Fakespot RoBERTa AI Detector — ONNX (Quantized INT8)

Upload Instructions

1. Create a new HuggingFace repository:
   bash

   1huggingface-cli login 2huggingface-cli repo create fakespot-roberta-ai-detector-onnx --type model
2. Clone and push all files from this directory:
   bash

   1git clone https://huggingface.co/YOUR_USERNAME/fakespot-roberta-ai-detector-onnx 2cp -r benchmark/hf-upload/* fakespot-roberta-ai-detector-onnx/ 3cd fakespot-roberta-ai-detector-onnx 4git lfs install 5git lfs track "*.onnx" 6git add . 7git commit -m "Add quantized ONNX model for Transformers.js" 8git push
3. Update offscreen.js with your repo ID:
   pipeline('text-classification', 'YOUR_USERNAME/fakespot-roberta-ai-detector-onnx', { dtype: 'q8' })

Model Details

• Source: fakespot-ai/roberta-base-ai-text-detection-v1
• Architecture: RoBERTa-base (125M params)
• Quantization: INT8 dynamic (avx512_vnni via Optimum)
• Size: ~120MB (q8) vs ~499MB (fp32)
• Labels: {0: "Human", 1: "AI"}
• License: Apache 2.0