Skip to main content
T3 Chips and hardware ZTE Round 2

ZTE AI Infra second round

Real ZTE AI Infra second-round interview questions covering inference, training, operator optimization, and high-performance computing

Inference optimization Training optimization Operator optimization High-performance computing

Project experience

  1. Walk through your main projects and their core technical designs.

Basics

  1. In distributed training, what are data, model, and pipeline parallelism? When does each fit?
  2. Core features and optimization strategies of DeepSpeed?
  3. Given a model’s parameter count, how do you estimate optimizer-state, gradient, and parameter memory during training?
  4. To train a 200B model on 16 GPUs, how would you design a reasonable distributed plan?
  5. Idea and implementation of mixed-precision training?
  6. Common techniques for LLM inference acceleration?
  7. Ways to optimize the KV Cache?
  8. Idea of operator fusion and its performance impact?
  9. Python basics (data structures, GIL, decorators, and so on).
  10. Common Linux commands.
  11. PyTorch-related questions.