$S^3$: Stratified Scaling Search for Test-Time in Diffusion Language Models
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866915921759567872 |
|---|---|
| author | Bilal, Ahsan Mohsin, Muhammad Ahmed Umer, Muhammad Aali, Asad Khanzada, Muhammad Usman Rafique, Muhammad Usman He, Zihao Fox, Emily Hougen, Dean F. |
| author_facet | Bilal, Ahsan Mohsin, Muhammad Ahmed Umer, Muhammad Aali, Asad Khanzada, Muhammad Usman Rafique, Muhammad Usman He, Zihao Fox, Emily Hougen, Dean F. |
| contents | Test-time scaling investigates whether a fixed diffusion language model (DLM) can generate better outputs when given more inference compute, without additional training. However, naive best-of-$K$ sampling is fundamentally limited because it repeatedly draws from the same base diffusion distribution, whose high-probability regions are often misaligned with high-quality outputs. We propose $S^3$ (Stratified Scaling Search), a classical verifier-guided search method that improves generation by reallocating compute during the denoising process rather than only at the final output stage. At each denoising step, $S^3$ expands multiple candidate trajectories, evaluates them with a lightweight reference-free verifier, and selectively resamples promising candidates while preserving diversity within the search frontier. This procedure effectively approximates a reward-tilted sampling distribution that favors higher-quality outputs while remaining anchored to the model prior. Experiments with LLaDA-8B-Instruct on MATH-500, GSM8K, ARC-Challenge, and TruthfulQA demonstrate that $S^3$ consistently improves performance across benchmarks, achieving the largest gains on mathematical reasoning tasks while leaving the underlying model and decoding schedule unchanged. These results show that classical search over denoising trajectories provides a practical mechanism for test-time scaling in DLMs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_06260 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | $S^3$: Stratified Scaling Search for Test-Time in Diffusion Language Models Bilal, Ahsan Mohsin, Muhammad Ahmed Umer, Muhammad Aali, Asad Khanzada, Muhammad Usman Rafique, Muhammad Usman He, Zihao Fox, Emily Hougen, Dean F. Machine Learning Artificial Intelligence Test-time scaling investigates whether a fixed diffusion language model (DLM) can generate better outputs when given more inference compute, without additional training. However, naive best-of-$K$ sampling is fundamentally limited because it repeatedly draws from the same base diffusion distribution, whose high-probability regions are often misaligned with high-quality outputs. We propose $S^3$ (Stratified Scaling Search), a classical verifier-guided search method that improves generation by reallocating compute during the denoising process rather than only at the final output stage. At each denoising step, $S^3$ expands multiple candidate trajectories, evaluates them with a lightweight reference-free verifier, and selectively resamples promising candidates while preserving diversity within the search frontier. This procedure effectively approximates a reward-tilted sampling distribution that favors higher-quality outputs while remaining anchored to the model prior. Experiments with LLaDA-8B-Instruct on MATH-500, GSM8K, ARC-Challenge, and TruthfulQA demonstrate that $S^3$ consistently improves performance across benchmarks, achieving the largest gains on mathematical reasoning tasks while leaving the underlying model and decoding schedule unchanged. These results show that classical search over denoising trajectories provides a practical mechanism for test-time scaling in DLMs. |
| title | $S^3$: Stratified Scaling Search for Test-Time in Diffusion Language Models |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2604.06260 |