Chinese model from OPUS-MT

Link (en-zh, zh-eh): OPUS-models – Google Drive

Seems to perform better than the current model (OPUS model on top):

Input: When people don't see moose as potentially dangerous, they may approach too closely and put themselves at risk.

当人们不认为麋鹿有潜在危险时, 他们可能会靠近太近,把自己置于危险之中

vs. 

当人们看作是潜在的危险的时候,他们可能非常接近,使他们面临危险.
When people don't think moose are potentially dangerous, they may get too close and put themselves in danger.

vs.

When people see them as potentially dangerous, they may come very close, putting them at risk.

Input: hello

你们好

vs. 

背 景
Hello.

vs.

BACKGROUND

This one made me giggle (like SALAD) :laughing:

1 Like

Also uploaded Urdu (en-ur, ur-en).

(en)> hi there!
(ur)> وہاں ہیلو!

Nice I’ll add these to argospm too

1 Like

The .argosmodel packages need a directory in the root of the zip archive.

For example, the zip archive on the left in this screenshot is correct:

With the current implementation of Argos Translate the model on the .argosmodel package on the right fails when you try to install it with this error:

Traceback (most recent call last):
  File "/home/pj/git/argos-translate-gui/argostranslategui/gui.py", line 396, in load_languages
    self.languages = translate.load_installed_languages()
  File "/home/pj/git/argos-translate-gui/env/lib/python3.10/site-packages/argostranslate/translate.py", line 627, in load_installed_languages
    return get_installed_languages()
  File "/home/pj/git/argos-translate-gui/env/lib/python3.10/site-packages/argostranslate/translate.py", line 512, in get_installed_languages
    packages = package.get_installed_packages()
  File "/home/pj/git/argos-translate-gui/env/lib/python3.10/site-packages/argostranslate/package.py", line 317, in get_installed_packages
    to_return.append(Package(path))
  File "/home/pj/git/argos-translate-gui/env/lib/python3.10/site-packages/argostranslate/package.py", line 191, in __init__
    raise FileNotFoundError(
FileNotFoundError: Error opening package at /home/pj/.local/share/argos-translate/packages/stanza/metadata.json no metadata.json
Aborted (core dumped)

The models work though!

The Urdu → English model doesn’t seem great but it’s useable. I tried to train an Urdu last year and there wasn’t very much data available.

1 Like

Ah! I will fix that. :pray:

1 Like

Hi @pierotofy , do you have also the [zh-en] pair model ? Thank you