README_en.md

Psychological Counselor's Digital Twin（SoulChat2.0）

[ English | 中文 ]

News

• 👏🏻 2024.12.19: Our paper has released on the arXiv: PsyDT: Using LLMs to Construct the Digital Twin of Psychological Counselor with Personalized Counseling Style for Psychological Counseling

Data Download

• PsyDTCorpus

Model Download

In order to facilitate further research in the mental health dialogue research community, we plan to open source a series of psychological counselor digital twin models that have undergone full parameter fine-tuning, as shown in the table below：

Models	Download Link	Foundation Model Download Link
SoulChat2.0-Qwen2-7B	download from modelscope	Qwen2-7B-Instruct
SoulChat2.0-internlm2-7b	download from modelscope	internlm2-chat-7b
SoulChat2.0-Baichuan2-7B	download from modelscope	Baichuan2-7B-Chat
SoulChat2.0-Llama-3.1-8B	download from modelscope	Meta-Llama-3.1-8B-Instruct
SoulChat2.0-Llama-3-8B	download from modelscope	Meta-Llama-3-8B-Instruct
SoulChat2.0-Yi-1.5-9B	download from modelscope	Yi-1.5-9B-Chat-16K
SoulChat2.0-glm-4-9b	download from modelscope	glm-4-9b-chat

Model Inference

Assuming the IP address of your server is 198.0.0.8

vllm inference

SERVER_MODEL_NAME=SoulChat2.0-Llama-3.1-8B
MODEL_NAME_OR_PATH=<local path>/SoulChat2.0-Llama-3.1-8B
GPU_MEMORY_UTILIZATION=0.8
PORT=8001
API_KEY=soulchat-rcEmrhVe6zWot67QkJSwqUnNI0EQxxFBMQSAXLtMNsD97PlyGQgjgjW-9jCdQD30
MAX_MODEL_LEN=20000

python -m vllm.entrypoints.openai.api_server \
    --served-model-name $SERVER_MODEL_NAME \
    --model $MODEL_NAME_OR_PATH \
    --gpu-memory-utilization $GPU_MEMORY_UTILIZATION \
    --port $PORT \
    --api-key $API_KEY \
    --max-model-len $MAX_MODEL_LEN

streamlit demo construction

pip install openai==1.7.1
pip install streamlit==1.27.0
pip install streamlit_authenticator==0.3.1
cd infer_demo
streamlit run soulchat2.0_app.py --server.port 8002

Citation

@misc{xie2024psydtusingllmsconstruct,
      title={PsyDT: Using LLMs to Construct the Digital Twin of Psychological Counselor with Personalized Counseling Style for Psychological Counseling}, 
      author={Haojie Xie and Yirong Chen and Xiaofen Xing and Jingkai Lin and Xiangmin Xu},
      year={2024},
      eprint={2412.13660},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2412.13660}, 
}