Labels order in pretrained models

I’ve trained a model for sequence classification and pushed it to the hub.
Now, when I load it, can I somehow restore the original labels order?

For instance, I trained it on 10 classes.
Then I load it:
model = AutoModelForSequenceClassification.from_pretrained(hub_name)
The problem is that new predicted labels don’t correspond to labels which were used in training.
For instance, when I trained, category ‘coffee’ was assigned label 3.
Now, when I make predictions on my test data, all coffee items are predicted with label 7.
How can I know which new labels correspond to old labels?