Rico Sennrich

thanks for reporting this issue. @yimmon , this is related to parallel support you contributed; could you have a look?

thank you for your pull request. I have two comments: - please test code before opening a PR. This PR contains a syntax error. - I'm happy to accept PRs...

we're currently looking into this, but migrating the code to Tensorflow 2 is a major undertaking that also carries risk (to stability, compatibility, efficiency).

There is a branch compatible with Tensorflow 2, which will be eventually merged into master: https://github.com/EdinburghNLP/nematus/tree/tf2 The branch doesn't yet support eager execution, and any contributions to make it the...

Hi Antonio, yes, this should be changed to 'or'. The reason why the current code doesn't crash is that there's another length check in prepare_data().

This was fixed in data_iterator.py in commit 52ea8803cc1aef7683ecf646de4d1cec2332f21d, but still remains TODO for domain_interpolation_data_iterator.py

Hello Amit, it looks like you don't have the permission to execute "$script_dir/validate.sh". Can you check the file permissions? If you let me know how you downloaded the files, I...

thank you for this report. I'm happy to accept PRs that improve efficiency.

thanks for your suggestion. There are pre-trained models available at the following locations: http://data.statmt.org/wmt16_systems/ http://data.statmt.org/wmt17_systems/ https://github.com/EdinburghNLP/tramooc-mt (models trained with Nematus; inference with Marian) only the last one is dockerized though,...

(Also note that all models were trained with the Theano version, and can be converted to run with the Tensorflow version of Nematus using `nematus/theano_tf_convert.py`.