Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing
This addresses the challenge of democratizing AI by providing scalable, high-quality alignment data without human labor, though it is incremental as it builds on existing LLM capabilities.
The paper tackles the problem of generating high-quality alignment data for large language models by proposing Magpie, a self-synthesis method that extracts instructions from aligned LLMs like Llama-3-Instruct, producing 4 million instructions and selecting 300K high-quality instances. The result shows that models fine-tuned with Magpie data perform comparably to official models in some tasks and surpass previous public datasets on benchmarks like AlpacaEval, ArenaHard, and WildBench.
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.