D5P4: Partition Determinantal Point Process for Diversity in Parallel Discrete Diffusion Decoding
Published in arXiv, 2026
Abstract: Discrete diffusion models are promising alternatives to autoregressive approaches for text generation, yet their decoding methods remain under-studied. Standard autoregressive search procedures, such as beam search, do not directly apply to iterative denoising, where hypotheses are complete intermediate sequences rather than left-to-right prefixes. Furthermore, existing diffusion decoding procedures only provide limited control over the diversity and coverage of retained hypotheses. In this work, we introduce D5P4, a beam-style decoding method tailored to discrete diffusion models, which casts intermediate beam selection as MAP inference under a partitioned Determinantal Point Process. This yields a model-internal batch objective that balances quality and diversity without external verifiers. Experiments on open-ended generation, question answering, and mathematical reasoning show that D5P4 improves diversity and pass@k coverage while matching or surpassing baseline quality and fidelity.
Recommended citation: Lys, et al. (2026). "D5P4: Partition Determinantal Point Process for Diversity in Parallel Discrete Diffusion Decoding." arXiv preprint arXiv:2603.19146.
Download Paper | Download Bibtex