Condor: Enhance LLM Alignment with Knowledge-Driven Data Synthesis and Refinement

TL;DR

Condor is a two-stage synthetic data generation framework that significantly improves LLM fine-tuning by leveraging knowledge-driven synthesis and self-refinement, reducing reliance on costly human annotations.

Contribution

Introduces Condor, a novel framework combining knowledge-driven data synthesis and self-refinement for scalable high-quality LLM fine-tuning.

Findings

A base model fine-tuned on 20K Condor samples outperforms counterparts.

Refinement stage enables iterative self-improvement across various model scales.

Scaling synthetic data in post-training shows significant untapped potential.

Abstract

The quality of Supervised Fine-Tuning (SFT) data plays a critical role in enhancing the conversational capabilities of Large Language Models (LLMs). However, as LLMs become more advanced, the availability of high-quality human-annotated SFT data has become a significant bottleneck, necessitating a greater reliance on synthetic training data. In this work, we introduce Condor, a novel two-stage synthetic data generation framework that incorporates World Knowledge Tree and Self-Reflection Refinement to produce high-quality SFT data at scale. Our experimental results demonstrate that a base model fine-tuned on only 20K Condor-generated samples achieves superior performance compared to counterparts. The additional refinement stage in Condor further enables iterative self-improvement for LLMs at various scales (up to 72B), validating the effectiveness of our approach. Furthermore, our investigation into the scaling for synthetic data in post-training reveals substantial unexplored potential for performance improvements, opening promising avenues for future research.