Direct Large Language Model Alignment Through Self-Rewarding Contrastive Prompt Distillation
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911988858224640 |
|---|---|
| author | Liu, Aiwei Bai, Haoping Lu, Zhiyun Kong, Xiang Wang, Simon Shan, Jiulong Cao, Meng Wen, Lijie |
| author_facet | Liu, Aiwei Bai, Haoping Lu, Zhiyun Kong, Xiang Wang, Simon Shan, Jiulong Cao, Meng Wen, Lijie |
| contents | Aligning large language models (LLMs) with human expectations without human-annotated preference data is an important problem. In this paper, we propose a method to evaluate the response preference by using the output probabilities of response pairs under contrastive prompt pairs, which could achieve better performance on LLaMA2-7B and LLaMA2-13B compared to RLAIF. Based on this, we propose an automatic alignment method, Direct Large Model Alignment (DLMA). First, we use contrastive prompt pairs to automatically generate preference data. Then, we continue to evaluate the generated preference data using contrastive prompt pairs and calculate a self-rewarding score. Finally, we use the DPO algorithm to effectively align LLMs by combining this self-rewarding score. In the experimental stage, our DLMA method could surpass the \texttt{RLHF} method without relying on human-annotated preference data. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_11907 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Direct Large Language Model Alignment Through Self-Rewarding Contrastive Prompt Distillation Liu, Aiwei Bai, Haoping Lu, Zhiyun Kong, Xiang Wang, Simon Shan, Jiulong Cao, Meng Wen, Lijie Computation and Language 68T50 I.2.7 Aligning large language models (LLMs) with human expectations without human-annotated preference data is an important problem. In this paper, we propose a method to evaluate the response preference by using the output probabilities of response pairs under contrastive prompt pairs, which could achieve better performance on LLaMA2-7B and LLaMA2-13B compared to RLAIF. Based on this, we propose an automatic alignment method, Direct Large Model Alignment (DLMA). First, we use contrastive prompt pairs to automatically generate preference data. Then, we continue to evaluate the generated preference data using contrastive prompt pairs and calculate a self-rewarding score. Finally, we use the DPO algorithm to effectively align LLMs by combining this self-rewarding score. In the experimental stage, our DLMA method could surpass the \texttt{RLHF} method without relying on human-annotated preference data. |
| title | Direct Large Language Model Alignment Through Self-Rewarding Contrastive Prompt Distillation |
| topic | Computation and Language 68T50 I.2.7 |
| url | https://arxiv.org/abs/2402.11907 |