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。
Pooyan Rahmanzadehgervi, Logan Bolton, Mohammad Reza Taesiri +1 authors
State-of-the-art large language models with vision capabilities perform poorly on simple visual tasks, indicating limitations in their visual understanding.
50 篇论文 · 按点赞排序
Lucas Beyer, Andreas Steiner, André Susano Pinto +32 authors
PaliGemma, a versatile Vision-Language Model based on SigLIP-So400m and Gemma-2B, demonstrates strong performance across numerous open-world tasks, including specialized areas like remote sensing and segmentation.
Pujiang He, Shan Zhou, Wenhuan Huang +7 authors
The paper presents a performance optimization solution for deploying large language models on CPUs by reducing KV cache size and implementing distributed inference, open-sourced for use.
Haiwen Diao, Yufeng Cui, Xiaotong Li +3 authors
EVE is an encoder-free vision-language model that achieves competitive performance on multiple benchmarks using a unified decoder and extra supervision.
Zhaorun Chen, Yichao Du, Zichen Wen +16 authors
MJ-Bench evaluates multimodal judges across alignment, safety, image quality, and bias for text-to-image models and confirms close-source VLMs generally provide better feedback.
Liang Zeng, Liangjun Zhong, Liang Zhao +9 authors
Skywork-Math model series achieves high performance in math reasoning by fine-tuning on a large and diverse dataset, demonstrating that data quantity significantly enhances LLM mathematical capabilities.
Arindam Mitra, Luciano Del Corro, Guoqing Zheng +11 authors
AgentInstruct, an agentic framework, generates high-quality synthetic data for post-training language models, leading to significant performance improvements across various benchmarks.
Mihir Prabhudesai, Russell Mendonca, Zheyang Qin +2 authors
Utilizing pre-trained reward models to adapt video diffusion models with gradient-based feedback enhances efficiency and performance compared to gradient-free methods.
Wenqi Zhang, Zhenglin Cheng, Yuanyu He +8 authors
A multi-modal self-instruct benchmark using large language models and code capabilities enhances understanding and reasoning for abstract images and visual tasks, improving performance metrics like chart understanding and map navigation.
Feng Li, Renrui Zhang, Hao Zhang +5 authors
LLaVA-NeXT-Interleave addresses multi-image, multi-frame, multi-view, and multi-patch scenarios in LMMs using the M4-Instruct dataset and LLaVA-Interleave Bench to achieve leading results while maintaining single-image performance.
Tongyi SpeechTeam
FunAudioLLM enhances voice interactions by integrating SenseVoice for multilingual speech recognition, emotion detection, and audio event detection with CosyVoice for natural speech generation across languages, timbres, and styles.
Yinquan Lu, Wenhao Zhu, Lei Li +2 authors
LLaMAX enhances multilingual translation by extensive continual pre-training, achieving superior performance in 100+ languages without sacrificing generalization.
Ivan Rodkin, Yuri Kuratov, Aydar Bulatov +1 authors
The Associative Recurrent Memory Transformer effectively handles long sequences by combining transformer self-attention with segment-level recurrence, achieving high accuracy in long-context tasks like BABILong.
Ali Hatamizadeh, Jan Kautz
MambaVision, a hybrid Mamba-Transformer backbone, enhances visual feature modeling with self-attention blocks, achieving state-of-the-art performance in image classification and downstream tasks.
Yu Sun, Xinhao Li, Karan Dalal +9 authors
Test-Time Training (TTT) layers achieve linear complexity and competitive performance on long context tasks, surpassing both Transformers and RNNs in certain scenarios.
Zhenyu Zhang, Ajay Jaiswal, Lu Yin +4 authors
Q-GaLore reduces memory usage for training large language models by combining quantization and low-rank projection, surpassing GaLore and LoRA in terms of efficiency and performance.
Petr Anokhin, Nikita Semenov, Artyom Sorokin +3 authors
AriGraph, a memory graph integrating semantic and episodic memories, enhances the exploratory and planning capabilities of an LLM-based agent, outperforming existing methods in complex zero-shot tasks.
Renrui Zhang, Xinyu Wei, Dongzhi Jiang +9 authors
The paper introduces MAVIS, a comprehensive framework for improving the visual mathematics problem-solving capabilities of multi-modal large language models through tailored datasets and specialized training stages.
Benno Krojer, Dheeraj Vattikonda, Luis Lara +4 authors
AURORA Dataset and benchmark improve general image editing by focusing on action, reasoning, and minimal visual changes, outperforming previous models with better automatic metrics.
Weize Chen, Ziming You, Ran Li +7 authors
Internet of Agents (IoA) framework facilitates seamless collaboration among heterogeneous large language models (LLMs) in distributed environments, outperforming existing baselines.
Orr Zohar, Xiaohan Wang, Yonatan Bitton +2 authors
Video Self-Training with augmented Reasoning (Video-STaR) enhances Large Vision Language Models by leveraging labeled video datasets for video instruction tuning, improving general video understanding and downstream task performance.
Bowen Zhang, Yiji Cheng, Chunyu Wang +6 authors
RodinHD generates high-fidelity 3D avatars from portraits by addressing catastrophic forgetting in triplanes and enhancing detail capture through hierarchical representation and cross-attention in 3D diffusion models.
Peng Xia, Kangyu Zhu, Haoran Li +5 authors
RULE improves factual accuracy in Medical Large Vision Language Models by calibrating retrieved context selection and fine-tuning model preference.
Ahmed Masry, Megh Thakkar, Aayush Bajaj +3 authors
ChartGemma, a novel chart understanding model, outperforms existing methods by training directly on chart images and using a vision-language backbone, achieving state-of-the-art results in summarization, question answering, and fact-checking tasks.
Tim R. Davidson, Viacheslav Surkov, Veniamin Veselovsky +3 authors
A study assesses self-recognition in language models using model-generated security questions, revealing no evidence of self-recognition and emphasizing models' preference for the "best" answer.
Shuai Yang, Yuying Ge, Yang Li +4 authors
SEED-Story uses a Multimodal Large Language Model with multimodal attention to generate extended multimodal stories from interleaved image-text sequences.
Zihao Zhou, Shudong Liu, Maizhen Ning +6 authors
MATHCHECK introduces a comprehensive checklist and tool for evaluating mathematical reasoning and generalization of large language models across various tasks, improving on traditional benchmarks.
Shaltiel Shmidman, Avi Shmidman, Amir DN Cohen +1 authors
DictaLM2.0 and DictaLM2.0-Instruct, derived from the Mistral model, are fine-tuned on Hebrew and English using novel methodologies to address challenges in low-resource language training, evaluated on a comprehensive Hebrew benchmark suite.
Ethan Chern, Jiadi Su, Yan Ma +1 authors
Anole, an autoregressive large multimodal model based on Meta AI's Chameleon, supports efficient and high-quality interleaved image-text generation using native integration and an innovative fine-tuning strategy.
Young-Jun Lee, Dokyong Lee, Junyoung Youn +4 authors
A large-scale dataset and framework are introduced to capture long-term multi-modal conversations, enhancing a model's visual imagination ability.
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