About the order of the context sentence and the auxiliary sentence
huhk-sysu opened this issue · comments
I notice that when processing the data (in processor.py), if the task is QA-M or NLI-M, the context sentence is assigned to text_a, and the auxiliary sentence is assigned to text_b. However, when dealing with the task QA-B and NLI-B, text_a is the auxiliary sentence and text_b is the context sentence, which is the opposite.
Could you explain your idea of doing this? Thanks!
Hello. In QA-B and NLI-B, the order of the sentences has little effect on the experimental results. We assign auxiliary sentence to text_a just to make it easier to make some other attempts (for example, take the final hidden state of label (text_a ) instead of [CLS] to do classification).
Thank you for the quick response.