Closed yc-song closed 4 weeks ago
Hello, thanks for your interest in our work! You should preprocess the original IIRC dataset using the method mentioned in Appendix D (more specifically, we follow the steps in IRCOT ), then you can train Beam Retrieval by editing the dataset_type in the bash file. The Retrieval EM is calculated in the same way as MuSiQue because the preprocessed IIRC shares the same format with MuSiQue.
Everything clear. Thanks.
Hi. Thanks for your interesting work.
I'm wondering if you can provide a minimal guideline for running your codebase on IIRC dataset (appendix D)
Also, I'm curious if the result from appendix D is applied to the official IIRC dataset or you went through another pre-processing.
Best regards.