CARVE: Verified Expansion for Variable-Length Generation in Diffusion Language Models
CARVE (Counterfactual-Aware Reveal with Verified Expansion), a training-free variable-length algorithm for masked diffusion LMs that consistently improves average performance over fixed-length baselines across all evaluated model families.
Wail Bouhedja, Amr Mohamed, Guo-Kan Shang
· 0 citations