2601.09421v2.pdf

{
  "dorsal/arxiv": [
    {
      "date_modified": "2026-02-17T05:53:20.116000Z",
      "hash": "2b70174c2f9629cf26f295402f11928e16b204d94cd6389d3f64496f54146235",
      "id": "038008a9-3a35-413a-b997-d7be9abea12c",
      "source": {
        "execution_id": "167ed49c-d013-4e85-83a5-877f4a45f757",
        "id": "arXiv Dataset",
        "type": "Model",
        "variant": "snapshot-2026-01-17",
        "version": "0.1.0"
      },
      "user_id": 1000002
    }
  ],
  "file/base": {
    "record": {
      "extension": ".pdf",
      "hash": "2b70174c2f9629cf26f295402f11928e16b204d94cd6389d3f64496f54146235",
      "media_type": "application/pdf",
      "media_type_prefix": "application",
      "name": "2601.09421v2.pdf",
      "size": 1429396
    },
    "source": {
      "execution_id": "4854ce70-9c3f-4529-a6ec-ab97cdb49bdb",
      "id": "file/base",
      "type": "Model",
      "version": "1.0.0"
    }
  },
  "file/pdf": {
    "private": false,
    "record": {
      "keywords": [],
      "page_count": 21,
      "producer": "pikepdf 8.15.1",
      "title": "Bias Dynamics in BabyLMs: Towards a Compute-Efficient Sandbox for Democratising Pre-Training Debiasing",
      "version": "1.7"
    },
    "source": {
      "execution_id": "24a0004f-05f1-4865-bef3-6eef1a1732de",
      "id": "dorsal/pdf",
      "type": "Model",
      "version": "1.1.0"
    }
  }
}