9 papers
ProphetKV: User-Query-Driven Selective Recomputation for Efficient KV Cache Reuse in Retrieval-Augmented Generation
Shihao Wang, Jiahao Chen, Yanqi Pan +7
The prefill stage of long-context Retrieval-Augmented Generation (RAG) is severely bottlenecked by computational overhead. To mitigate this, recent methods assemble pre-calculated…
Reasoning-Style Poisoning of LLM Agents via Stealthy Style Transfer: Process-Level Attacks and Runtime Monitoring in RSV Space
Xingfu Zhou, Pengfei Wang
Large Language Model (LLM) agents relying on external retrieval are increasingly deployed in high-stakes environments. While existing adversarial attacks primarily focus on content…
DreamActor-H1: High-Fidelity Human-Product Demonstration Video Generation via Motion-designed Diffusion Transformers
Lizhen Wang, Zhurong Xia, Tianshu Hu +6
In e-commerce and digital marketing, generating high-fidelity human-product demonstration videos is important for effective product presentation. However, most existing frameworks…
GLAD: Generalizable Tuning for Vision-Language Models
Yuqi Peng, Pengfei Wang, Jianzhuang Liu +1
Pre-trained vision-language models, such as CLIP, show impressive zero-shot recognition ability and can be easily transferred to specific downstream tasks via prompt tuning, even w…
Exploring Implicit Visual Misunderstandings in Multimodal Large Language Models through Attention Analysis
Pengfei Wang, Guohai Xu, Weinong Wang +3
Recent advancements have enhanced the capability of Multimodal Large Language Models (MLLMs) to comprehend multi-image information. However, existing benchmarks primarily evaluate…
Qwen2.5-VL Technical Report
Shuai Bai, Keqin Chen, Xuejing Liu +24
We introduce Qwen2.5-VL, the latest flagship model of Qwen vision-language series, which demonstrates significant advancements in both foundational capabilities and innovative func…