engine = "pg_vectorsearch" [dataset] name = "cohere" size = "10m" table = "cohere_wiki" expected_rows = 10000000 dims = 1024 vector_field = "emb" s3_base_path = "s3://paradedb-benchmark/datasets/cohere/v1" [postgres.build] shared_buffers = "8GB" maintenance_work_mem = "8GB" effective_cache_size = "24GB" work_mem = "256MB" max_parallel_maintenance_workers = 7 max_parallel_workers_per_gather = 4 max_parallel_workers = 8 max_worker_processes = 32 [postgres.search] shared_buffers = "8GB" maintenance_work_mem = "8GB" effective_cache_size = "24GB" work_mem = "256MB" max_parallel_maintenance_workers = 7 max_parallel_workers_per_gather = 4 max_parallel_workers = 8 max_worker_processes = 32 [index] nlist = 0 [search] queries = 100 runs = 1 warmup_runs = 1 explain_sample = 100 [search.session] "cohere.titles_1pct" = "battle" "cohere.titles_10pct" = "year" enable_seqscan = "off" enable_bitmapscan = "off" enable_sort = "off" max_parallel_workers_per_gather = "4" jit = "off" plan_cache_mode = "force_custom_plan" [search.sweep] param = "prism.nprobe" integer = true min = 1 max = 4096 start = 64 targets = [0.9, 0.95, 0.99] tolerance = 0.000001 max_probes = 20 [search.fast] pilot_queries = 25 max_probes = 64 bracket_ratio = 1.1 query_timeout_seconds = 300 budget_seconds = 10800