Modernmt: How to extend an already-built engine?

[mzeidhassan]

Hi David,

I am wondering if there is a way to train more tmx or bilingual files and add them to an already trained engine? How can this be achieved?

Thanks,
Mohamed

[letconex]

https://github.com/ModernMT/MMT/wiki/API-Add-to-Domain

[nicolabertoldi]

The current example refers explain how to add a single sentence pairs.

I would suggest to add more details and an example to clarify how to add an entire file in TMX format.

On 23 Mar 2017, at 10:20, Leon [e-mail] wrote:

https://github.com/ModernMT/MMT/wiki/API-Add-to-Domain

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.

[AndRossi]

Hi @memoday,

Yes, you can execute the ./mmt import command to upload a TMX to your model under a certain domain. The command syntax is:
./mmt import -d -x -e -p
(where as usual engine name and api port have default values "default" and 8045 respectively)

However it is also possible to upload a TMX file through the REST APIs, too, e.g. using curl and sending the TMX with form encoding.

@nicolabertoldi
I think you are right, that wiki page can be expanded with a TMX upload example.
I'll take care of that :)

[letconex]

Thank you, AndRossi, can you please also provide the right syntax for doing this with the API?

[AndRossi]

Hi @letconex, sure. Here is the syntax:

curl -X PUT -H "Content-Type: multipart/form-data"
-F "content=@
-F "content_type=tmx"
http://localhost:8045/domains//corpus
| python -mjson.tool

Please note that this command uploads the tmx to an already existing domain.
If you want to upload it on a new domain, first create the new domain with command:

curl -X POST http://localhost:8045/domains -d "name=" | python -mjson.tool

Then check which ID was assigned to the new domain, and finally use it in the curl -X PUT command above.

[AndRossi]

@nicolabertoldi
I have just updated the wiki page with an example of TMX file upload.
I think it is clearer now.