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。
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.
42 篇论文 · 按点赞排序
Shilong Liu, Hao Cheng, Haotian Liu +10 authors
LLaVA-Plus, a general-purpose multimodal assistant, enhances large multimodal models by integrating pre-trained vision and vision-language models, performing tool-assisted tasks and improving interaction through direct image grounding.
Yicong Hong, Kai Zhang, Jiuxiang Gu +7 authors
A Large Reconstruction Model using a transformer-based architecture predicts 3D neural radiance fields from single images using massive multi-view training data.
Meredith Ringel Morris, Jascha Sohl-dickstein, Noah Fiedel +5 authors
The framework proposes a classification system for AGI capabilities and behavior using levels of performance and generality, with principles to assess risks and measure progress.
Hanoona Rasheed, Muhammad Maaz, Sahal Shaji +7 authors
GLaMM is a multimodal model that generates visually grounded language responses with object segmentation masks from both text and optional visual prompts.
Bo Li, Peiyuan Zhang, Jingkang Yang +3 authors
OtterHD-8B, an advanced multimodal model from Fuyu-8B, excels in processing high-resolution inputs and discerning detailed spatial relationships through MagnifierBench, an evaluation framework highlighting the importance of vision encoder flexibility.
Shiwei Zhang, Jiayu Wang, Yingya Zhang +6 authors
The cascaded I2VGen-XL approach uses hierarchical encoders and text-image pairs to improve the semantic accuracy, continuity, and clarity of generated videos.
In Gim, Guojun Chen, Seung-seob Lee +3 authors
Prompt Cache improves inference speed for large language models by reusing precomputed attention states of common text segments in user prompts.
Ying Sheng, Shiyi Cao, Dacheng Li +9 authors
S-LoRA is a system that allows for efficient and scalable serving of numerous LoRA adapters using a unified memory pool, tensor parallelism, and custom CUDA kernels.
Weihan Wang, Qingsong Lv, Wenmeng Yu +13 authors
CogVLM, a visual language foundation model, uses a trainable visual expert module to deeply integrate vision and language without compromising NLP performance.
Qinghao Ye, Haiyang Xu, Jiabo Ye +7 authors
mPLUG-Owl2, a versatile multi-modal large language model, uses a modularized network design to enhance performance in both text and multi-modal tasks through modality collaboration and modality-adaptive modules.
Yuvanesh Anand, Zach Nussbaum, Adam Treat +6 authors
GPT4All is an open-source project aimed at democratizing access to large language models, addressing infrastructure costs and limited accessibility of state-of-the-art models.
Yiduo Guo, Zekai Zhang, Yaobo Liang +2 authors
The PowerPoint Task Completion (PPTC) benchmark assesses LLMs' ability to complete multi-turn, multi-modal instructions for editing PPT files and identifies challenges such as error accumulation, template processing, and multi-modality perception.
Zhen Yang, Yingxue Zhang, Fandong Meng +1 authors
TEAL, a method that tokenizes and jointly embeds multi-modal inputs, enables existing frozen LLMs to handle multi-modal understanding and generation tasks with improved performance.
Ruihang Lai, Junru Shao, Siyuan Feng +16 authors
Relax is a compiler abstraction that optimizes dynamic shape computations in machine learning workloads, enabling competitive performance and broader deployment of large language models.
Ruyi Gan, Ziwei Wu, Renliang Sun +8 authors
Ziya2, a 13-billion-parameter language model built on LLaMA2 and further pre-trained on 700 billion tokens, achieves superior performance across multiple benchmarks using data-centric optimization techniques.
Linning Xu, Vasu Agrawal, William Laney +10 authors
The system captures, reconstructs, and renders high-fidelity walkable spaces using neural radiance fields, achieving high-quality real-time VR rendering with a custom dataset and multi-camera rig.
Ching-Feng Yeh, Po-Yao Huang, Vasu Sharma +2 authors
FLAP is a self-supervised method using masking, contrastive learning, and reconstruction to align audio and language representations, achieving state-of-the-art performance in audio-text retrieval tasks.
Ruomeng Ding, Chaoyun Zhang, Lu Wang +7 authors
XoT, combining pretrained reinforcement learning and Monte Carlo Tree Search, enhances Large Language Models' capabilities for flexible, efficient problem-solving with minimal interactions.
Jiawei Yang, Boris Ivanovic, Or Litany +8 authors
EmerNeRF uses neural fields with self-bootstrapping to learn and represent static and dynamic components of driving scenes efficiently, achieving state-of-the-art performance in sensor simulation and semantic generalization.
Tony Lee, Michihiro Yasunaga, Chenlin Meng +15 authors
A new benchmark evaluates text-to-image models across 12 aspects, revealing no single model excels comprehensively in all areas.
Ao Zhang, Liming Zhao, Chen-Wei Xie +3 authors
The pixel2emb method enables large multimodal models to effectively handle object location tasks by outputting location embeddings, leading to superior performance in multimodal conversations and various visual processing tasks.
Jinjin Xu, Liwu Xu, Yuzhe Yang +4 authors
A framework u-LLaVA integrates visual information into LLMs using modality alignment and multi-task modules to address hallucinations and task interference, achieving state-of-the-art performance.
Norman Mu, Sarah Chen, Zifan Wang +5 authors
RuLES is a framework for evaluating rule-following behaviors in LLMs using programmatically defined scenarios and adversarial inputs, revealing vulnerabilities across different models and attack strategies.
Mason Wang, Samuel Clarke, Jui-Hsien Wang +2 authors
SoundCam, a large dataset of real-world room impulse responses and music recordings with varying object positions, enables tasks like detecting and tracking human positions.
Qingru Zhang, Chandan Singh, Liyuan Liu +4 authors
PASTA enhances LLM performance by directing model attention to user-specified parts through post-hoc attention reweighting without altering model parameters.
Chenfeng Xu, Huan Ling, Sanja Fidler +1 authors
3DiffTection enhances 3D object detection from single images using a 3D-aware diffusion model with specialized tuning strategies, achieving state-of-the-art performance and robust generalization.
Joseph Suárez, Phillip Isola, Kyoung Whan Choe +15 authors
A flexible reinforcement learning environment with procedurally generated maps supports generalization to new tasks, maps, and opponents, featuring improved performance and compatibility with CleanRL.
Licheng Wen, Xuemeng Yang, Daocheng Fu +14 authors
Visual Language Models (VLMs) demonstrate superior scene understanding and causal reasoning in autonomous driving but face challenges in specific vision tasks like direction discernment and traffic light recognition.
Bingchang Liu, Chaoyu Chen, Cong Liao +9 authors
MFTcoder, a multi-task fine-tuning framework, improves the coding capabilities of LLMs through parallel fine-tuning on multiple tasks, achieving better performance than single-task fine-tuning and GPT-4 in zero-shot evaluation.
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