The accuracy doesn't match the output

why always when I train the model I can see the accuracy of the training is much higher than the one plotted in the model outcome what is the reason for that? It does not make sense I wait for the training for too long and eventually I found the accuracy is not high and did not match the one in the training outcome .

Hi @Joker147,

The training loop shows the float32 accuracy. In your screenshot you have selected the quantized model which is usually a bit lower than float32.


I should disable it to show the right accuracy?

Just select the float32 in the UI: