Hugging Face Overview
Nifty50GPT-Final: India's First Offline Financial SQL LLM
Nifty50GPT-Final is a lightweight, 1.1 billion parameter transformer model, based on TinyLLaMA, developed by Shubham Sood at Student One. It is uniquely fine-tuned for generating SQL queries from natural language prompts related to the Indian stock market. Designed for transparency and local usability, this model operates entirely offline, eliminating dependencies on APIs or cloud services.
Key Capabilities
- SQL Query Generation: Translates financial questions into executable SQL queries.
- Bundled Data: Includes a
student_data.duckdbfile with 10+ years of historical data for 50+ NIFTY stocks, Indian indices (e.g., NIFTY50, SENSEX), and global indices (e.g., S&P 500, Nasdaq 100). - Offline Operation: Supports both CPU and GPU inference without internet access or live data feeds.
- Diverse Financial Queries: Capable of handling fundamental metric lookups, year-over-year growth, CAGR calculations, OHLCV data retrieval, rolling metrics, and NIFTY-wide aggregates.
Good for
- Research and Experimentation: Ideal for academic or personal projects requiring structured financial data analysis.
- Educational Purposes: Provides a hands-on tool for learning about financial data querying and LLM applications.
- Local Financial Analysis: Enables users to perform detailed analysis of Indian stock market data without external dependencies.
- SQL Generation for DuckDB: Specifically designed to generate queries compatible with the included DuckDB database, ensuring instant usability.