Sorry to bother u again.
During the inference,after the queries is encoded,the encoded queries is setted to save under the "split_save"=False.While the encoded corpus is setted to save under the "split_save"=True,which means during finding the topk relevent documents,each query will own topk relevant documents for each encoded corpus file because u use for i, part in enumerate(corpus_all_partitions):.
I don't know if I get wrong understanding. But for the later computation of ***@10 metric,is this right?
chenbuhuu
commented
2 weeks ago
Sorry to bother u again. During the inference,after the queries is encoded,the encoded queries is setted to save under the "split_save"=False.While the encoded corpus is setted to save under the "split_save"=True,which means during finding the topk relevent documents,each query will own topk relevant documents for each encoded corpus file because u use
for i, part in enumerate(corpus_all_partitions):. I don't know if I get wrong understanding. But for the later computation of ***@10 metric,is this right?