fela-moderator / config.json
itstheraj's picture
initial commit
4751195
Raw
History Blame Contribute Delete
6.18 kB
{
"model_type": "fela-moderation",
"library_name": "transformers",
"architectures": [
"FelaModeratorForSequenceClassification"
],
"auto_map": {
"AutoConfig": "configuration_moderation.FelaModeratorConfig",
"AutoModelForSequenceClassification": "modeling_moderation.FelaModeratorForSequenceClassification"
},
"arch": "FELAModerationV2",
"note": "byte level FNO+GLA moderation and PII model, 18.9M params, vocab 259; seven heads on one shared encoder: 19 way taxonomy, 6 way Jigsaw toxicity, per token PII BIO (113 tags, 56 entity types), plus spam, jailbreak, nsfw, and identity.",
"vocab_size": 259,
"max_len": 512,
"d_model": 448,
"n_layers": 8,
"n_heads": 7,
"fno_modes": 128,
"gla_chunk": 32,
"ffn_hidden": 1280,
"layer_pattern": "SSSL",
"dropout": 0.1,
"pad_id": 256,
"cls_id": 257,
"sep_id": 258,
"n_tox_labels": 6,
"n_pii_tags": 113,
"n_tax": 19,
"n_spam": 3,
"n_jailbreak": 4,
"n_nsfw": 2,
"n_identity": 7,
"taxonomy": [
"harassment",
"harassment_threatening",
"hate",
"hate_threatening",
"self_harm",
"self_harm_intent",
"self_harm_instructions",
"sexual",
"sexual_minors",
"violence",
"violence_graphic",
"identity_attack",
"insult",
"profanity_obscene",
"threat",
"sexual_explicit",
"severe_toxicity",
"dehumanize",
"incitement_violence"
],
"jigsaw_labels": [
"toxic",
"severe_toxic",
"obscene",
"threat",
"insult",
"identity_hate"
],
"spam_labels": [
"spam",
"scam",
"phishing"
],
"jailbreak_labels": [
"jailbreak",
"instruction_override",
"role_hijack",
"data_exfiltration"
],
"nsfw_labels": [
"sexual_suggestive",
"sexual_explicit"
],
"identity_labels": [
"race",
"religion",
"gender",
"sexuality",
"disability",
"origin",
"age"
],
"pii_entity_types": [
"PERSON",
"FIRSTNAME",
"LASTNAME",
"USERNAME",
"EMAIL",
"PHONE",
"FAX",
"STREET_ADDRESS",
"CITY",
"STATE",
"ZIPCODE",
"COUNTRY",
"BUILDINGNUMBER",
"SECONDARYADDRESS",
"NEARBYGPSCOORDINATE",
"DOB",
"AGE",
"SSN",
"PASSPORT",
"DRIVERLICENSE",
"IDCARD",
"CREDITCARD",
"CREDITCARDCVV",
"CREDITCARDEXP",
"IBAN",
"BIC_SWIFT",
"ACCOUNTNUMBER",
"ROUTINGNUMBER",
"BITCOINADDRESS",
"ETHEREUMADDRESS",
"IPV4",
"IPV6",
"MACADDRESS",
"URL",
"USERAGENT",
"COMPANYNAME",
"JOBTITLE",
"MRN",
"MEDICALLICENSE",
"HEALTHINSURANCEID",
"VEHICLEVIN",
"VEHICLEPLATE",
"TIME",
"DATE",
"PIN",
"PASSWORD",
"APIKEY",
"ORDINALDIRECTION",
"CURRENCY",
"AMOUNT",
"GENDER",
"ETHNICITY",
"RELIGION",
"SEXUALITY",
"POLITICAL",
"MARITALSTATUS"
],
"pii_tags": [
"O",
"B-PERSON",
"I-PERSON",
"B-FIRSTNAME",
"I-FIRSTNAME",
"B-LASTNAME",
