angelahzyuan's picture
Update README.md
2076437 verified
metadata
license: apache-2.0
datasets:
  - openbmb/UltraFeedback
language:
  - en
pipeline_tag: text-generation

Self-Play Preference Optimization for Language Model Alignment (https://arxiv.org/abs/2405.00675)

Llama-3-Instruct-8B-SPPO-Iter1

This model was developed using Self-Play Preference Optimization at iteration 1, 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.

Links to Other Models

Model Description

  • Model type: A 8B parameter GPT-like model fine-tuned on synthetic datasets.
  • Language(s) (NLP): Primarily English
  • License: Apache-2.0
  • Finetuned from model: meta-llama/Meta-Llama-3-8B-Instruct

AlpacaEval Leaderboard Evaluation Results

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

Open LLM Leaderboard Evaluation Results

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

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 5e-07
  • eta: 1000
  • per_device_train_batch_size: 8
  • gradient_accumulation_steps: 1
  • seed: 42
  • distributed_type: deepspeed_zero3
  • num_devices: 8
  • optimizer: RMSProp
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_train_epochs: 6.0 (stop at epoch=1.0)

Citation

@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}
}