SciAssist a multimodal question answering system for NLP Papers that is implemented using Retrieval Augmented Generation. It can be deployed using
Gemini API or locally using an open source model such as a backbone to answer questions on scientific NLP Papers. The paper pool that the underlines
the rag system contains 72k papers from the ACL Anthology. You can also supply your own Papers by adding a folder that contains the papers to the configuration under path_dataset_own_domainand setting own_domain to True.
The system is based on a central configuration where parameters such as the size of the paper chunks and the size of retrieved paper chunks are stored.
./run_server.sh
Then you can access the server as follows
http://127.0.0.1:8585/chat?q=What are typical design patterns of RAG?
The server returns a dictionary as follows
{
asnwer: "The answer
}
Locally using CLI To deploy the Retrieval Augmented Generation pipeline via clli which allows you to chat with the interface via command line you can run the following
python scirag/setup_rag.py --path-index data/index --path-dataset data/acl-publication-info.74k.parquet
--path-model /bigwork/nhwpajjy/pre-trained-models/DeepSeek-R1-Distill-Qwen-1.5B
The chat ui client is implemented using streamlit and is deployed as a single docker container
to build run the following
./run_client.sh
Then you should be able to access the web client using 127.0.0.1:8501
./scirag/deploy_sciarg_cpu.sh
The web service can be then accessed as follows
http://127.0.0.1:80/chat?q= What are typical software designs of RAG
docker build -t sciassist-client -f chatui/docker/Dockerfile .
To run the client use
docker run -p 8081:8051 sciassist-client
1) you can start an EC2 instance g4dn.xlarge.
2) docker run -it --rm --gpus all -p 80:80 --name "sciassist-cnt" --tty "yamenajjour:/sciassist-img:17"
