I created an rt-table in plain mode with an embedded vector, the conf file for the table creation is attached (manti.conf.rtf), when I connect to the manticore server and describe the table it looks ok (describe.info.rtf) I then insert data into the table and all data seems ok and loaded except that all the embedded vector values (auto-calculated) have the same value (?) see file (select_all_output.rtf). As the data is different I would expect the vectors to be different.
Other problem is when I query the data, see file (query_output.rtf) the knn query returns all rows with the same knn_dist while the fulltext query returns correct results.
To me it does not look right, I am new to rt-tables and auto-embeddings but I would expect different vector values in the rows and different knn distances in the query results.
Any ideas ?
Update
works fine on latest version (25) on my Mac, the problem may be related to #4314 as I was using an empty “from” field list.
Bug confirmed on version 17.5.1 if you leave the from field empty as in:
knn = {“attrs”:[{“name”:“emb_vector”,“type”:“hnsw”,“hnsw_similarity”:“L2”,“hnsw_m”:16,“hnsw_ef_construction”:200,“model_name”:“sentence-transformers/all-MiniLM-L6-v2”,“from”:“”}]}
it will not default to using all rt_fields, what is actually using is a mystery but all the emb_vector values for us were the same. Now we are listing the field names and it is working fine (this should be fixed in V25 though).