Textual Entailment on entire retrieved document using pretrained model with decomposable attention

[j6mes]

Textual Entailment on entire retrieved document using pretrained model with decomposable attention

[j6mes]

@andreasvlachos

This can be run with:
PYTHONPATH=src:lib/allennlp/:lib/DrQA/ python src/scripts/rte/da/train.py data/fever/drqa.db config/fever.json logs

You might have to adjust the batch size depending on your GPU. (I had to turn it down from 64 to 4 to fit on Maki's 1080Ti).

Currently training and should have an indication of the dev accuracy soon

@glampouras - thanks for letting me borrow your PC. Without the GPU it takes 1.5 hours per epoch. With the GPU it takes 6 mins

[andreasvlachos]

Sounds promising! Glad to hear money well spent on that GPU! Will give it a spin myself too when I am back!

…

On Wed, 6 Dec 2017 at 10:32 James Thorne ***@***.***> wrote: @andreasvlachos <https://github.com/andreasvlachos> This can be run with: PYTHONPATH=src:lib/allennlp/:lib/DrQA/ python src/scripts/rte/da/train.py data/fever/drqa.db config/fever.json logs You might have to adjust the batch size depending on your GPU. (I had to turn it down from 64 to 4 to fit on Maki's 1080Ti). At epoch 3, we're getting dev accuracy of >.79 @glampouras <https://github.com/glampouras> - thanks for letting me borrow your PC. Without the GPU it takes 1.5 hours per epoch. With the GPU it takes 6 mins — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABbUhWANST4dgFqgfyU_F5n_xIcp5zdzks5s9m02gaJpZM4QrfH0> .