#mlsys

I Wrote My First GPU Kernel — Here’s What Changed

1.49x faster softmax. But only 1.074x faster forward pass. Amdahl’s Law explains the gap — and understanding it matters more than the…

I Profiled LLM Inference From First Principles — Here’s What I Found

“Why is this so slow?” My colleague was frustrated. We work at the same company, and he was building a RAG system with a local LLM running…

Im attending MLSys ’26!

How Big Is an LLM? Count the Facts It Remembers

Frontier labs rarely disclose parameter counts. As a result, practitioners rely on indirect proxies such as API latency, pricing, and…

ML Systems — TVM & Triton

https://www.usenix.org/system/files/osdi18-chen.pdf

ML Systems — Tensorflow & PyTorch

This week we move on to the most famous ML frameworks— Tensorflow’s original paper and PyTorch 2!

ML Systems — Introduction

This quarter, I am enrolled in a Machine Learning Systems course. The following series of articles will serve as my summaries and…

ELI5: Flash Attention

Step by step explanation of how one of the most important MLSys breakthroughs work — in gory detail.