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。
Yaowei Zheng, Richong Zhang, Junhao Zhang +2 authors
LlamaFactory is a unified framework enabling efficient fine-tuning of large language models across various tasks using a web-based user interface.
50 篇论文 · 按点赞排序
Zhengqing Yuan, Ruoxi Chen, Zhaoxu Li +4 authors
A new multi-agent framework Mora replicates Sora's generalist video generation capabilities using several advanced visual AI agents, achieving similar performance across various tasks.
Tianjun Zhang, Shishir G. Patil, Naman Jain +4 authors
Retrieval Augmented FineTuning (RAFT) enhances pre-trained large language models' in-domain performance by filtering out irrelevant documents and citing relevant passages.
Axel Sauer, Frederic Boesel, Tim Dockhorn +3 authors
Latent Adversarial Diffusion Distillation (LADD) simplifies the training and improves the performance of diffusion models by using generative features from latent diffusion models, enabling fast and high-resolution image synthesis.
Hakim Sidahmed, Samrat Phatale, Alex Hutcheson +16 authors
Parameter-efficient reinforcement learning using LoRA in RLHF matches conventional RLHF performance with reduced computational cost and memory usage.
Takuya Akiba, Makoto Shing, Yujin Tang +2 authors
An evolutionary algorithm automates model merging, discovering effective combinations of open-source models without additional training data or compute, achieving state-of-the-art performance on benchmarks and culturally-aware tasks.
Hengxing Cai, Xiaochen Cai, Shuwen Yang +14 authors
Uni-SMART, a multimodal analysis model, outperforms LLMs in understanding scientific literature with diverse content and has applications in patent infringement and chart analysis.
Renrui Zhang, Dongzhi Jiang, Yichi Zhang +8 authors
MathVerse is a comprehensive visual math benchmark with CoT evaluation to assess MLLMs' true understanding and reasoning quality from visual diagrams.
Xiaohan Wang, Yuhui Zhang, Orr Zohar +1 authors
VideoAgent, an agent-based system using a large language model, significantly enhances long-form video understanding through interactive reasoning and efficient zero-shot accuracy on EgoSchema and NExT-QA benchmarks.
Junliang Ye, Fangfu Liu, Qixiu Li +5 authors
A framework called DreamReward uses human preference feedback to enhance text-to-3D generation, producing high-fidelity and aligned 3D models.
Han Zhao, Min Zhang, Wei Zhao +3 authors
Cobra, a linear-complexity multimodal large language model, integrates the Mamba language model with visual modality, achieving competitive performance and faster inference speed compared to existing state-of-the-art models.
Payel Das, Subhajit Chaudhury, Elliot Nelson +10 authors
Larimar, a brain-inspired architecture with distributed episodic memory, enhances LLMs for efficient, fast, and accurate knowledge updates, fact editing, and context length generalization.
Anwen Hu, Haiyang Xu, Jiabo Ye +8 authors
A Unified Structure Learning framework is introduced to enhance Visual Document Understanding by integrating structure-aware parsing and multi-grained text localization tasks using a vision-to-text module H-Reducer, achieving state-of-the-art results on multiple benchmarks.
Swapnaja Achintalwar, Ioana Baldini, Djallel Bouneffouf +16 authors
An Alignment Studio architecture empowers developers to tailor language models to specific values, social norms, laws, and guidelines by integrating Framers, Instructors, and Auditors.
Armen Avetisyan, Christopher Xie, Henry Howard-Jenkins +11 authors
SceneScript generates full scene models using structured language commands via an encoder-decoder architecture, achieving top performance in architectural layout estimation and competitive results in 3D object detection.
Max Ku, Cong Wei, Weiming Ren +2 authors
AnyV2V, a training-free framework, leverages existing image editing and generation models to perform a wide range of video editing tasks, including prompt-based editing and novel reference-based style transfer, with improved performance and versatility.
Baifeng Shi, Ziyang Wu, Maolin Mao +2 authors
Running smaller vision models at multiple image scales can achieve state-of-the-art performance on various tasks, often surpassing larger models.
Zhuoshi Pan, Qianhui Wu, Huiqiang Jiang +10 authors
A task-agnostic prompt compression method uses data distillation and a Transformer encoder to achieve efficient compression with minimal information loss and improved generalization.
Siying Cui, Jiankang Deng, Jia Guo +4 authors
IDAdapter, a tuning-free method, enhances the diversity and identity preservation in personalized image generation using a single face image by integrating personalized textual and visual injections and a face identity loss.
Aonan Zhang, Chong Wang, Yi Wang +2 authors
Improved speculative decoding uses a single lightweight draft head with recurrent dependency, combining simplicity and efficiency of single-model design while avoiding Medusa's complexity.
Nathan Lambert, Valentina Pyatkin, Jacob Morrison +9 authors
A benchmark dataset and code-base, RewardBench, evaluates reward models to understand alignment with human preferences, revealing capabilities and limitations of different training methods.
Mengting Wan, Tara Safavi, Sujay Kumar Jauhar +11 authors
TnT-LLM is a two-phase framework using Large Language Models to automate label generation and classification with minimal human effort, demonstrating superior accuracy and efficiency in text mining tasks.
Daniel Garibi, Or Patashnik, Andrey Voynov +2 authors
ReNoise method improves image inversion accuracy in text-guided diffusion models without increasing operations, demonstrating effectiveness and editability.
Sihyun Yu, Weili Nie, De-An Huang +3 authors
A new content-motion latent diffusion model (CMD) enhances video generation quality and efficiency by decoupling content and motion and leveraging pretrained image diffusion models.
Wenqiao Zhang, Tianwei Lin, Jiang Liu +10 authors
HyperLLaVA improves multimodal language model performance through adaptive tuning and dynamic experts derived from HyperNetworks.
Vikram Voleti, Chun-Han Yao, Mark Boss +6 authors
Stable Video 3D (SV3D) achieves high-resolution, multi-view 3D object generation by adapting image-to-video diffusion models and improving 3D optimization techniques.
Michael Niemeyer, Fabian Manhardt, Marie-Julie Rakotosaona +7 authors
RadSplat combines radiance fields and Gaussian splatting for efficient real-time rendering of complex scenes, achieving state-of-the-art performance.
Yi Wu, Ziqiang Li, Heliang Zheng +2 authors
Infinite-ID decouples identity and semantics in image generation, using enhanced training with image cross-attention and feature interaction to preserve identity fidelity and enhance semantic consistency.
Ming Gui, Johannes S. Fischer, Ulrich Prestel +6 authors
A flow matching approach using pre-trained image diffusion models achieves state-of-the-art monocular depth estimation with high quality and low computational cost, leveraging synthetic data and surface normals loss.
Vincent Tao Hu, Stefan Andreas Baumann, Ming Gui +4 authors
Zigzag Mamba, an improvement over Mamba, enhances visual data generation with better spatial continuity, speed, and memory usage, and is scalable with Stochastic Interpolant on large-resolution datasets.
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