Simpo

Reference-free preference alignment, simpler than DPO.

By Orchestra Research · Official Docs · Date unknown

SimPO is a reference-free preference optimization method that outperforms DPO without needing a reference model.

Platforms: linux, macos, windows.

hermes skills install official/mlops/simpo

• #skills
• #optional
• #mlops
• #post-training
• #simpo
• #preference optimization
• #alignment