Prerequisites

To successfully implement the kernels in Mini-YAIE, you should be familiar with:

Programming Languages

• Python (Intermediate): Understanding of classes, inheritance, type hinting, and PyTorch tensors.
• C++ (Basic): For reading and writing the CUDA kernels (though much of the boilerplate is provided).
• CUDA (Basic): Understanding of the GPU execution model (blocks, threads, shared memory).

Machine Learning Concepts

• Transformer Architecture: Queries, Keys, Values, Attention mechanism.
• Tensors: Shapes, dimensions, matrix multiplication.

Tools

• Git: For version control.
• Linux/Unix Shell: For running commands.