Real-time multimodal inference on resource-constrained edge devices is essential for applications such as autonomous driving, human-computer interaction, and mobile health. However, prior work often overlooks the tight coupling between sensing dynamics and model execution, as well as the complex inter-modality dependencies. In this paper, we propose MMEdge, an new on-device multi-modal inference framework based on pipelined sensing and encoding. Instead of waiting for complete sensor inputs, MMEdge decomposes the entire inference process into a sequence of fine-grained sensing and encoding units, allowing computation to proceed incrementally as data arrive. MMEdge also introduces a lightweight but effective temporal aggregation module that captures rich temporal dynamics across different pipelined units to maintain accuracy performance. Such pipelined design also opens up opportunities for fine-grained cross-modal optimization and early decision-making during inference. To further enhance system performance under resource variability and input data complexity, MMEdge incorporates an adaptive multimodal configuration optimizer that dynamically selects optimal sensing and model configurations for each modality under latency constraints, and a cross-modal speculative skipping mechanism that bypasses future units of slower modalities when early predictions reach sufficient confidence. We evaluate MMEdge using two public multimodal datasets and deploy it on a real-world unmanned aerial vehicle (UAV)-based multimodal testbed. The results show that MMEdge significantly reduces end-to-end latency while maintaining high task accuracy across various system and data dynamics.

Chain-of-Thought (CoT) prompting is a key technique for enabling complex reasoning in large language models. However, generating full, fixed-length rationales is computationally wasteful, inflating both token usage and latency. We introduce LEASH: Logit-Entropy Adaptive Stopping Heuristic, a training-free decoding algorithm that adaptively halts rationale generation. LEASH monitors two intrinsic signals: the slope of token-level entropy and the improvement in the top-logit margin. It terminates the generation once both signals plateau, indicating the model has reached a stable reasoning state. Across four instruction-tuned models on the GSM8K and AQuA-RAT benchmarks, LEASH reduces average token generation by 30--35% and latency by 27%, while incurring a 10 p.p. accuracy drop relative to CoT. LEASH is model-agnostic and requires no additional training or supervision, offering a simple and efficient alternative to CoT decoding.

Given the significant advances in machine learning techniques on mobile devices, particularly in the domain of computer vision, in this work we quantitatively study the performance characteristics of 190 real-world vision transformers (ViTs) on mobile devices. Through a comparison with 102 real-world convolutional neural networks (CNNs), we provide insights into the factors that influence the latency of ViT architectures on mobile devices. Based on these insights, we develop a dataset including measured latencies of 1000 synthetic ViTs with representative building blocks and state-of-the-art architectures from two machine learning frameworks and six mobile platforms. Using this dataset, we show that inference latency of new ViTs can be predicted with sufficient accuracy for real-world applications.

Despite their strong performance on reasoning tasks, large reasoning models (LRMs) often suffer from overthinking, producing unnecessarily long outputs and incurring high end-to-end latency, a significant limitation to their real-world deployment. To address overthinking, early-exit mechanisms have been proposed to terminate reasoning before typical completion, showing that this approach can effectively shorten generation length with minimal impact on accuracy. However, their reliance on probing mechanisms introduces a detection overhead that limits their end-to-end latency gains and compromises their generalizability across diverse problems. Inspired by the use of hidden states in speculative decoding, we propose SpecExit, a novel framework that predicts both future tokens and an early-exit signal directly from a lightweight draft model without probing overhead. Our method offers significant improvements, reducing average generation length by 66% and achieving a 2.5x speedup in end-to-end latency compared to the speculative decoding baseline, without compromising accuracy. Our method leverages the inherent signals from hidden states to provide effective early-exit signals, suggesting broader use of hidden states for efficient reasoning. Large reasoning models (LRMs) such as OpenAI-o1 (OpenAI, 2024), DeepSeek-R1 (DeepSeek-AI et al., 2025) and Qwen (Qwen et al., 2025) have recently achieved state-of-the-art performance in complex tasks.

