{ "model": "ft-v6", "source": "experiments/finetune/out/small-ft-v6/best", "labels": [ "medical", "student_records", "id_numbers", "financial", "credentials", "legal_hr", "none" ], "label_descriptions": { "medical": "Health, medical, or mental-health details about a specific person", "student_records": "Grades, conduct, advising, accommodations, or other FERPA records about a named student", "id_numbers": "Social Security, passport, driver's license, visa, or other ID numbers", "financial": "Bank accounts, card numbers, salaries, income, or payroll for a specific person", "credentials": "Real passwords, API keys, private keys, or login secrets", "legal_hr": "Legal, HR, disciplinary, harassment, criminal, or immigration matters about a person", "none": "None of these: general, public, aggregate, fictional, redacted, or placeholder content" }, "prefix_ids": [ 287, 128003, 10562, 128010, 1159, 294, 1516, 261, 1159, 261, 289, 2338, 271, 14901, 1045, 314, 266, 996, 604, 128010, 1234, 616, 74658, 294, 40198, 261, 3360, 261, 15500, 261, 11120, 261, 289, 340, 120623, 2227, 314, 266, 1784, 1234, 128010, 9764, 616, 63452, 294, 2491, 2590, 261, 9878, 261, 1986, 280, 268, 2989, 261, 6828, 261, 289, 340, 2819, 1687, 128010, 999, 294, 2122, 2681, 261, 1081, 1687, 261, 11661, 261, 1766, 261, 289, 11473, 270, 266, 996, 604, 128010, 10104, 294, 2858, 13405, 261, 5804, 5258, 261, 1094, 5258, 261, 289, 6607, 7595, 128010, 1334, 616, 9512, 294, 7121, 261, 6530, 261, 17343, 261, 8930, 261, 3030, 261, 289, 4854, 2973, 314, 266, 604, 128010, 3083, 294, 7422, 265, 378, 294, 1042, 261, 613, 261, 9401, 261, 12967, 261, 56991, 261, 289, 11736, 822, 287, 128007, 1159, 128007, 1234, 616, 74658, 128007, 9764, 616, 63452, 128007, 999, 128007, 10104, 128007, 1334, 616, 9512, 128007, 3083, 1263, 1263, 128002 ], "marker_positions": [ 137, 139, 143, 147, 149, 151, 155 ], "none_index": 6, "scoring": "multi-label: P(label) = sigmoid(logit). A block's score is 1 - P(none) = sigmoid(-logit_none).", "threshold": 0.5, "block_words": 100, "text_encoding": "split the block with the word regex (URL | email | @handle | word(-word)* | any non-space), lower-case each piece, tokenize each piece alone without special tokens, append after prefix_ids", "recommended_file": "onnx/model_fp16.onnx" }