From 3690ad3b2e0b2338481ee2f95689392b681b908b Mon Sep 17 00:00:00 2001 From: xiamengzhou Date: Wed, 17 Jul 2024 13:37:04 -0400 Subject: [PATCH] Update README.md --- README.md | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index b042795..6a2d75b 100644 --- a/README.md +++ b/README.md @@ -5,7 +5,7 @@ This repository contains the code and released models for our paper [SimPO: Simp ## 🆕 Changelog -- [2024.07.17] We released a new SimPO model [gemma-2-9b-it-SimPO](https://huggingface.co/princeton-nlp/gemma-2-9b-it-SimPO) by fine-tuning Google's gemma-2 9B model using on-policy [UltraFeedback data](https://huggingface.co/datasets/princeton-nlp/gemma2-ultrafeedback-armorm) annotated by [ArmoRM](https://huggingface.co/RLHFlow/ArmoRM-Llama3-8B-v0.1), achieving a **72.4** LC win rate on AlpacaEval 2 (**#1 on the Leaderboard**🎉🎉) and a **59.1** win rate on Arena-Hard! Please find the training script ([here](https://github.com/princeton-nlp/SimPO/blob/main/training_configs/gemma-2-9b-it-simpo.yaml))! +- [2024.07.17] We released a new SimPO model [gemma-2-9b-it-SimPO](https://huggingface.co/princeton-nlp/gemma-2-9b-it-SimPO) by fine-tuning Google's gemma-2 9B model using on-policy [UltraFeedback data](https://huggingface.co/datasets/princeton-nlp/gemma2-ultrafeedback-armorm) annotated by [ArmoRM](https://huggingface.co/RLHFlow/ArmoRM-Llama3-8B-v0.1), achieving a **72.4** LC win rate on AlpacaEval 2 (**#1 on the Leaderboard**🎉🎉) and a **59.1** win rate on Arena-Hard! Please find the training script [here](https://github.com/princeton-nlp/SimPO/blob/main/training_configs/gemma-2-9b-it-simpo.yaml)! - [2024.07.08] We updated our paper ([v2](https://arxiv.org/abs/2405.14734v2)) - Additional baselines (RRHF, SLiC-HF, CPO) - New Llama3-Instruct setting (v0.2) with [ArmoRM](https://huggingface.co/RLHFlow/ArmoRM-Llama3-8B-v0.1) as the preference label annotator, yielding a better-performing model, [Llama-3-Instruct-8B-SimPO-v0.2](https://huggingface.co/princeton-nlp/Llama-3-Instruct-8B-SimPO-v0.2), with a **53.7** LC win rate on AlpacaEval 2 and a **36.5** win rate on Arena-Hard ([training script](https://github.com/princeton-nlp/SimPO/blob/main/training_configs/llama-3-8b-instruct-simpo-v2.yaml))! @@ -78,7 +78,8 @@ We release the following two models that are built on top of the strong [google/ | [princeton-nlp/gemma-2-9b-it-DPO](https://huggingface.co/princeton-nlp/gemma-2-9b-it-DPO) | 67.8 | 65.4 | 2016 | 58.9 | 717 | 88.5 | 392 | 72.2 | 624 | | [princeton-nlp/gemma-2-9b-it-SimPO](https://huggingface.co/princeton-nlp/gemma-2-9b-it-SimPO) | 72.4 | 65.9 | 1833 | 59.1 | 693 | 88.0 | 341 | 72.2 | 441 | -Compared to the llama3 models, we found that the gemma models exhibit significantly less catastrophic forgetting on math tasks (e.g., GSM) and MMLU, despite the ultrafeedback dataset having limited math-related data. This demonstrates that the [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) model is more suitable for continued preference optimization. +- Compared to the llama3 models, we found that the gemma models exhibit significantly less catastrophic forgetting on math tasks (e.g., GSM) and MMLU, despite the ultrafeedback dataset having limited math-related data. This demonstrates that the [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) model is more suitable for continued preference optimization. +- SimPO and DPO perform comparably across all benchmarks, but SimPO is inherently simpler and less resource-intensive. ### v0.2