{
  "id": 19301757,
  "title": "Phrase-Level Speculative Decoding on a Frozen Target",
  "url": "https://urgent.news/2026/11/09/phrase-level-speculative-decoding-on-a-frozen-target",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-09T14:13:32.000Z",
  "source": {
    "name": "arXiv cs.AI",
    "slug": "arxiv-cs-ai",
    "url": "https://arxiv.org/abs/2611.12464v1"
  },
  "original_language": "en",
  "account": null,
  "summary": "Speculative decoders raise throughput by banking more target tokens per forward pass, and Medusa, EAGLE-3, DFlash and DSpark all do it the same way: more candidates that survive verification. All of them draft in the target's own vocabulary, so one node of the candidate tree is one target token, and the tokenizer makes that unit look like a constant. It is not. BPE's pretokenizer splits on…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}