This app is for learning/fun purpose only. Its use-case is pretty much non-existent, and the data is definitely not the best for the job. But it is easily and freely accessible so... I used it. As you might see, this app is encountering one of the few problems of RAG technique - the app will only be as good as the context it gets from the vector store, and if the user's query is vogue, the similarity search will fail.
Download and install Ollama, through Ollama official website.
After installation, pull mistral and/or llama2 models with the following terminal command:
ollama pull mistral
ollama pull llama2
Make sure the ollama local server is up and running using ollama serve. The following error means the server is on, and you are ready for the next step: Error: listen tcp 127.0.0.1:11434: bind: address already in use.
Clone repository and cd into it
git clone [repo's url]
cd cs_comps
(Not mandatory)
Create a virtual environment, based on python3.10 or above, activate it and install all requirements:
Mac/Linux:
python -m venv venv
source venv/bin/activate
Windows:
python -m venv venv
cd venv
Scripts/activate
cd ../
Install requirements:
pip install -r requirements.txt
Run the main app file:
streamlit run app.py
The first run can take some long long time, since the app builds the vector-store with local resources. After that, it will always be loaded instead of created.