Projects

No description yet.

GPU kernels explained through indexing, synchronization, and numerical correctness.

First-principles notes on attention, distributed training, and GPU performance models.

LLM training performance case studies, negative results, and small diagnostic tools.

Explainable FLOPs accounting for LLM training, with a kernel-invariant MFU numerator.

A Quirky Assortment of CuTe Kernels