eval_model function results differ from test_step function.

[SailSabnis]

Hi all,

From 03. PyTorch Computer Vision where we create a CNN model for FashionMNIST data -

We first train and test the model using the train_step and test_step function. I understand we later run it for two epochs to begin with to get the test loss and test accuracy. After that we run the eval_model function to capture model loss and accuracy in a variable. In the video, Daniel gets the model loss and accuracy (i.e. eval_model results) same as the test_step function. But I get different results. Why is that so. What am I missing?

Screenshot 1 - My results

**Daniels results from the video

[SailSabnis]

FIgured it out. In my eval_model function -

model_loss /= len(dataloader)
model_acc /= len(dataloader)

should be outside the for loop.

[mrdbourke]

Glad you figured it out!

Those kind of errors get me too!