Ilya Kuznetsov
Ilya Kuznetsov
I made a minor fix in MCL (2249b2ccdff80a5db1d951c2358d653d9fea9c91). Now embeddings generation seems to work fine.
Example: ``` drop table if exists test; CREATE TABLE test ( title1 TEXT, title2 TEXT, strattr STRING, image_vector FLOAT_VECTOR KNN_TYPE='hnsw' HNSW_SIMILARITY='l2' MODEL_NAME = 'sentence-transformers/all-MiniLM-L6-v2' FROM = 'title1, title2, strattr' )...
Daemon branch is `embeddings_generation`
I added an error on CREATE TABLE and model loading on table prealloc in 40afbd0e2bc332d8a934d544020ae7c3fdec895d. I also made some changes to cmake build scripts in MCL so it I could...
Implemented improvements in `oversampling_improvements` branches (daemon/MCL). The result on my test dataset (dbpedia): ``` No oversampling: 0.004s With 3x oversampling: 1.17s After improvements, with 3x oversampling: 0.028s ``` Performance has...
Done in https://github.com/manticoresoftware/manticoresearch/commit/b83863e29cebc1a3de6f1ae4b0c1cd92a0c9c2dc / https://github.com/manticoresoftware/columnar/commit/d65abb5a99f82a1fcd272c450960c03c723bd872
> Could you please update [the spreadsheet](https://docs.google.com/spreadsheets/d/1iyRR4Qv-TarS99J0WvFQbwhbvZBloTfAqpRyUmRg6T4/edit?gid=0#gid=0), so we can see the current status? I am still benchmarking different option combos on perf3.
Updated table to include recent results
Updated table with include recent results after changing columnar attribute fetch stage
Test suite already has RoundFloatValues that is used for some cases, why not use it specifically for knn_dist case?