The paper reports a reproducible GPU power benchmark for 18 open‑source LLMs (0.5B–7B parameters) run on a single consumer RTX 4060ti GPU using the Ollama inference engine. Energy metrics such as mean/peak power, total energy per prompt, energy per output token, and throughput were measured, revealing that model architecture and quantization strategy, rather than parameter count alone, drive energy efficiency. The most efficient models were qwen2.5:0.5b and tinyllama:1.1b, while the 7B‑Mistral model consumed up to 8.6× more energy per token, and qwen3.5:0.8b(on) showed unusually high per‑prompt energy due to extended internal reasoning.
By Philipp M. Z\"ahl, Elja Dalipaj, Anika Hennig, Timon Bayer
The paper introduces an agentic AI-driven immersive simulation for High Dose Rate vaginal cylinder brachytherapy, combining Virtual Reality, mobile computing, and a knowledge-aware assistant powered by Retrieval-Augmented Generation. The system offers a high‑fidelity, risk‑free training environment that provides natural language interfaces and real‑time guidance during complex medical maneuvers. A prototype using Meta Quest 3 and a GPU‑accelerated backend demonstrates acceptable latency and high context precision, answer completeness, and relevance in the pedagogical support.
By Ronghua Xu, Kepha Barasa, Manoj Kumal, Xinyun Liu, Weihua Zhou, Xin Qian
T-LoopFormer introduces token-level elastic-depth looped transformers that allow each token to decide its own number of loop iterations based on its hidden state, improving token generation accuracy. It also adds a recursion-wise key‑value cache so tokens at different depths only attend to their corresponding cached states, speeding up autoregressive decoding. Experiments demonstrate strong performance on language modeling and zero‑shot reasoning, achieving the lowest decoding latency among comparable models.
By Mingqian Yu, Wenpeng Zhang, Shaobo Cui, Peilin Zhao
LEAD is a gated temporal‑spatial Transformer foundation model designed for EEG‑based Alzheimer's disease detection. It was trained on the world’s largest EEG‑AD corpus of 2,238 subjects and uses a subject‑regularized strategy and medical contrastive learning across 13 datasets. LEAD outperforms existing methods on five downstream AD datasets, achieving the best average ranking across 20 evaluations.
By Yihe Wang, Nan Huang, Nadia Mammone, Marco Cecchi, Xiang Zhang
BabelCoder is an agentic framework for automatic code translation that splits the task into specialized agents for translation, testing, and refinement. Each agent focuses on a specific aspect—generating code, validating correctness, or repairing errors—allowing collaborative improvement of translation quality. Evaluated on four benchmark datasets, BabelCoder outperforms four state‑of‑the‑art baselines, achieving an average accuracy of 94.16% and surpassing existing methods in 94% of cases.
By Fazle Rabbi, Soumit Kanti Saha, Tri Minh Triet Pham, Song Wang, Jinqiu Yang
Consist‑Retinex introduces a one‑step noise‑emphasized consistency training framework for Retinex‑based low‑light image enhancement. It first decomposes images into reflectance and illumination maps using a Retinex Transformer Decomposition Network, then trains two conditional consistency models with a dual objective that blends trajectory consistency and ground‑truth alignment. The method employs adaptive noise‑emphasized fixed‑point sampling to focus supervision near the inference endpoint, achieving state‑of‑the‑art VE‑LOL‑L scores on paired and unpaired low‑light benchmarks while reducing sampling and training costs.
By Jian Xu, Wei Chen, Shigui Li, Delu Zeng, John Paisley, Qibin Zhao
RAPTOR (Ridge-Adaptive Logistic Probe) is a lightweight, L2‑regularized logistic probe designed to extract accurate, directionally stable concept vectors from frozen large language model (LLM) layer representations. By tuning the ridge penalty on validation data, RAPTOR produces concept vectors from normalized weights that match or surpass strong baselines in accuracy, maintain stability under ablation, and require significantly less training cost. The authors also provide a mechanistic analysis using the Convex Gaussian Min‑max Theorem to explain how penalty strength influences probe accuracy and stability in a high‑dimensional few‑shot setting, with predictions that align with empirical observations on real LLM embeddings.
