Hi, we have an implementation of Approximate kNN search, with 768-dimension embedding vector in the ANN search. We observed that the total hits are not consistent when running the SAME query multiple times, or paginate using from and size. Sometimes the hits/total could vary by 100s when the total is thousands. We use k=100 and there are 5 shards. Any idea to make the total not changing?
Hi @Dazhi_Jiao , that is surprising! To help us debug, could you please send the API request you're using, as well as some information on the dataset?
It is also always a good idea to specify which version of Elasticsearch you are using.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.