LLM with RAG

Stage One

1. RAG File
   • Large Language Model:
     • Language Model: "databricks/dbrx-instruct": https://huggingface.co/databricks/dbrx-instruct
     • Nvidia Client: https://build.nvidia.com/databricks/dbrx-instruct
   • Vector Database:
     • Milvus: https://milvus.io/
     • Embedding model: https://huggingface.co/thenlper/gte-base
     • Support OS: Linux
       • Currently Does not support Windows OS because Milvus_lite does not support Windows OS
       • Will choose different database in the future in order to fix this issue
2. pdf_to_txt File
   • Current Handle:
     • pdf(text) to txt
     • Need to improve preprocessing inorder to feed to RAG model
3. Progress(10/01/2024): Simplified version works on Linux, with one query ability
   • (10/02/2024): Able to reuse collection for query

Stage Two (Current)

1. Creating pdf reader using OCR
   • accept uploaded pdf
   • read using EasyOCR
   • store results in files, preferably one file for each pdf
2. RAG File supports recursive question and answer
3. Able to store historical QA in corresponding files

Stage Three

1. Combine RAG with pdf reader
2. Support LLM long term memory
   • Add QA history to another storage
   • Combine answers with history

Cleanup implementatio and Optimize

1. optimization, speed and memory

Deployment?