Deployments coming soon!

1. FAISS - Vector database
2. Google Colab - Development/ Inference using T4 GPU
3. Gradio - Web UI, inference using free-tier Colab T4 GPU
4. HuggingFace - Transformer, Sentence transformers (for creating vector embeddings), Mistral7b quantized model
5. LangChain - Retrieval augmented generation (RAG) using RetrievalQA chain functionality

Colab	Info
	Creating FAISS vector database from Kaggle dataset
	Mistral7b (4bit) RAG Inference of Bhagavad Gita using Gradio

• Store the vector database in your Google Drive in the following format "vectorstore/db_faiss". The db_faiss contains the following: index.faiss and index.pkl.
• Mount the Google Drive to load the vector embeddings for inference. Mistral7b (4bit) RAG Inference of Bhagavad Gita using Gradio
• Using BitandBytes configurations (load_in_4bit) for quantization - A bit loss in precision, but performance is almost at par with the Mistral7b (base) model.
• HuggingFace pipeline for "text-generation".
• AutoTokenizer and AutoModelforCasualLM from "transformers" for tokenization and loading Mistral7b model from HuggingFace Spaces.

• See - Kaggle Dataset
• 150 books, 37486 pages are there.

• Using sentence-transformers/all-MiniLM-L6-V2 from Huggingface
• Vector database - Google Drive

https://github.com/mistralai/mistral-src

• https://mistral.ai/news/announcing-mistral-7b/
  
• https://arxiv.org/abs/2212.04356
  
• https://mistral.ai/

Shrimanta Satpati