Initial commit

benchmarks/profiling/build-from-npy.sql

@@ -0,0 +1,17 @@
+.timer on
+pragma page_size = 32768;
+--pragma page_size = 16384;
+--pragma page_size = 16384;
+--pragma page_size = 4096;
+
+create virtual table vec_items using vec0(
+  embedding float[1536]
+);
+
+-- 65s (limit 1e5), ~615MB on disk
+insert into vec_items
+  select
+    rowid,
+    vector
+  from vec_npy_each(vec_npy_file('examples/dbpedia-openai/data/vectors.npy'))
+  limit 1e5;

benchmarks/profiling/query-k.sql

@@ -0,0 +1,31 @@
+.timer on
+
+select rowid, distance
+from vec_items
+where embedding match (select embedding from vec_items where rowid = 100)
+  and k = :k
+order by distance;
+
+select rowid, distance
+from vec_items
+where embedding match (select embedding from vec_items where rowid = 100)
+  and k = :k
+order by distance;
+
+select rowid, distance
+from vec_items
+where embedding match (select embedding from vec_items where rowid = 100)
+  and k = :k
+order by distance;
+
+select rowid, distance
+from vec_items
+where embedding match (select embedding from vec_items where rowid = 100)
+  and k = :k
+order by distance;
+
+select rowid, distance
+from vec_items
+where embedding match (select embedding from vec_items where rowid = 100)
+  and k = :k
+order by distance;