{
 "id": "catt-bench/vtok101-distr-attribution-baselines",
 "owner": "catt-bench",
 "name": "vtok101-distr-attribution-baselines",
 "downloads": 9869,
 "likes": 4,
 "last_modified": "2026-10-29T18:11:38.000Z",
 "stated_licence": "mit",
 "stated_from": "tag",
 "licence_name": null,
 "licence_link": null,
 "file_licence": null,
 "licence_files": [],
 "sources": [],
 "gated": false,
 "tasks": [],
 "citation": "The attribution scores of \"Elucidating the Design Space of LM Data Attribution\"\non **Function Bindings with distractors**, for the LoRA adapters in\n[catt-bench/Qwen3.5-4B-d0-vtok101-distr-lora-seeds](https://huggingface.co/catt-bench/Qwen3.5-4B-d0-vtok101-distr-lora-seeds). The CATT code release reads them\nwith `catt fetch results --task fb-distr` and rebuilds the paper's figures and tables\nfrom them with `catt report results`.\n\nEvery real document has a decoy beside it: a shadow function `<Axx>` taught the\nsame constant in the same template as `<Bxx>`. The pool is twice as large and\nqueries only ask about real functions, so AUC keeps its meaning while Recall@N's\nchance line halves. `metrics.json` adds `decoy_auc` (a real document against the\nquery's own decoys) and `decoy_top1`.",
 "signal": "permissive",
 "reason": "The declared licence (mit) is classified as permissive.",
 "revision": "8ae29298ef11d1bfbc3fed943a1cf2e50382f9b1",
 "stated_evidence": "https://huggingface.co/datasets/catt-bench/vtok101-distr-attribution-baselines/blob/8ae29298ef11d1bfbc3fed943a1cf2e50382f9b1/README.md",
 "file_evidence": "https://huggingface.co/datasets/catt-bench/vtok101-distr-attribution-baselines/tree/8ae29298ef11d1bfbc3fed943a1cf2e50382f9b1",
 "card_nc": false
}