Simple Image Classification Model gives different confidence level (Between Coreml UI and iOS App)

[VipulArvind]

Here is what I have done

1. Create a model using Create ML Tool
2. Test the model and it passes 100% times
3. Verify by adding more images in UI tool and it passes
4. Generate the .nlmodel file from CreateML UI
5. use the model in iOS app and check the result of VNClassificationObservation.
6. The confidence level is a lot different.
7. And running this request on same image in program tends to finally put the correct confidence level (like in 3rd or 4th attempt)
8. There is NOTHING fancy and I am using the simple app (https://iosexample.com/classifying-images-with-vision-and-core-ml/) to test my model