Spaces:
Sleeping
Sleeping
File size: 9,101 Bytes
1d14767 4573c7e 1d14767 4573c7e 1d14767 4a34671 1d14767 225d2b6 1d14767 225d2b6 1d14767 225d2b6 05d98c3 4a34671 225d2b6 05d98c3 1d14767 4573c7e 05d98c3 56e15e0 4573c7e 05d98c3 4573c7e 61db6eb 4573c7e 56e15e0 1d14767 61db6eb 56e15e0 4573c7e | 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 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 | # import gradio as gr
# from transformers import AutoTokenizer, AutoModelForSequenceClassification
# import torch
# tokenizer = AutoTokenizer.from_pretrained("sourabh5500/hate-speech-muril")
# model = AutoModelForSequenceClassification.from_pretrained("sourabh5500/hate-speech-muril")
# model.eval()
# # BLOCKLIST = [
# # # English profanity
# # "fuck", "fucking", "fucker", "shit", "bitch", "asshole",
# # "dick", "pussy", "whore", "slut", "cock", "cunt", "bastard",
# # "damn", "crap", "piss", "nigger", "nigga", "faggot",
# # # Hindi / Hinglish slang & profanity
# # "bsdk", "bhosdike", "bhosdi", "mc", "madarchod", "madar",
# # "bc", "behenchod", "behen", "chutiya", "chutiye", "chuti",
# # "gand", "gaand", "lund", "loda", "lauda", "laude",
# # "harami", "kaminey", "kamine", "bewakoof", "bhak",
# # "aabe", "abe", "suar", "kutty", "kutte", "kutta", "kuttiya",
# # "randi", "randi ke", "bache", "saale", "saala", "gandu",
# # "gandi", "bhen", "behen ke", "chod", "chodu", "chodne",
# # "tatte", "tatti", "bhadwa", "bhadve", "pataak", "chinal",
# # ]
# BLOCKLIST = [
# # Original Hinglish / Hindi Slang
# "bsdk", "bhosdike", "bhosdi", "mc", "madarchod", "madar",
# "bc", "behenchod", "behen", "chutiya", "chutiye", "chuti",
# "gand", "gaand", "lund", "loda", "lauda", "laude",
# "harami", "kaminey", "kamine", "bewakoof", "bhak",
# "aabe", "abe", "suar", "kutty", "kutte", "kutta", "kuttiya",
# "randi", "randi ke", "bache", "saale", "saala", "gandu",
# "gandi", "bhen", "behen ke", "chod", "chodu", "chodne",
# "tatte", "tatti", "bhadwa", "bhadve", "pataak", "chinal",
# # Additional Hinglish / Hindi
# "jhaat", "jhant", "jhantu", "bakland", "gashti", "gasti", "ghasti",
# "chut", "chutiyo", "chutia", "chutiyappa",
# "madarchodh", "madrchod", "maderchod", "madarchhod",
# "behenchodh", "bhenchod", "behenchhod",
# "bhosadike", "bhosdika", "bhosdiki",
# "gaandu", "gaandfat", "gaandmasti",
# "kameena", "kaminee", "kamin", "kamina", "kamini",
# "lavda", "lavde", "lawda", "lounde", "lunda",
# "haramzada", "haramkhor", "haraami",
# "najayaz", "nalayak",
# "teri maa", "teri behen", "maa ka bhosda",
# "ullu ke pathe", "kutte ki jat", "bhains ki aulad",
# "saale kutta", "saali kutti",
# # English Profanity (Cleaned of false positives)
# "fuck", "fucking", "fucker", "fucked", "fck", "fuk", "fuuck",
# "shit", "shitty", "shyt", "shithead",
# "bitch", "bitches", "bytch", "b!tch", "biatch",
# "asshole", "assholes", "ashole", "dumbass", "jackass",
# "bastard", "bastards",
# "dick", "dickhead", "d!ck", "dickweed",
# "pussy", "cunt", "kunt",
# "slut", "slutty", "sluts",
# "whore",
# "nigger", "nigga",
# "faggot", "fag",
# "retard", "retarded",
# "cock", "cockhead",
# "twat",
# "damn", "dammit", "goddamn", "goddammit",
# "piss", "pissed", "pissoff",
# "bollocks", "bugger",
# "bullshit", "crap", "crappy",
# "cum", "cumshot",
# "dyke", "homo",
# "jackoff", "jerkoff", "jizz",
# "milf", "mofo", "motherfucker",
# "prick", "scrotum", "semen",
# "tit", "tits", "titty",
# "turd", "wank", "wanker",
# ]
# def predict(text: str, check_type: str = "note") -> dict:
# if not text or not text.strip():
# return {"is_abusive": False, "confidence": 0}
# text_lower = text.lower()
# # Check blocklist first (instant detection)
# for word in BLOCKLIST:
# if word in text_lower:
# return {"is_abusive": True, "confidence": 1.0}
# # ✅ For names, only use blocklist (skip AI to avoid false positives)
# if check_type == "name":
# return {"is_abusive": False, "confidence": 0}
# # Model inference for notes
# inputs = tokenizer(
# text,
# return_tensors="pt",
# max_length=128,
# truncation=True,
# padding=True
# )
# with torch.no_grad():
# outputs = model(**inputs)
# probs = torch.softmax(outputs.logits, dim=1)[0]
# hof_score = probs[1].item()
# return {
# "is_abusive": hof_score > 0.5,
# "confidence": round(hof_score, 3)
# }
# iface = gr.Interface(
# fn=predict,
# inputs=[
# gr.Textbox(lines=3, placeholder="Enter text to check..."),
# gr.Radio(["note", "name"], value="note", label="Check type")
# ],
# outputs="json",
# title="SafeShield AI - Hate Speech Detection",
# description="Detects hate speech in Hinglish/English text"
# )
# iface.launch(server_name="0.0.0.0")
import os
