Alert button
Picture for Guangxuan Xiao

Guangxuan Xiao

Alert button

BitDelta: Your Fine-Tune May Only Be Worth One Bit

Feb 15, 2024
James Liu, Guangxuan Xiao, Kai Li, Jason D. Lee, Song Han, Tri Dao, Tianle Cai

Viaarxiv icon

InfLLM: Unveiling the Intrinsic Capacity of LLMs for Understanding Extremely Long Sequences with Training-Free Memory

Feb 07, 2024
Chaojun Xiao, Pengle Zhang, Xu Han, Guangxuan Xiao, Yankai Lin, Zhengyan Zhang, Zhiyuan Liu, Song Han, Maosong Sun

Viaarxiv icon

Efficient Streaming Language Models with Attention Sinks

Sep 29, 2023
Guangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han, Mike Lewis

Figure 1 for Efficient Streaming Language Models with Attention Sinks
Figure 2 for Efficient Streaming Language Models with Attention Sinks
Figure 3 for Efficient Streaming Language Models with Attention Sinks
Figure 4 for Efficient Streaming Language Models with Attention Sinks
Viaarxiv icon

FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention

May 21, 2023
Guangxuan Xiao, Tianwei Yin, William T. Freeman, Frédo Durand, Song Han

Figure 1 for FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention
Figure 2 for FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention
Figure 3 for FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention
Figure 4 for FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention
Viaarxiv icon

Sparse and Local Networks for Hypergraph Reasoning

Mar 09, 2023
Guangxuan Xiao, Leslie Pack Kaelbling, Jiajun Wu, Jiayuan Mao

Figure 1 for Sparse and Local Networks for Hypergraph Reasoning
Figure 2 for Sparse and Local Networks for Hypergraph Reasoning
Figure 3 for Sparse and Local Networks for Hypergraph Reasoning
Figure 4 for Sparse and Local Networks for Hypergraph Reasoning
Viaarxiv icon

Offsite-Tuning: Transfer Learning without Full Model

Feb 09, 2023
Guangxuan Xiao, Ji Lin, Song Han

Figure 1 for Offsite-Tuning: Transfer Learning without Full Model
Figure 2 for Offsite-Tuning: Transfer Learning without Full Model
Figure 3 for Offsite-Tuning: Transfer Learning without Full Model
Figure 4 for Offsite-Tuning: Transfer Learning without Full Model
Viaarxiv icon

ReFresh: Reducing Memory Access from Exploiting Stable Historical Embeddings for Graph Neural Network Training

Jan 19, 2023
Kezhao Huang, Haitian Jiang, Minjie Wang, Guangxuan Xiao, David Wipf, Xiang Song, Quan Gan, Zengfeng Huang, Jidong Zhai, Zheng Zhang

Figure 1 for ReFresh: Reducing Memory Access from Exploiting Stable Historical Embeddings for Graph Neural Network Training
Figure 2 for ReFresh: Reducing Memory Access from Exploiting Stable Historical Embeddings for Graph Neural Network Training
Figure 3 for ReFresh: Reducing Memory Access from Exploiting Stable Historical Embeddings for Graph Neural Network Training
Figure 4 for ReFresh: Reducing Memory Access from Exploiting Stable Historical Embeddings for Graph Neural Network Training
Viaarxiv icon

SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models

Nov 28, 2022
Guangxuan Xiao, Ji Lin, Mickael Seznec, Julien Demouth, Song Han

Figure 1 for SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
Figure 2 for SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
Figure 3 for SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
Figure 4 for SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
Viaarxiv icon

Red Alarm for Pre-trained Models: Universal Vulnerabilities by Neuron-Level Backdoor Attacks

Jan 19, 2021
Zhengyan Zhang, Guangxuan Xiao, Yongwei Li, Tian Lv, Fanchao Qi, Zhiyuan Liu, Yasheng Wang, Xin Jiang, Maosong Sun

Figure 1 for Red Alarm for Pre-trained Models: Universal Vulnerabilities by Neuron-Level Backdoor Attacks
Figure 2 for Red Alarm for Pre-trained Models: Universal Vulnerabilities by Neuron-Level Backdoor Attacks
Figure 3 for Red Alarm for Pre-trained Models: Universal Vulnerabilities by Neuron-Level Backdoor Attacks
Figure 4 for Red Alarm for Pre-trained Models: Universal Vulnerabilities by Neuron-Level Backdoor Attacks
Viaarxiv icon