PLATO-XL: Exploring The Large-scale Pre-training Of Dialogue Generation

Bao Siqi, He Huang, Wang Fan, Wu Hua, Wang Haifeng, Wu Wenquan, Wu Zhihua, Guo Zhen, Lu Hua, Huang Xinxian, Tian Xin, Xu Xinchao, Lin Yingzhan, Niu Zheng-yu. Arxiv 2021

[Paper]    
Efficiency And Optimization Model Architecture Pretraining Methods Training Techniques Transformer

To explore the limit of dialogue generation pre-training, we present the models of PLATO-XL with up to 11 billion parameters, trained on both Chinese and English social media conversations. To train such large models, we adopt the architecture of unified transformer with high computation and parameter efficiency. In addition, we carry out multi-party aware pre-training to better distinguish the characteristic information in social media conversations. With such designs, PLATO-XL successfully achieves superior performances as compared to other approaches in both Chinese and English chitchat. We further explore the capacity of PLATO-XL on other conversational tasks, such as knowledge grounded dialogue and task-oriented conversation. The experimental results indicate that PLATO-XL obtains state-of-the-art results across multiple conversational tasks, verifying its potential as a foundation model of conversational AI.

Similar Work