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。
Keivan Alizadeh, Iman Mirzadeh, Dmitry Belenko +5 authors
Efficient inference for large language models on devices with limited DRAM by optimizing data transfer and access from flash memory.
403 篇论文 · 按点赞排序
Hugo Touvron, Louis Martin, Kevin Stone +65 authors
Llama 2, a series of pretrained and fine-tuned large language models, achieves superior performance in dialogue tasks compared to open-source alternatives and offers safety enhancements.
Grégoire Mialon, Clémentine Fourrier, Craig Swift +3 authors
GAIA benchmarks general AI assistants using real-world questions that challenge both reasoning and multi-modality handling, showcasing a significant gap between human and AI performance.
Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler +1 authors
A method using 3D Gaussians for scene representation and optimized rendering allows high-quality, real-time novel-view synthesis at 1080p resolution.
Dongsheng Wang, Natraj Raman, Mathieu Sibue +6 authors
DocLLM extends large language models for visual document understanding by focusing on bounding box information and disentangled matrices for multimodal attention, achieving state-of-the-art performance on document intelligence tasks.
Yutao Sun, Li Dong, Shaohan Huang +5 authors
Retentive Network (RetNet) offers a foundation for large language models with parallel training, low-cost inference, and efficient long-sequence modeling through a retention mechanism supporting parallel, recurrent, and chunkwise recurrent paradigms.
Jade Copet, Felix Kreuk, Itai Gat +5 authors
MusicGen, a single-stage transformer language model, generates high-quality music conditioned on text or melodic features using efficient token interleaving patterns, outperforming existing models on text-to-music benchmarks.
Suriya Gunasekar, Yi Zhang, Jyoti Aneja +16 authors
A new compact Transformer-based large language model for code, phi-1, achieves high accuracy on coding benchmarks despite having fewer parameters than competing models.
Albert Gu, Tri Dao
Mamba, a novel SSM-based model, outperforms Transformers in inference speed and scalability across various modalities by selectively propagating information and using efficient hardware-aware algorithms.
Peter Belcak, Roger Wattenhofer
FastBERT achieves significant inference speedup over BERT by selectively engaging a small fraction of neurons using fast feedforward networks.
Shuai Yang, Yifan Zhou, Ziwei Liu +1 authors
A novel framework adapts image diffusion models for video by generating key frames with hierarchical constraints and propagating them using patch matching and blending, achieving high-quality and temporally-coherent videos.
Hongyu Wang, Shuming Ma, Li Dong +7 authors
BitNet, a 1-bit Transformer architecture, reduces memory and energy consumption while achieving competitive performance in language modeling.
Yichun Shi, Peng Wang, Jianglong Ye +3 authors
MVDream generates geometrically consistent multi-view images from text prompts using pre-trained image diffusion models and Score Distillation Sampling, improving 3D generation stability and supporting personalized generation.
Yujia Qin, Shihao Liang, Yining Ye +15 authors
ToolLLM is a framework that enhances open-source LLMs with tool-use capabilities through a comprehensive dataset, efficient decision tree, automatic evaluation, and neural API retrieval.
Bin Xiao, Haiping Wu, Weijian Xu +6 authors
A new prompt-based vision foundation model, Florence-2, is introduced for diverse vision and vision-language tasks, achieving strong zero-shot and fine-tuning capabilities with comprehensive annotations.
Dustin Podell, Zion English, Kyle Lacey +5 authors
SDXL, a latent diffusion model using a larger UNet with additional text encoders and attention mechanisms, improves text-to-image synthesis significantly.
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.
Yukang Chen, Shengju Qian, Haotian Tang +4 authors
LongLoRA extends LLM context sizes efficiently using sparse local attention and parameter-efficient fine-tuning, demonstrating strong results with minimal computational cost.
Thuat Nguyen, Chien Van Nguyen, Viet Dac Lai +5 authors
CulturaX is a large multilingual dataset designed to improve the training and quality of large language models through rigorous data cleaning and deduplication.
Timothée Darcet, Maxime Oquab, Julien Mairal +1 authors
Additional input tokens in Vision Transformers mitigate artifacts in feature maps, enhancing performance and enabling smoother visual processing.
Simian Luo, Yiqin Tan, Suraj Patil +6 authors
LCMs enhance text-to-image generation by distilling from LDMs using LoRA for reduced memory and superior quality, and introduce LCM-LoRA as a plug-in accelerator for various tasks.
Grégoire Delétang, Anian Ruoss, Paul-Ambroise Duquenne +9 authors
Large language models demonstrate strong compression capabilities, outperforming domain-specific compressors and offering new insights into scaling laws, tokenization, and in-context learning.
Bowen Peng, Jeffrey Quesnelle, Honglu Fan +1 authors
YaRN extends the context window of transformer-based language models like LLaMA with improved efficiency and performance.
Liang Wang, Nan Yang, Xiaolong Huang +3 authors
A method using synthetic data and contrastive loss achieves high-quality text embeddings without labeled data, and sets state-of-the-art results when combined with labeled data.
Yuxiang Wei, Zhe Wang, Jiawei Liu +2 authors
Magicoder, using OSS-Instruct to incorporate open-source code snippets, achieves superior performance on coding benchmarks while reducing bias in synthetic data generation.
Daixuan Cheng, Shaohan Huang, Furu Wei
Transforming raw corpora into reading comprehension texts enhances large language models' question-answering abilities across various domains and improves performance on general benchmarks.
Jiayu Ding, Shuming Ma, Li Dong +4 authors
LongNet, a Transformer variant with dilated attention, enables scaling to over 1 billion tokens with linear computational complexity and maintains performance on shorter sequences.
Ramakrishna Bairi, Atharv Sonwane, Aditya Kanade +6 authors
CodePlan automates repository-level coding tasks, such as package migration and temporal code edits, using a planning framework that leverages LLMs with context derived from code repositories and change analysis.
Chengrun Yang, Xuezhi Wang, Yifeng Lu +4 authors
OPRO, a method using large language models to optimize tasks described in natural language, outperforms human-designed prompts on various benchmark datasets.
Anton Razzhigaev, Arseniy Shakhmatov, Anastasia Maltseva +7 authors
Kandinsky1, a latent diffusion architecture, achieves high-quality text-to-image generation by integrating image prior models and modified MoVQ autoencoders, outperforming other open-source models.
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