yangjianxin1/GPT2-chitchat
A Chinese casual chat system implementing a GPT2-based dialogue model with DialoGPT's Mutual Mutual Information decoding strategy.

This repository provides a Chinese chitchat system built on GPT2 architecture, implemented using HuggingFace Transformers. The project adopts the Mutual Mutual Information (MMI) approach from DialoGPT to improve dialogue response quality. It supports multiple generation strategies including Temperature, Top-k Sampling, and Nucleus Sampling. The author is active in Chinese NLP research, maintaining multiple related projects covering large language models, multimodal models, and text generation systems.