Open GarfieldHuang opened 1 month ago
I don't see any error and my test with Seamless M4T as Text translator worked.
So i think it failed to load the M4T model alltogether.
Can you maybe post the full log here in case there is already an error at loading time when starting? For that go to "Advanced -> Logs" and press "Copy Log" and paste it here, or enable "Write Logfile", restart the app and upload the log.txt.
Its possible that it failed loading to CUDA since M4T uses the transformers library which might require the CUDA Toolkit driver from NVIDIA, compared to CTranslate2 which does not need it and is used by M2M100 and Faster NLLB200
When using the whispering-tigeri-ui.exe and selecting the text-translation type: Seamless M4T with the following settings:
CUDA Large float 16 precision English to Chinese The following error occurs:
Transcribe (OSC): One million, two million, three million. Error: 'NoneType' object has no attribute 'text_translate' Traceback (most recent call last): File "Models\TextTranslation\texttranslate.py", line 100, in TranslateLanguage AttributeError: 'NoneType' object has no attribute 'text_translate'
However, when using M2M100, there is no error, but the translation quality is not good enough.