Tenne: AI-Generated Content Detection Model

📖 Product Overview

Tenne is a binary classification model specifically designed to distinguish between human-written and AI-generated text. The model effectively identifies content produced by large language models such as ChatGPT, providing technical support for content moderation, academic integrity detection, and related applications.

✅ Core Capabilities

• High Accuracy: 97%+ accuracy on validation set
• Zero False Negatives: 100% recall for AI-generated text
• Chinese Optimized: Built on BERT-base-chinese architecture
• Fast Inference: Millisecond-level response time

🎯 Performance Metrics

Overall Metrics

Metric	Score
Accuracy	97.17%
F1 Score	0.965
AUC	1.000

Classification Performance

Category	Precision	Recall
Human Text	100%	95.35%
AI Text	93.26%	100%

🚀 Quick Start

Requirements

pip install torch transformers

Load Model from Hugging Face

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from huggingface_hub import hf_hub_download
import torch

tokenizer = AutoTokenizer.from_pretrained("MAHE-model/Tenne")
model = AutoModelForSequenceClassification.from_pretrained("MAHE-model/Tenne")
model.eval()

Single Text Detection

def detect(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)

    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.softmax(outputs.logits, dim=-1)
    
    label = "AI Generated" if probs[0][1] > 0.5 else "Human Written"
    confidence = probs[0][1].item() if label == "AI Generated" else probs[0][0].item()
    
    return {
        "label": label, 
        "confidence": confidence, 
        "ai_probability": probs[0][1].item()
    }

# Example usage
result = detect("Text content to be analyzed")
print(f"Result: {result['label']}, Confidence: {result['confidence']:.4f}")

Batch Detection

def batch_detect(texts, batch_size=16):
    results = []
    for i in range(0, len(texts), batch_size):
        batch = texts[i:i+batch_size]
        inputs = tokenizer(
            batch, 
            return_tensors="pt", 
            truncation=True, 
            max_length=512, 
            padding=True
        )

        with torch.no_grad():
            outputs = model(**inputs)
            probs = torch.softmax(outputs.logits, dim=-1)
        
        for j, prob in enumerate(probs):
            results.append({
                "text": batch[j],
                "label": "AI Generated" if prob[1] > 0.5 else "Human Written",
                "ai_score": prob[1].item()
            })
    return results

💼 Use Cases

Scenario	Description
Academic Integrity	Detect AI-generated student essays and papers
Content Moderation	Identify AI-generated content on platforms
Recruitment Screening	Evaluate authenticity of resumes and cover letters
News Verification	Determine if news articles are AI-generated

📊 Model Specifications

Parameter	Specification
Model Size	~400 MB
Max Input Length	512 tokens
Inference Time	< 100ms (GPU)
Language	Chinese (Simplified)

📞 Business Inquiries

For model files, API keys, or enterprise deployment solutions, please contact: Website: https://maherjon.github.io/MAHE/

🔒 Copyright Notice

This model and related technologies are proprietary. Commercial use without authorization is prohibited.