EasyDeepLearn

Learning-rate schedule for PPO / SAC — what works?

medium

Answer

  • PPO: linear anneal from 3e-4 to 0 over training.
  • Adam optimizer, β2  =  0.999{\beta}_{2}\; = \;0.999.
  • SAC: fixed 3e-4 typically.
  • LLM RLHF: much smaller LR (1e-6 to 1e-5) since starting from a well-trained model; sometimes with cosine schedule.
  • Never use aggressive schedules that spike — RL is fragile to LR.
  • Warmup helps prevent initial destabilization when starting fresh.
Check yourself — multiple choice
  • Any high LR
  • PPO: linear anneal 3e-4→0 with Adam; SAC: fixed 3e-4; RLHF: 1e-6 to 1e-5 with cosine; warmup helps fresh starts
  • Random
  • Not real

LR: PPO 3e-4 linear anneal; SAC fixed; RLHF 1e-6 to 1e-5.

#engineering

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions