-
Notifications
You must be signed in to change notification settings - Fork 92
New issue
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
Eng to Chinese translation weird behaviour when translating single word #13
Comments
Sorry, no I don't know. Do you know if the same happens with the native MarianNMT model? |
I have not tried on native Marian model, as I have been using Python and huggingface mainly. Is there a major difference between the two? |
You could try this: Seq2SeqSharp Release Package |
this bug seems to only repredoce on ">>cmn_Hans<<", won't repreduce on ">>cmn_Hant<<". |
@zhongkaifu Can you give a demo for showing us how to use your model to translation? |
@iDustPan For code, you can look Seq2SeqConsole project in Seq2SeqSharp repo. I copy and paste some code here which include functions for training, validation and test. Let me know if you have any additional questions. ``c#
`` |
I am using Marian translation via transformers library in python, most of the time it works great, but in certain cases it takes a long time and returns weird results.
Most of the time it works fine and returns within a second, for example:
But if I enter certain words it takes more than 30 seconds and give a weird result:
where the correct translation should be just a single "狗"
Any idea what is wrong or if there are any ways to handle exceptional cases?
The text was updated successfully, but these errors were encountered: