SPPO-Llama-3-8B-Instruct-GPM-2B/README.md

language, license, datasets, pipeline_tag, model-index

language	license	datasets	pipeline_tag	model-index
en	apache-2.0	openbmb/UltraFeedback	text-generation	name results SPPO-Llama-3-8B-Instruct-GPM-2B task dataset metrics source type name text-generation Text Generation name type args IFEval (0-Shot) HuggingFaceH4/ifeval num_few_shot 0 type value name inst_level_strict_acc and prompt_level_strict_acc 60.24 strict accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=general-preference/SPPO-Llama-3-8B-Instruct-GPM-2B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args BBH (3-Shot) BBH num_few_shot 3 type value name acc_norm 27.89 normalized accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=general-preference/SPPO-Llama-3-8B-Instruct-GPM-2B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args MATH Lvl 5 (4-Shot) hendrycks/competition_math num_few_shot 4 type value name exact_match 8.01 exact match url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=general-preference/SPPO-Llama-3-8B-Instruct-GPM-2B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args GPQA (0-shot) Idavidrein/gpqa num_few_shot 0 type value name acc_norm 1.23 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=general-preference/SPPO-Llama-3-8B-Instruct-GPM-2B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type args MuSR (0-shot) TAUR-Lab/MuSR num_few_shot 0 type value name acc_norm 3.19 acc_norm url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=general-preference/SPPO-Llama-3-8B-Instruct-GPM-2B Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args MMLU-PRO (5-shot) TIGER-Lab/MMLU-Pro main test num_few_shot 5 type value name acc 29.53 accuracy url name https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=general-preference/SPPO-Llama-3-8B-Instruct-GPM-2B Open LLM Leaderboard

General Preference Modeling with Preference Representations for Aligning Language Models (https://arxiv.org/abs/2410.02197)

SPPO-Llama-3-8B-Instruct-GPM-2B

This model was developed using SPPO at iteration 3 and the General Preference representation Model (GPM) (specifically, using GPM-Gemma-2B), 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

• SPPO-Llama-3-8B-Instruct-GPM-2B
• GPO-Llama-3-8B-Instruct-GPM-2B

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
SPPO-Llama-3-8B-Instruct-GPM-2B	35.30	45.44	2490

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
SPPO-Llama-3-8B-Instruct-GPM-2B	62.03	52.95	76.56	75.36	78.57	65.66	68.52

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

@article{zhang2024general,
  title={General Preference Modeling with Preference Representations for Aligning Language Models},
  author={Zhang, Yifan and Zhang, Ge and Wu, Yue and Xu, Kangping and Gu, Quanquan},
  journal={arXiv preprint arXiv:2410.02197},
  year={2024}
}