index is out of bounds for axis 1 about finetune

[madisonmay]

Hi @nico1as -- will take a more in depth look at this when I have some spare cycles, but my first guess is that this is a pd.Series() problem. Try casting to a numpy array or calling df_train['text'].values, etc. and I think things will function fine for you. This is kind of an insidious error -- I think we should look into either supporting pandas objects as inputs or raising a more explicit error when they are received.

--Madison

[nico1as]

Thanks @madisonmay for you prompt answer. I figure out that by replacing chung_long_sequences=True by max_length=1000 and using df_train['label'].values.tolist() it is working!

[madisonmay]

Interesting -- does the problem persist if chunk_long_sequences=False.

[nico1as]

Yes, but as my inputs are long (ie: up to 5000 some times) I have to play with max_length...