Haoran Shi

@amioranza Just a supplement. On mac, keep the option key instead of SHIFT key when selecting. Also, use command+v to paste.

I'm also wondering why these codes are missing. Seems that the html page structure is tidy so it should be crawled correctly. Could you share your fix? @22csnyder Thanks very...

I checked the source code, and found that the first levelFactory only considers 'lvl1.chapter',ignoring the 'lvl1.tabular'. And the format of source website is inconsistent, which is the reason why '740-759'...

Any update on this? I also need an overwrite mode especially when the input S3 path is the output from a spark job.

Also wondering what's the purpose of these query masking and key masking. Can anyone help? Thanks very much. @duyvuleo @Kyubyong

@duyvuleo Hi, the query masking and key masking are utilized to filtered out the attention allocated to `padding' in the sequence.

@gitfourteen @jiangxinyang227 Yes, this issue has been posted here https://github.com/Kyubyong/transformer/issues/33 before. So this repo can only serve as a toy example, not the same as the original implementation in tensor2tensor....

@sorear According to your procedures, I build a 32bit bbl on 64bit ubuntu. And the bbl can run correctly to print out the LOGO. Then I build vmlinux according to...

The original sentence is actually the self-reconstructed sentence from the decoder, not the input sentence.