# Following the previous study record, I do not have 20 days to learn, light to play, dying ah.
1. MTB is still looking at code, some of which pre-trained, many do not understand, where:
The use of pre-training data set is cnn.txt, I do not know whether the QA data set used https://cs.nyu.edu/~kcho/DMQA/ .
It's the whole format is like this:
Is divided into two parts: one is short, and the other is 4 @highlight, stressed part of the character does not appear in a standard short article is a summary of the.
From the above link can know, these are incidental issue, the lack of a word or phrase can be found in the essay. (That is also a cloze type?)
cnn dataset has about 90,000 documents, there are 380,000 problem.
# No, the above should be misunderstood, is the top story, the following is the problem question:
Download format is to look inside, but found not read ah, yes @entity used to replace the above it? Well, this is why?
Well, since there is the question papers to story also dim?
2. The author gives pre-trained model file, so happy ah!