EasyDeepLearn

How does batch size / rollout length affect PPO?

medium

Answer

  • Larger rollout length → lower variance advantage estimates (more MC-like), but slower feedback.
  • Batch size (nactors  ×  rolloutlength)(n_{\mathrm{actors}}\; \times \;\mathrm{rollout}_{\mathrm{length}}): must be big enough for a stable gradient estimate; typical 2048-16384.
  • Too small → noisy gradients + unstable KL.
  • Too big → wasted computation.
  • Modern LLM RLHF uses very large batch sizes (millions of tokens per update).
Check yourself — multiple choice
  • Random
  • Longer rollout = lower advantage variance but slower feedback; batch (actors × rollout) = 2048-16384 typical for control; RLHF batches are millions of tokens
  • Not important
  • Same as SL

PPO batch: 2048-16384 for control; RLHF batches much larger.

#policy-methods#engineering

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions