[
https://issues.apache.org/jira/browse/TIKA-3329?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17303669#comment-17303669
]
ASF GitHub Bot commented on TIKA-3329:
--------------------------------------
thammegowda opened a new pull request #419:
URL: https://github.com/apache/tika/pull/419
RTG translator integration to tika-translate
This service requires running RTG model for 500-eng translation as a REST
service
By default, the RTGTranslator uses `http://localhost:6060` as base URL
Which can be obtained from a prebuilt docker image:
* can be run on CPU (i.e. without GPU, for testing)
docker run --rm -i -p 6060:6060 tgowda/rtg-model:500toEng-v1
* but GPU (e.g. Device 0) is recommended for translating a lot of documents.
docker run --rm -i -p 6060:6060 --gpus '"device=0"'
tgowda/rtg-model:500toEng-v1
`rtg.base.url` property from `translator.rtg.properties` file can be used
to customize base URL if needed.
CC @chrismattmann @tballison
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> RTG Translator with many-to-eng translation
> -------------------------------------------
>
> Key: TIKA-3329
> URL: https://issues.apache.org/jira/browse/TIKA-3329
> Project: Tika
> Issue Type: Improvement
> Components: translation
> Reporter: Thamme Gowda
> Assignee: Thamme Gowda
> Priority: Major
>
> The existing translation services in tika-translate are either
> commercial/paid engines (e.g. Google, Microsoft etc ) or not state of the
> art (such as Joshua, Moses etc).
> Reader Translator Generator () is a neural machine translation toolkit
> [https://isi-nlp.github.io/rtg/]
> and has the implementation of Transformer NMT model (current state of the
> art).
> It also has massively multilingual pretrained NMT model ( many-to-English
> translation direction)
> [https://hub.docker.com/repository/docker/tgowda/rtg-model]
> in which about 500 source languages are represented, with atleast ~300 source
> languages have good enough quality (For a comparison Google translate has
> ~106 languages, and Microsoft has about 80 languages).
> This issue is for integrating RTG Translator into tika-translate
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
