drafting

DeLS-Spec decouples long- and short-context modeling in speculative decoding by adding a lightweight local head to DFlash, achieving consistent speedups without full retraining. It requires only standard next-token prediction training for the local head and improves acceptance length on Qwen3 benchmarks.

Grok by SpaceXAI is a Microsoft Word add-in that helps users draft, restructure, and tighten wording directly within the Word panel.

This paper proposes Speculative Pipeline Decoding (SPD), a framework that uses pipeline parallelism within a single LLM to enable parallel token speculation, avoiding the latency bubbles and accuracy degradation of multi-token prediction in traditional speculative decoding.