Large Language Models (LLMs), such as OpenAI-o1 and DeepSeek-R1, have demonstrated strong reasoning capabilities. To further enhance LLM capabilities, recent agentic systems, such as Deep Research, incorporate web interactions into LLM reasoning to mitigate uncertainties and reduce potential errors. However, existing research predominantly focuses on reasoning performance, often neglecting the efficiency of agentic systems. In this work, we present a comprehensive empirical study that identifies efficiency bottlenecks in web-interactive agentic systems. We decompose end-to-end latency into two primary components: LLM API latency and web environment latency. We conduct a comprehensive empirical study across 15 models and 5 providers to demonstrate high variability in API-based agentic systems. We observe that web environment latency can contribute as much as 53.7% to the overall latency in a web-based agentic system. To improve latency, we propose SpecCache, a caching framework augmented with speculative execution that can reduce web environment overhead. Extensive evaluations on two standard benchmarks show that our approach improves the cache hit rate by up to 58x compared to a random caching strategy, while reducing web environment overhead by up to 3.2x, without degrading agentic system performance.

W e will release all the source code and data . Meanwhile, please find our detailed responses below. The key idea of Shi et al. is predicting DARTS search space and will include the results in the final paper. We ran ablation studies based on the comments. LatBench with FP32/FP16/INT8 on supported platforms by November.

Future 6G radio access networks (RANs) will be artificial intelligence (AI)-native: observed, reasoned about, and re-configured by autonomous agents cooperating across the cloud-edge continuum. We introduce MX-AI, the first end-to-end agentic system that (i) instruments a live 5G Open RAN testbed based on OpenAirInterface (OAI) and FlexRIC, (ii) deploys a graph of Large-Language-Model (LLM)-powered agents inside the Service Management and Orchestration (SMO) layer, and (iii) exposes both observability and control functions for 6G RAN resources through natural-language intents. On 50 realistic operational queries, MX-AI attains a mean answer quality of 4.1/5.0 and 100 % decision-action accuracy, while incurring only 8.8 seconds end-to-end latency when backed by GPT-4.1. Thus, it matches human-expert performance, validating its practicality in real settings. We publicly release the agent graph, prompts, and evaluation harness to accelerate open research on AI-native RANs. A live demo is presented here: https://www.youtube.com/watch?v=CEIya7988Ug&t=285s&ab_channel=BubbleRAN

Efficient on-device neural network (NN) inference has various advantages over cloud-based processing, including predictable latency, enhanced privacy, greater reliability, and reduced operating costs for vendors. This has sparked the recent rapid development of microcontroller-scale NN accelerators, often referred to as neural processing units ($\mu$NPUs), designed specifically for ultra-low-power applications. In this paper we present the first comparative evaluation of a number of commercially-available $\mu$NPUs, as well as the first independent benchmarks for several of these platforms. We develop and open-source a model compilation framework to enable consistent benchmarking of quantized models across diverse $\mu$NPU hardware. Our benchmark targets end-to-end performance and includes model inference latency, power consumption, and memory overhead, alongside other factors. The resulting analysis uncovers both expected performance trends as well as surprising disparities between hardware specifications and actual performance, including $\mu$NPUs exhibiting unexpected scaling behaviors with increasing model complexity. Our framework provides a foundation for further evaluation of $\mu$NPU platforms alongside valuable insights for both hardware designers and software developers in this rapidly evolving space.

Large Language Model (LLM) inference, where a trained model generates text one word at a time in response to user prompts, is a computationally intensive process requiring efficient scheduling to optimize latency and resource utilization. A key challenge in LLM inference is the management of the Key-Value (KV) cache, which reduces redundant computations but introduces memory constraints. In this work, we model LLM inference with KV cache constraints theoretically and propose novel batching and scheduling algorithms that minimize inference latency while effectively managing the KV cache's memory. We analyze both semi-online and fully online scheduling models, and our results are threefold. First, we provide a polynomial-time algorithm that achieves exact optimality in terms of average latency in the semi-online prompt arrival model. Second, in the fully online case with a stochastic prompt arrival, we introduce an efficient online scheduling algorithm with constant regret. Third, we prove that no algorithm (deterministic or randomized) can achieve a constant competitive ratio in fully online adversarial settings. Our empirical evaluations on a public LLM inference dataset, using the Llama-70B model on A100 GPUs, show that our approach significantly outperforms benchmark algorithms used currently in practice, achieving lower latency while reducing energy consumption. Overall, our results offer a path toward more sustainable and cost-effective LLM deployment.