Result on ReadMe ?

[dylanee2]

Hi @tobyyouup ,

I thought it would be good for people like me (who want to use your implementation) if you have result of running your implementation on the dataset.

Were you able to reproduce the result of original conv seq2seq?

Thanks!

[anglil]

@tobyyouup could you report the result in terms of BLEU scores in the wmt'14 en-de translation task? Thanks!

[tobyyouup]

@dylanee2 @anglil I have run the iwslt14 de-en task, the command is shown in the README, and I can get a BLEU score 25 at step 60k with batchsize 32.

[anglil]

@tobyyouup thanks, that looks promising. Could you comment on how many GPUs you used and what dataset (newstest 2014 or newstest 2015 or something else) you got a BLEU score of 25 on?

[tobyyouup]

Hi anglil, just one tesla K40 GPU. The traning data is IWSLT Germain-English, I concatenate dev2010, tst2010, tst2011 and tst2012 as the test set.

[anglil]

Thanks.

[cauivy]

@tobyyouup Hi, tobyyouup, do you remember how long it took to train step 60k with batchsize 32 on iwslt14 de-en?