LLM System Design Interview, issue 65, Sep 9, 2026
The Hyperparameter Scaling Trap
Principal ML Engineer interview at Anthropic, and the interviewer asks:
“StepFun says optimal batch size depends only on total tokens D, and optimal LR goes up with more data. DeepSeek says LR just decreases with compute. Both trained real frontier models. Which one do you use for our next run?”
Don’t say: “I’d follow the more recent paper, StepFun has the bigger grid.”
Why two frontier labs arrived at seemingly opposite scaling rules for the exact same surface, and how to isolate the true sign on your corpus for the compute cost of a single run.
The full answer, with the mechanism and the arithmetic, is free on Substack.