Skip to main content
T3 Chips and hardware Phytium New grad Round 2

Phytium AI Infra campus recruiting second round

Real Phytium AI Infra campus recruiting second-round interview questions covering inference optimization, training optimization, and operator optimization

Inference optimization Training optimization Operator optimization

Basics

  1. What is the underlying implementation of C++ virtual functions? How does the vtable work?
  2. Compilation has which four stages? What does each stage do?
  3. What common design patterns are there? Describe use cases for the singleton and factory patterns.
  4. What are the main differences between multi-process and multi-thread?

Operator and GPU optimization

  1. For operator tuning on a specific shape, how do you set an optimization strategy to beat the official library?
  2. What is the design idea behind fused operators?
  3. What key issues does mixed-precision training and inference involve?
  4. In GPU performance optimization, how do you use the Nsight toolchain? Which compute-efficiency and bandwidth-efficiency metrics should you watch?

Inference optimization

  1. How does KV Cache work? What is the core memory-optimization idea in FlashAttention?