[Adapt] [seminar] Length Reduction for Efficient Transformer Inference

[任思宇]

Hi Adapters,


Large-scale pre-trained language models have significantly advanced the whole NLP community. Nevertheless, their excellent performance comes with intensive computation and tremendous parameter counts, which hinder the deployment of PLMs in resource-limited environments. There are mainly two ways to make PLMs more efficient so as to meet the practical requirement: static model compression and dynamic computation. Today I will introduce one line of research falling under the dynamic computation category: length reduction. By reducing the input length, the run-time memory and speed of PLMs can be improved. Hope you find this talk useful.


Time: Wed 4:00 pm
Venue: SEIEE 3-414
Best Regards,
Roy

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.sjtu.edu.cn/pipermail/adapt/attachments/20221019/7a941aab/attachment.htm>