EVA2.0: Investigating Open-Domain Chinese Dialogue Systems with Large-Scale Pre-Training

TL;DR

EVA2.0 is a large-scale Chinese dialogue model that leverages extensive experiments to optimize data quality, architecture, training, and decoding, significantly advancing human-like chatbot capabilities.

Contribution

The paper introduces EVA2.0, a 2.8-billion-parameter pre-trained Chinese dialogue system, with comprehensive analysis of key factors affecting performance and public release of models and code.

Findings

EVA2.0 outperforms existing open-source Chinese dialogue models.

Extensive experiments reveal critical factors influencing dialogue system performance.

Human and automatic evaluations confirm EVA2.0's superior conversational abilities.

Abstract

Large-scale pre-training has shown remarkable performance in building open-domain dialogue systems. However, previous works mainly focus on showing and evaluating the conversational performance of the released dialogue model, ignoring the discussion of some key factors towards a powerful human-like chatbot, especially in Chinese scenarios. In this paper, we conduct extensive experiments to investigate these under-explored factors, including data quality control, model architecture designs, training approaches, and decoding strategies. We propose EVA2.0, a large-scale pre-trained open-domain Chinese dialogue model with 2.8 billion parameters, and will make our models and codes publicly available. Automatic and human evaluations show that EVA2.0 significantly outperforms other open-source counterparts. We also discuss the limitations of this work by presenting some failure cases and pose some future research directions on large-scale Chinese open-domain dialogue systems.