KeyError: ‘1.decoder.bias’ in convert_weights_with_prefix

[abedkhooli]

While running the train_clas.py script, I get
convert_weights_with_prefix. I also tried the refactored version of the same fn.

    dec_bias, enc_wgts = wgts[prefix+'1.decoder.bias'], wgts[prefix+'0.encoder.weight']
KeyError: '1.decoder.bias'

If run with fine_tune off, there is another error:
Error(s) in loading state_dict for SequentialRNN:
Missing key(s) in state_dict: "0.fwd_lm.encoder.weight", "0.fwd_lm.encoder_dp.emb.weight", ...
Unexpected key(s) in state_dict: "fwd_lm.0.encoder.weight", "fwd_lm.0.encoder_dp.emb.weight", ...

[PiotrCzapla]

I've noticed that it is due to the changes to the fastai, I will push a fix soon enough, stay tuned

[abedkhooli]

Great. I think the fix did it. I managed to test pretrain_lm on Arabic wiki tokens. The model (only 1 cycle) was read successfully by fastai's language_model_learner (implemented IMDB sample lesson - using machine translation to convert texts to Arabic). Will work on train_clas next.
As for drop_mult, may be it can be passed as a hyberparameter (the 0.1 default fits large tokens).