ChatGSPP prototype

Technical overview

• using privateGPT's fully local setup configuration
• configured chatGSPP to use Mistral-7B for the LLM and tokenization
• currently using BAAI for embedding and Qdrant as the vector database
• the LLM's queried data is scraped from the main GSPP website using the web-scraper.py file (mentioned below)
• the demo custom UI resources are located within the ~/ui/gspp-ui directory, and uses PrivateGPT's Python SDK to make API calls
  • the demo custom UI is built using Flask and a barebones HTML file

How to run chatGSPP

• pull the repo
• create a virtual environment (highly recommended)
• export your hugging face API auth token
  • export HF_TOKEN=your-token
• install dependencies and LLM/etc
  • pip install make flask poetry pgpt_python docx2txt==0.8
  • poetry install --extras "ui llms-llama-cpp embeddings-huggingface vector-stores-qdrant"
  • poetry run python scripts/setup
• run the web-scraper.py file
  • python web-scraper.py
• ingest the scraped.csv file using this documentation, specifically the "bulk local ingestion" part
  • ingestion script would look something like this:
    • make ingest ~/chatGSPP/ingested_data -- --log-file ~/chatGSPP/ingested_data/ingestion_errors.log
• reference this documentation to load the dependencies and start the API server, specifically the "local, llama-cpp powered setup" part
  • PGPT_PROFILES=local make run
• now you can start the demo UI server by
  • navigating to ~/ui/gspp-ui
  • running python gspp-ui-app.py
• navigate to the demo custom UI via http://localhost:5000 and ask the bot some questions
  • can also access the built-in Gradio UI for testing via http://localhost:8001
• to-do: create a cool UI so we can add this functionality to GSPP's website :)

More documentation!

Full documentation on installation, dependencies, configuration, running the server, deployment options, ingesting local documents, API details and UI features can be found here: https://docs.privategpt.dev/

Project Summary

ChatGSPP is built on top of PrivateGPT. PrivateGPT is a production-ready AI project that allows you to ask questions about your documents using the power of Large Language Models (LLMs), even in scenarios without an Internet connection. 100% private, no data leaves your execution environment at any point.

The project provides an API offering all the primitives required to build private, context-aware AI applications. It follows and extends the OpenAI API standard, and supports both normal and streaming responses.

The API is divided into two logical blocks:

High-level API, which abstracts all the complexity of a RAG (Retrieval Augmented Generation) pipeline implementation:

• Ingestion of documents: internally managing document parsing, splitting, metadata extraction, embedding generation and storage.
• Chat & Completions using context from ingested documents: abstracting the retrieval of context, the prompt engineering and the response generation.

Low-level API, which allows advanced users to implement their own complex pipelines:

• Embeddings generation: based on a piece of text.
• Contextual chunks retrieval: given a query, returns the most relevant chunks of text from the ingested documents.

In addition to this, a working Gradio UI client is provided to test the API, together with a set of useful tools such as bulk model download script, ingestion script, documents folder watch, etc.

Architecture

This project is built on top of PrivateGPT. Conceptually, PrivateGPT is an API that wraps a RAG pipeline and exposes its primitives.

• The API is built using FastAPI and follows OpenAI's API scheme.
• The RAG pipeline is based on LlamaIndex.
• The API has SDK's available for building custom workflows (like this demo)

The design of PrivateGPT allows to easily extend and adapt both the API and the RAG implementation. Some key architectural decisions are:

• Dependency Injection, decoupling the different components and layers.
• Usage of LlamaIndex abstractions such as LLM, BaseEmbedding or VectorStore, making it immediate to change the actual implementations of those abstractions.
• Simplicity, adding as few layers and new abstractions as possible.
• Ready to use, providing a full implementation of the API and RAG pipeline.

Main building blocks:

• APIs are defined in private_gpt:server:<api>. Each package contains an <api>_router.py (FastAPI layer) and an <api>_service.py (the service implementation). Each Service uses LlamaIndex base abstractions instead of specific implementations, decoupling the actual implementation from its usage.
• Components are placed in private_gpt:components:<component>. Each Component is in charge of providing actual implementations to the base abstractions used in the Services - for example LLMComponent is in charge of providing an actual implementation of an LLM (for example LlamaCPP or OpenAI).

🤗 Partners & Supporters

PrivateGPT is actively supported by the teams behind:

• Qdrant, providing the default vector database
• Fern, providing Documentation and SDKs
• LlamaIndex, providing the base RAG framework and abstractions

This project has been strongly influenced and supported by other amazing projects like LangChain, GPT4All, LlamaCpp, Chroma and SentenceTransformers.