Skip to main content
T1 Large companies and unicorns Meituan New grad

Meituan Beidou AI Infra campus hire

Real Meituan AI Infra campus-hire interview questions covering inference optimization, operator optimization, and high-performance computing

Inference optimization Operator optimization High-performance computing

Interview topics for Meituan Beidou AI Infra campus hire.

Theory

  1. Explain the overall Transformer architecture, and what advantages it has over classic RNNs
  2. Where do Transformer parameters mainly sit? Which part has the most parameters? Which part has the most compute, and why
  3. Explain the difference between CUDA Core and Tensor Core on GPUs; list common GPU models with memory capacity and bandwidth
  4. Introduce the basic idea of large-model quantization, and describe how quantization operators are implemented
  5. Detail the design idea of Prefill-Decode separation (PD disaggregation)
  6. Detail how Paged Attention works

Live coding

  1. LeetCode 25: reverse nodes in k-group
  2. Write a CUDA operator: basic prefix sum, and explain feasible optimizations
  3. Write a CUDA operator: basic GEMM, and explain feasible optimizations