Code / benchmarks/promptsets/manifest_v2.json

benchmarks/promptsets/manifest_v2.json 42 lines
{
  "version": "v2",
  "seed": 54321,
  "n_per_class": 120,
  "author": "Simon-Pierre Boucher",
  "contact": "contact@spboucher.ai",
  "website": "https://modelmap.io",
  "design": "structure-borne properties, token-balanced classes (response to expA run #1 validity-gate failure)",
  "files": {
    "agreement_A.jsonl": {
      "sha256": "622c5e0966d2b8e8abc9fb7b964a8e5de8df5f8d41c37a0ed4ed8c8894f1f0ea",
      "n": 240,
      "class_token_overlap": 1.0
    },
    "agreement_B.jsonl": {
      "sha256": "243b85e2c207be6cff860a4660400c9c90859d5014636177b4475959b872542f",
      "n": 240,
      "class_token_overlap": 1.0
    },
    "arith_valid_A.jsonl": {
      "sha256": "a4bb4c645b3f1797e562f62765d04e7603d0e0ac0286e013cfa90f774de27159",
      "n": 240,
      "class_token_overlap": 0.6735
    },
    "arith_valid_B.jsonl": {
      "sha256": "adc68ef876a823ec43c4cf3c64a408f46b2613e2f3e7386fa1d20e3f80e9cf93",
      "n": 240,
      "class_token_overlap": 0.9674
    },
    "word_order_A.jsonl": {
      "sha256": "9052b924930aa5cde7ce434439c418cbd3e5a0731d421cd370bd46d60abcdb66",
      "n": 240,
      "class_token_overlap": 0.5294
    },
    "word_order_B.jsonl": {
      "sha256": "3012db3b5053ba5d113d0fe6529c4b88b361b9e4572044b9e5e64a3480929a1f",
      "n": 240,
      "class_token_overlap": 0.5072
    }
  }
}