Skip to main content
T5 Others Beike

KE Holdings AI Infra

Real KE Holdings AI Infra interview questions covering inference optimization, training optimization, and related topics

Inference optimization Training optimization

Large-model fine-tuning and training

  1. Basic idea of LoRA fine-tuning
  2. How the two low-rank matrices in LoRA are initialized; why you cannot initialize matrix A to zero
  3. How P-Tuning v2 is implemented and how it differs from earlier methods
  4. Goals and content of LLM pretraining versus post-training
  5. Basic GRPO flow in reinforcement learning
  6. Designing and evaluating reward functions
  7. How do you use an LLM as a judge?

Inference and retrieval

  1. What Tokenizer optimization methods exist?
  2. What does TF-IDF validation specifically do?
  3. RAG (retrieval-augmented generation) implementation flow
  4. Common vector-retrieval methods and how you pick among them

Quantization

  1. Basic idea of quantization and the concrete workflow

Coding

  1. String to integer (LeetCode 8)