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。
Yuanzhi Li, Sébastien Bubeck, Ronen Eldan +3 authors
A new 1.3 billion parameter Transformer-based language model, phi-1.5, demonstrates comparable performance to much larger models on common sense reasoning and complex tasks despite the absence of web data.
36 篇论文 · 按点赞排序
Shengqiong Wu, Hao Fei, Leigang Qu +2 authors
NExT-GPT, an any-to-any Multimodal Large Language Model, combines LLMs with multimodal adaptors and diffusion decoders to generate content across various modalities, enhanced by modality-switching instruction tuning and a curated dataset.
Woosuk Kwon, Zhuohan Li, Siyuan Zhuang +6 authors
PagedAttention algorithm and vLLM system enhance the throughput of large language models by efficiently managing memory and reducing waste in the key-value cache.
Brendan Walsh, Mark Hamilton, Greg Newby +8 authors
A system leverages neural text-to-speech to automatically generate high-quality, customizable audiobooks from e-books, significantly expanding access to literature.
Zhengqi Li, Richard Tucker, Noah Snavely +1 authors
A frequency-coordinated diffusion sampling process is used to predict long-term motion representations for still images, enabling dynamic video creation and interactive scene manipulation.
Li Chen, Mengyi Zhao, Yiheng Liu +8 authors
PhotoVerse uses a dual-branch conditioning mechanism and facial identity loss to generate high-quality, identity-preserving images efficiently with minimal tuning.
Wangchunshu Zhou, Yuchen Eleanor Jiang, Long Li +14 authors
Agents is an open-source library that facilitates the creation, customization, and deployment of autonomous language agents with features like planning, memory, and tool usage, aiming to make advances in large language models accessible to a broader audience.
Xingchao Liu, Xiwen Zhang, Jianzhu Ma +2 authors
Rectified Flow is used to develop an ultra-fast one-step text-to-image generator named InstaFlow, achieving high image quality with significantly reduced inference time compared to existing methods.
Griffin Adams, Alexander Fabbri, Faisal Ladhak +2 authors
CoD prompt in GPT-4 improves summary density, informativeness, and human preference over vanilla prompts, with tradeoffs between readability and detail.
Haohe Liu, Ke Chen, Qiao Tian +2 authors
A diffusion-based generative model, AudioSR, achieves robust audio super-resolution across various audio types and bandwidths, enhancing generation quality for different audio models.
Junha Hyung, Jaeyo Shin, Jaegul Choo
MagiCapture personalizes text-to-image models by integrating subject and style concepts using limited references and an Attention Refocusing loss, enhancing realism and quality in portrait image generation.
Dave Van Veen, Cara Van Uden, Louis Blankemeier +12 authors
Domain-adapted large language models outperform human experts in various clinical text summarization tasks, improving completeness, correctness, and aligning with physician preferences.
Sneha Kudugunta, Isaac Caswell, Biao Zhang +8 authors
A multilingual translation model trained on a large dataset of 250 billion tokens across over 450 languages outperforms larger models, with competitive results in various domains and effective few-shot translation performance.
Chris Cummins, Volker Seeker, Dejan Grubisic +8 authors
A transformer model trained to optimize LLVM assembly code size significantly improves compiler performance with auxiliary learning tasks, achieving better results than baselines and demonstrating strong code reasoning abilities.
Qingxiu Dong, Li Dong, Ke Xu +4 authors
LLMs, using Socratic reasoning, generate a proof schema and engage in rigorous dialogue to address the P vs. NP problem, aligning with existing research.
Tuan Dung Nguyen, Yuan-Sen Ting, Ioana Ciucă +21 authors
AstroLLaMA, a fine-tuned large language model using astronomy abstracts, significantly reduces perplexity and improves text completions and embeddings in specialized astronomical domains.
Elena Voita, Javier Ferrando, Christoforos Nalmpantis
The analysis of the OPT family of large language models reveals that early network layers are sparse, with many neurons never activating, and some neurons functioning as token detectors that also remove information about triggering tokens.
Sacha Lewin, Maxime Vandegar, Thomas Hoyoux +2 authors
Dynamic NeRFs show promise for photorealistic novel view synthesis of soccer scenes, though they do not yet meet the quality standards required by the broadcast industry.
Max Marion, Ahmet Üstün, Luiza Pozzobon +3 authors
Perplexity outperforms more complex methods for data quality estimation in pruning large language model training datasets, resulting in performance improvement with reduced data.
Wenhua Cheng, Weiwei Zhang, Haihao Shen +3 authors
SignRound, a lightweight block-wise tuning method using signed gradient descent, optimizes weight rounding in weight-only quantization of Large Language Models, outperforming RTN and other recent methods without inference overhead.
Erik Daxberger, Floris Weers, Bowen Zhang +7 authors
Sparse Mobile Vision MoEs are applied to reduce resource requirements in Vision Transformers, achieving improved performance and efficiency compared to dense models.
Tianqi Liu, Yao Zhao, Rishabh Joshi +4 authors
A novel approach called Statistical Rejection Sampling Optimization (RSO) enhances preference data sourcing for aligning language models with human preferences, outperforming previous methods like SLiC and DPO.
Johannes von Oswald, Eyvind Niklasson, Maximilian Schlegel +9 authors
Transformers' superior performance might be due to mesa-optimization, an internal learning objective and its solution during the forward pass, which can be repurposed for few-shot tasks and potentially enhances language models' in-context learning.
Namhyuk Ahn, Junsoo Lee, Chunggi Lee +4 authors
Introduction of DreamStyler, a framework combining text-to-image synthesis and style transfer, enhances artistic image quality and flexibility using context-aware text prompts.
Youwei Liang, Ruiyi Zhang, Li Zhang +1 authors
A graph neural network and large language model integrated system, DrugChat, enables interactive queries about drug compounds,加速ing pharmaceutical research.
Kenneth Shaw, Ananye Agarwal, Deepak Pathak
LEAP Hand, a low-cost anthropomorphic robotic hand, outperforms more expensive alternatives in real-world manipulation tasks including visual teleoperation and learning from video data.
Haotian Yang, Mingwu Zheng, Wanquan Feng +5 authors
A tracking-free framework captures and reconstructs high-fidelity 3D avatars with realistic relighting and real-time animation using a novel network architecture optimized for capturing dynamic lighting conditions.
Benjamin Elizalde, Soham Deshmukh, Huaming Wang
A Contrastive Language-Audio Pretraining model using innovative encoders achieves state-of-the-art results in various audio and language tasks through joint multimodal representation learning.
Geng Lin, Chen Gao, Jia-Bin Huang +4 authors
A novel video matting technique, OmnimatteRF, combines 2D foreground layers with a 3D background model to improve scene reconstruction in dynamic real-world videos.
Yi-Ling Qiao, Alexander Gao, Yiran Xu +3 authors
Rendering and simulating mesh assets within NeRF volumes integrates mesh and NeRF light transport in an efficient, GPU-accelerated system for enhanced visual realism.
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