By Ziqi Gao, Yaotian Zhu, Qingcheng Zeng, Xu Zhao, Ziqing Wang, Feng Ruan, Kaize Ding
The paper introduces the Affective Flow Language Model (AFlow), which treats multi‑turn emotional support conversations as an evolving affective utility flow along dialogue trajectories. AFlow searches diverse support paths, estimates utilities of intermediate states, and employs Affective Flow Preference Optimization (AFPO) to propagate downstream preference signals to earlier states, enabling consistent strategy transitions. Experiments on ExTES and ESConv demonstrate improved strategy alignment, response diversity, and generation quality across various model settings.
By Chenghui Zou, Ning Wang, Tiesunlong Shen, Luwei Xiao, Chuan Ma, Xiangpeng Li, Rui Mao, Erik Cambria
The paper presents VLANeXt, a Vision‑Language‑Action (VLA) model built from a unified framework that systematically analyzes design choices across foundational components, perception essentials, and action modeling. By dissecting these dimensions, the authors identify 12 key findings that form a practical recipe for strong VLA models, and demonstrate that VLANeXt outperforms state‑of‑the‑art methods on LIBERO and LIBERO‑plus benchmarks while also excelling in real‑world experiments. The study further extends VLANeXt into a family of variants—scaled models, latent‑action pretraining, latent predictive representation learning, and world action modeling—showing that the core recipe remains effective across different scales and emerging paradigms.
By Xiao-Ming Wu, Kang Liao, Yihang Luo, Bin Fan, Jian-Jian Jiang, Runze Yang, Zhonghua Wu, Wei-Shi Zheng, Chen Change Loy
VLAA-GUI is a modular framework for autonomous GUI agents that addresses early stopping and repetitive loops by integrating three core components: a Completeness Verifier, a Loop Breaker, and an on-demand Search Agent. The framework also includes optional Coding and Grounding Agents for specialized tasks. Evaluations on five backbones across Linux and Windows benchmarks show strong performance, with some models surpassing human results and the Loop Breaker significantly reducing wasted steps.
By Qijun Han, Haoqin Tu, Zijun Wang, Haoyue Dai, Yiyang Zhou, Nancy Lau, Alvaro A. Cardenas, Yuhui Xu, Ran Xu, Caiming Xiong, Zeyu Zheng, Huaxiu Yao, Yuyin Zhou, Cihang Xie
StraTA introduces Strategic Trajectory Abstraction, a framework that samples a compact strategy from the initial task state and conditions subsequent actions on that strategy, training strategy generation and action execution jointly with a hierarchical GRPO-style rollout design. The method enhances exploration and credit assignment over long horizons by incorporating diverse strategy rollouts and critical self-judgment. Experiments on ALFWorld, WebShop, and SciWorld demonstrate that StraTA consistently improves sample efficiency and final performance, achieving success rates of 93.1% on ALFWorld, 84.2% on WebShop, and a 63.5% overall score on SciWorld, surpassing frontier closed‑source models.
By Xiangyuan Xue, Yifan Zhou, Zidong Wang, Shengji Tang, Philip Torr, Wanli Ouyang, Lei Bai, Zhenfei Yin
arXiv:2605. 08475v3 Announce Type: replace-cross Abstract: In this paper, we study in-context kernel ridge regression (KRR) with Gaussian kernels and show, both theoretically and empirically, that a standard softmax-attention transformer can approximate the KRR predictor during its forward pass.
By Mingsong Yan, Dongyang Li, Charles Kulick, Sui Tang
Grid‑Orch is a framework that connects Large Language Models (LLMs) to power system simulation via the Model Context Protocol (MCP), allowing engineers to conduct complex distribution grid analyses using natural language. It offers 36 domain‑specific tools across eleven categories—including power flow, voltage analysis, quasi‑static time‑series simulation, and automated optimization—implemented with OpenDSS as the reference engine. The platform supports both cloud‑hosted and locally deployed LLMs, enabling air‑gapped operation, and demonstrates that tasks such as DER interconnection screening can be completed in under two minutes with results identical to traditional scripting.
