Diverse, not Short: A Length-Controlled Data Selection Strategy for Improving Response Diversity of Language Models
This addresses a specific bottleneck in language model training for creative and open-ended tasks, offering an incremental improvement in response diversity.
The paper tackled the problem of language models producing short, less diverse responses due to biases in diversity metrics and reward models, and introduced Diverse-NS, a length-controlled data selection strategy that improved lexical and semantic diversity while maintaining length parity, using only 3,000 preference pairs and showing consistent gains on creative generation tasks.
Diverse language model responses are crucial for creative generation, open-ended tasks, and self-improvement training. We show that common diversity metrics, and even reward models used for preference optimization, systematically bias models toward shorter outputs, limiting expressiveness. To address this, we introduce Diverse, not Short (Diverse-NS), a length-controlled data selection strategy that improves response diversity while maintaining length parity. By generating and filtering preference data that balances diversity, quality, and length, Diverse-NS enables effective training using only 3,000 preference pairs. Applied to LLaMA-3.1-8B and the Olmo-2 family, Diverse-NS substantially enhances lexical and semantic diversity. We show consistent improvement in diversity with minor reduction or gains in response quality on four creative generation tasks: Divergent Associations, Persona Generation, Alternate Uses, and Creative Writing. Surprisingly, experiments with the Olmo-2 model family (7B, and 13B) show that smaller models like Olmo-2-7B can serve as effective "diversity teachers" for larger models. By explicitly addressing length bias, our method efficiently pushes models toward more diverse and expressive outputs.