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。
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.
50 篇论文 · 按点赞排序
Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux +23 authors
Mixtral 8x7B, a Sparse Mixture of Experts language model, achieves superior performance across benchmarks by using a selective architecture that leverages fewer active parameters.
Weizhe Yuan, Richard Yuanzhe Pang, Kyunghyun Cho +3 authors
A study on Self-Rewarding Language Models shows that using LLM-as-a-Judge prompting for iterative DPO training enhances both instruction-following and self-reward generation, leading to superior performance compared to existing systems.
Peiyuan Zhang, Guangtao Zeng, Tianduo Wang +1 authors
TinyLlama, a compact 1.1B language model, leverages FlashAttention to achieve high performance in downstream tasks with enhanced computational efficiency.
Omer Bar-Tal, Hila Chefer, Omer Tov +11 authors
A text-to-video diffusion model using Space-Time U-Net architecture generates realistic, diverse, and coherent videos through a single pass, achieving state-of-the-art results and supporting various content creation and editing tasks.
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.
Fanghua Yu, Jinjin Gu, Zheyuan Li +6 authors
SUPIR, an advanced image restoration method combining generative prior and model scaling, achieves high-quality results and flexible restoration guided by textual prompts.
Daya Guo, Qihao Zhu, Dejian Yang +10 authors
DeepSeek-Coder, a series of open-source code models trained from scratch, outperforms both open-source and closed-source models across multiple benchmarks.
Maciej Pióro, Kamil Ciebiera, Krystian Król +2 authors
A combination of Mixture of Experts (MoE) with State Space Models (SSMs) improves model efficiency and performance compared to both standalone SSMs and MoE-enhanced Transformers.
Saleh Ashkboos, Maximilian L. Croci, Marcelo Gennari do Nascimento +2 authors
SliceGPT, a novel post-training sparsification method, reduces model parameters by up to 25% without significant performance decrease, optimizing inference speed and GPU usage on large language models.
Lichao Sun, Yue Huang, Haoran Wang +64 authors
This study assesses the trustworthiness of large language models across various dimensions, including truthfulness, safety, fairness, robustness, privacy, and machine ethics, finding a positive correlation with utility and highlighting differences between proprietary and open-source models.
Zixiang Chen, Yihe Deng, Huizhuo Yuan +2 authors
A novel self-play fine-tuning method (SPIN) enhances Large Language Models by generating its own training data and refining its policy, achieving performance on par with models trained using direct preference optimization and additional human data.
Mehmet Saygin Seyfioglu, Karim Bouyarmane, Suren Kumar +2 authors
Diffuse to Choose is a diffusion-based inpainting model that balances real-time performance with high-fidelity detail preservation and semantic accuracy.
Yiheng Liu, Hao He, Tianle Han +18 authors
The paper reviews techniques for cost-efficient training and deployment of large language models, covering aspects like data preprocessing, parallel training, model fine-tuning, and inference optimizations including model compression and memory scheduling.
Lihe Yang, Bingyi Kang, Zilong Huang +3 authors
Depth Anything is a robust monocular depth estimation model built on a large-scale dataset with data augmentation and auxiliary supervision strategies, achieving state-of-the-art results on various datasets and enhancing depth-conditioned ControlNet.
Damai Dai, Chengqi Deng, Chenggang Zhao +14 authors
The DeepSeekMoE architecture improves expert specialization in Mixture-of-Experts models by segmenting experts and isolating shared ones, achieving better performance and computational efficiency compared to GShard and other models.
Lianghui Zhu, Bencheng Liao, Qian Zhang +3 authors
A new vision backbone using bidirectional Mamba blocks (Vim) outperforms established vision transformers like DeiT in terms of performance and efficiency on high-resolution image tasks.
Junxiong Wang, Tushaar Gangavarapu, Jing Nathan Yan +1 authors
MambaByte, a token-free byte-level autoregressive model, demonstrates computational efficiency and competitive performance compared to subword token-based models, with the added benefit of fast inference due to linear length scaling.
Tianle Cai, Yuhong Li, Zhengyang Geng +4 authors
Medusa enhances Large Language Model inference by adding parallel decoding heads and using tree-based attention to predict multiple tokens simultaneously, achieving significant speedup with minimal latency.
Qixun Wang, Xu Bai, Haofan Wang +2 authors
InstantID is a diffusion model-based solution for personalized image synthesis that uses a single facial image and integrates with pre-trained models while maintaining high face fidelity and efficiency.
DeepSeek-AI, Xiao Bi, Deli Chen +84 authors
DeepSeek LLM, an open-source language model project, develops a large dataset and employs SFT and DPO to achieve performance surpassing LLaMA-2 70B and GPT-3.5 in various benchmarks and open-ended evaluations.
Xiaoyi Dong, Pan Zhang, Yuhang Zang +20 authors
InternLM-XComposer2, a vision-language model, excels in free-form text-image creation and understanding using Partial LoRA, surpassing existing models and matching GPT-4V and Gemini Pro in multimodal benchmarks.
Bin Lin, Zhenyu Tang, Yang Ye +6 authors
MoE-LLaVA, a sparse Large Vision-Language Model architecture, improves performance and reduces hallucinations with fewer active parameters compared to larger models.
Chengyue Wu, Yukang Gan, Yixiao Ge +5 authors
A new post-pretraining method using expanded Transformer blocks for Large Language Models improves knowledge without catastrophic forgetting, yielding LLaMA Pro-8.3B that excels in general tasks, programming, and mathematics.
Jun Zhao, Zhihao Zhang, Qi Zhang +2 authors
A comprehensive study on transferring English-dominant LLMs to non-English languages using LLaMA demonstrates that comparable performance can be achieved with minimal pretraining data across various language capabilities and response quality metrics.
Pratyush Maini, Skyler Seto, He Bai +3 authors
Web Rephrase Augmented Pre-training (WRAP) improves large language model pre-training efficiency and performance by incorporating synthetic rephrases styled like Wikipedia or question-answer format into the training dataset.
Xiaoding Lu, Adian Liusie, Vyas Raina +2 authors
A combination of smaller chat AI models, through a method called blending, can achieve performance comparable to or better than large single models with lower computational resources.
Duzhen Zhang, Yahan Yu, Chenxing Li +4 authors
A survey examining the design, performance, and future directions of MultiModal Large Language Models (MM-LLMs) to enhance reasoning and MM task capabilities.
Moab Arar, Andrey Voynov, Amir Hertz +5 authors
Prompt-aligned personalization improves text alignment and enhances personalized image creation with complex prompts by using score distillation sampling.
Junsong Chen, Yue Wu, Simian Luo +5 authors
PIXART-{\delta} integrates Latent Consistency Model (LCM) and ControlNet into PIXART-{\alpha} for accelerated, high-quality text-to-image synthesis with efficient training and inference.
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