Micro GPT written in Excel formulas (github.com)

A new project has successfully implemented a minimal version of the GPT architecture using only Excel formulas, eschewing traditional programming languages like Python or any external dependencies. Inspired by Andrej Karpathy's microGPT, this "Micro GPT in Excel" generates plausible-sounding names through a complex yet entirely formula-based model, showcasing the potential of Excel as a computational tool for machine learning concepts. The implementation uses key elements such as RMSNorm instead of LayerNorm, and features an unrolled loop for each output token, integrating attention mechanisms and multi-layer perceptron (MLP) blocks—all within Excel's capabilities. This novel approach is significant for the AI/ML community as it makes the inner workings of a GPT-like model accessible in a familiar spreadsheet environment, allowing users to explore and understand fundamental machine learning concepts without needing to write code. While the complexity of the model is higher than simply invoking a function in Python, this Excel implementation provides a unique educational tool that can enhance learning and experimentation. As Excel continues to evolve with features like LAMBDA functions, such projects highlight innovative ways to democratize access to AI technologies, broadening the audience engaged with machine learning.