hub-dollars / sample-summary.json
thomwolf's picture
thomwolf HF Staff
Publish Hub Dollars V2: sourced estimates and data visualizations
dc570a9 verified
Raw History Blame Contribute Delete
813 Bytes
{
"sample_size": 160,
"counts": {
"derived_or_adapted_with_provenance": 17,
"card_unavailable": 56,
"unresolved": 75,
"specific_scratch_training_report": 1,
"conversion_with_provenance": 6,
"template_scratch_claim_unverified": 5
},
"design": "Fixed-seed MD5-order sample among snapshot repos with weight-looking file and without parsed baseModels.parents. Not all Hub; no replacement of inaccessible cards.",
"valid_inference": "Frequency of discoverable provenance in this stratum only. Unresolved cases cannot be treated as zero-cost or scratch.",
"new_training_cost_observations": 1,
"conclusion": "Mean incremental cost and number of independent training events not identifiable from this sample. Retire V1 global1%\u00d7$1000 point estimate; keep as sensitivity only."
}