🔥 Chat with Magpie Here!

🐦 Llama-3.1-8B-Magpie-Align-v0.1

Project Web: https://magpie-align.github.io/

Online Model Demo: https://huggingface.co/spaces/flydust/Chat-with-Magpie

Arxiv Technical Report: https://arxiv.org/abs/2406.08464

Codes: https://github.com/magpie-align/magpie

🧐 About This Model

This model is an aligned version of meta-llama/Meta-Llama-3.1-8B. We apply the following pipeline:

We first perform SFT using:

• Magpie-Align/Magpie-Pro-MT-300K-v0.1, and
• Magpie-Align/Magpie-Reasoning-150K
• SFT Model Checkpoint: Magpie-Align/Llama-3.1-8B-Magpie-Align-SFT-v0.1

We then perform DPO on the princeton-nlp/llama3-ultrafeedback-armorm dataset.

The overall performance is much better than the official Llama-3.1-8B-Instruct Model!

• MT-Bench: 8.375 (1st Turn), 7.650 (Second Turn), 8.013 (Average)
• Alpaca Eval 2 (GPT-4-Turbo-1106): 45.73 (LC), 52.79 (WR)
• Arena Hard: 42.4

👀 Other Information

License: Please follow Meta Llama 3 Community License (Data) and Meta Llama 3.1 Community License (Model).

Conversation Template: Please use Llama 3 official chat template for the best performance.

How to use it? Please check the official Llama 3.1 repository for detailed instructions. Simply replace the original model_id with this model id.

---

Alignment Pipeline

The detailed alignment pipeline is as follows.

Stage 1: Supervised Fine-tuning

We use Axolotl for SFT. Please refer to the model card of SFT checkpoint for detailed configurations.

Stage 2: Direct Preference Optimization

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-06
• train_batch_size: 2
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 16
• total_train_batch_size: 256
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.4439	0.4275	100	0.4168	-4.9964	-6.3086	0.8145	1.3123	-829.7151	-704.6570	-0.5150	-0.5001
0.343	0.8549	200	0.3298	-4.9310	-6.7966	0.8952	1.8655	-878.5105	-698.1248	-0.5776	-0.5622

Framework versions

• Transformers 4.43.2
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1

See alignment handbook config

# Customized Configs
model_name_or_path: Magpie-Align/Llama-3.1-8B-Magpie-Align-SFT-v0.1
hub_model_id: Magpie-Align/Llama-3.1-8B-Magpie-Align-v0.1
output_dir: /data/zhangchen_xu/alignment_handbook_out/Llama-3.1-8B-Magpie-Align-v0.1
run_name: Llama-3.1-8B-Magpie-Align-v0.1

dataset_mixer:
  princeton-nlp/llama3-ultrafeedback-armorm: 1.0
dataset_splits:
- train
- test
preprocessing_num_workers: 64

# DPOTrainer arguments
bf16: true
beta: 0.01
learning_rate: 1.0e-6
gradient_accumulation_steps: 16
per_device_train_batch_size: 2
per_device_eval_batch_size: 4
num_train_epochs: 1
max_length: 2048
max_prompt_length: 1800
warmup_ratio: 0.1
logging_steps: 1
lr_scheduler_type: cosine
optim: adamw_torch

torch_dtype: null
use_flash_attention_2: true
do_eval: true
evaluation_strategy: steps
eval_steps: 100
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: False
log_level: info
push_to_hub: true
save_strategy: "steps"
save_steps: 100
save_total_limit: 1
seed: 42
report_to:
- wandb

---

Paper Abstract

Click Here

High-quality instruction data is critical for aligning large language models (LLMs). Although some models, such as Llama-3-Instruct, have open weights, their alignment data remain private, which hinders the democratization of AI. High human labor costs and a limited, predefined scope for prompting prevent existing open-source data creation methods from scaling effectively, potentially limiting the diversity and quality of public alignment datasets. Is it possible to synthesize high-quality instruction data at scale by extracting it directly from an aligned LLM? We present a self-synthesis method for generating large-scale alignment data named Magpie. Our key observation is that aligned LLMs like Llama-3-Instruct can generate a user query when we input only the left-side templates up to the position reserved for user messages, thanks to their auto-regressive nature. We use this method to prompt Llama-3-Instruct and generate 4 million instructions along with their corresponding responses. We perform a comprehensive analysis of the extracted data and select 300K high-quality instances. To compare Magpie data with other public instruction datasets, we fine-tune Llama-3-8B-Base with each dataset and evaluate the performance of the fine-tuned models. Our results indicate that in some tasks, models fine-tuned with Magpie perform comparably to the official Llama-3-8B-Instruct, despite the latter being enhanced with 10 million data points through supervised fine-tuning (SFT) and subsequent feedback learning. We also show that using Magpie solely for SFT can surpass the performance of previous public datasets utilized for both SFT and preference optimization, such as direct preference optimization with UltraFeedback. This advantage is evident on alignment benchmarks such as AlpacaEval, ArenaHard, and WildBench.

📚 Citation

If you find the model, data, or code useful, please cite our paper:

@article{xu2024magpie,
    title={Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing}, 
    author={Zhangchen Xu and Fengqing Jiang and Luyao Niu and Yuntian Deng and Radha Poovendran and Yejin Choi and Bill Yuchen Lin},
    year={2024},
    eprint={2406.08464},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Please also cite the creators of preference datasets:

SimPO paper:

@article{meng2024simpo,
  title={{SimPO}: Simple preference optimization with a reference-free reward},
  author={Meng, Yu and Xia, Mengzhou and Chen, Danqi},
  journal={arXiv preprint arXiv:2405.14734},
  year={2024}
}

UltraFeedback paper:

@article{cui2023ultrafeedback,
  title={{UltraFeedback}: Boosting language models with high-quality feedback},
  author={Cui, Ganqu and Yuan, Lifan and Ding, Ning and Yao, Guanming and Zhu, Wei and Ni, Yuan and Xie, Guotong and Liu, Zhiyuan and Sun, Maosong},
  journal={arXiv preprint arXiv:2310.01377},
  year={2023}
}

ArmoRM paper:

@article{wang2024interpretable,
  title={Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts},
  author={Wang, Haoxiang and Xiong, Wei and Xie, Tengyang and Zhao, Han and Zhang, Tong},
  journal={arXiv preprint arXiv:2406.12845},
  year={2024}
}

Questions? Please contact Zhangchen by email.