ZenGen is a comprehensive framework designed to develop socially intelligent large language models (LLMs). It introduces SoMBench, a psychology‑grounded benchmark with 3 primary and 17 secondary dimensions, and shows that current LLMs perform poorly on social cognition tasks. ZenGen combines supervised fine‑tuning, on‑policy distillation, and rubric‑based reinforcement learning to improve internalization, while Actio provides deployment‑time grounding with typed supports such as PRISM, Starling, SAGE, and gated RAG, yielding significant performance gains across benchmarks.
By ZenGen Team, Ao Xiang, Bi Jingping, Chen Jiahui, Chen Lehan, Chen Yilin, Cheng Xueqi, Fan Yixing, Gan Kairong, Gao Haowen, Gao Jinhua, Gao Shuxuan, Gong Chang, Guo Jiafeng, Guo Ruijie, Han Zhouyu, He Guangfu, He Yichun, Jiang Shuo, Jing Shaoling, Jing Ya, Lei Chenhao, Lei Yan, Li Anqi, Li Chengao, Li Haoyu, Li Shitian, Liang Xinjian, Liu Zhaoge, Lyu Xingyu, Nie Zhuwei, Pang Liang, Quan Zeping, Shan Shiguang, Shen Huawei, Tang Xinran, Tian Feng, Wang Qian, Wang Ruiping, Wang Xiaohong, Xia Zaiyu, Xiao Yi, Xu Jiayuan, Xu Kehan, Xu Qianqian, Xu Tianyu, Xu Yongjun, Yang Haoming, Yang Jun, Yao Di, Yu Xiaoming, Zhang Futong, Zhang Jie, Zhang Shixuan, Zhang Yuxuan, Zhao Xinyu, Zhao Zhuoran, Zhong Yunfei, Zhu Shengyu
Swift-Image is a compact unified model that performs text-to-image generation, single-image editing, and multi-image editing using a 6B parameter DiT architecture. It employs a progressive training pipeline, parallel expert reinforcement learning, and multi-teacher distillation to balance diverse objectives, while a Prompt Enhancer decouples high-level reasoning from pixel-level rendering. After training, structural pruning and few-step distillation produce efficient 3B and accelerated variants that maintain near‑lossless performance and improve editing efficiency.
By Taihang Hu, Zhao Wang, Zuan Gao, Tao Liu, Hao Yan, Zhengze Xu, Yuhang Yu, Yongchao Du, Xingjian Wang, Jun Zheng, Qinye Zhou, Yaqi Cai, Zhengrui Chen, Chao Lin, Yefeng Shen, Yuan Wang, Zhengtao Wu, Ge Wu, Xiaoli Xu, Denghui Yang, Huayu Zhang, Mingzhou Zhang, Mengting Chen
The paper introduces Semantic Localization-Enhanced Teacher (SLE‑T), a knowledge‑distillation framework that aligns spatial‑scale and semantic features between a Vision Foundation Model (VFM) teacher and a student detector for cross‑domain object detection. SLE‑T employs a lightweight SLE Adapter that injects pretrained local‑texture priors into DINOv2 and reformulates its features into dense, spatially and semantically compatible representations, enabling effective pseudo‑label learning or feature alignment. Experiments on three domain‑adaptive object detection benchmarks show that SLE‑T with DINOv2‑B achieves state‑of‑the‑art performance while using only a quarter of the training time and less GPU memory compared to the larger DINOv2‑G teacher.
By Qifeng Zhang, Ting Xiang, Zeyuan Bai, Changjian Chen
The paper introduces a framework for evaluating how large language model agents revise their success criteria after failures, defining five non‑compensatory conditions that must be met for a criterion revision to be considered valid. Using the CMB‑0.1 protocol, the authors test twelve cross‑domain scenarios across four system configurations, finding that no model trial satisfies all five conditions and highlighting specific failure modes such as zero‑state reconstruction and inadequate intervention sensitivity. They propose a more stringent trace‑anchored CMB‑0.4 protocol to better isolate and measure criterion revision in future studies.
By Guodong Xu
MentorPulse introduces a method for refreshing cross‑model latent guidance during long‑form generation. By compressing mentor states into a capped slot memory and updating it incrementally as new tokens are generated, the approach keeps the guidance signal fresh without resetting the student’s KV cache. Across thirteen datasets, MentorPulse closes more than half of the mentor‑student performance gap, outperforming existing techniques such as C2C, T2T, and LoRA, especially on long outputs.
By Ziwu Liu, Guozhong Li, Chen Qiu, Weiyang Kong, Panos Kalnis
arXiv:2608.20530v1 Announce Type: new
Abstract: Speculative decoding accelerates language-model inference by drafting future tokens that the target model verifies in parallel. A diffusion-style block...
By Matan Rusanovsky, Yoav Miron, Roy Uziel, Omer Belhasin, Ran Zilberstein, Maor Ashkenazi, Michael Elad
The paper introduces Preference‑Based Self‑Distillation (PBSD), a new on‑policy self‑distillation method that replaces traditional KL matching with a reward‑regularized objective. PBSD derives a reward‑reweighted teacher distribution, optimizing preference gaps between teacher and student samples while keeping on‑policy sampling. Experiments on mathematical reasoning and tool‑use tasks show PBSD achieves stronger average performance, improved training stability, and maintains token efficiency compared to prior self‑distillation baselines.