By Boming Liu, Jin Dong, Jianming Lian
The paper introduces OASIS, a method designed to stabilize dual‑normalized attention‑residual architectures by employing explicit null routing and token‑to‑depth null coupling. OASIS mitigates attention sinks and activation outliers, improving low‑bit quantization performance across several language‑model backbones. Empirical results show significant reductions in attention norms and perplexity, with notable gains on long‑context benchmarks.
By Haozheng Luo, Haoran Dai, Jingyuan Huang, Shaoyang Zhang, Xi Chen, Eric Hanchen Jiang, Yijiang Li, Chenghao Qiu, Chenwei Xu, Zhenyu Pan, Haotian Zhang, Binghui Wang, Yan Chen
The paper demonstrates that Item Response Theory (IRT) can uncover meaningful structure in safety benchmarks for language models, allowing adaptive item selection to approximate full benchmark rankings with Spearman’s ρ > 0.90 while cutting evaluation costs by at least 80% and up to 99.9% on some suites. It also proposes a static method to extract a small, informative subset of items that can be reused across models, achieving 80–99.8% cost savings. These findings show that psychometric techniques can make safety evaluation more efficient without sacrificing ranking accuracy.
By Fabio Spagliardi, M\'irian Silva, Ayan Datta, Aiden Zhou, Vamshi Bonagiri, Diogo Cruz
The paper presents a preprocessor that recovers and decodes encoded content in vision‑language models to close the decode gap that allows harmful requests to bypass safety classifiers. Evaluated against eleven encoding attacks, the preprocessor raises block rates from 0 % to 67‑90 % but also increases benign over‑refusal, and no configuration achieves an ensemble attack‑success rate below 40 % while keeping benign over‑refusal under 70 %. The study shows that closing one encoding channel merely relocates success rather than eliminating it, highlighting the limits of recovery‑based defenses.
By Haoyu Zhang, Zhuoxi Wang, Shibo Zheng, Hanwen Liu, Yi Feng, Haowen Xu, Xiangchen Guan, Yang Chen, Zijian Xiao, Xiao Luo, Mohammad Zandsalimy, Shanu Sushmita
The paper demonstrates that vision‑language models’ refusal behavior is heavily influenced by whether an image is attached to a request, even when the image is blank or unreadable. Attaching such an image shifts refusal scores by large margins for borderline‑benign prompts while leaving genuinely neutral instructions largely unchanged. This effect varies with image properties, persists across checkpoints, and is not mitigated by explicit instructions to ignore the image.
By Haoyu Zhang, Yi Feng, Hanwen Liu, Shibo Zheng, Zhuoxi Wang, Yang Chen, Haowen Xu, Xiangchen Guan, Mohammad Zandsalimy, Shanu Sushmita
arXiv:2609.30316v1 Announce Type: cross
Abstract: Language models used in financial backtests suffer from look-ahead bias, as a model trained on text published after the study period has already obse...
By Seunghan Lee, Jun Seo, Jaehoon Lee, Junhyeok Kang, Sangjun Han, Sungdong Yoo, Minjae Kim, Tae Yoon Lim, Dongwan Kang, Hwanil Choi, Soonyoung Lee, Wonbin Ahn
arXiv:2609.30465v1 Announce Type: cross
Abstract: Mixture-of-experts (MoE) models activate few experts per token but store the full expert pool. Expert pruning reduces this storage burden; at a fixed...
By Mingyang Song, Mao Zheng
arXiv:2609.30572v1 Announce Type: cross
Abstract: Large language models are often post-trained on expert demonstrations using cross-entropy (CE), even when the downstream objective is not to imitate...
By Mihir Dhanakshirur, Adam Ousherovitch, Ambuj Tewari