"I-LASTNAME",
"B-USERNAME",
"I-USERNAME",
"B-EMAIL",
"I-EMAIL",
"B-PHONE",
"I-PHONE",
"B-FAX",
"I-FAX",
"B-STREET_ADDRESS",
"I-STREET_ADDRESS",
"B-CITY",
"I-CITY",
"B-STATE",
"I-STATE",
"B-ZIPCODE",
"I-ZIPCODE",
"B-COUNTRY",
"I-COUNTRY",
"B-BUILDINGNUMBER",
"I-BUILDINGNUMBER",
"B-SECONDARYADDRESS",
"I-SECONDARYADDRESS",
"B-NEARBYGPSCOORDINATE",
"I-NEARBYGPSCOORDINATE",
"B-DOB",
"I-DOB",
"B-AGE",
"I-AGE",
"B-SSN",
"I-SSN",
"B-PASSPORT",
"I-PASSPORT",
"B-DRIVERLICENSE",
"I-DRIVERLICENSE",
"B-IDCARD",
"I-IDCARD",
"B-CREDITCARD",
"I-CREDITCARD",
"B-CREDITCARDCVV",
"I-CREDITCARDCVV",
"B-CREDITCARDEXP",
"I-CREDITCARDEXP",
"B-IBAN",
"I-IBAN",
"B-BIC_SWIFT",
"I-BIC_SWIFT",
"B-ACCOUNTNUMBER",
"I-ACCOUNTNUMBER",
"B-ROUTINGNUMBER",
"I-ROUTINGNUMBER",
"B-BITCOINADDRESS",
"I-BITCOINADDRESS",
"B-ETHEREUMADDRESS",
"I-ETHEREUMADDRESS",
"B-IPV4",
"I-IPV4",
"B-IPV6",
"I-IPV6",
"B-MACADDRESS",
"I-MACADDRESS",
"B-URL",
"I-URL",
"B-USERAGENT",
"I-USERAGENT",
"B-COMPANYNAME",
"I-COMPANYNAME",
"B-JOBTITLE",
"I-JOBTITLE",
"B-MRN",
"I-MRN",
"B-MEDICALLICENSE",
"I-MEDICALLICENSE",
"B-HEALTHINSURANCEID",
"I-HEALTHINSURANCEID",
"B-VEHICLEVIN",
"I-VEHICLEVIN",
"B-VEHICLEPLATE",
"I-VEHICLEPLATE",
"B-TIME",
"I-TIME",
"B-DATE",
"I-DATE",
"B-PIN",
"I-PIN",
"B-PASSWORD",
"I-PASSWORD",
"B-APIKEY",
"I-APIKEY",
"B-ORDINALDIRECTION",
"I-ORDINALDIRECTION",
"B-CURRENCY",
"I-CURRENCY",
"B-AMOUNT",
"I-AMOUNT",
"B-GENDER",
"I-GENDER",
"B-ETHNICITY",
"I-ETHNICITY",
"B-RELIGION",
"I-RELIGION",
"B-SEXUALITY",
"I-SEXUALITY",
"B-POLITICAL",
"I-POLITICAL",
"B-MARITALSTATUS",
"I-MARITALSTATUS"
],
"toxicity_thresholds": {
"toxic": 0.9029,
"severe_toxic": 0.8966,
"obscene": 0.9394,
"threat": 0.7783,
"insult": 0.8721,
"identity_hate": 0.9185
},
"input_shape": [
1,
512
],
"input_desc": "one UTF-8 text string, byte tokenized to [CLS] b0 b1 ... truncated to max_len, int64 ids in [0,258]",
"head_thresholds": {
"spam": {
"spam": 0.999,
"scam": 0.957,
"phishing": 0.999
},
"jailbreak": {
"jailbreak": 0.957,
"instruction_override": 0.936,
"role_hijack": 0.997,
"data_exfiltration": 0.994
},
"nsfw": {
"sexual_suggestive": 0.664,
"sexual_explicit": 0.999
},
"identity": {
"race": 0.97,
"religion": 0.95,
"gender": 0.924,
"sexuality": 0.936
}
},
"head_gate": {
"spam": "ship",
"jailbreak": "ship",
"nsfw": "ship",
"identity": "hold",
"taxonomy_sexual": "defer"
},
"head_gate_note": "ship = held-out AUROC >= 0.8. Thresholds in head_thresholds are calibrated at 90% precision on held-out eval; tune per your traffic. identity (0.71) and the taxonomy 'sexual' category (0.67) are below the ship line and off/masked."
}