Fine-tune MT5ConditionalGeneration for question generation

I trained MT5ConditionalGeneration for an end2end question generation task. Now I have a problem with inference. for example, when I test the model with sample A, the output of MODEl.generate() is B. After that, when I test the model with sample B (which was the output of A), the output of the model is again B. can someone tell me why??

1 Like