CritiQ: Mining Data Quality Criteria from Human Preferences
This addresses the need for efficient and interpretable data selection in language model training, reducing reliance on expert heuristics and human annotation, though it appears incremental as it builds on existing preference-based and agent-based approaches.
The paper tackles the problem of data selection for language models by introducing CritiQ, a method that automatically mines data quality criteria from human preferences using only about 30 annotated pairs, achieving improved performance in code, math, and logic domains and enhancing downstream task results for Llama 3.1 models compared to uniform sampling.
Language model heavily depends on high-quality data for optimal performance. Existing approaches rely on manually designed heuristics, the perplexity of existing models, training classifiers, or careful prompt engineering, which require significant expert experience and human annotation effort while introduce biases. We introduce CritiQ, a novel data selection method that automatically mines criteria from human preferences for data quality with only ~30 human-annotated pairs and performs efficient data selection. The main component, CritiQ Flow, employs a manager agent to evolve quality criteria and worker agents to make pairwise judgments. We build a knowledge base that extracts quality criteria from previous work to boost CritiQ Flow. Compared to perplexity- and classifier- based methods, verbal criteria are more interpretable and possess reusable value. After deriving the criteria, we train the CritiQ Scorer to give quality scores and perform efficient data selection. We demonstrate the effectiveness of our method in the code, math, and logic domains, achieving high accuracy on human-annotated test sets. To validate the quality of the selected data, we continually train Llama 3.1 models and observe improved performance on downstream tasks compared to uniform sampling. Ablation studies validate the benefits of the knowledge base and the reflection process. We analyze how criteria evolve and the effectiveness of majority voting.