metadata

license: llama3
base_model: Magpie-Align/Llama-3-8B-Magpie-Align-SFT-v0.2
tags:
  - alignment-handbook
  - axolotl
  - trl
  - dpo
  - sft
  - generated_from_trainer
datasets:
  - princeton-nlp/llama3-ultrafeedback-armorm
  - Magpie-Align/Magpie-Pro-MT-300K-v0.1
  - Magpie-Align/Magpie-Reasoning-150K
model-index:
  - name: Magpie-Align/Llama-3-8B-Magpie-Align-v0.2
    results: []
language:
  - en

🔥 Chat with Magpie Here!

🐦 Llama-3-8B-Magpie-Align-v0.2

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-3-8B. We apply the following pipeline:

We first use Magpie-Align/Magpie-Pro-MT-300K-v0.1 and Magpie-Align/Magpie-Reasoning-150K dataset and perform SFT -> Magpie-Align/Llama-3-8B-Magpie-Align-SFT-v0.2
We then perform DPO on the princeton-nlp/llama3-ultrafeedback-armorm dataset.

The overall performance is even better than the official Llama-3-8B-Instruct Model!

Alpaca Eval 2 (vs GPT-4-Turbo-1106): 49.86 (LC), 51.98 (WR)
Alpaca Eval 2 (vs Llama-3-8B-Instruct): 75.17 (LC), 78.20 (WR)
Arena Hard: 37.5
WildBench WB-Score: 42.7
Zero-Eval GSM: 46.70

🔥 Model Performance

We compare our Llama-3-8B-Magpie-Align with official and other open-aligned LLMs that have been fine-tuned from base models and have publicly released their training datasets. The results are as follows:

+---------------------------------------------+--------------------+--------------------+-----------------------+------------+
|               Aligned Model ID              |      MT-Bench      |   Alpaca Eval 2    |     Alpaca Eval 2     | Arena Hard |
|                                             |                    | (GPT-4-Turbo-1106) | (Llama-3-8B-Instruct) |            |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
|                                             |  R1  |  R2  |  AVG |   LC WR  |    WR   |   LC WR   |     WR    |    Score   |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| meta-llama/Meta-Llama-3-8B-Instruct         | 8.31 | 7.65 | 7.98 |   22.92  |  22.57  |     50    |     50    |    20.6    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| princeton-nlp/Llama-3-Base-8B-SFT-DPO       | 8.12 | 7.23 | 7.67 |   17.71  |  15.34  |   43.73   |   38.80   |    14.8    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| NousResearch/Hermes-2-Pro-Llama-3-8B        | 8.05 | 7.35 | 7.70 |   15.60  |  12.86  |   36.37   |   30.52   |    11.5    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| allenai/llama-3-tulu-2-dpo-8b               | 7.71 | 7.15 | 7.43 |   14.89  |  14.80  |   35.43   |   35.42   |    11.7    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| cognitivecomputations/dolphin-2.9-llama3-8b | 7.97 | 6.98 | 7.47 |   12.50  |   8.79  |   32.67   |   22.80   |     8.2    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| openchat/openchat-3.6-8b-20240522           | 7.83 | 7.23 | 7.53 |   17.70  |  12.53  |   41.30   |   30.79   |     6.7    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| Magpie-Align/Llama-3-8B-Magpie-Align-v0.1   | 8.01 | 7.63 | 7.82 |   38.52  |  38.47  |   69.37   |   70.05   |    32.4    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+
| Magpie-Align/Llama-3-8B-Magpie-Align-v0.2   | 7.81 | 7.64 | 7.73 |   49.86  |  51.98  |   75.17   |   78.20   |    37.5    |
+---------------------------------------------+------+------+------+----------+---------+-----------+-----------+------------+

👀 Other Information

License: Please follow Meta Llama 3 Community License.

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

How to use it? Please check the official Llama 3 repository for detailed instructions. Simply replace the original model_id with Magpie-Align/Llama-3-8B-Magpie-Align-SFT-v1.0.

The detailed training pipeline is as follows.

Stage 1: Supervised Fine-tuning

We use Axolotl for SFT.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 32
total_train_batch_size: 128
total_eval_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 79
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
0.8241	0.0024	1	0.8068
0.5623	0.2007	85	0.5087
0.4704	0.4014	170	0.4326
0.4478	0.6020	255	0.4079
0.4256	0.8027	340	0.3948
0.4261	1.0034	425	0.3867
0.3662	1.1844	510	0.3850
0.363	1.3851	595	0.3823
0.357	1.5858	680	0.3813
0.3677	1.7865	765	0.3813

Framework versions

Transformers 4.42.3
Pytorch 2.3.1+cu121
Datasets 2.19.1
Tokenizers 0.19.1

Internal name for identification: Llama-3-8B-Magpie-Mix-300KMT-150KR. Please change the model name in the below Axolotl config.

See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3-8B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: Magpie-Align/Magpie-Reasoning-150K
    type: sharegpt
    conversation: llama3
  - path: Magpie-Align/Magpie-Pro-MT-300K-v0.1
    type: sharegpt
    conversation: llama3
dataset_prepared_path: last_run_prepared
val_set_size: 0.001
output_dir: axolotl_out/Llama-3-8B-Magpie-Mix-300KMT-150KR

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

wandb_project: SynDa
wandb_entity:
wandb_watch:
wandb_name: Llama-3-8B-Magpie-Mix-300KMT-150KR
wandb_log_model:
hub_model_id: Magpie-Align/Llama-3-8B-Magpie-Mix-300KMT-150KR

gradient_accumulation_steps: 32
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_ratio: 0.1
evals_per_epoch: 5
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

Stage 2: Direct Preference Optimization

We use alignment handbook for DPO.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 8e-07
train_batch_size: 2
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 8
total_train_batch_size: 128
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
0.5793	0.2137	100	0.5183	-2.6521	-3.4319	0.7460	0.7798	-616.7447	-544.7192	-1.2844	-1.2700
0.5011	0.4275	200	0.4428	-3.5015	-4.6329	0.7903	1.1314	-736.8406	-629.6548	-1.3147	-1.2977
0.3663	0.6412	300	0.4012	-3.8886	-5.3500	0.8387	1.4613	-808.5509	-668.3669	-1.3327	-1.3138
0.3856	0.8549	400	0.3841	-4.5601	-6.2488	0.8589	1.6887	-898.4371	-735.5136	-1.3606	-1.3413

Framework versions

Transformers 4.42.3
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-3-8B-Magpie-Mix-300KMT-150KR
hub_model_id: Magpie-Align/Llama-3-8B-Magpie-Align-v0.2-RC
output_dir: alignment_handbook_out/Llama-3-8B-Magpie-Align-v0.2-RC
run_name: Llama-3-8B-Magpie-Align-v0.2-RC

dataset_mixer:
  princeton-nlp/llama3-ultrafeedback-armorm: 1.0
dataset_splits:
- train
- test
preprocessing_num_workers: 24

# DPOTrainer arguments
bf16: true
beta: 0.01
learning_rate: 0.8e-6
gradient_accumulation_steps: 8
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

Downstream Performance

Datasets	Llama-3-8B-Magpie-Align-v0.2
MMLU (5)	65.42
ARC (25)	63.91
HellaSwag (25)	81.66
TruthfulQA (0)	60.97
Winogrande (5)	73.40

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.