Skip to main content
T0 Big tech Alibaba Round 1

Alibaba Taotian AI Infra first round (2)

Real Alibaba AI Infra first-round interview questions covering AI Infra and related topics

Fundamentals

  1. Core differences between GRPO and PPO.
  2. What does the Clip operation in PPO specifically do?
  3. What does importance sampling do? Besides pairing with Clip to limit update size, what other uses? How does it differ from constraining updates with KL divergence?
  4. What is the Markov property? Give a formal definition.
  5. From early policy-gradient methods to GRPO, which modules were kept and which dropped? Sketch the evolution and why.

Project experience

  1. Share screen and walk through a paper: research background, task setup, and matching real business scenarios.
  2. Detail the workflow, training method, and derivation of the formulas used.

Coding

  1. Binary Search Tree Iterator (LeetCode 173); first state the idea, then give solutions from brute force to optimal.