better readme

README.md

@@ -16,6 +16,28 @@ and the optimized model is converted to the onnx format.
 
 ## How to use
 
+```python
+from optimum.onnxruntime.modeling_ort import ORTModelForImageClassification
+from transformers import AutoFeatureExtractor
+
+from PIL import Image
+import requests
+
+preprocessor = AutoFeatureExtractor.from_pretrained("fxmarty/levit-256-onnx")
+ort_model = ORTModelForImageClassification.from_pretrained("fxmarty/levit-256-onnx")
+
+url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
+image = Image.open(requests.get(url, stream=True).raw)
+
+inputs = preprocessor(images=image, return_tensors="pt")
+outputs = model(**inputs)
+
+predicted_class_idx = outputs.logits.argmax(-1).item()
+print("Predicted class:", model.config.id2label[predicted_class_idx])
+```
+
+To be safe, check as well that the onnx model returns the same logits as the PyTorch model:
+
 ```python
 from optimum.onnxruntime.modeling_ort import ORTModelForImageClassification
 from transformers import AutoModelForImageClassification
@@ -65,6 +87,8 @@ Optimum-onnxruntime runtime:
 
 ```
 
+Run on your own machine with:
+
 ```python
 from optimum.runs_base import TimeBenchmark