HealthCareQA - a Vietnamese health Q&A chatbot based on Retrieval-Augmented Generation (RAG) architecture. The chatbot uses the SeaLLM-7B-v2.5 large language model, fine-tuned by LoRA technique, combined with the ColBERTv2 information retrieval model to provide accurate and reliable medical information to users.
git clone https://github.com/xuantruong118/HeathcareQA.git
pip install -r requirements.txt
I use bkai-foundation-models/vietnamese-bi-encoder as the pretrained model to train ColBERTv2 Retriever. For the training llm stage, I use SeaLLMs/SeaLLM-7B-v2.5 as base model for continued pretraining. Then use this pretrained model to train supervised finetuning.
Data for fine-tuning SeaLLM-7B-v2.5 and ColBERTv2 can be downloaded from the following link: Data
After downloading, extract the data and place it in the data folder.
Please download the supervised-finetuned SeaLLM-7B-v2.5 model and the ColBERTv2 model from the following link (or Drive) and place them in the models folder.
| Model | Link |
|---|---|
| SeaLLM-7B-v2.5 | https://huggingface.co/truongxl/med-seallm-7b-v2.5-sft |
| ColBERTv2 | https://huggingface.co/truongxl/colbert-retriever |
The data used to Retriever for the chatbot can be downloaded from the following link and placed in the data folder.
Run the following command to start the server:
bash retriever.sh
vLLM server
bash server.sh
Run the following command to start the chatbot:
python app.py
After running complete, open your web browser and navigate to http://localhost:8001 to access the chatbot.
If you have any questions or suggestions, please feel free to contact me via email: [email protected]
HealthCareQA - a Vietnamese health Q&A chatbot based on Retrieval-Augmented Generation (RAG) architecture. The chatbot uses the SeaLLM-7B-v2.5 large language model, fine-tuned by LoRA technique, combined with the ColBERTv2 information retrieval model to provide accurate and reliable medical information to users.
git clone https://github.com/xuantruong118/HeathcareQA.git
pip install -r requirements.txt
I use bkai-foundation-models/vietnamese-bi-encoder as the pretrained model to train ColBERTv2 Retriever. For the training llm stage, I use SeaLLMs/SeaLLM-7B-v2.5 as base model for continued pretraining. Then use this pretrained model to train supervised finetuning.
Data for fine-tuning SeaLLM-7B-v2.5 and ColBERTv2 can be downloaded from the following link: Data
After downloading, extract the data and place it in the data folder.
Please download the supervised-finetuned SeaLLM-7B-v2.5 model and the ColBERTv2 model from the following link (or Drive) and place them in the models folder.
| Model | Link |
|---|---|
| SeaLLM-7B-v2.5 | https://huggingface.co/truongxl/med-seallm-7b-v2.5-sft |
| ColBERTv2 | https://huggingface.co/truongxl/colbert-retriever |
The data used to Retriever for the chatbot can be downloaded from the following link and placed in the data folder.
Run the following command to start the server:
bash retriever.sh
vLLM server
bash server.sh
Run the following command to start the chatbot:
python app.py
After running complete, open your web browser and navigate to http://localhost:8001 to access the chatbot.
If you have any questions or suggestions, please feel free to contact me via email: [email protected]