What it does
SimPO (Simple Preference Optimization) is a reference-free method for aligning large language models, presented as an alternative to DPO. The publisher reports a 6.4-point improvement on AlpacaEval 2.0.
It suits preference alignment work where simpler, faster training than DPO or PPO is wanted. The publisher describes it as more efficient than DPO.
Training needs no reference model, which is the main setup difference from DPO. The source gives no further setup details. The resource is released under the MIT license.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- AI Templates (aitmpl) first seen 13 Nov 2026 · last seen 18 Nov 2026 · its page on the source · figures reported by the source: aitmpl downloads 27
post-training-simpo
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.
v1.0.0As shown on www.aitmpl.com when we last read it