Using examples in metrics function.

[Ali1858]

Hi,

I fine-tuned t5 models and now want to evaluate the trained model on a specific dataset/Task which can be easily done using eval mode. But the problem is, I also need other information (like row_id and other important columns) including targets and predictions. Is there a way to pass the whole example into the metrics function?

Currently, I am getting predictions from seqio.Task using eval mode and reading and evaluating the prediction file outside the seqio pipeline. Is there a way I can do my custom evaluation inside the pipeline only instead of reading the generated prediction file separately?

Thanks.