---
dataset_info:
features:
- name: uuid
dtype: string
- name: model
dtype: string
- name: gen_input_config
struct:
- name: temperature
dtype: float64
- name: top_p
dtype: float64
- name: input
dtype: string
- name: output
dtype: string
- name: conversations
list:
- name: from
dtype: string
- name: value
dtype: string
- name: task_category
dtype: string
- name: difficulty
dtype: string
- name: intent
dtype: string
- name: knowledge
dtype: string
- name: input_quality
dtype: string
- name: quality_explanation
dtype: string
- name: llama_guard_2
dtype: string
- name: reward_model
dtype: string
- name: instruct_reward
dtype: float64
- name: base_output
dtype: string
- name: base_reward
dtype: float64
- name: reward_difference
dtype: float64
- name: min_neighbor_distance
dtype: float64
- name: repeat_count
dtype: int64
- name: min_similar_uuid
dtype: string
- name: input_length
dtype: int64
- name: output_length
dtype: int64
splits:
- name: train
num_bytes: 19031408037
num_examples: 3000000
download_size: 9936635779
dataset_size: 19031408037
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
Project Web: [https://magpie-align.github.io/](https://magpie-align.github.io/)
Arxiv Technical Report: [https://arxiv.org/abs/2406.08464](https://arxiv.org/abs/2406.08464)
Codes: [https://github.com/magpie-align/magpie](https://github.com/magpie-align/magpie)
## 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.