Výsledky vyhledávání - "HUANG Wenxuan"

Report

Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsification

Autor: Huang, Wenxuan, Zhai, Zijie, Shen, Yunhang, Cao, Shaosheng, Zhao, Fei, Xu, Xiangfeng, Ye, Zheyu, Lin, Shaohui

Multimodal Large Language Models (MLLMs) have achieved remarkable success in vision understanding, reasoning, and interaction. However, the inference computation and memory increase progressively with the generation of output tokens during decoding,

Externí odkaz: http://arxiv.org/abs/2412.00876

Zobrazit plný text záznamu

Report

A General and Efficient Training for Transformer via Token Expansion

Autor: Huang, Wenxuan, Shen, Yunhang, Xie, Jiao, Zhang, Baochang, He, Gaoqi, Li, Ke, Sun, Xing, Lin, Shaohui

The remarkable performance of Vision Transformers (ViTs) typically requires an extremely large training cost. Existing methods have attempted to accelerate the training of ViTs, yet typically disregard method universality with accuracy dropping. Mean

Externí odkaz: http://arxiv.org/abs/2404.00672

Zobrazit plný text záznamu

Report

Filter Pruning for Efficient CNNs via Knowledge-driven Differential Filter Sampler

Autor: Lin, Shaohui, Huang, Wenxuan, Xie, Jiao, Zhang, Baochang, Shen, Yunhang, Yu, Zhou, Han, Jungong, Doermann, David

Filter pruning simultaneously accelerates the computation and reduces the memory overhead of CNNs, which can be effectively applied to edge devices and cloud services. In this paper, we propose a novel Knowledge-driven Differential Filter Sampler~(KD

Externí odkaz: http://arxiv.org/abs/2307.00198

Zobrazit plný text záznamu