|
Yixin Song 宋奕欣 Bringing AGI to life with local GPU 💓 Ph.D., SJTU yixinsong@sjtu.edu.cn / jeremysyx@gmail.com
I am a Ph.D. candidate at Shanghai Jiao Tong University (SJTU),
focusing on Large Language Models (LLMs).
I am a member of the Institute of Parallel and Distributed Systems (IPADS),
where I am fortunate to be supervised by Prof. Zeyu
Mi and Prof. Haibo Chen.
My research focuses on algorithm and hardware co-design for efficient model serving and training systems. CV / Google Scholar / HuggingFace / Github |
|
News
|
Selected Publications
*Equal contribution. |
|
|
Powerinfer: Fast large language model serving with a consumer-grade gpu
Yixin Song, Zeyu Mi, Haotong Xie, Haibo Chen SOSP, 2024 project page / arXiv / code 
PowerInfer is a CPU/GPU LLM inference engine leveraging activation locality for your device. |
|
|
PowerInfer-2: Fast Large Language Model Inference on a Smartphone
Zhenliang Xue*, Yixin Song*, Zeyu Mi, Le Chen, Yubin Xia, Haibo Chen arxiv, 2024 project page / arXiv PowerInfer-2 is a highly optimized inference framework designed specifically for smartphones. |
|
ReLU^2 Wins: Discovering Efficient Activation Functions for Sparse LLMs
Zhengyan Zhang* Yixin Song*, Guanghui Yu, Xu Han Yankai Lin, Chaojun Xiao, Chenyang Song, Zhiyuan Liu, Zeyu Mi, Maosong Sun arxiv, 2024 aixiv We introduce a general method that defines neuron activation through neuron output magnitudes and a tailored magnitude threshold, demonstrating that non-ReLU LLMs also exhibit sparse activation. |
|
|
Bamboo: A New 7B Mistral-level Open LLM with High Sparsity
Yixin Song, Haotong Xie, Zeyu Mi, Li Ma, Haibo Chen project, 2023 github We introduce Bamboo-v0.1, a new 7B LLM that boasts high sparsity while delivering performance equivalent to Mistral-7B. This repo provides the details of the model. |
Working Experience |
|
Shanghai AI Laboratory
Researcher Intern, 2022.10 ~ Present |
Miscellanea |