AI Text Detector (XLM-R, ONNX)
browser-based generated text classification
1 min readupdated
This model is a multilingual text classifier based on the XLM-RoBERTa architecture, optimized for client-side execution using ONNX Runtime Web. It classifies text snippets as either human-written or AI-generated (Human and ChatGPT labels). Thanks to int8 quantization and weight sharding, the model runs entirely on the user's device without sending data to a server, making it ideal for web applications and browser extensions.
Key features
- Multilingual support based on the XLM-RoBERTa architecture for text classification.
- Int8 quantization and external weight shards for efficient browser execution.
- Entirely client-side computation via onnxruntime-web with no text data sent to servers.
- Document scoring via score averaging across segments of around 1,200 characters.
What you can do
- Integrate the detector into a web service or browser extension to verify text authorship.
- Use in applications for client-side user-generated content analysis.
- Check large documents by splitting them into segments followed by score averaging.
Example
The model loads via onnxruntime-web and evaluates text segments on the client device, returning probabilities for the Human and ChatGPT labels.
Read more
DetailsArchitecture: XLM-RoBERTa base
Format: ONNX with int8 quantization and weight shards
Labels: Human and ChatGPT
Limitations: serves as a statistical helper rather than absolute proof of authorship










Comments