{
  "id": 9256659,
  "title": "Train Where the Quantized Model Goes: On-Policy Distillation for Low-Bit Reasoning",
  "url": "https://urgent.news/2026/09/22/train-where-the-quantized-model-goes-on-policy-distillation-for-low",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-22T17:01:09.000Z",
  "source": {
    "name": "arXiv cs.AI",
    "slug": "arxiv-cs-ai",
    "url": "https://arxiv.org/abs/2609.26708v1"
  },
  "original_language": "en",
  "account": null,
  "summary": "Quantization-aware distillation (QAD) restores much of the short-form question-answering performance lost to sub-3-bit quantization, yet leaves mathematical and code reasoning substantially impaired. Long generations often degenerate into repetitive loops, exhausting the decoding budget without completing a solution. We trace this gap to quantization-amplified exposure bias: QAD trains on fixed…",
  "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."
}