A repository for a RAG model project in CS 579 at ISU

We used llamaIndex as the process pipeline and pinecone as vector storage.

We have a <data_lovecraft> folder which contains a collection of pdf files of H.P. Lovecraft's work that will be loaded into the vector storage.

The text chunk size is set to <512>

The embedding model is set to <bge-small-en-v1.5>

To Use: after adding a valid PineCone API key to the source code, call python load.py <your/path/filename.pdf> --index "index_name" to perform a commandline file upload.

We used Ollama's llama2 model as the llm backbone that handles the answering.

We used VectorIndexRetriever to retrive top-k similar chunk that's relevent to the query and generated the answer using RetrieverQueryEngine based on the retrived information.

To Use: after adding a valid PineCone API key to the source code, call python query.py --query "What do you want to know?" --index "index_name" to perform an answering generation based on the files contained in the pinecone vector store index.

Additionally, the user will need to either be running a local instance of Ollama with a llama2 model loaded, or otherwise take steps to modify the LLM used by the llama_index library for our query engine to run.

We used funix from https://github.com/TexteaInc/funix as the ui library for this project. To test, you have to intsall Ollama locally and install funix via pip.