NMT Benchmark platform

evaluation

(jean.senellart) #1

We are releasing a benchmark platform here: http://nmt-benchmark.net.

The idea is simple: we provide training corpus, test corpora (from open competitions and internal in-domain test set), you have of course OpenNMT code base - or you can use any other framework, and your mission is to submit the best system! The training data is on purpose small (one million sentence) - so that training of a system takes less than one day.

An restfull API is available here to automate system creation and test translation submission and you have commandline tools here.

We have already submitted some baseline results for 13 language pairs and will submit more baseline and advanced systems.

Who will be on top of the leaderboard :crown: ?