{
  "schema_version": 2,
  "title": "Reproduction: How much can language models memorize?",
  "emoji": "🎯",
  "space_id": "Varshith19/repro-how-much-can-language-models-memorize",
  "paper": null,
  "tags": [
    "icml2026-repro",
    "paper-bA6BgSbaUi"
  ],
  "updated_at": "2026-07-22T12:49:11+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction: How much can language models memorize?",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "executive-summary",
        "title": "Executive summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1-gpt-style-transformers-trained-on-uniform-random-data-show-an-empirical-memorization-capacity-plateau-of-about-3-6-bits-per-parameter-figure-1",
        "title": "Claim 1: GPT-style transformers trained on uniform random data show an empirical memorization-capacity plateau of about 3.6 bits per parameter (Figure 1)",
        "file": "pages/claim-1-gpt-style-transformers-trained-on-uniform-random-data-show-an-empirical-memorization-capacity-plateau-of-about-3-6-bits-per-parameter-figure-1/page.md",
        "children": []
      },
      {
        "slug": "claim-2-capacity-estimates-across-model-widths-and-depths-support-a-roughly-linear-bits-per-parameter-scaling-law-with-bfloat16-to-float32-increasing-capacity-only-modestly-table-1",
        "title": "Claim 2: Capacity estimates across model widths and depths support a roughly linear bits-per-parameter scaling law, with bfloat16 to float32 increasing capacity only modestly (Table 1)",
        "file": "pages/claim-2-capacity-estimates-across-model-widths-and-depths-support-a-roughly-linear-bits-per-parameter-scaling-law-with-bfloat16-to-float32-increasing-capacity-only-modestly-table-1/page.md",
        "children": []
      },
      {
        "slug": "claim-3-on-text-data-unintended-memorization-rises-with-model-size-but-decreases-once-models-begin-generalizing-relative-to-an-oracle-reference-model-figure-2",
        "title": "Claim 3: On text data, unintended memorization rises with model size but decreases once models begin generalizing relative to an oracle reference model (Figure 2)",
        "file": "pages/claim-3-on-text-data-unintended-memorization-rises-with-model-size-but-decreases-once-models-begin-generalizing-relative-to-an-oracle-reference-model-figure-2/page.md",
        "children": []
      },
      {
        "slug": "claim-4-double-descent-begins-when-dataset-information-content-exceeds-estimated-model-capacity-in-both-synthetic-bitstrings-and-text-experiments-figures-3-and-4",
        "title": "Claim 4: Double descent begins when dataset information content exceeds estimated model capacity in both synthetic bitstrings and text experiments (Figures 3 and 4)",
        "file": "pages/claim-4-double-descent-begins-when-dataset-information-content-exceeds-estimated-model-capacity-in-both-synthetic-bitstrings-and-text-experiments-figures-3-and-4/page.md",
        "children": []
      },
      {
        "slug": "claim-5-the-paper-derives-and-evaluates-scaling-law-predictions-for-membership-inference-as-a-function-of-model-capacity-and-dataset-size-figure-7",
        "title": "Claim 5: The paper derives and evaluates scaling-law predictions for membership inference as a function of model capacity and dataset size (Figure 7)",
        "file": "pages/claim-5-the-paper-derives-and-evaluates-scaling-law-predictions-for-membership-inference-as-a-function-of-model-capacity-and-dataset-size-figure-7/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  },
  "traces": [],
  "workspace": {
    "file": "workspace.json",
    "file_count": 0,
    "total_size": 0,
    "bucket_id": null
  },
  "agent_view_tokens": 5591,
  "trace_view_tokens": 49,
  "workspace_view_tokens": 8,
  "revision": "bec1d8cc4213c1dd707a",
  "traces_ref": {
    "repo_id": "Varshith19/repro-how-much-can-language-models-memorize-traces",
    "repo_type": "dataset",
    "repo_url": "https://huggingface.co/datasets/Varshith19/repro-how-much-can-language-models-memorize-traces",
    "private": true
  },
  "trace_dataset": "https://huggingface.co/datasets/Varshith19/repro-how-much-can-language-models-memorize-traces"
}