Trained on 5% of 'structs_token_size_4_pd_False_reduced_labelled' for 10 epochs.

Foundation model: 'distilbert-heaps-masked'

Training time: 19h

Eval Loss: 0.133 (pretty stable from the first epoch on)