RewardBench: Evaluating Reward Models for Language Modeling
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866929378091335680 |
|---|---|
| author | Lambert, Nathan Pyatkin, Valentina Morrison, Jacob Miranda, LJ Lin, Bill Yuchen Chandu, Khyathi Dziri, Nouha Kumar, Sachin Zick, Tom Choi, Yejin Smith, Noah A. Hajishirzi, Hannaneh |
| author_facet | Lambert, Nathan Pyatkin, Valentina Morrison, Jacob Miranda, LJ Lin, Bill Yuchen Chandu, Khyathi Dziri, Nouha Kumar, Sachin Zick, Tom Choi, Yejin Smith, Noah A. Hajishirzi, Hannaneh |
| contents | Reward models (RMs) are at the crux of successfully using RLHF to align pretrained models to human preferences, yet there has been relatively little study that focuses on evaluation of those models. Evaluating reward models presents an opportunity to understand the opaque technologies used for alignment of language models and which values are embedded in them. Resources for reward model training and understanding are sparse in the nascent open-source community around them. To enhance scientific understanding of reward models, we present RewardBench, a benchmark dataset and code-base for evaluation. The RewardBench dataset is a collection of prompt-chosen-rejected trios spanning chat, reasoning, and safety, to benchmark how reward models perform on challenging, structured and out-of-distribution queries. We create specific comparison datasets for RMs that have subtle, but verifiable reasons (e.g. bugs, incorrect facts) why one answer should be preferred to another. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods, such as the direct MLE training of classifiers and the implicit reward modeling of Direct Preference Optimization (DPO). We present many findings on propensity for refusals, reasoning limitations, and instruction following shortcomings of various reward models towards a better understanding of the RLHF process. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_13787 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | RewardBench: Evaluating Reward Models for Language Modeling Lambert, Nathan Pyatkin, Valentina Morrison, Jacob Miranda, LJ Lin, Bill Yuchen Chandu, Khyathi Dziri, Nouha Kumar, Sachin Zick, Tom Choi, Yejin Smith, Noah A. Hajishirzi, Hannaneh Machine Learning Reward models (RMs) are at the crux of successfully using RLHF to align pretrained models to human preferences, yet there has been relatively little study that focuses on evaluation of those models. Evaluating reward models presents an opportunity to understand the opaque technologies used for alignment of language models and which values are embedded in them. Resources for reward model training and understanding are sparse in the nascent open-source community around them. To enhance scientific understanding of reward models, we present RewardBench, a benchmark dataset and code-base for evaluation. The RewardBench dataset is a collection of prompt-chosen-rejected trios spanning chat, reasoning, and safety, to benchmark how reward models perform on challenging, structured and out-of-distribution queries. We create specific comparison datasets for RMs that have subtle, but verifiable reasons (e.g. bugs, incorrect facts) why one answer should be preferred to another. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods, such as the direct MLE training of classifiers and the implicit reward modeling of Direct Preference Optimization (DPO). We present many findings on propensity for refusals, reasoning limitations, and instruction following shortcomings of various reward models towards a better understanding of the RLHF process. |
| title | RewardBench: Evaluating Reward Models for Language Modeling |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2403.13787 |