paperx

Feed papers to get your desired results by few clicks!

Introduction

Using llama and grobid tools, I have built an application where understanding papers and processing it would be easier, while keeping your privacy safe :)!

the chatbot is provided by replicate api.

Getting started

you got two option in order to deploy and use this app

install requirements and setup python env

• First of all set up a python environment (or just run it on your base python env):
  1. install venv to create virtual python environment:

python3 -m pip install venv

• Now let's install all the requirements in order to get started with the project.

pip install -r requirements

bring up the grobid editor

• now using docker we will bring up the grobid server in a container:

docker run -d --rm --init --ulimit core=0 -p 8070:8070 lfoppiano/grobid:0.8.0

llama the language model

• Downlaod the models from the source (huggingface) using the download_model.sh script, this will take the url from the .env_model and then download the llm model accordingly, you can skip downlaoding the remaining models by removing their url from that file. this script will save the models in ./models directory.

./download_model.sh

• now bring up the container that has the llama model as a server, the model i choose is llama-2-7b-chat.Q4_K_M.gguf:

docker run -p 8080:8080 -v ./models:/models ghcr.io/ggerganov/llama.cpp:server -m models/llama-2-7b-chat.Q4_K_M.gguf -c 512 --host 0.0.0.0 --port 8080

• or another option is to use an api key from https://replicate.com an put is in the .env file.(that's what i did)

streamlit app

• you can run the app directly on your machine using the below command on port 8051.

streamlit run streamlit_app.py

resources

• structure-vision: https://github.com/lfoppiano/structure-vision
• grobid: https://github.com/kermitt2/grobid
• grobid_client_python: https://github.com/kermitt2/grobid_client_python
• How to build a Llama 2 chatbot: https://blog.streamlit.io/how-to-build-a-llama-2-chatbot/
• streamlit-pdf-viewer: https://github.com/lfoppiano/streamlit-pdf-viewer
• models source: https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF
• llama + streamlit blog post: https://medium.com/@odhitom09/running-openais-server-locally-with-llama-cpp-5f29e0d955b7