Etienne Monneret
Etienne Monneret
Using this PR https://github.com/modernmt/modernmt/pull/553 I wrote a small NMT live-tuning demonstrator: http://biliedit.cubaix.com/ Here is a short video in action: https://youtu.be/et3tyhMegKE I think it's a nice and funny way to demonstrate...
I would like to be able to mark a Very Important Part (VIP) on a source text and get it back precisely either after a MMT translation, or after a...
MMT tag projection can change the order of the sequence of tags in a sentence. This may cause a DTD-invalid result. This is especially damageable in full-automatic process. Here is...
One of the major problem we encountered, is the fact that some well-known terms are not well translated, even after having post-edited them a lot of times in a lot...
Hi all, I have an existing memory 1: ``` curl "http://localhost:10001/memories" {"status":200,"data":[{"id":1,"name":"USERS.fr-en"}]} ``` I'm feeding a sentence pair to memory 1, getting what seems to be a valid result: ```...
When trying to install latest MMT421, I got this error: `ERROR: torch-1.0.1.post2-cp36-cp36m-linux_x86_64.whl is not a supported wheel on this platform.` This is due to the fact that your installation procedure...
I have 2 trainings running on a machine, where I do not notice any strong disk access. This is just for a comparison with the next situation. I have an...
When looking at the neural translation log, it appears to me that, very often, for the live re-training, MMT is using large (or very large) sentence, where only a very...
Working on #436, I'm studying MMT code. I discovered that MMT is strictly serialising sentence processing, one sentence at a time. This is done using a decoder list in a...
What was proposed in #447 is very interesting, but brings several bad behaviours, because of the way the neural model is handling the live-tuning. Here is a new version, reverting...