SPPO
Collection
Self-Play Preference Optimization
β’
10 items
β’
Updated
β’
12
Self-Play Preference Optimization for Language Model Alignment (https://arxiv.org/abs/2405.00675)
This model was developed using Self-Play Preference Optimization at iteration 3, based on the meta-llama/Meta-Llama-3-8B-Instruct architecture as starting point. We utilized the prompt sets from the openbmb/UltraFeedback dataset, splited to 3 parts for 3 iterations by snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset. All responses used are synthetic.
Model | LC. Win Rate | Win Rate | Avg. Length |
---|---|---|---|
Llama-3-8B-SPPO Iter1 | 31.73 | 31.74 | 1962 |
Llama-3-8B-SPPO Iter2 | 35.15 | 35.98 | 2021 |
Llama-3-8B-SPPO Iter3 | 38.77 | 39.85 | 2066 |
Results are reported by using lm-evaluation-harness v0.4.1
arc_challenge | truthfulqa_mc2 | winogrande | gsm8k | hellaswag | mmlu | average | |
---|---|---|---|---|---|---|---|
Llama-3-8B-SPPO Iter1 | 63.82 | 54.96 | 76.40 | 75.44 | 79.80 | 65.65 | 69.35 |
Llama-3-8B-SPPO Iter2 | 64.93 | 56.48 | 76.87 | 75.13 | 80.39 | 65.67 | 69.91 |
Llama-3-8B-SPPO Iter3 | 65.19 | 58.04 | 77.11 | 74.91 | 80.86 | 65.60 | 70.29 |
Detailed results can be found here
Metric | Value |
---|---|
Avg. | 23.68 |
IFEval (0-Shot) | 68.28 |
BBH (3-Shot) | 29.74 |
MATH Lvl 5 (4-Shot) | 7.33 |
GPQA (0-shot) | 2.01 |
MuSR (0-shot) | 3.09 |
MMLU-PRO (5-shot) | 29.38 |
The following hyperparameters were used during training:
@misc{wu2024self,
title={Self-Play Preference Optimization for Language Model Alignment},
author={Wu, Yue and Sun, Zhiqing and Yuan, Huizhuo and Ji, Kaixuan and Yang, Yiming and Gu, Quanquan},
year={2024},
eprint={2405.00675},
archivePrefix={arXiv},
primaryClass={cs.LG}
}