neurons persist from one token to the next. This means that most sparse weights needed for the current token are already resident, and only the newly needed rows must be fetched from storage. We present NeuroPrefetcher, a storage-backed LLM inference system that exploits this property through predictive delta prefetching. After layer 0, a single GPU-resident predictor, occupying 2.86% of base model parameters, predicts sparse activity for all downstream MLP layers in one forward pass. The runtime compares these predictions against resident GPU buffers and issues application-scheduled NVMe reads only for incoming delta rows, replacing reactive operating-system demand paging with explicit, model-aware weight movement. On real unified-memory edge hardware, NeuroPrefetcher achieves 7.9-12.0x speedup over llama.cpp across constrained memory budgets.
Alish Kanani, Layan Badawi, Umit Y. Ograscs.AR cs.AI cs.LG
Mixture-of-Experts (MoE) models are attractive for edge deployment because they provide high model capacity while activating only a small subset of parameters per token, improving compute efficiency. However, MoE inference at the edge is fundamentally limited by memory. Expert parameters are large and often reside in off-chip memory due to capacity, cost, and power constraints, putting expert loading to the critical path. We present APEX: Adaptive Expert Prefetching, a predictive resource management framework that overlaps expert loading with useful computation. APEX introduces a lightweight prefetch router that predicts candidate experts before the attention block to dynamically fetch additional experts using a learned confidence model. This adaptive strategy achieves over 99% overlap accuracy, significantly outperforming fixed top-k prefetching techniques. APEX supports two execution modes: a correctness-preserving mode that guarantees exact routing semantics, and a stall-free mode that eliminates residual stalls by operating on available experts with negligible impact on application accuracy. Across multiple MoE models, the correctness-preserving mode reduces per-token latency by up to 26% and improves energy-delay product (EDP) by up to 41% over state-of-the-art baselines, while the stall-free mode provides additional efficiency gains with negligible impact on application accuracy. These results establish adaptive, confidence-driven expert prefetching as an effective approach for efficient MoE inference on edge systems.
Multi-agent LLM systems have emerged as an important deployment paradigm for AI services, where each user request is decomposed into a sequence of specialized agents. Across these workflows, every agent repeatedly executes a fixed context consisting of system prompts, tool definitions, and few-shot examples, creating substantial opportunities for KV-cache reuse. Existing LLM serving systems, however, manage KV-cache reactively using prefix caching and recency-based replacement, causing reusable agent contexts to be evicted before their next invocation and forcing repeated recomputation. We present CacheScout, an agent-aware KV-cache runtime layer for multi-agent LLM serving. The key insight is that future KV-cache reuse is governed by agent execution semantics rather than cache recency alone. CacheScout captures these semantics by learning agent execution transitions online, without requiring predefined workflow graphs or offline training, and uses the learned execution model to guide both cache eviction and proactive prefetching while leaving the serving critical path unchanged. We implement CacheScout on top of vLLM. Across representative real-world multi-agent workloads, CacheScout improves KV-cache hit rate by 10-18 percentage points, reduces mean TTFT by 18-45%, lowers mean per-turn latency by 29-38%, and increases peak throughput by up to 57%. These benefits also generalize to larger models, reducing TTFT by up to 54% while sustaining 37% higher throughput.
Large Language Models (LLMs) store factual knowledge and domain-specific patterns implicitly in dense Transformer parameters, making knowledge expansion costly through pretraining, fine-tuning, retrieval augmentation, or longer contexts. Engram-style memory offers a compact hidden-state injection pathway, but existing GPU-resident designs often rely on hash-based compression, causing unrelated phrases to collide in shared slots and weakening phrase-level semantic fidelity. We present TF-Engram, a train-free Engram system that constructs phrase-specific semantic memory offline from external corpora, stores large memory tables across a GPU--DRAM--SSD hierarchy, and uses Early-Exit Guided Predictive Prefetching to hide external-memory latency during autoregressive decoding. On Qwen3-0.6B, TF-Engram improves the average downstream score from 57.6 to 59.4, outperforming both the frozen backbone and a parameter-matched LoRA baseline. System evaluation shows that large TF-Engram tables can be built with moderate offline cost, SSD-backed storage substantially reduces GPU memory demand, and predictive prefetching recovers much of the throughput loss caused by external memory access. These results demonstrate that static phrase memory can be integrated into LLM inference as a scalable, train-free, and low-overhead system component.
Sparse Mixture-of-Experts (MoE) models expand foundation model capacity through conditional expert activation, but their full expert pools remain difficult to deploy under limited accelerator memory. Although expert offloading alleviates memory pressure by moving inactive experts to host memory or storage, it introduces a routing-dependent transfer bottleneck: required experts are known only after native top-\(K\) routing, which serializes routing, expert loading, and expert execution during inference. To address this bottleneck, we propose SpecPrefetch, a parameter-efficient prefetching framework for offloaded MoE inference. SpecPrefetch uses a shared lightweight adapter to predict next-layer expert candidates only for asynchronous transfer, while the frozen native router still determines the final executed experts. By separating transfer prediction from execution routing, SpecPrefetch reduces exposed expert-loading latency without changing pretrained routing semantics, so prediction errors affect transfer efficiency rather than model outputs. In addition, a window-aware scheduler prioritizes feasible transfers under cache and bandwidth constraints. Across Qwen3-VL-30B-A3B and DeepSeek-VL2-Tiny, SpecPrefetch achieves the best average expert recall in 9 out of 10 model-benchmark settings with substantially fewer trainable parameters than learned predictor baselines. On a Snapdragon 8 Elite device, SpecPrefetch further improves decoding throughput by up to \(20\%\) over a compute-optimized offloading runtime, demonstrating practical benefits for storage-constrained MoE deployment. The code and model weights are available at https://github.com/wei390/SpecPrefetch.