DeLS-Spec improves block-parallel speculative decoding by fusing DFlash logits with an independently trained lightweight local head and a unigram prior correction.
For each training sequence, we count only tokens whose loss mask equals one, i.e., assistant-response tokens that contribute to the super- vised objective
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
-
DeLS-Spec: Decoupled Long-Short Contexts for Parallel Speculative Drafting
DeLS-Spec improves block-parallel speculative decoding by fusing DFlash logits with an independently trained lightweight local head and a unigram prior correction.