LayoutLM training after following the tutorial gives different results

[arpytoth]

I tried to follow the tutorial from here: https://github.com/NielsRogge/Transformers-Tutorials/blob/master/LayoutLM/Fine_tuning_LayoutLMForTokenClassification_on_FUNSD.ipynb but I've got very bad results. I'm quite certain I followed all the steps but the evaluation always returns something like this: {'loss': 1.1183262513234065, 'precision': 0.49382716049382713, 'recall': 0.5802901450725363, 'f1': 0.5335786568537259}

Did anything changed? What might be the problem?

[Admiral-Monferrat]

Me too, I got the results {'loss': 1.141750035377649, 'precision': 0.4488888888888889, 'recall': 0.6063031515757878, 'f1': 0.5158544371142795}.

[tianchiguaixia]