Skip to main content
T5 Others Hikvision

Hikvision AI Infra

Real Hikvision AI Infra interview questions covering inference, operator optimization, and HPC

Inference optimization Operator optimization High-performance computing

Operator optimization

  1. What is a bank conflict in CUDA? How do you avoid it?
  2. How do you implement Softmax in CUDA?
  3. What are the core ideas of Online Softmax and FlashAttention?

Inference optimization

  1. What challenges does on-device large-model deployment face? What common optimization techniques exist?
  2. How do you handle dynamic dimensions when exporting a model?
  3. What quantization techniques exist, and when does each apply?

High-performance computing

  1. In multithreaded programming, how do OpenMP and MPI differ, and when does each apply?

Engineering practice

  1. What does the ncnn cross-compile flow look like?
  2. How familiar are you with the ncnn source?

Project experience

  1. Self-introduction and project walkthrough.
  2. What do you know about perception, planning, and control algorithms?