Neura Market
CDial-GPT logo

CDial-GPT

Free

A Large-scale Chinese Short-Text Conversation Dataset and Chinese pre-training dialog models

Writing AssistantsFreeFree tier
Inputs: textOutputs: text
Type
Open Source
GitHub (1.9k stars)

About CDial-GPT

CDial-GPT is an open-source project from Tsinghua University's COAI group that provides a large-scale Chinese short-text conversation dataset and pre-trained dialog models based on the GPT architecture. The dataset, called LCCC (Large-scale Cleaned Chinese Conversation), contains over 12 million Chinese conversation pairs sourced from social media and other platforms, which have been cleaned and filtered to improve quality. The project offers pre-trained models (CDial-GPT, CDial-GPT2, and CDial-GPT2-large) that can be fine-tuned for various Chinese dialogue tasks, such as response generation and conversational AI development. As an open-source tool, it is freely available on GitHub for research and non-commercial use, with the code and models accessible for download and modification.

Key Features

Large-scale Chinese short-text conversation dataset (LCCC) with over 12 million cleaned dialogue pairs
Pre-trained Chinese dialogue models based on GPT and GPT-2 architectures
Multiple model sizes available (CDial-GPT, CDial-GPT2, CDial-GPT2-large)
Open-source code and model weights on GitHub
Designed for Chinese language dialogue generation and conversational AI research
Dataset cleaning and filtering to reduce noise from social media sources

Pros & Cons

Pros
  • Open-source and freely available for research and non-commercial use
  • Provides a large, cleaned Chinese conversation dataset that is otherwise scarce
  • Offers pre-trained models that can be fine-tuned, saving training time and resources
  • Backed by a reputable academic group (Tsinghua University COAI)
  • Multiple model sizes allow flexibility based on computational resources
Cons
  • Primarily designed for Chinese language; limited utility for other languages
  • Requires significant computational resources for fine-tuning larger models
  • Dataset may contain biases or noise from social media sources despite cleaning
  • Documentation and community support may be limited compared to commercial tools
  • Not intended for production deployment without further adaptation and testing

Best For

Research in Chinese dialogue generation and conversational AIBuilding Chinese chatbots or virtual assistantsFine-tuning for domain-specific Chinese conversation tasksAcademic study of pre-trained language models for dialogueDeveloping Chinese language understanding and generation systems

Alternatives to CDial-GPT

ChatLM-mini-Chinese

中文对话0.2B小模型（ChatLM-Chinese-0.2B），开源所有数据集来源、数据清洗、tokenizer训练、模型预训练、SFT指令微调、RLHF优化等流程的全部代码。支持下游任务sft微调，给出三元组信息抽取微调示例。

Chat-Haruhi-Suzumiya

Chat凉宫春日, An open sourced Role-Playing chatbot Cheng Li, Ziang Leng, and others.

DialoGPT

Large-scale pretraining for dialogue

GPT2-chitchat

GPT2 for Chinese chitchat/用于中文闲聊的GPT2模型(实现了DialoGPT的MMI思想)

GPT2-Chinese

Chinese version of GPT2 training code, using BERT tokenizer.

gpt-2-simple

Python package to easily retrain OpenAI's GPT-2 text-generating model on new texts

FAQ

What is CDial-GPT?
CDial-GPT is an open-source project that provides a large-scale Chinese short-text conversation dataset (LCCC) and pre-trained dialog models based on GPT, developed by Tsinghua University's COAI group.
Is CDial-GPT free to use?
Based on available information, CDial-GPT is open-source and appears to be free for research and non-commercial use. Users should verify the license terms on the GitHub repository for specific usage conditions.
What models are available?
The project offers pre-trained models including CDial-GPT, CDial-GPT2, and CDial-GPT2-large, which are based on the GPT and GPT-2 architectures and trained on the LCCC dataset.
Can I use CDial-GPT for commercial applications?
The licensing terms should be checked on the GitHub repository. As an open-source project, it may have restrictions on commercial use; users should verify this before deploying in a commercial product.
What is the LCCC dataset?
LCCC (Large-scale Cleaned Chinese Conversation) is a dataset of over 12 million Chinese short-text conversation pairs, cleaned and filtered from social media sources to improve quality for dialogue research.
How do I get started with CDial-GPT?
Users can download the code and models from the GitHub repository (https://github.com/thu-coai/CDial-GPT) and follow the provided documentation for installation and fine-tuning instructions.