domains / families / research / simpo

simpo

separate

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

When to use

• Want simpler training than DPO (no reference model)
• Have preference data (chosen/rejected pairs)
• Need better performance than DPO
• Limited compute resources
• Single-node training sufficient
• **SimPO**: Simplest, best performance, no reference model
• **DPO**: Need reference model baseline, more conservative
• **PPO**: Maximum control, need reward model, complex setup

When not to use

• Multi-node distributed training
• PPO/GRPO training
• Using multiple methods in one framework
• Established baseline comparison with DPO

Skill page

Source