Megvii AI Infra campus hire

Project experience

1. What preprocessing is done on LLaMA model files (.bin)?
2. Is the model split across multiple .bin files or a single one?
3. How are all layers initialized?
4. How is CPU data copied to the GPU?
5. What is the model forward call flow?
6. Detail the overall approach and process for operator optimization.
7. How is the memory-management scheme designed?
8. What data structure does the GPU memory pool use?

Paper discussion

9. Deep discussion around related papers.

Coding

10. 3Sum.