Remove ONNX files to reduce download size for Spaces

Replace with float32 ONNX (matches PyTorch exactly)

Remove duplicate fp16 file

Add fp16 ONNX (676MB, same accuracy as PyTorch, faster on CPU)

Remove int8 quantized ONNX

Add fp16 ONNX model (676MB, same accuracy as float32)

Remove unquantized ONNX (1.4GB), keep quantized only

Fix model: save 200 representative examples per class for better predictions

Remove third-party citations, keep only adaptive-classifier

Update model card with results, dataset link, and proper metadata

AI text detector trained on EditLens ICLR 2026 dataset

initial commit