Qarvexium/NoInsult

Quick Start

Run inference:

from NoInsult import predict

result = predict("I will fuck you up, you piece of shit.")

print(result["flags"])   # Boolean predictions
print(result["scores"])  # Confidence scores

Example output:

Loading NoInsult by Qarvexium
Loading weights: 100%|████████████████████| 201/201 [00:00<00:00, 330.76it/s]

Labels:
['toxicity', 'severe_toxicity', 'obscene', 'threat', 'insult', 'identity_attack', 'sexual_explicit']

Thresholds:
[0.909860372543335, 0.5, 0.983537495136261, 0.9544731974601746,
 0.9325506091117859, 0.9784075617790222, 0.9820615649223328]

Flags:
{
    "toxicity": true,
    "severe_toxicity": true,
    "obscene": false,
    "threat": false,
    "insult": false,
    "identity_attack": false,
    "sexual_explicit": false
}

Scores:
{
    "toxicity": 0.9529,
    "severe_toxicity": 0.8466,
    "obscene": 0.9799,
    "threat": 0.7640,
    "insult": 0.9251,
    "identity_attack": 0.4795,
    "sexual_explicit": 0.9042
}

Output Format

predict() returns a dictionary with two fields:

• flags – Boolean predictions after applying class-specific thresholds.
• scores – Raw sigmoid probabilities for each category.

Supported labels:

• toxicity
• severe_toxicity
• obscene
• threat
• insult
• identity_attack
• sexual_explicit