ishaanshettigar / text2sql-query-bot

Creating a bot that harnesses the power of LLM's like Llama 2 to convert natural language queries to SQL queries using various techniques like in context learning and PEFT

Ishaan Shettigar

👋 Hello! I'm Ishaan Shettigar, I'm currently in my final year studying CSE at PES University, Bangalore.

• 📫 You can reach me at [email protected]
• 💬 Connect with me on LinkedIn

About Me

• 🔭 I'm currently working on a computer vision project that involves detecting osteoporosis and osteoarthritis from knee X-ray images as well as a project that converts natural language queries into valid SQL queries using generative AI.
• 🌱 I'm continuously learning and exploring, my fields of interest include GenerativeAI, Crypto/Web3, AI/ML and FullStack Web Dev.
• 👯 I'm currently learning React and would love to contribute to any open-source repositories to further my understanding in it.

Technologies I Love

• Languages: Python, JavaScript, C/C++, Solidity (Ethereum), Java
• Frameworks: React (currently learning), Node.js, MERN, FastAPI, Django
• Tools: Git, OpenCV, pandas, Keras
• Databases: MariaDB, MongoDB, PostgreSQL, MySQL,

My GitHub Stats

Thank you for visiting my GitHub profile. Feel free to explore my projects, get in touch, or collaborate on exciting projects. 🚀

text2sql-query-bot's Issues

Add a "Supported Databases" section in the homepage

Come up with a cost estimate based on number of tokens for using the openAI API

OpenAI API information:

• GPT-4: 0.03$/1k tokens for input into the prompt and 0.06$/1k tokens for the GPT-4's output.
• GPT-3.5 Turbo: 0.0015$/1k tokens for input sent to the LLM and 0.02$/1k tokens for the LLM's output.
• OpenAI embeddings: 0.0001$/1k tokens

---

We basically need to estimate how many tokens we feed into the model. We need to see how many tokens the schema linking is gonna take. How many tokens the query classification (easy, medium, hard) is going to take and how many tokens the SQL query classification and the self-correction phase is going to take.

Take 100 examples from the Spider dataset and come up with the estimate.
As a continuation you can even take 100 examples each from both, the BIRD dataset and the WikiSQL dataset.

Check out this article by OpenAI: How to count tokens with tiktoken