DecIF: Improving Instruction-Following through Meta-Decomposition
This work addresses the need for flexible and generalizable instruction-following data synthesis in LLMs, representing an incremental improvement over existing methods that depend on pre-existing documents.
The paper tackles the problem of generating diverse and high-quality instruction-following data for large language models (LLMs) by introducing DecIF, a fully autonomous framework that uses meta-decomposition to produce such data without relying on external resources, resulting in superior performance in instruction-following tasks as demonstrated through extensive experiments.
Instruction-following has emerged as a crucial capability for large language models (LLMs). However, existing approaches often rely on pre-existing documents or external resources to synthesize instruction-following data, which limits their flexibility and generalizability. In this paper, we introduce DecIF, a fully autonomous, meta-decomposition guided framework that generates diverse and high-quality instruction-following data using only LLMs. DecIF is grounded in the principle of decomposition. For instruction generation, we guide LLMs to iteratively produce various types of meta-information, which are then combined with response constraints to form well-structured and semantically rich instructions. We further utilize LLMs to detect and resolve potential inconsistencies within the generated instructions. Regarding response generation, we decompose each instruction into atomic-level evaluation criteria, enabling rigorous validation and the elimination of inaccurate instruction-response pairs. Extensive experiments across a wide range of scenarios and settings demonstrate DecIF's superior performance on instruction-following tasks. Further analysis highlights its strong flexibility, scalability, and generalizability in automatically synthesizing high-quality instruction data.