Algorithm Configuration

Configure the RL algorithm.

Basic Settings

algorithm:
  name: grpo              # Algorithm name
  clip_range: 0.2         # PPO clip range
  entropy_coef: 0.01      # Entropy bonus
  value_coef: 0.5         # Value loss weight
  max_grad_norm: 1.0      # Gradient clipping

Available Algorithms

Name	Description
grpo	Group Relative Policy Optimization (default)
ppo	Proximal Policy Optimization
dpo	Direct Preference Optimization
reinforce	Vanilla REINFORCE
dapo	Decoupled PPO
rloo	Leave-One-Out baseline

Custom Algorithms

algorithm:
  name: my_custom_loss
  adv_estimator: my_advantage
  clip_range: 0.2

See Also

• Algorithms Guide
• Custom Algorithms