Self-Host LibreTranslate Compact Version on Cloudron
-
By default, LibreTranslate is large (16GB) due to all the modules.
There is a yaml file where you can specify the arguments for what happens during installation.
If Cloudron could provide a list of boxes to be ticked in the first run section, or something like that, it might be a nice way of selecting which of all the languages you want available for translation.
This could reduce the storage requirement for LibreTranslate, and make the time from launching it on its first run to being usable much quicker.
-
Agree that the package is quite large. @nebulon and @vladimir-d can correct me, but from what I remember, even without any language models at all, the docker image was ~7GB.
@girish said in Self-Host LibreTranslate Compact Version on Cloudron:
Agree that the package is quite large. @nebulon and @vladimir-d can correct me, but from what I remember, even without any language models at all, the docker image was ~7GB.
Because the default configuration for the first run includes downloading every module, I think a lot of people (like myself) are going to think that LibreTranslate doesn't work, as it takes a while to retrieve everything. It might be a good idea to include a message in "first run" indicating that there is likely to be a delay while it gets ready.
-
@girish said in Self-Host LibreTranslate Compact Version on Cloudron:
Agree that the package is quite large. @nebulon and @vladimir-d can correct me, but from what I remember, even without any language models at all, the docker image was ~7GB.
Because the default configuration for the first run includes downloading every module, I think a lot of people (like myself) are going to think that LibreTranslate doesn't work, as it takes a while to retrieve everything. It might be a good idea to include a message in "first run" indicating that there is likely to be a delay while it gets ready.
@LoudLemur actually all language models are currently part of the image itself, so the initial startup is faster. Also half the image size is actually the app itself, mostly python modules like torch. The goal for the package was basically to have an app which needs no further configuration and has everything included, with the tradeoff of disk space essentially.
-
@LoudLemur actually all language models are currently part of the image itself, so the initial startup is faster. Also half the image size is actually the app itself, mostly python modules like torch. The goal for the package was basically to have an app which needs no further configuration and has everything included, with the tradeoff of disk space essentially.
@nebulon That is good. I am still trying to work out why it didn't run for when I tried it. Perhaps Cloudron indicated that LibreTranslate was "Running" prior to it having fully got going properly. A few other applications, like the forum software, are similar. In the case of LibreTranslate, it would probably take even longer.
Anyway, I have it running on my desktop and have been having a look around for additional translation tools.
https://opennmt.net/
https://github.com/maidis/awesome-machine-translation#applications- -
@nebulon That is good. I am still trying to work out why it didn't run for when I tried it. Perhaps Cloudron indicated that LibreTranslate was "Running" prior to it having fully got going properly. A few other applications, like the forum software, are similar. In the case of LibreTranslate, it would probably take even longer.
Anyway, I have it running on my desktop and have been having a look around for additional translation tools.
https://opennmt.net/
https://github.com/maidis/awesome-machine-translation#applications-
