DBLast augments a block diffusion drafter with a low-rank latent mixture and an acceptance-oriented training objective, increasing accepted draft length in non-greedy speculative decoding.
Proceedings of the 41st International Conference on Machine Learning , year =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DBLAST: Dependent Block Drafting for Stochastic Speculative Decoding
DBLast augments a block diffusion drafter with a low-rank latent mixture and an acceptance-oriented training objective, increasing accepted draft length in non-greedy speculative decoding.