waymo machine learning system interview: runtime optimization and debugging
Interview Experience
本帖最后由 匿名 于 2026-5-22 15:13 编辑
面的是ML system方向 偏runtime/optimization 四轮
三轮system design/ML accelerator and efficiency/ML framework
Inference system design based on 100M DAU, how to do back of envelope study估计memory，bandwidth, optimize the latency/OOM
QAT, distillation, model eval(), contrastive learning
Kernel optimization, memory layout, kernel fusion, etc
一轮美国大姐debug
coding numpy, tensor, distributed,大概率挂在这里 bug没找出来，相关的比如 Matrix.zeros aliasing; to_ndarray是不是miss axis=1和fr...
Full Details
本帖最后由 匿名 于 2026-5-22 15:13 编辑
面的是ML system方向 偏runtime/optimization 四轮
三轮system design/ML accelerator and efficiency/ML framework
Inference system design based on 100M DAU, how to do back of envelope study估计memory，bandwidth, optimize the latency/OOM
QAT, distillation, model eval(), contrastive learning
Kernel optimization, memory layout, kernel fusion, etc
一轮美国大姐debug
coding numpy, tensor, distributed,大概率挂在这里 bug没找出来，相关的比如 Matrix.zeros aliasing; to_ndarray是不是miss axis=1和from_ndarray 用法truncates remainders etc
请大家狠狠加米，我还有其他面筋分享
About This Question
This is a candidate experience report from a waymo interview for a mle role (newgrad level) during the phone screen round reported in 2025.
It covers the following topics: System Design, Matrix .