New Paper Proposes Mixture Block Drafters for Tree-Verified Speculative Decoding
An arXiv paper introduces Multi-Future Drafting, a method using mixture block drafters to draft blocks of future tokens in one forward pass, with draft trees verifying the highest-ranked prefixes.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-11 05:00 · arXiv cs.AI
Multi-Future Drafting: Mixture Block Drafters for Tree-Verified Speculative Decoding