The Curious Engineer

All blogs / AI / ML

Thonk From First Principles

thonking.ai

ML Systems from first principles. Aims to be better than a ChatGPT summary. Click to read Thonk From First Principles, by Horace He, a Substack publication with thousands of subscribers.

AI / MLPerformance / optimization

’24’25’26
Posts
6 from 26 Feb 2024 to 4 Mar 2025
Posting
now and then
Kind
personal blog
Feed
https://www.thonking.ai/feed

2025 1

  1. Why PyTorch is an amazing place to work... and Why I'm Joining Thinking Machines

2024 5

  1. FlexAttention: The Flexibility of PyTorch with the Performance of FlashAttention [external]
  2. Strangely, Matrix Multiplications on GPUs Run Faster When Given "Predictable" Data! [short]
  3. Solutions: What Shapes Do Matrix Multiplications Like?
  4. What Shapes Do Matrix Multiplications Like? [medium]
  5. Supporting Mixtral in gpt-fast through torch.compile [short]