This repository contains the source code for the Retrieval-augmented Generation (RAG) technique, as described in the following articles:
- Part 1: Getting started, Chain of Thought
- Part 2: Reason-Act, multi-turn conversation
- Part 3: PDF ingestion, vector search
- Bonus: RAG with SLM (Small Language Model)
First, you need to run the API server of llama.cpp with Phi 2:
./server -m /path/to/phi-2.Q4_K_M.gguf
(For a slower response but with improved accuracy, consider using Mistral 7B OpenOrca).
To launch Pico Jarvis, you need Node.js v18 or later:
npm install
npm start
and then open localhost:5000.
Ask the following questions:
- Who wrote the Canon of Medicine?
- Is ramen typically eaten in Egypt?
- Who directed the Dark Knight movie?
- Name Indonesia #1 tourist destination!
(and some more to try)
- What is the native language of Mr. Spock?
- Which US state starts with G?
- What is the atomic number of Magnesium?
- Where do we find kangoroo?
- Who is the father of Luke Skywalker?
- In which country Mandarin is spoken?
- What is the longest river in Latin America?
- Who authored the special theory of relativity?
- Which fictional metal is infused into Wolverine body?
- Who sailed with the flagship Santa Maria?
- Name the big desert close to Mongolia
- Which is closer to Singapor: Vietnam or Australia?
- Who is the fictional spy 007?
- Which country is known for IKEA?
Meanwhile, questions related the solar system will be answered by searching the PDF document:
- What is a dwarf planet?
- Which planet known as the red one?
- What materials compose the gas giants?
- How about the ice giants?
- Explain the heliopause
- When did Voyager 2 enter the interstellar space?
- How about Voyager 1?
If you get an API key for OpenWeatherMap and supply it as OPENWEATHERMAP_API_KEY environment variable, try to ask the following:
- How is the weather in Jakarta?
- What is the current temperature in Palo Alto?
- Is it currently cloudy in Seattle?