By Xin Yu, Liuchen Liao, Yiwen Zhang, Yingchen Yu, Lingzhou Xue, Qinzhen Guo
The paper introduces Sparse Token Routing in Efficient Transformers, evaluating a two-stream Transformer (SEWN) that routes tokens through either lightweight or full-capacity processing via a learned gate. Experiments show that routing causes negligible accuracy change compared to parameter-matched baselines, and that the effectiveness of the gate’s token-importance signal depends on its learning method. A static lexicon-seeded prior fails a counterfactual faithfulness test on BoolQ, whereas a fully contextual gate achieves highly significant separation ($p<10^{-10}$) on both evaluated tasks without altering task accuracy.
By Sai Krishna Arthanari, JaeHyeong Chang, Chengzhe Sun, Siwei Lyu
The paper presents a generative AI approach that enhances galaxy images from ground‑based sky surveys to match the detail level of space‑based telescopes. By training on space‑based data, the model converts weak signals into clear, detailed galaxy images, enabling the combination of high survey throughput with superior image quality. The authors provide source code, training data, a catalog of 63,202 enhanced galaxy images, and a software tool that encapsulates the entire pipeline.
By Sai Teja Erukude, Lior Shamir
arXiv:2605.15508v3 Announce Type: replace
Abstract: The quadratic complexity of attention imposes severe memory and computational bottlenecks on Large Language Model (LLM) inference. This challenge i...
By Jiangnan Yu, Ceyu Xu, Yongji Wu, Yuan Xie
Tetris is a video object tracking system that uses tile-level sampling to efficiently extract high‑fidelity tracks. It partitions videos into tile‑based polyominoes, classifies relevant tiles, prunes redundant ones with an ILP under a user‑defined accuracy constraint, and packs the remaining polyominoes to minimize detector calls. On seven stationary‑video datasets, Tetris maintains less than a 5% loss in tracking accuracy while achieving up to 17.4× higher throughput than prior systems and up to 68.8× higher than a full‑frame reference pipeline.
By Chanwut Kittivorawong, Alena Chao, Charlie Si, Alvin Cheung
arXiv:2608.20554v1 Announce Type: cross
Abstract: The critical failure modes in deployed large language models (LLMs) are cross-dimensional: a model can score 99.3 in safety alignment while refusing...
By Fatih Deniz, Yazan Boshmaf, Dorde Popovic, Issa Khalil
The paper proposes replacing a static multi‑level small semantic codebook with a dynamic single‑level large semantic codebook for generative recommendation. It introduces an exposure‑aware update mechanism and an offline evaluation framework, achieving significant improvements in recall, NDCG, decoding efficiency, and online consumption metrics on public datasets and production traffic.
By Tianlu Xie, Xin Ku, Mingjie Sun, Yunhao Sha, Lixiang Wang, Peng Wang, Yiyu Wang, Wenjin Wu, Zhaojie Liu, Peng Jiang, Wenwu Ou
arXiv:2608.20988v1 Announce Type: cross
Abstract: Quantization of Large Language Models (LLMs) is often hindered by the sensitivity of the self-attention mechanism to discretization errors. We identi...
By Deepanshu Pandey, Arnav Chavan, Nahush Lele, Sankalp Dayal, Deepak Gupta
arXiv:2608.21019v1 Announce Type: cross
Abstract: Quantization is widely used to deploy large language models, but its effect on uncertainty behavior, such as confidence, margins, and abstention, is...
By Zhen Yang, Sizai Hou, Kaiwen Zheng, Yaofang Liu, Liang He, Yixuan Chen, Kangning Cui
The paper introduces Llama-Mobile, a framework that quantizes vision‑language models for efficient mobile deployment. It uses a quantization pipeline that generates training data from the model itself, eliminating the need for the original training setup, and employs a novel 2.7‑bit‑per‑parameter format optimized for Arm CPUs. Applying this method, the authors compress the Llama 3.2 11B Vision Instruct model to 3.7 GB with 8‑bit activations while maintaining strong performance on visual question answering tasks.
By Luka Ribar, Jeevan Bhoot, Douglas Orr
The paper investigates whether distilling reasoning traces from large teacher models is worth the extra compute compared to standard instruction fine‑tuning (IFT). By generating paired IFT and reasoning outputs from the same teacher and training student models at five scales, the authors find that, at matched FLOPs, IFT generally lies on or near the Pareto frontier across most configurations. Reasoning traces only reach the frontier on open‑ended tasks for models 7B and larger, and a curriculum mixing 25–50% reasoning data with IFT can capture most of the accuracy benefit at a lower compute cost.
By Nicolas Boizard, Hippolyte Gisserot-Boukhlef, Kevin El Haddad, C\'eline Hudelot, Pierre Colombo
arXiv:2608.21308v1 Announce Type: new
Abstract: Test-Time Training (TTT) enables long-context processing via continuous weight updates during inference, but current methods struggle to balance the ex...
By Zeyun Zhong, Joya Chen, Manuel Martin, Frederik Diederichs, Juergen Gall, Juergen Beyerer
arXiv:2608.21066v1 Announce Type: new
Abstract: Deploying autonomous systems in safety-critical domains demands guaranteed robustness against physically plausible geometric perturbations rather than...
By Gregoire Theau, Melanie Ducoffe
arXiv:2608.20515v1 Announce Type: new
Abstract: Generative video compression can recover rich visual details at low bitrates, but simultaneously achieving high temporal consistency and low inference...
By Wenzhuo Ma, Zhenzhong Chen