Skip to main content
T5 Others miHoYo Internship Round 2

miHoYo AI Infra internship second round

Real miHoYo AI Infra internship second-round interview questions covering AI Infra and related topics

Below is what miHoYo AI Infra internship second round covers.

Project experience

  1. Deep questions on the project

Fundamentals

  1. Outline the main structural traits of large models
  2. How attention works in large models and what it does
  3. Common optimization algorithms in large-model training, with pros and cons of each
  4. How do you handle vanishing or exploding gradients during training
  5. In model design, how do you balance model complexity and performance
  6. Facing the large compute needed for training and inference, what practical solutions exist

Model evaluation and optimization

  1. Common metrics when evaluating large-model performance
  2. Concrete methods to optimize large-model performance and efficiency
  3. Describe a case where you used or built a large model, from real experience

Alignment and safety

  1. What categories of Reward Model appear on Reward Bench
  2. What is the Reward Model training flow and training objective
  3. DPO training loss and objective; how DPO improves on RLHF
  4. How do you understand large-model safety, and which aspects does it cover