Astariul

It's weird because when I download the data, I have `tgt` key inside the dictionary. Try to download again the data and check if `tgt` is inside this time ?

Sorry @hi-wangyan I thought you are someone else ^^ Your error is about pyrouge. To check if your pyrouge installation is working : `python -m pyrouge.test` --- If some tests...

Did you use `-bert_data_path ../bert_data` ? `-bert_data_path` option should also contain the prefix of your data. --- So instead of using `-bert_data_path ../bert_data`, try : `-bert_data_path ../bert_data/cnndm`

I changed the hard-coded value (`10`) to something much higher..

While I agree BertTokenizer is compulsory for subtokens, Stanford CoreNLP does not only split sentence. For example, it convert `(` into `-lrb-`. Another example is `$10,000` which will be tokenized...

Hey thanks for the fast answer ! Quick question, how did you display the boxes in blue like this ? In my browser, the bounding boxes are fine. Is it...

Thanks for your answer ! There must be a way, because my browser is fine :  I will search more...

I'm using version `20200726`. After upgrading to `20201018`, I still can reproduce the error... >MJEENE+YGO12 5.9500000000000455 MJEENE+YGO12 5.9500000000000455 MJEENE+YGO12 5.9500000000000455 MJEENE+YGO12 5.9500000000000455 MJEENE+YGO12 5.9500000000000455 [...] MJEENE+YGO12 5.0 MJEEOF+YGO11 8.0 MJEEOF+YGO11...

I could make it converge : - By using `BertAdam` from [pytorch-pretrained-BERT](https://github.com/huggingface/pytorch-pretrained-BERT) - By **not** sharing the weights of the final target embeddings (`tgt_emb_prj_weight_sharing = False`) --- Now I have...

> $ pyrouge_set_rouge_path -bash: syntax error near unexpected token `newline' You need to replace `` with your actual path to `ROUGE-1.5.5`. Check [my answer on Stack Overflow](https://stackoverflow.com/a/57686103/9494790) for more details.