1. I want to know whether "Preprocessed Data" is generated through "Generate Candidate Summaries". 2. If "Own Data" is processed through preprocess.py, then the test.source, test.source.tokenized and other files in src_dir Where did it come from, was it created manually?

[cq-cdy]

No description provided.

[thaokimctu]

I think you have to created source and target files manually from raw data by using the code here facebookresearch/fairseq#1391 which was modified from the code provided by the author https://github.com/abisee/cnn-dailymail. After that, gen_candidate.py was used to create out files and then you create tokenized files by following the instruction of Evaluate section from README. Finally, use preprocess.py to create new dataset.