After utilizing the convert_marian_pytorch.py script to convert an OPUS translation model to a format compatible with Hugging Face Transformers, I uploaded the resulting model to Hugging Face's model hub here. However, upon testing, the translated outputs from the model appear to be a jumble of random words and lack coherence.
I'm seeking assistance in resolving this issue or exploring alternative methods to utilize Marian models directly in Python without relying on the OPUS cat mt engine. Any insights or suggestions would be greatly appreciated.
upon testing, the translated outputs from the model appear to be a jumble of random words and lack coherence.