Don't put volumes

NLG Task: <br> <br> This paper describes Language Agnostic Delexicalisation an SP for Multilingual RDF to text generation <br> Training Data: <br> <br> The training data used is provided by the organizer of the WebNLG Challenge 2020. The Russian part of the data has been created by translating English with a Machine Translation system and then post edited by crowdsourcing then spell checked. The data can be found here <a href=”https://gitlab.com/shimorina/webnlg-dataset/-/tree/master/release_v3.0”>https://gitlab.com/shimorina/webnlg-dataset/-/tree/master/release_v3.0</a><br> <br> Model Description:<br> <br> Use a transformer as architecture. fairseq toolkit is used. There is 2 encoder and 1 decoder. 4 layers 256 hidden size, 3072 size for the feed forward layers. Trained with 0.4 dropout and 0.1 attention dropout <br> <br> Key Contribution:<br> <br> The LAD approach outperforms other solutions for English but cannot be tested for Russian because they didn’t have enough data. <br> <br> Result: <br> <be> <image class="img-fluid" src="https://puu.sh/Ik0l7/782b4f264d.png"alt="Result">