Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
Qwen, An Yang, Baosong Yang +39 authors
Qwen2.5, an enhanced series of large language models, demonstrates superior performance across various benchmarks and use cases through extensive pre-training and advanced post-training techniques.
50 篇论文 · 按点赞排序
Benjamin Warner, Antoine Chaffin, Benjamin Clavié +11 authors
ModernBERT, an optimized encoder-only transformer model, demonstrates superior performance and efficiency in diverse classification and retrieval tasks compared to previous models.
Orr Zohar, Xiaohan Wang, Yann Dubois +9 authors
A comprehensive study reveals effective design principles for video understanding in Large Multimodal Models, introducing Apollo, which achieves superior performance across different model sizes.
Artidoro Pagnoni, Ram Pasunuru, Pedro Rodriguez +11 authors
A Byte Latent Transformer (BLT) matches tokenization-based LLM performance at scale with improved inference efficiency and robustness, using entropy-based byte patching.
Taiming Lu, Tianmin Shu, Junfei Xiao +8 authors
GenEx generates 3D environments from a single image, enabling AI agents to explore and interact with a consistent, expansive space through guided generative imagination.
Junnan Liu, Hongwei Liu, Linchen Xiao +6 authors
G-Pass@k and LiveMathBench assess Large Language Models' reasoning capabilities and consistency using complex mathematical problems.
Pavan Kumar Anasosalu Vasu, Fartash Faghri, Chun-Liang Li +8 authors
FastVLM, using FastViTHD, optimizes Vision Language Models by scaling image resolution to reduce latency and token count, improving time-to-first-token and maintaining performance.
Guanting Dong, Chenghao Zhang, Mengjie Deng +3 authors
AR-MCTS enhances multimodal large language models' reasoning capabilities through active retrieval, Monte Carlo Tree Search, and a process reward model, improving performance across multimodal reasoning tasks.
Junjie Zhou, Zheng Liu, Ze Liu +6 authors
MegaPairs, a data synthesis method using vision language models and open-domain images, generates high-quality multimodal retrieval data, outperforming larger existing datasets and achieving state-of-the-art zero-shot performance.
Yihao Meng, Hao Ouyang, Hanlin Wang +6 authors
AniDoc uses video diffusion models to automate colorization and in-betweening in 2D animation, improving efficiency by leveraging correspondence matching.
Xuekai Zhu, Daixuan Cheng, Hengli Li +7 authors
The use of synthetic data in language model training leads to model collapse, which is mitigated by token-level editing of human-produced data to create semi-synthetic data.
Frank F. Xu, Yufan Song, Boxuan Li +18 authors
A benchmark platform evaluates AI agents' autonomy in performing professional tasks in a simulated workplace environment, showing that a significant portion of simpler tasks can be automated but complex, long-term tasks remain challenging.
Shuting Wang, Jiejun Tan, Zhicheng Dou +1 authors
OmniEval is a comprehensive benchmark for Retrieval-Augmented Generation (RAG) in the financial domain, offering a structured evaluation framework with automatic generation, human annotation, and robust metrics.
YiFan Zhang, Shanglin Lei, Runqi Qiao +10 authors
The Multi-Dimensional Insights (MDI) benchmark evaluates large multimodal models across different scenarios and age groups to assess their real-world applicability, revealing areas for improvement in personalization.
Yushi Bai, Shangqing Tu, Jiajie Zhang +9 authors
A benchmark evaluates large language models' performance on long-context problems across various tasks, showing that enhanced reasoning and increased computational resources improve accuracy beyond human performance.
Hao Li, Changyao Tian, Jie Shao +8 authors
SynerGen-VL, an encoder-free MLLM using token folding and vision-expert-based progressive alignment, performs comparably to task-specific models on image understanding and generation with reduced training complexity.
Yaowei Li, Yuxuan Bian, Xuan Ju +3 authors
BrushEdit addresses image editing limitations by combining multimodal large language models and dual-branch inpainting models for autonomous and interactive free-form instruction editing.
Fan Zhang, Shulin Tian, Ziqi Huang +2 authors
The Evaluation Agent framework efficiently evaluates visual generative models using human-like strategies, reducing time to 10% compared to traditional methods while maintaining quality.
Xiaoxi Li, Jiajie Jin, Yujia Zhou +4 authors
RetroLLM is a unified framework that integrates retrieval and generation for LLMs, using constraints and decoding strategies to enhance performance and accuracy in evidence generation.
Lu Wang, Fangkai Yang, Chaoyun Zhang +15 authors
A framework is proposed for developing Large Action Models (LAMs) that generate and execute actions in dynamic environments, moving beyond traditional Large Language Models (LLMs) towards intelligent agents capable of task completion.
Minghao Xu, Lichuan Xiang, Xu Cai +1 authors
SGD-SaI, an enhancement to stochastic gradient descent with momentum, outperforms AdamW in training Transformer-based tasks by scaling learning rates based on gradient signal-to-noise ratios and reducing memory usage.
Jeffrey Cheng, Benjamin Van Durme
Compressed Chain-of-Thought (CCoT) enhances language model reasoning by generating efficient, variable-length contemplation tokens during inference, improving accuracy without fixed-length tokens.
Sahal Shaji Mullappilly, Mohammed Irfan Kurpath, Sara Pieri +8 authors
BiMediX2, a bilingual Arabic-English biomedical multimodal model, achieves state-of-the-art performance in medical applications across text-based and image-based tasks.
Dang Nguyen, Jian Chen, Yu Wang +26 authors
A survey of graphical user interface agents powered by large foundation models, detailing benchmarks, metrics, architectures, training methods, and future challenges in automating human-computer interaction.
Tingfeng Hui, Lulu Zhao, Guanting Dong +3 authors
Smaller language models can generate more effective and diverse instructions than larger models, challenging the assumption that model size directly correlates with instruction synthesis quality.
Qihao Liu, Xi Yin, Alan Yuille +2 authors
CrossFlow proposes a direct mapping between modalities for cross-modal media generation using flow matching, outperforming standard methods in various tasks.
Junhao Zhuang, Xuan Ju, Zhaoyang Zhang +4 authors
ColorFlow, a three-stage diffusion-based framework with a dual-branch design, achieves superior image sequence colorization by incorporating retrieval augmented colorization and self-attention mechanisms, outperforming existing models.
Jihan Yang, Shusheng Yang, Anjali W. Gupta +3 authors
MLLMs trained on large video datasets show competitive but subhuman visual-spatial intelligence, with spatial reasoning as a key bottleneck that can be improved by generating cognitive maps.
Chaorui Deng, Deyao Zh, Kunchang Li +2 authors
Causal Diffusion combines autoregressive models with diffusion techniques to improve generation performance and multimodal capabilities, achieving state-of-the-art results in image generation and zero-shot image manipulations.
Haonan Qiu, Shiwei Zhang, Yujie Wei +5 authors
FreeScale, a tuning-free inference paradigm, enables high-resolution visual generation by fusing information from different scales, overcoming the limitations of existing methods.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号