{
  "id": 13616,
  "url": "https://arxiv.org/abs/2607.22334v1",
  "title": "Cross-Tokenizer On-Policy Distillation via Byte-Prefix Marginalization",
  "summary": "Open-weight language models from different families exhibit complementary capabilities, motivating their consolidation into a compact student through on-policy distillation (OPD). However, full-vocabulary OPD typically assumes a shared tokenizer, while existing cross-tokenizer methods may discard teacher probability mass or assign it to student tokens with unrelated content. We introduce Byte-Prefix Marginalization (BPM), which re-expresses the teacher's next-token distribution over the student",
  "authors": "Hao Wang, Kun Yuan, Wenlin Zhong, Minglei Zhang, Han Xiao, Ming Sun et al.",
  "category": "research",
  "topics": "regulation,children-education",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-24T14:12:54.000Z",
  "fetched_at": "2026-07-27T05:10:06.638Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/13616",
  "original_url": "https://arxiv.org/abs/2607.22334v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}