When I load a model:
model = transformers.AutoModelForSequenceClassification.from_pretrained("german-nlp-group/electra-base-german-uncased")
And then save it 2 times:
model.save_pretrained("saved_model-01")
model.save_pretrained("saved_model-02")
Both pytorch_model.bin
files are equal (when I do sha256sum). When I load the model again both sha256sums are equal again but are changend from the first iteration.
Why is that the case? Is the saving should be deterministic or is there a date in the searialized data?