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。
Tim Dettmers, Artidoro Pagnoni, Ari Holtzman +1 authors
QLoRA enables efficient finetuning of large language models using 4-bit quantization and Low Rank Adapters, achieving high performance with reduced memory usage.
50 篇论文 · 按点赞排序
Chunting Zhou, Pengfei Liu, Puxin Xu +12 authors
A 65B parameter LLaMa language model trained with minimal instructional data matches or outperforms models with extensive human preference modeling in most cases, indicating that pretraining is predominantly responsible for knowledge acquisition.
Bo Peng, Eric Alcaide, Quentin Anthony +27 authors
A new architecture, RWKV, combines the parallelizable training of Transformers with the efficient inference of RNNs, achieving linear scaling and matching performance.
Guanzhi Wang, Yuqi Xie, Yunfan Jiang +5 authors
Voyager is an LLM-powered agent in Minecraft that autonomously explores, learns, and discovers skills through a curriculum, skill library, and prompting mechanism, demonstrating superior lifelong learning and task-solving capabilities.
Zhengyi Wang, Cheng Lu, Yikai Wang +4 authors
Variational score distillation addresses issues in text-to-3D generation by modeling 3D parameters as random variables, improving sample quality and diversity.
Zhibin Gou, Zhihong Shao, Yeyun Gong +4 authors
CRITIC, a framework that uses external tools to validate and improve LLM outputs, consistently enhances performance across various tasks.
Ning Ding, Yulin Chen, Bokai Xu +6 authors
A large-scale dataset, UltraChat, and a fine-tuned model, UltraLLaMA, are created to improve open-source chat language models' performance.
Yabo Zhang, Yuxiang Wei, Dongsheng Jiang +3 authors
ControlVideo, a training-free framework adapted from ControlNet, improves text-to-video generation by ensuring appearance coherence, reducing flicker, and efficiently producing long videos.
Arnav Gudibande, Eric Wallace, Charlie Snell +5 authors
Finetuning open-source language models on outputs from proprietary models like ChatGPT shows improved instruction-following but fails to close the performance gap on unsupported tasks.
Liying Cheng, Xingxuan Li, Lidong Bing
GPT-4 demonstrates comparable performance to human data analysts in end-to-end data analysis tasks across various domains.
Tiedong Liu, Bryan Kian Hsiang Low
A fine-tuned LLaMA model, Goat, outperforms GPT-4 in arithmetic tasks through supervised learning and task decomposition.
Hiroki Furuta, Ofir Nachum, Kuang-Huei Lee +3 authors
An instruction-following multimodal agent named WebGUM, trained offline using vision transformers and language models, significantly outperforms existing methods in web navigation and multi-step reasoning tasks.
Bo-Kyeong Kim, Hyoung-Kyu Song, Thibault Castells +1 authors
Classical architectural compression and knowledge distillation reduce the size and computational cost of Stable Diffusion models while maintaining competitive text-to-image synthesis performance.
Weifeng Chen, Jie Wu, Pan Xie +5 authors
Video-ControlNet is a controllable T2V diffusion model that uses spatial-temporal self-attention and residual-based noise initialization to generate high-quality, consistent videos conditioned on control signals.
Michael Hassid, Tal Remez, Tu Anh Nguyen +9 authors
TWIST improves SpeechLMs by leveraging pretrained textual language models, demonstrating superior performance across various evaluations and introducing new benchmarks for spoken data.
Kevin Black, Michael Janner, Yilun Du +2 authors
Reinforcement learning method DDPO optimizes diffusion models directly for downstream objectives such as image quality and feedback-driven alignment.
Byungjun Kim, Patrick Kwon, Kwangho Lee +4 authors
A 3D avatar generation pipeline utilizes diffusion models for dual normal map creation and 3D reconstruction, enhanced with text-based identity control.
Zineng Tang, Ziyi Yang, Chenguang Zhu +2 authors
CoDi is a generative model that can produce and condition on any combination of modalities by aligning them through a shared multimodal space in the diffusion process.
Jooyoung Choi, Yunjey Choi, Yunji Kim +2 authors
Custom-Edit enhances text-guided image editing by fine-tuning only language-relevant diffusion model parameters to improve reference similarity while preserving source similarity.
Yufan Zhou, Ruiyi Zhang, Tong Sun +1 authors
A novel framework for customizing text-to-image generation without regularization demonstrates superior performance and fine detail preservation.
Muru Zhang, Ofir Press, William Merrill +2 authors
Language models frequently generate incorrect answers and offer false explanations, exacerbating errors due to a phenomenon called hallucination snowballing.
Guy Yariv, Itai Gat, Lior Wolf +2 authors
A novel method using latent diffusion models and an audio encoding model generates images from audio recordings, outperforming baseline methods in both objective and subjective metrics.
Ronak Pradeep, Kai Hui, Jai Gupta +5 authors
Generative retrieval scales to large corpora with significant findings on the importance of synthetic queries and the limits of parameter scaling.
Jingbo Zhang, Xiaoyu Li, Ziyu Wan +2 authors
Text2NeRF generates high-fidelity, photo-realistic 3D scenes from text prompts by integrating NeRF with a text-to-image diffusion model and monocular depth estimation.
Liangchen Song, Liangliang Cao, Hongyu Xu +4 authors
RoomDreamer synthesizes 3D indoor scenes with desired styles by aligning geometry and texture using Geometry Guided Diffusion and Mesh Optimization.
Max W. Y. Lam, Qiao Tian, Tang Li +10 authors
MeLoDy is an LM-guided diffusion model that generates high-quality music with reduced computational cost and improved sampling speed compared to MusicLM.
Ahmed A. Elhag, Joshua M. Susskind, Miguel Angel Bautista
Manifold Diffusion Fields (MDF) learns generative models of continuous functions on Riemannian manifolds using eigen-functions of the Laplace-Beltrami Operator, achieving better diversity and fidelity.
Eric Hedlin, Gopal Sharma, Shweta Mahajan +4 authors
Text-to-image diffusion models leverage semantic knowledge to find correspondences between images without training, achieving state-of-the-art results on multiple datasets.
Philippe Laban, Wojciech Kryściński, Divyansh Agarwal +4 authors
A new benchmark called SummEdits shows that most large language models struggle to detect factual inconsistencies effectively, highlighting gaps in their reasoning abilities.
Vijayaraghavan Murali, Chandra Maddila, Imad Ahmad +5 authors
CodeCompose, an AI-assisted coding tool using the InCoder LLM, enhances code authoring by offering suggestions and improving documentation while achieving high user acceptance and positive feedback at Meta.
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