patw / ragtag

A tool for testing RAG functionality with Atlas Vector Search

RAGTAG

A tool for manual RAG chunk entry for question/answer systems. Create, search or edit text chunks paired up with questions to ensure good retrieval for embeddings.

Takes advantage of Atlas Mongo and Atlas Vector Search

RAGTAG allows you to:

• Create Q/A chunks for use with your chatbots
• Vectorize chunks with open source embedding models (Instructor-large)
• Search existing chunks, edit chunk, update embedding and update chatbot in real time
• Test your chunks for recall by using real questions

Installation

pip install -r requirements.txt

Downloading the Mistral 7b model (with dolphin fine tune)

wget https://huggingface.co/TheBloke/dolphin-2.1-mistral-7B-GGUF/resolve/main/dolphin-2.1-mistral-7b.Q5_K_S.gguf

Running App

Copy sample.env to .env and modify with connection string to your Atlas instance

flask run

WARNING: You will need about 20 gigs of ram to run this process! Mistral-7b requires 14 gig with the Q5 quantization, and instructor needs 4 gig on it's own

Atlas Search Index

Create and Atlas Search index, in the Atlas UI under the Search tab for the "chunks" collection under the "ragtag" database.

{
  "analyzer": "lucene.english",
  "searchAnalyzer": "lucene.english",
  "mappings": {
    "dynamic": false,
    "fields": {
      "chunk_answer": {
        "type": "string"
      },
      "chunk_embedding": {
        "dimensions": 768,
        "similarity": "cosine",
        "type": "knnVector"
      },
      "chunk_enabled": {
        "type": "boolean"
      },
      "chunk_question": {
        "type": "string"
      }
    }
  }
}

ragtag's Issues

Add token count to chunk list page

Add a column with token count so as to highlight when it may be a problem

Also should put it on edit page