When I using pipeline to inference the valid dataset, I got the different results with evaluation result during training! · Issue #37526 · huggingface/transformers · GitHub
You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am working with three classes image classification task.
I got the confusion matrix at 30th epoch during training like this confusion matrix:
[[230, 0, 1], [1, 28, 2], [1, 10, 3727]]
And I used the pipeline with the model's checkpoint at 30th epoch to infer the same test dataset, I got the confusion matrix like this:
[[230, 1, 0], [0, 31, 0], [8, 166, 3564]]
the model is timm/mobilenetv3_small_100.lamb_in1k
Anybody knows why?