princeton-nlp/SimPO
A reference-free preference optimization algorithm for aligning large language models, published at NeurIPS 2024.
Not currently ranked — collecting fresh signals.
star history
SimPO is a preference alignment method for LLMs that improves upon DPO by removing the need for a reference model. It uses a reward-free objective with length-normalized scoring to achieve better alignment. The project includes training code, released model checkpoints on HuggingFace, and demonstrates strong results on AlpacaEval 2 and Arena-Hard benchmarks across multiple base models including Llama3 and Gemma2.
Frequently asked
- What is princeton-nlp/SimPO?
- A reference-free preference optimization algorithm for aligning large language models, published at NeurIPS 2024.
- Is SimPO open source?
- Yes — princeton-nlp/SimPO is open source, released under the MIT license.
- What language is SimPO written in?
- princeton-nlp/SimPO is primarily written in Python.
- How popular is SimPO?
- princeton-nlp/SimPO has 956 stars on GitHub.
- Where can I find SimPO?
- princeton-nlp/SimPO is on GitHub at https://github.com/princeton-nlp/SimPO.