{
  "source_file": "0930",
  "source_platform": "研究机构",
  "author": "OpenAI Blog",
  "title_listed": "Scaling Laws for Neural Language Models",
  "keywords": "缩放定律,语言模型,规模",
  "domain": "大模型算法",
  "content_type": "深度技术解析",
  "is_paid_flag": "否",
  "url": "https://openai.com/research/scaling-laws-for-neural-language-models",
  "publish_date": "2020-01",
  "corpus_stage": "预训练/SFT",
  "notes": "神经语言模型缩放定律",
  "id": "e6901bb4c72c",
  "collected_at": "2026-10-02T15:28:53",
  "title": "Scaling laws for neural language models",
  "body_markdown": "Abstract\nWe study empirical scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training, with some trends spanning more than seven orders of magnitude. Other architectural details such as network width or depth have minimal effects within a wide range. Simple equations govern the dependence of overfitting on model/dataset size and the dependence of training speed on model size. These relationships allow us to determine the optimal allocation of a fixed compute budget. Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence.",
  "metadata": {
    "title": "Just a moment...",
    "description": null,
    "keywords": null,
    "author": null
  },
  "paywall": {
    "listed_paid": false,
    "detected": false,
    "marker": "",
    "public_only": false,
    "policy": "付费/会员内容只保留页面公开可见部分，不绕过付费墙"
  },
  "crawl": {
    "success": true,
    "status_code": 200,
    "error": "",
    "retried": true,
    "engine": "curl_cffi+trafilatura",
    "fetched_from": "https://openai.com/index/scaling-laws-for-neural-language-models/"
  },
  "code_blocks": [],
  "images": [],
  "audio": [],
  "videos": [],
  "links": [],
  "retry_at": "2026-10-03T08:11:57",
  "final_url": "https://openai.com/index/scaling-laws-for-neural-language-models/",
  "source_note": "recovered from relocated public URL or Wayback snapshot; not a paywall bypass",
  "repair_version": 2,
  "repaired_at": "2026-10-20T17:34:51Z",
  "source_url": "https://openai.com/research/scaling-laws-for-neural-language-models",
  "delivery_status": "unverified",
  "discovery_status": "http_403",
  "missing_reason": "http_403",
  "repair_error": "",
  "actual_video_files": [],
  "source_body_markdown": "Abstract\nWe study empirical scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training, with some trends spanning more than seven orders of magnitude. Other architectural details such as network width or depth have minimal effects within a wide range. Simple equations govern the dependence of overfitting on model/dataset size and the dependence of training speed on model size. These relationships allow us to determine the optimal allocation of a fixed compute budget. Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence.",
  "body_base_path": "attachments/e6901bb4c72c/pages/",
  "body_markdown_path": "attachments/e6901bb4c72c/pages/archived-body.md",
  "local_html": "attachments/e6901bb4c72c/pages/archived-body.html",
  "body_source": "prior_capture"
}