New paper proposes multi-draft on-policy distillation for diffusion drafters in speculative decoding
An arXiv paper introduces multi-draft on-policy distillation for one-step diffusion drafters, which propose blocks of multiple tokens in parallel to reduce language model inference cost via speculative decoding.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-10-30 04:00 · arXiv cs.CL
Multi-Draft On-Policy Distillation for Diffusion Drafters