a baby giant panda near a large giant panda
Try input "a baby giant panda near a large giant panda" and see what the model outputs.
I directly use the sample code, and the model outputs: A baby giant pandanda near a large giant pandemic.
Haha, good catch. I guess it's a child of its time.
To prevent these kinds of issues, I guess we would need more data and a bigger model. However, I think that is the fundamental flaw of this sequence to sequence approach, you can never be sure that the model does not make up texts.
Hey
@huihuixu
,
I just uploaded an improved multilingual version of my model. It uses a different base model and an improved training routine.
Your example works now: https://huggingface.co/oliverguhr/spelling-correction-multilingual-base?text=fix%3Aa+baby+giant+panda+near+a+large+giant+panda
Let me know what you think.