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
tokenizer = AutoTokenizer.from_pretrained("sourabh5500/hate-speech-muril")
model = AutoModelForSequenceClassification.from_pretrained("sourabh5500/hate-speech-muril")
model.eval()
BLOCKLIST = [
# Original Hinglish / Hindi Slang
"bsdk", "bhosdike", "bhosdi", "mc", "madarchod", "madar",
"bc", "behenchod", "behen", "chutiya", "chutiye", "chuti",
"gand", "gaand", "lund", "loda", "lauda", "laude",
"harami", "kaminey", "kamine", "bewakoof", "bhak",
"aabe", "abe", "suar", "kutty", "kutte", "kutta", "kuttiya",
"randi", "randi ke", "bache", "saale", "saala", "gandu",
"gandi", "bhen", "behen ke", "chod", "chodu", "chodne",
"tatte", "tatti", "bhadwa", "bhadve", "pataak", "chinal",
# Additional Hinglish / Hindi
"jhaat", "jhant", "jhantu", "bakland", "gashti", "gasti", "ghasti",
"chut", "chutiyo", "chutia", "chutiyappa",
"madarchodh", "madrchod", "maderchod", "madarchhod",
"behenchodh", "bhenchod", "behenchhod",
"bhosadike", "bhosdika", "bhosdiki",
"gaandu", "gaandfat", "gaandmasti",
"kameena", "kaminee", "kamin", "kamina", "kamini",
"lavda", "lavde", "lawda", "lounde", "lunda",
"haramzada", "haramkhor", "haraami",
"najayaz", "nalayak",
"teri maa", "teri behen", "maa ka bhosda",
"ullu ke pathe", "kutte ki jat", "bhains ki aulad",
"saale kutta", "saali kutti",
# English Profanity (Cleaned of false positives)
"fuck", "fucking", "fucker", "fucked", "fck", "fuk", "fuuck",
"shit", "shitty", "shyt", "shithead",
"bitch", "bitches", "bytch", "b!tch", "biatch",
"asshole", "assholes", "ashole", "dumbass", "jackass",
"bastard", "bastards",
"dick", "dickhead", "d!ck", "dickweed",
"pussy", "cunt", "kunt",
"slut", "slutty", "sluts",
"whore",
"nigger", "nigga",
"faggot", "fag",
"retard", "retarded",
"cock", "cockhead",
"twat",
"damn", "dammit", "goddamn", "goddammit",
"piss", "pissed", "pissoff",
"bollocks", "bugger",
"bullshit", "crap", "crappy",
"cum", "cumshot",
"dyke", "homo",
"jackoff", "jerkoff", "jizz",
"milf", "mofo", "motherfucker",
"prick", "scrotum", "semen",
"tit", "tits", "titty",
"turd", "wank", "wanker",
]
def predict(text: str, check_type: str = "note", secret_key: str = "") -> dict:
# 🔒 Secret key validation
if secret_key != os.environ.get("HOPIN_API_SECRET"):
return {"is_abusive": False, "confidence": 0, "error": "Unauthorized"}
if not text or not text.strip():
return {"is_abusive": False, "confidence": 0}
text_lower = text.lower()
# Check blocklist first (instant detection)
for word in BLOCKLIST:
if word in text_lower:
return {"is_abusive": True, "confidence": 1.0}
# ✅ For names, only use blocklist (skip AI to avoid false positives)
if check_type == "name":
return {"is_abusive": False, "confidence": 0}
# Model inference for notes
inputs = tokenizer(
text,
return_tensors="pt",
max_length=128,
truncation=True,
padding=True
)
with torch.no_grad():
outputs = model(**inputs)
probs = torch.softmax(outputs.logits, dim=1)[0]
hof_score = probs[1].item()
return {
"is_abusive": hof_score > 0.5,
"confidence": round(hof_score, 3)
}
# iface = gr.Interface(
# fn=predict,
# inputs=[
# gr.Textbox(lines=3, placeholder="Enter text to check..."),
# gr.Radio(["note", "name"], value="note", label="Check type"),
# gr.Textbox(visible=False, label="Secret Key") # Hidden from UI but required for API
# ],
# outputs="json",
# title="SafeShield AI - Hate Speech Detection",
# description="Detects hate speech in Hinglish/English text"
# )
iface = gr.Interface(
fn=predict,
inputs=[
gr.Textbox(lines=3, placeholder="Enter text to check..."),
gr.Radio(["note", "name"], value="note", label="Check type"),
gr.Textbox(label="🔒 Secret Key (required)", placeholder="Enter your API secret key") # Made visible!
],
outputs="json",
title="SafeShield AI - Hate Speech Detection",
description="Detects hate speech in Hinglish/English text"
)
iface.launch(server_name="0.0.0.0") |