File size: 1,417 Bytes
264b834 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 | {
"thresholds": {
"lo": 0.05784,
"hi": 0.736825,
"rule": "lo = 1st percentile of harmful p, hi = 95th percentile of safe p, on confident half B of the head's fit pool"
},
"routing": "p >= hi -> block; p < lo -> pass; otherwise delegate",
"piece_calibration": {
"a": 0.6507,
"b": 0.0792,
"form": "sigmoid(a * logit(s) + b)"
},
"piece_hot_threshold": 0.26094,
"substance_hot_threshold": 0.5,
"head": {
"model": "sklearn HistGradientBoostingClassifier(max_iter=200, learning_rate=0.1, max_leaf_nodes=15, random_state=42)",
"fit_rows": "confident rows (y <= 0.2 or y >= 0.8) of half A",
"features": [
"core",
"req_max",
"req_mean",
"mat_max",
"n_req",
"n_mat",
"n_frame",
"has_frame",
"mixed",
"has_decoded",
"has_joined",
"core_is_whole",
"pair_max",
"deframed_max",
"clause_max",
"all_max",
"top2",
"hot",
"frame_max",
"s_core",
"s_max",
"s_top2",
"s_mean",
"s_hot",
"s_frame_max"
]
},
"head_fit_features": "gate1_head_fit.csv.gz",
"classifiers": {
"piece_classifier": {
"encoder": "microsoft/deberta-v3-large",
"max_tokens": 256,
"truncation": "head + tail",
"pooling": "[CLS]"
},
"substance_classifier": {
"encoder": "microsoft/deberta-v3-base",
"max_tokens": 256,
"truncation": "head + tail",
"pooling": "[CLS]"
}
}
} |