Tempered remasking heuristics increase diversity in diffusion language model sampling, closing the pass@k gap with autoregressive methods at equivalent computational cost.
That is, the fork token’s value and the semantic outcome are tightly coupled: knowing one constrains the other to a high degree of certainty
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
A Tale of Two Temperatures: Simple, Efficient, and Diverse Sampling from Diffusion Language Models
Tempered remasking heuristics increase diversity in diffusion language model sampling, closing the pass@k gap with autoregressive methods at equivalent computational cost.