I get this error when I use Example code for huggingface ,how can i fix it?

Question

I get this error when I use Example code for huggingface ,how can i fix it?

sammichenVV opened this issue 2 years ago · comments

this is my code:

the error is:Expected format: {'predictions': Value(dtype='int32', id=None), 'references': Value(dtype='int32', id=None)},

Skander Hellal · Answer 1 · Mon Jan 02 2023 20:52:42 GMT+0800 (China Standard Time)

Hello,
Can you give us further details on the task you are working on? is it a simple classification task? In that case , if you flatten your tensors, you will get the good result.

xiaochenchen · Answer 2 · Tue Jan 03 2023 13:49:52 GMT+0800 (China Standard Time)

it's fixed by flatten tensors. my task is translation . I use xlm-roberta to fine tune. The way to load the model is XLMRobertaForCausalLM.from_pretrained('xlm-roberta-base'). and I fix it by flatten tensor "predictions=predictions.view(5120) references=batch["labels"].view(5120)".

Leandro von Werra · Answer 3 · Fri Jan 06 2023 15:26:44 GMT+0800 (China Standard Time)

Closing this as it seems resolved.

cspink · Answer 4 · Wed Jul 05 2023 16:25:48 GMT+0800 (China Standard Time)

Could you elaborate on the flattening, @SkanderHellal? I get the same error for my setup and don't understand why it doesn't work, as training works with the same preparation of the dataset. Where can I read more on this?