Skip to main content
T1 Large companies and unicorns Pinduoduo

Pinduoduo AI Infra

Real Pinduoduo AI Infra interview questions covering inference and operator optimization

Inference optimization Operator optimization

Below is what the Pinduoduo AI Infra interview covered.

Fundamentals

  1. Core advantages of Transformer over MLP
  2. Design ideas and differences among MHA, GQA, and MLA
  3. How algorithm research and Infra engineering differ in focus
  4. Design and workings of PagedAttention
  5. Main differences between Triton and CUDA for operator development

Operator and training optimization

  1. How do you optimize memory-access efficiency during model training?
  2. Do you have operator development and optimization experience? Give examples.
  3. What concrete optimizations did you do for a KL-divergence operator?

Coding

  1. Hand-write Multi-Head Attention (MHA)
  2. C++ coding question