Etienne Monneret

Results 71 comments of Etienne Monneret

My problem occurred using the URL API. Doing the same with the `mmt` command line, as you did, worked. Now, the URL API translation is properly using the memory. It's...

What is strange is that, now, I get a memory score, even when no suggestion is used: ``` contextVector":{"entries":[{"memory":{"id":1,"name":"USERS.fr-en"},"score":0.080684796}]} LOG: suggestions = [ ] ``` Before the manual test, as...

Possibly an explanation... In your documentation, the `http://localhost:8045/memories/2/corpus` URL is using a POST. https://github.com/modernmt/modernmt/wiki/API-Add-to-Memory Historically, I used to send the values on the URL itself, thus as GET values. It...

I think I finally found the bug: the context as string is no more working! When I send a `context-vector` param in both the URL API or the command line,...

In experiments reported in past posts here, I used some new memories adding only one sentence pair in it, and then use it to translate right after this. Most of...

I did a `git pull` and was surprised by the number of impacted files on this 4.5.4 release. I did a `python3 setup.py` making `cassandra` and `kafka` to be downloaded...

I tried with older MMT44, MMT421, MMT412. Starting with a fresh new model just after training. Manually creating a memory and feeding it with the `mmt` command line. Can't get...

An idea: in your test above, you are using the default model, while in my case I'm using several named models. If the `context` parameter is only evaluating the default...

@davidecaroselli, as I said above, for me, there is really still a problem. And this problem might be an old one.

All is explain above about my own case, mainly here: https://github.com/modernmt/modernmt/issues/537#issuecomment-631251346 https://github.com/modernmt/modernmt/issues/537#issuecomment-631318270 To be sure your memory is used, you have to start `mmt` with option `-v 3`, translate something,...