Skip to main content
T1 Large companies and unicorns JD.com Internship

JD.com AI Infra internship

Real JD.com AI Infra internship interview questions covering inference optimization, training optimization, and high-performance computing

Inference optimization Training optimization High-performance computing

Fundamentals

  1. Introduce the architecture design and training methods of the Qwen model series.
  2. Explain the differences among PPO, DPO, and GRPO reinforcement learning algorithms.
  3. How are entropy, cross-entropy, and KL divergence related and different?
  4. Why is MSE usually not used as the loss for classification tasks?

Training optimization

  1. Introduce the core capabilities and principles of DeepSpeed, DDP, and FlashAttention.
  2. How is the Advantage function computed in PPO? How is the Critic model updated?

Inference optimization

  1. Estimate how much GPU memory Qwen3-8B needs for inference.
  2. What is the full RAG (retrieval-augmented generation) pipeline? What optimization strategies are viable?

Internship experience

  1. How was the reward function designed? Was there a cold-start stage before GRPO training?

Project experience

  1. Deep dive on project experience.

Coding

  1. Hand-write GQA (Grouped Query Attention).