nicoamoretti/roberta-openai-detector-onnx

RoBERTa OpenAI Detector (ONNX for Transformers.js)

AI-generated text detector converted to ONNX format for browser-based inference with Transformers.js.

Model Details

• Base Model: openai-community/roberta-base-openai-detector
• Format: ONNX (quantized to ~120MB)
• Task: Binary text classification (Real vs Fake)

Usage with Transformers.js

import { pipeline } from '@xenova/transformers';

const classifier = await pipeline('text-classification', 'nicoamoretti/roberta-openai-detector-onnx');

const result = await classifier('Your text here');
console.log(result);
// [{ label: 'Real', score: 0.95 }] - Human written
// [{ label: 'Fake', score: 0.85 }] - AI generated

Labels

Label	Meaning
Real	Human-written text
Fake	AI-generated text (GPT-2 era models)

Notes

This model was originally trained to detect GPT-2 generated text. It may have reduced accuracy on text from newer models like GPT-4 or Claude, but still provides useful signal for AI detection.