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。
Marah Abdin, Sam Ade Jacobs, Ammar Ahmad Awan +81 authors
Phi-3-mini, a compact 3.8 billion parameter language model, achieves competitive performance with larger models through an enhanced training dataset and alignment.
44 篇论文 · 按点赞排序
Sachin Mehta, Mohammad Hossein Sekhavat, Qingqing Cao +8 authors
OpenELM, an open language model, uses layer-wise scaling to enhance accuracy and is released with comprehensive training and evaluation frameworks, enabling transparent and reproducible research.
Mostafa Elhoushi, Akshat Shrivastava, Diana Liskovich +10 authors
LayerSkip, an end-to-end method, accelerates large language model inference through layer dropout and self-speculative decoding without auxiliary layers.
Xun Wu, Shaohan Huang, Wenhui Wang +1 authors
MH-MoE enhances SMoE by splitting tokens into sub-tokens processed by diverse experts in parallel, improving expert activation, context understanding, and overfitting, and demonstrating effectiveness across various modeling tasks.
Zhe Chen, Weiyun Wang, Hao Tian +24 authors
InternVL 1.5, an open-source multimodal language model, improves visual understanding and high-resolution image processing through continuous learning, dynamic tile division, and a high-quality bilingual dataset, achieving state-of-the-art results in several benchmarks.
Shengnan An, Zexiong Ma, Zeqi Lin +2 authors
FILM-7B enhances long-context processing by using information-intensive training with synthesized datasets, improving retrieval and performance on real-world tasks without compromising short-context performance.
Wei Huang, Xudong Ma, Haotong Qin +7 authors
LLaMA3 performance degrades significantly when quantized to low bit-width, highlighting challenges in low-bit quantization for large language models.
Wenhao Huang, Chenghao Peng, Zhixu Li +4 authors
AutoCrawler combines LLMs with crawlers to efficiently handle diverse and changing web environments using a hierarchical HTML structure for progressive learning.
Eric Wallace, Kai Xiao, Reimar Leike +3 authors
A hierarchical instruction framework for LLMs increases robustness against malicious prompts by selectively prioritizing system instructions over user inputs.
Zhen Zeng, William Watson, Nicole Cho +4 authors
FlowMind uses Large Language Models with a generic prompt recipe to generate automatic workflows, addressing spontaneous tasks and ensuring data integrity, and it is evaluated using a new financial dataset NCEN-QA.
Raehyuk Jung, Hyojun Go, Jaehyuk Yi +41 authors
Pegasus-1 is a multimodal language model designed for video content comprehension, handling spatiotemporal information and demonstrated in benchmarks for video conversation, zero-shot video question answering, and video summarization.
Zhen Ye, Zeqian Ju, Haohe Liu +10 authors
FlashSpeech, using latent consistency model and adversarial consistency training, achieves fast and high-quality zero-shot speech synthesis with enhanced prosody generation.
Chuofan Ma, Yi Jiang, Jiannan Wu +2 authors
Groma, a Multimodal Large Language Model, utilizes region-level visual tokenization to enhance visual understanding and localization accuracy in image-grounded tasks.
Jingqun Tang, Chunhui Lin, Zhen Zhao +13 authors
Text-centric VQA performance is significantly improved by a large, high-quality instruction-tuning dataset, Square-10M, generated through a four-step process, leading to state-of-the-art results on multiple benchmarks.
Yuxi Ren, Xin Xia, Yanzuo Lu +5 authors
Hyper-SD distills Diffusion Models by preserving and reformulating ODE trajectories, incorporating human feedback, and using score distillation, achieving state-of-the-art performance in step compression.
Sachin Mehta, Maxwell Horton, Fartash Faghri +5 authors
A novel weakly supervised pre-training method for vision models on web-scale data reframes the task as classification to avoid pairwise similarity computations, significantly speeding up training while maintaining high representation quality.
Yuhong Li, Yingbing Huang, Bowen Yang +6 authors
SnapKV reduces KV cache size and computational overhead in LLMs by compressing the cache based on observed attention patterns, improving decoding speed and memory efficiency without significant performance loss.
Zinan Guo, Yanze Wu, Zhuowei Chen +2 authors
PuLID combines a Lightning T2I branch with a diffusion model to provide ID customization in text-to-image generation, maintaining ID fidelity and consistency in image elements.
Tianyuan Zhang, Hong-Xing Yu, Rundi Wu +5 authors
PhysDreamer utilizes video generation models to simulate realistic 3D object dynamics based on learned action-conditioned priors, enhancing immersive virtual interactions.
Amirmojtaba Sabour, Sanja Fidler, Karsten Kreis
A novel method optimizes sampling schedules for diffusion models, improving their performance across various generative tasks beyond hand-crafted heuristics.
Tamar Rott Shaham, Sarah Schwettmann, Franklin Wang +4 authors
MAIA, a multimodal automated interpretability agent, uses neural models to perform feature interpretation and failure mode discovery for other models, demonstrating comparable results to human experimenters and aiding in reducing sensitivity to spurious features and identifying potential misclassifications.
Shaocong Dong, Lihe Ding, Zhanpeng Huang +3 authors
Interactive3D framework enhances 3D object generation by allowing precise user control through direct interaction capabilities and iterative refinement.
Anej Svete, Ryan Cotterell
Transformer models using hard or sparse attention mechanisms can represent any $n$-gram model, providing insight into their probabilistic representational capacity.
Yuying Ge, Sijie Zhao, Jinguo Zhu +6 authors
SEED-X is a versatile multimodal foundation model that enhances vision-language understanding and generation by comprehending diverse image sizes and enabling multi-granularity image generation.
Jiehui Huang, Xiao Dong, Wenhui Song +10 authors
ConsistentID enables high-fidelity personalized facial generation with improved identity consistency through multimodal facial prompts and facial attention localization strategy.
Xiang Li, Yiqun Yao, Xin Jiang +17 authors
An open-sourced 52B multilingual large language model, Tele-FLM, features efficient pre-training and superior language capabilities, and is comparable to larger models in terms of performance metrics.
An Yan, Zhengyuan Yang, Junda Wu +8 authors
A new paradigm, "list items one by one," enhances multimodal large language models' visual reasoning capability through the use of visual tags in training, even when these tags are omitted during inference.
Olivia Wiles, Chuhan Zhang, Isabela Albuquerque +8 authors
This study evaluates the alignment of text-to-image models with prompts by introducing a skills-based benchmark, collecting extensive human ratings, and developing a new QA-based auto-evaluation metric.
Ruilong Li, Sanja Fidler, Angjoo Kanazawa +1 authors
NeRF-XL enables the scalable training and rendering of large-capacity NeRFs across multiple GPUs, demonstrating improved reconstruction quality and speed on extensive datasets.
Zhengcong Fei, Mingyuan Fan, Junshi Huang
Music Consistency Models (MusicCM) efficiently synthesize high-quality music by leveraging consistency models, reducing computational requirements and enabling real-time application.
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