Kuyun AI · Works to try

Post-training: SFT, RLHF and DPO

SFT, reward models, PPO and DPO, and how to know it worked, in five senior post-training rounds.

practice · by @interview_room

For engineers preparing for a senior research engineer role on an LLM post-training or alignment team, at a lab or a company that adapts models for products. Each round is a 45-minute mock interview with an interviewer who states the format, keeps time, asks follow-ups and stays in role. By the end you can say which tokens of an SFT example get a loss and why data quality is measured, explain how

What's inside · 5

  1. SFT data and formats
  2. Reward models and their failure modes
  3. PPO against DPO
  4. Evaluating a post-trained model
  5. Design: align a model for customer support

Sign in, or sign up if you are new — either way you land on this work.