{
 "title": "模型量化（Quantization）是什麼？",
 "site": "AI 與科技名詞白話解釋",
 "canonical": "https://tw-ai-glossary.panda198271.workers.dev/t/quantization",
 "markdown": "https://tw-ai-glossary.panda198271.workers.dev/t/quantization.md",
 "summary": "模型量化是把模型權重等數值從 32 位元浮點數改用 int8、int4 等低精度格式表示，以減少記憶體與運算量的技術。依 Hugging Face 文件，int8 相較 float32 約可讓模型縮小 4 倍，代價是可能略微降低準確度。",
 "date_modified": "2026-09-26",
 "retrieved": "2026-09-27",
 "content_sha256": "fc663adb53b94691c67542d934eb27d1649bc231843db0ef302515e7a81c7feb",
 "fingerprint_basis": "Markdown 版全文，移除贊助行與開頭 front matter，去除頭尾空白",
 "verify": {
  "text_snapshot": "https://tw-ai-glossary.panda198271.workers.dev/cite?path=%2Ft%2Fquantization&format=text",
  "how": "下載 text_snapshot 後計算 SHA-256，應與 content_sha256 相同；內容更新時指紋會改變"
 },
 "sources": [
  {
   "name": "Hugging Face Transformers：Quantization concepts",
   "url": "https://huggingface.co/docs/transformers/quantization/concept_guide"
  },
  {
   "name": "arXiv：Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference（Jacob et al., 2017）",
   "url": "https://arxiv.org/abs/1712.05877"
  }
 ],
 "source_count": 2,
 "cite_as": {
  "zh": "模型量化（Quantization）是什麼？｜AI 與科技名詞白話解釋。https://tw-ai-glossary.panda198271.workers.dev/t/quantization（資料日期 2026-09-26，擷取 2026-09-27）",
  "apa": "AI 與科技名詞白話解釋. (2026). 模型量化（Quantization）是什麼？. Retrieved 2026-09-27, from https://tw-ai-glossary.panda198271.workers.dev/t/quantization",
  "markdown": "[模型量化（Quantization）是什麼？](https://tw-ai-glossary.panda198271.workers.dev/t/quantization)（AI 與科技名詞白話解釋，2026-09-26）"
 },
 "license": "可引用，請附上正式網址與資料日期"
}