The list as it stood that day: the papers researchers were upvoting on Hugging Face's daily papers. Papers stay on arXiv.
Since visual tokens make up 89-97% of this input on the benchmarks we study, most of the decoding computation is spent repeatedly processing the same image.
Donghyeon Cho, Joowon Kim, Seungho Shin and 1 moreWe present Youtu-Parsing-Omni, an omni-modal parser that converts documents, images, audio, and video into structured records grounded in space and time, providing a perception layer for agents that read, locate, and cite evidence across modalities.
Yunhang Shen, Zhangquan Chen, Xiaotian Li and 27 moreWe address these two questions separately: training dynamics determine when to acquire new experience, while the answers to new tasks remain independently checkable.
Yu Liu, Youxing Li, Yuedong Tan and 17 moreWe turn the hypothesis into a theorem. First, we prove a Platonic Representation Theorem: Joint-Embedding Predictive Architectures (JEPAs) trained separately on any number of modalities learn the same representation up to one rotation whenever their positive pairs are two…
Randall Balestriero, Yann LeCunIn this paper, we introduce a Latent 3D Gaussian Memory (L3G) that combines the efficiency of feed-forward reconstruction with a compact, rapidly renderable scene representation.
Jinhyeok Choi, JoungBin Lee, Jongmin Lee and 5 moreWe observe that finer detail does not necessarily require more global communication: low-frequency structure benefits from global exchange, while high-frequency detail can be modeled locally.
Jiongfan Yang, Weichen Fan, Haiwen Diao and 2 moreWe propose WorkFactory, a scalable framework for constructing coherent and executable professional tasks.
Ziyang Zeng, Tingyang Chen, Jia Deng and 10 moreLocating relevant evidence in long inputs is a core capability of Multimodal Large Language Models.
Jia Wang, Yuqing Li, Jiangnan Li and 2 moreWe introduce VibeGame, a framework for creating complete 2D game projects from natural-language prompts and optional reference images.
Wenbo Hu, Ken Li, Jiazhe Wei and 13 moreIn this work, we develop this measure-based view of benchmark evaluation. This perspective allows us to identify and formalize two central failure modes: measure mismatch, where the benchmark difficulty distribution differs from the intended difficulty distribution, and coverage…
Yifan Sun, Naicheng Yu, Jiawen Gong and 2 moreObserving that a card hairstyle is an unordered set of artist-defined primitives, we propose HairDeck, which generates it as one: each card is a single token, and a set diffusion transformer denoises all cards together.
Yuze He, Yixiang Wei, Wang Zhao and 3 moreFollowing prior work, we estimate this difference with internal velocity branches, avoiding a separate fake-score network.
Deyuan Liu, Yansen Han, Kaiwen Zheng and 7 moreWe introduce Co-Design Bench, a reference-free benchmark of 273 tasks curated from 37,463 deduplicated queries across 26,314 real user coding sessions.
Xilin Wei, Shengyuan Ding, Xinyu Fang and 13 moreWe address this gap with WeVisCut, a long-form script-to-video system that connects tool execution with policy learning.
Chenyu Mu, Wenming Weng, Qinrui Wang and 8 moreIn this paper, we present REALCLI, a paired benchmark that unifies the evaluation of task execution and human-agent collaboration quality on realistic CLI tasks.
Yunqi Liu, Peiyu Wang, Bingli Wang and 21 moreWe introduce PersonaCUABench, which holds 74 tasks across 11 domains fixed and varies only who asks.
Haoming Li, Defu Cao, Xu Hu and 23 moreWe identify a flaw in standard MTP training, in which a residual component of the MTP gradient interferes with both next-token and multi-token learning, increasingly so with scale.
Jayden TeohTo this end, we present IntReal, comprising two complementary components: IntReal-100K, a real-world dataset of multimodal interleaved samples spanning four major categories, and the IntReal Evaluation Suite, which evaluates model performance across seven complementary…
Shaohan Yu, Jie Qin, Wende Tan and 4 moreWe present PhysInsert, a framework for physically plausible reference-based object insertion that jointly adapts objects and their surrounding scenes.
Yue Gong, Le Zhang, Rui Chen and 5 moreWe present WebAnything, a fully automated framework that reconstructs source-grounded, reusable Web environments from real URLs without per-site manual implementation or repair after initial setup and authorization.
Tianci Xie, Zikun Cai, Chaoyun Zhang and 4 more