Starting a blog
Why I am writing things down in public, and what I plan to put here.
Most of what I learn at work never gets written down. It lives in a scratch file, a benchmark log, or a thread that scrolls away. Then six months later I hit the same problem and rediscover the same answer from scratch.
So: this site. The plan is to write up the things I would want to reread.
What goes here
Mostly systems work around machine learning:
- Inference performance. Where the time actually goes, measured rather than assumed.
- Long-context serving. KV cache growth, memory pressure, and what breaks first.
- Hardware. Memory hierarchies, bandwidth limits, and why the same model behaves differently on different silicon.
- Papers. Notes on things I read, with an honest note on what I did not follow.
How I plan to write it
Short, specific, and wrong sometimes. I would rather post a measurement with a caveat than wait until I have a complete theory. Posts get edited after publishing when I learn that something in them is off — I will note the change rather than quietly fix it.
If you find an error, tell me. That is the fastest way for this to be useful.