All assessments
DL Coding

Post-Training and Alignment

GRPO, PPO, DPO, and RLHF foundations: losses, advantages, and reward modeling primitives.

Unlock this assessment

Upgrade to Pro to take any set.

Upgrade to Pro
30 min
Set 1Problems
1Sequence Log ProbabilityEasy
2GRPO Clipped Surrogate LossMedium
3PPO Value Function LossEasy

What this assessment covers

Post-Training and Alignment has 4 sets and 12 coding problems in total: 5 easy, 6 medium and 1 hard. The problems are:

  • Sequence Log Probability (easy)
  • GRPO Clipped Surrogate Loss (medium)
  • PPO Value Function Loss (easy)
  • PPO Clipped Surrogate Loss (medium)
  • Token Log-Probability Gather (medium)
  • GRPO Group Advantage Normalization (easy)
  • Reward Model Pairwise Loss (medium)
  • DPO Implicit Reward (easy)
  • GRPO k3 KL Estimator (medium)
  • DPO Loss (medium)
  • Generalized Advantage Estimation (hard)
  • RLHF KL Penalty Reward (easy)

Format

Each set is a separate timed sitting of 30 minutes with 3 problems. Pick a set, start the timer and solve the problems in the browser. When the sitting ends you get your score and a report on how you did, and you can take another set to practise again.