Use float32 ONNX (exact PyTorch match, faster CPU inference)

app.py

@@ -10,7 +10,7 @@ from adaptive_classifier import AdaptiveClassifier
 # Load model once at startup
 print("Loading model...")
 classifier = AdaptiveClassifier.from_pretrained(
-    "adaptive-classifier/ai-detector", use_onnx=False
+    "adaptive-classifier/ai-detector", use_onnx=True, prefer_quantized=False
 )
 print("Model loaded!")