{
 "manifest_version": 1,
 "role": "statement_tagger",
 "name": "statement-tagger",
 "version": "statements-v1/C-minilm-mix",
 "task": "multilabel_classification",
 "runtime": {
  "engine": "onnxruntime",
  "requires": [
   "onnxruntime",
   "tokenizers",
   "numpy"
  ],
  "opset": 17,
  "executes_code_on_load": false
 },
 "files": [
  {
   "path": "onnx/bibr_onnx.json",
   "bytes": 2550,
   "sha256": "37eb7225f10156c4e68679f22dff63ca62c3b9d5a949483be15eec5f75596ee2",
   "kind": "legacy_manifest"
  },
  {
   "path": "onnx/model.onnx",
   "bytes": 90411056,
   "sha256": "24e0fdb7b084170197e85226edf63029d0fecabcf7f4a67f11fd4d383110c4bf",
   "kind": "model"
  },
  {
   "path": "onnx/tokenizer.json",
   "bytes": 527292,
   "sha256": "325e6539e09081160e94f36b13f0869734587b16be2add1a6848ba7eb4c10c38",
   "kind": "tokenizer"
  }
 ],
 "encoder": {
  "id": "sentence-transformers/all-MiniLM-L6-v2",
  "revision": null
 },
 "labels": {
  "types": [
   "coi",
   "funding",
   "ethics",
   "data_availability",
   "author_contributions",
   "acknowledgment",
   "supplementary"
  ]
 },
 "input": {
  "layout": "text_pair",
  "add_special_tokens": true,
  "pair": {
   "first": "the previous block's text, first 200 characters (empty for the first block)",
   "second": "the block's text"
  },
  "max_length": 192,
  "truncation": "longest_first",
  "normalization": "NFKC; U+00AD, U+200B, U+FEFF removed; whitespace runs to one space",
  "special_tokens": {
   "pad_token": "[PAD]",
   "pad_token_id": 0
  },
  "tokenizer_file": "onnx/tokenizer.json",
  "window": null,
  "stride": null,
  "template": null,
  "features": null,
  "model_inputs": [
   {
    "name": "input_ids",
    "dtype": "int64",
    "shape": [
     "batch",
     "sequence"
    ]
   },
   {
    "name": "attention_mask",
    "dtype": "int64",
    "shape": [
     "batch",
     "sequence"
    ]
   }
  ],
  "model_outputs": [
   {
    "name": "probs",
    "dtype": "float32",
    "shape": [
     "batch",
     7
    ]
   }
  ]
 },
 "decode": {
  "method": "sigmoid_thresholds",
  "crf": null,
  "calibration": null,
  "when_nothing_found": "no_type",
  "thresholds": {
   "coi": 0.5,
   "funding": 0.4,
   "ethics": 0.825,
   "data_availability": 0.95,
   "author_contributions": 0.55,
   "acknowledgment": 0.925,
   "supplementary": 0.775
  }
 },
 "quantization": {
  "weights": "fp32",
  "note": null
 },
 "provenance": {
  "ledger_key": "statement-tagger",
  "run_record_sha256": "8befc900d89ce0d2db94675e16815847ba008f8985c3217c5349e13e45646f50",
  "platform_commit": "8521e69f6bb540e13e87fa27f785fa518477ed2e",
  "registry_build": "2026-11-10T21:38:59Z#d8d427de3ea7",
  "split_id": "corpus-split-v1@c6fc246571f4"
 },
 "licence": "apache-2.0",
 "compat": {
  "bibr_min_version": null,
  "export_schema": null
 },
 "self_test": {
  "tolerance": {
   "probability_abs": 0.01,
   "logit_abs": 0.05,
   "reason": "onnxruntime builds, CPUs and BLAS kernels give slightly different float results; the largest difference measured between onnxruntime 1.27.0 and 1.31.0 on the eight bundles normalised so far was 8e-05 (stored values are rounded to 5 places), so these limits are about two orders of magnitude looser than the drift seen, and far tighter than any change of a decision; tags and label lists are compared exactly"
  },
  "cases": [
   {
    "name": "three blocks",
    "input": {
     "pairs": [
      [
       "",
       "Funding"
      ],
      [
       "Funding",
       "This work was supported by the Invented Science Fund, grant 000-00."
      ],
      [
       "This work was supported by the Invented Science Fund, grant 000-00.",
       "The authors declare no competing interests."
      ]
     ]
    },
    "expect": {
     "probs": [
      [
       0.00117,
       0.86977,
       0.00012,
       3e-05,
       0.00016,
       0.00222,
       0.00014
      ],
      [
       0.00528,
       0.99652,
       0.0001,
       1e-05,
       4e-05,
       0.00341,
       5e-05
      ],
      [
       0.88738,
       0.04079,
       0.00058,
       0.00022,
       0.00014,
       0.02025,
       0.00062
      ]
     ],
     "types": [
      [
       "funding"
      ],
      [
       "funding"
      ],
      [
       "coi"
      ]
     ]
    }
   }
  ]
 },
 "spec": {
  "join": "each run of consecutive blocks of one type is one statement; its score is the run's highest probability",
  "context_chars": 200,
  "input_rule": "pair:previous_block"
 }
}