Thonk From First Principles

In my opinion, there is a sparsity of great technical writing online. Very often, I find that technical blogs written from a surface-level understanding of the content. For example, they may explain what Fully Sharded Data Parallelism and Tensor Parallelism are but not when you should use each one.

Understanding is achieved merely remembering lists of facts but instead when you have mental frameworks through which the facts become obvious.

In other words, I aim to provide something better than a ChatGPT summary for as long as I am able to. :)

Some pieces of previous writing include:

As my day job, I work on PyTorch Compilers at Meta.

Testimonials

"My learning style is Horace twitter threads" - @typedfemale


You should follow Horace He (@cHHillee). Super passionate about torch, GPUs, sysml. Answers questions at all levels of expertise - Sasha Rush

Paid Subscriptions

Unfortunately, I’m a slow writer and I often procrastinate on writing. If you enjoy my writing and want to see more of it, you can pay me money and I will feel pressured to write more. It’s a win-win!

However, I will never paywall an article. My tentative plan is that for longer-form articles, I will add a set of questions to test understanding. I will restrict the answer key to paid subscribers (or those who email me their solutions!)

Subscribe to Thonk From First Principles

ML Systems from first principles. Aims to be better than a ChatGPT summary.

People