Skip to main content
T5 Others Nanhu Laboratory Round 1

Nanhu Research Institute AI Infra first round

Nanhu Research Institute AI Infra first-round interview questions covering operator optimization

Operator optimization

The following topics were covered in the Nanhu Research Institute AI Infra first-round interview.

C++/Python fundamentals

  1. How does Python garbage collection work?
  2. How does Python GC handle cyclic references (mark-and-sweep)?
  3. When can function inlining hurt performance?
  4. How does the inline keyword affect scope?
  5. Besides inline, what else can trigger inlining (for example template functions)?
  6. What is perfect forwarding, and how is it implemented?
  7. What do rvalue references mean, and where are they used?

AI frameworks and CUDA

  1. What are the core features of FasterTransformer?
  2. Overview of common AI inference/training frameworks
  3. Common mechanisms for thread synchronization
  4. Requirements for using CUDA streams (for example overlapping device-side work)
  5. Common methods for CUDA operator performance tuning

Engineering practice

  1. Git workflow for multi-person collaboration

Coding

  1. Implement a smart pointer
  2. Find the top-K largest elements in an array