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。
598 篇论文 · 按点赞排序
Kexun Zhang, Weiran Yao, Zuxin Liu +13 authors
A DEI framework enhances software engineering problem-solving by leveraging a collective of LLM agents, achieving higher resolve rates compared to individual agents and closed-source solutions.
Daoguang Zan, Zhirong Huang, Ailun Yu +17 authors
A Java version of SWE-bench, SWE-bench-java, has been developed and released to evaluate the issue resolution capabilities of large language models in Java, along with a Docker-based environment and leaderboard.
Yunhao Fang, Ligeng Zhu, Yao Lu +6 authors
A novel data augmentation approach iteratively improves visual language model data quality and performance using self-augmentation and specialist-augmentation, leading to state-of-the-art results on MMMU tasks.
Di Zhang, Jingdi Lei, Junxian Li +10 authors
The Critic-V framework leverages an Actor-Critic paradigm to enhance multimodal reasoning in vision-language models by decoupling reasoning and critique processes, significantly improving performance and reliability.
Aitor Ormazabal, Che Zheng, Cyprien de Masson d'Autume +22 authors
Reka Core, Flash, and Edge are multimodal language models that achieve state-of-the-art performance across text, image, video, and audio tasks, offering high efficiency and competitive results relative to their compute class.
Haoran Que, Feiyu Duan, Liqun He +11 authors
A comprehensive benchmark, HelloBench, and evaluation method, HelloEval, are introduced to assess the long text generation capabilities of Large Language Models, revealing limitations and providing human-aligned evaluations.
Shengran Hu, Cong Lu, Jeff Clune
A new research area, Automated Design of Agentic Systems (ADAS), uses a meta agent to automatically create powerful agentic system designs, demonstrating superior performance and robustness across various domains.
Yue Liu, Yunjie Tian, Yuzhong Zhao +5 authors
VMamba, a novel state space model architecture, combines global receptive fields and dynamic weights from ViTs with linear complexity, outperforming established models as image resolution increases.
Jooyoung Choi, Chaehun Shin, Yeongtak Oh +2 authors
The Style-friendly SNR sampler modifies the noise level distribution during fine-tuning to improve style alignment in diffusion models, enabling better capture of unique artistic styles.
Fei Wang, Wenxuan Zhou, James Y. Huang +4 authors
mDPO, a multimodal preference optimization method, addresses the unconditional preference problem and enhances model performance in multimodal large language models.
Eric Wallace, Kai Xiao, Reimar Leike +3 authors
A hierarchical instruction framework for LLMs increases robustness against malicious prompts by selectively prioritizing system instructions over user inputs.
Joshua Engels, Isaac Liao, Eric J. Michaud +2 authors
Research explores multi-dimensional features in language models, discovering interpretable circular representations in GPT-2, Mistral 7B, and Llama 3 8B, which are used for modular arithmetic tasks.
Guoli Yin, Haoping Bai, Shuang Ma +21 authors
MMAU benchmark evaluates large language models' capabilities across multiple domains and essential skills using comprehensive offline tasks to enhance interpretability and reliability.
Ziyang Ma, Yakun Song, Chenpeng Du +5 authors
A novel listening-while-speaking language model (LSLM) enhances real-time, full-duplex speech interaction by integrating speech generation and real-time audio input with an optimal fusion strategy.
Zhenliang Xue, Yixin Song, Zeyu Mi +3 authors
PowerInfer-2 is a framework that enhances the high-speed inference of large language models on smartphones by utilizing heterogeneous resources and optimizing matrix computations through neuron cluster processing.
Liliang Ren, Yang Liu, Yadong Lu +3 authors
Samba, a hybrid architecture combining State Space Models and Sliding Window Attention, efficiently handles long-range context with lower computational complexity and better memory recall than pure attention or SSM models.
Xingguang Yan, Han-Hung Lee, Ziyu Wan +1 authors
Object Images convert 3D models into a manageable 2D format using image generation models like Diffusion Transformers, providing realistic 3D shapes and PBR materials.
Kai Chen, Yunhao Gou, Runhui Huang +27 authors
EMOVA integrates speech, vision, and language capabilities, achieving top performance in benchmarks and enabling emotive spoken dialogue using a disentangled speech tokenizer and style module.
Lingyun Zhang, Bin jin, Gaojian Ge +7 authors
A principle for leveraging human priors in data construction is proposed to improve small language models, demonstrating favorable performance on large benchmarks in resource-constrained settings.
Zayne Sprague, Fangcong Yin, Juan Diego Rodriguez +7 authors
Chain-of-thought prompting enhances large language model performance primarily in math or logic tasks by improving symbolic reasoning, but it is less beneficial in other domains and can be selectively applied to save inference costs.
Chaehun Shin, Jooyoung Choi, Heeseung Kim +1 authors
Diptych Prompting is a zero-shot text-to-image generation method that enhances subject alignment and attention between panels during inpainting, resulting in visually preferable images across various applications.
Dawei Li, Bohan Jiang, Liangjie Huang +10 authors
A survey of LLM-based judgment and assessment explores definitions, taxonomy, benchmarks, challenges, and future directions for the LLM-as-a-judge paradigm.
Tianbin Li, Yanzhou Su, Wei Li +15 authors
GMAI-VL, a vision-language model trained on the GMAI-VL-5.5M dataset, achieves top performance in multimodal medical tasks by integrating visual and textual information.
Zhangchen Xu, Fengqing Jiang, Luyao Niu +2 authors
The Larger Models' Paradox reveals that larger models are not always better teachers for fine-tuning smaller models, and a new metric, Compatibility-Adjusted Reward (CAR), is introduced to measure and improve the effectiveness of response generators.
Mouxiang Chen, Lefei Shen, Zhuo Li +3 authors
The VisionTS model, leveraging a visual masked autoencoder pre-trained on ImageNet, achieves state-of-the-art zero-shot and fine-tuned forecasting performance in time series forecasting without adapting specifically to the time-series domain.
Weiquan Huang, Aoqi Wu, Yifan Yang +8 authors
LLM2CLIP leverages large language models to enhance CLIP's multimodal representation learning, enabling better handling of complex image captions by fine-tuning the LLM and using contrastive learning.
Bo Peng, Daniel Goldstein, Quentin Anthony +24 authors
Eagle and Finch, improvements over RWKV-4, use multi-headed matrix-valued states and dynamic recurrence for enhanced expressivity and efficiency, trained on a multilingual corpus.
Fanqi Wan, Ziyi Yang, Longguang Zhong +3 authors
FuseChat extends the FuseLLM framework for knowledge fusion of chat LLMs through lightweight fine-tuning and parameter merging, achieving superior performance across various domains.
Lin Xu, Yilin Zhao, Daquan Zhou +3 authors
Pooling LLaVA improves video-language pre-trained models by smoothing high-norm visual features, achieving state-of-the-art performance in video question-answering, captioning, and benchmark tasks.
Xiaoyu Shi, Zhaoyang Huang, Fu-Yun Wang +9 authors
Motion-I2V, a diffusion-based framework for image-to-video generation, achieves consistency and controllability through explicit motion modeling and trajectory prediction.
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