ChatPLUG: Chinese Personalized Large Language Model

This is the repo for the ChatPLUG project, which aims to build and share a Chinese open-domain dialogue system.

爱用emoji的萌妹子小婉	富有智慧的得道高僧	会说古文的的三国NPC关羽

News

ChatPLUG-13B will soon be released on ModelScope for research.
[2023/05/23] Add guides about retrieval and role-play
[2023/05/10] Add training code, which helps train custom models and build chatbots quickly and easily.
[2023/04/26] Try our Role-Play-Chat Online Demo in ModelScope Now!
[2023/04/19] Add content including spotlights, results and limitations. Upload models to ModelScope.
[2023/04/16] Initialize project.

Online Demo

Role-Play-Chat

Spotlights

spotlights

Compared with existed open-source models, we highlight three feaures of ChatPLUG as follows:

Knowledge Augmentation

It's flexible to integrate external knowledge during inference, and this is an optional input. You can utilize a search engine to acquire up-to-date information or use a local knowledge base to obtain domain knowledge.

Personalization

It's easy to customize the style of conversations and characters by setting bot profiles or using role-paly instructions.

Multi Skills

It exhibits its proficiency in open-domain dialogue through mulit-turn conversation, while also displaying impressive multi-task abilities on a wide range of NLP tasks.

How to run

We offer three methods to use or continue developing ChatPLUG as follows:

	Getting Started	Inference	Train	Deploy
ModelScope	Easy	✔️ Cli	Not Ready	Not Ready
HuggingFace	Medium	✔️ Cli	Not Ready	Not Ready
XDPX	Hard	✔️ Cli	✔️ Support	✔️ Serving

ModelScope

You can download and use ChatPLUG models from ModelScope.

Model Name	URL
ChatPLUG-240M	ChatPLUG-开放域对话模型-240M
ChatPLUG-3.7B	ChatPLUG-开放域对话模型-3.7B

HuggingFace

Coming soon.

XDPX

XDPX is an easy-to-use library, that allows researchers and developers to train custom models and build own chatbots in a streamlined manner. Its all-in-one functionality allows for a one-stop solution that simplifies complex processes. quick start

One-Click Inference

When using ChatPLUG-3.7B, you can set core_chat_half_precision : true to save memory.

# Requirement
# in the dir of XDPX
cd XDPX
pip install -e .

# Download checkpoints
# in the same dir as the download.sh
cd ..
sh download.sh

# Inference
# in the dir of XDPX
cd XDPX
CUDA_VISIBLE_DEVICES=0 x-script fidchat_new chat_pipeline/chatplug_3.7B_sftv2.6.0_instruction.hjson
# input `#exit` and exit the terminal

One-Click Train

If your GPU(e.g. A100、A10) support bf16, set deepspeed_bf16: true and deepspeed_fp16: false, otherwise set deepspeed_bf16: false and deepspeed_fp16: true

# 1. Download dataset from belle
# in ChatPLUG/data/belle dir
cd data/belle
git lfs install
git clone https://huggingface.co/datasets/BelleGroup/train_0.5M_CN

python process_belle_0.5M.py 
# $ls data/belle 
# train_0.jsonl dev.jsonl ...

# 2. Preprocess Data 
# in XDPX dir
x-prepro chat_pipeline/chatplug_prepro_sft_instruction.hjson
# $ls data/dialogue/sft/chatplug/belle_instruction 
# train_0.pt dev.pt

# 3. Training
# in XDPX dir
x-train chat_pipeline/chatplug_3.7B_train_sftv2.6.0_instruction.hjson

One-Click Deploy

Coming soon.

Installation

Please refer to Installation for installation instructions.

For detailed user guides, please refer to our documentation:

User Guides
- Quick Start
- Retrieval
- Role-Play
Advanced Guides
- Intro
- Training
- Develop
- F.A.Q

Citations

If you find our project useful in your work, please cite:

  @misc{tian2023chatplug,
        title={ChatPLUG: Open-Domain Generative Dialogue System with Internet-Augmented Instruction Tuning for Digital Human}, 
        author={Junfeng Tian and Hehong Chen and Guohai Xu and Ming Yan and Xing Gao and Jianhai Zhang and Chenliang Li and Jiayi Liu and Wenshen Xu and Haiyang Xu and Qi Qian and Wei Wang and Qinghao Ye and Jiejing Zhang and Ji Zhang and Fei Huang and Jingren Zhou},
        year={2023},
        eprint={2304.07849},
        archivePrefix={arXiv},
        primaryClass={cs.CL}
  }

@misc{plug2021,
  title = {{PLUG: Pre-training for Language Understanding and Generation}},
  author={ModelScope},
  publisher = {ModelScope},
  journal = {ModelScope repository},
  year = {2021},
  howpublished = {url{https://modelscope.cn/models/damo/nlp_plug_text-generation_27B/summary}},
}