{
  "schema_version": 1,
  "title": "Repro - Midtraining Bridges Pretraining and Posttraining Distributions",
  "emoji": "\ud83d\udd2c",
  "space_id": "bertfil/5PfEQzE9bf",
  "paper": {
    "arxiv_id": "2510.14865",
    "openreview_id": "5PfEQzE9bf",
    "title": "Midtraining Bridges Pretraining and Posttraining Distributions",
    "url": "https://arxiv.org/abs/2510.14865"
  },
  "tags": [
    "icml2026-repro",
    "paper-5PfEQzE9bf"
  ],
  "updated_at": "2026-07-20T03:52:36+00:00",
  "root": {
    "slug": "index",
    "title": "Repro - Midtraining Bridges Pretraining and Posttraining Distributions",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "00-scored-evidence-summary",
        "title": "00 - Scored evidence summary",
        "file": "pages/00-scored-evidence-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1-the-paper-studies-midtraining-as-distributional-br",
        "title": "Claim 1 - The paper studies midtraining as distributional bridging tha\u2026",
        "file": "pages/claim-1-the-paper-studies-midtraining-as-distributional-br/page.md",
        "children": []
      },
      {
        "slug": "claim-2-controlled-experiments-pretrain-pythia-family-mode",
        "title": "Claim 2 - Controlled experiments pretrain Pythia-family models from 70\u2026",
        "file": "pages/claim-2-controlled-experiments-pretrain-pythia-family-mode/page.md",
        "children": []
      },
      {
        "slug": "claim-3-code-focused-midtraining-yields-the-largest-gains",
        "title": "Claim 3 - Code-focused midtraining yields the largest gains on code ta\u2026",
        "file": "pages/claim-3-code-focused-midtraining-yields-the-largest-gains/page.md",
        "children": []
      },
      {
        "slug": "claim-4-proximity-advantage-between-midtraining-and-target",
        "title": "Claim 4 - Proximity advantage between midtraining and target SFT data \u2026",
        "file": "pages/claim-4-proximity-advantage-between-midtraining-and-target/page.md",
        "children": []
      },
      {
        "slug": "claim-5-midtraining-mixtures-outperform-continued-pretrain",
        "title": "Claim 5 - Midtraining mixtures outperform continued pretraining on 100\u2026",
        "file": "pages/claim-5-midtraining-mixtures-outperform-continued-pretrain/page.md",
        "children": []
      },
      {
        "slug": "claim-6-for-code-midtraining-early-specialized-data-introd",
        "title": "Claim 6 - For code midtraining, early specialized-data introduction su\u2026",
        "file": "pages/claim-6-for-code-midtraining-early-specialized-data-introd/page.md",
        "children": []
      },
      {
        "slug": "reproduction-protocol-and-provenance",
        "title": "Reproduction protocol and provenance",
        "file": "pages/reproduction-protocol-and-provenance/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  },
  "agent_view_tokens": 7412,
  "revision": "1784519556067153100"
}