Open jind11 opened 3 years ago
Hi, could you also help provide the code for getting and processing the training data? BTW, in the train data, there are many weird tokens like "<#bleu1#>", which are also not in the UNILM vocab and can cause errors.
<#bleu1#> is just a placeholder for latent variable Z_a in implementation, just append such tokens in the sequence like the training samples is ok.
Hi, could you also help provide the code for getting and processing the training data? BTW, in the train data, there are many weird tokens like "<#bleu1#>", which are also not in the UNILM vocab and can cause errors.