When I follow the instruction for MS Marco, I obtain the expected results for the dev set but not for eval set. All the evaluation metrics are 0 after evaluating each 1k samples. I also tried the code with the TFRecord files provided but the same results for that. What am I missing?
That is because the ground-truth of the eval set is not public. The only way to get the MRR@10 for the eval set is to submit your run to the MS MARCO leaderboard.
Hi,
When I follow the instruction for MS Marco, I obtain the expected results for the dev set but not for eval set. All the evaluation metrics are 0 after evaluating each 1k samples. I also tried the code with the TFRecord files provided but the same results for that. What am I missing?