From e6c3d771a2e2b50a93a22c8ed75ad4e2ced79241 Mon Sep 17 00:00:00 2001 From: Yu Meng Date: Wed, 17 Jul 2024 11:34:06 -0400 Subject: [PATCH] Update README.md --- README.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index deee07a..26da29e 100644 --- a/README.md +++ b/README.md @@ -69,13 +69,13 @@ The [CPO_SIMPO](https://github.com/fe1ixxu/CPO_SIMPO/tree/main) repository did p ## Released Models ### Gemma -We release the following two models that are built on top of the strong [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) model by training DPO and SimPO on the on-policy dataset [princeton-nlp/gemma2-ultrafeedback-armorm](https://huggingface.co/datasets/princeton-nlp/gemma2-ultrafeedback-armorm). For GSM and MMLU, we use the [EvalZero](https://github.com/yuchenlin/ZeroEval) reporistory which aims to evaluate instruction-tuned LLMs (i.e., chat models instead of base models) for their zero-shot performance on reasoning and knowledge heavy tasks. More results on [WildBench](https://huggingface.co/spaces/allenai/WildBench) are coming soon. +We release the following two models that are built on top of the strong [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) model by training DPO and SimPO on the on-policy dataset [princeton-nlp/gemma2-ultrafeedback-armorm](https://huggingface.co/datasets/princeton-nlp/gemma2-ultrafeedback-armorm). For GSM and MMLU, we use the [EvalZero](https://github.com/yuchenlin/ZeroEval) repository which aims to evaluate instruction-tuned LLMs (i.e., chat models instead of base models) for their zero-shot performance on reasoning and knowledge heavy tasks. More results on [WildBench](https://huggingface.co/spaces/allenai/WildBench) are coming soon. | models | AE2 LC | AE2 WR | AE2 Length | AH | AH Length | GSM | GSM Length | MMLU | MMLU Length | |-----------------------------------|:------:|:------:|:----------:|:----:|:---------:|:----:|:----------:|:----:|:-----------:| | [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) | 51.1 | 38.1 | 1571 | 40.8 | 545 | 87.4 | 395 | 72.7 | 515 | -| [princeton-nlp/gemma-2-9b-it-DPO](https://huggingface.co/princeton-nlp/gemma-2-9b-it-DPO) | 69.6 | 67.2 | 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) | 73.2 | 66.7 | 1833 | 59.1 | 693 | 88.0 | 341 | 72.2 | 441 | +| [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.