I have been trying to create a language model and have been following the documentation. However, I am running into problems. The error occurs when I run the following line
learn = language_model_learner(data=data_lm, arch=AWD_LSTM)
However, I get a key error on ‘tie_weights’ when it calls get_model_weights() and runs this line:
tie_weights,output_p,out_bias = map(config.pop, ['tie_weights', 'output_p', 'out_bias'])
I have also tried to specify config but I run into the same problem. I just use
config=awd_lstm_lm_config.copy(). Even if I manually set a key/value pairing in the config dict, the problem is still there.
I must be missing something really obvious. I did some search and it seems that nobody else was having this problem. I found one topic which was rather similar, but nobody had replies on that topic.
I am running this using the Kaggle kernels. They have version 1.0.51 right now. But I have also tried it on a local environment using the latest version.
In case I did something wrong with my DataBunch, I have also provided the code I used to create it.
data_lm = TextLMDataBunch.from_df(path="", train_df=train, valid_df=val, text_cols="comment_text", bs=64)