Clean First, Align Later: Benchmarking Preference Data Cleaning for Reliable LLM Alignment
This research evaluates 13 data cleaning methods to standardize benchmarking for reliable LLM alignment. It explores how modular preprocessing strategies dir...