Compatible with pgvector
pgvecto.rs can be configured to be compatible with pgvector
at:
- Create table
- Create vector indexes
- Search vectors
For create table
and search vectors
, pgvecto.rs supports this feature natively.
For create vector indexes' this feature should be enabled with
SET vectors.pgvector_compat=on;`.
Examples
It's easy to enable compatibility mode and run a search.
DROP TABLE IF EXISTS t;
SET vectors.pgvector_compat=on;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) SELECT ARRAY[random(), random(), random()]::real[] FROM generate_series(1, 1000);
CREATE INDEX hnsw_l2_index ON t USING hnsw (val vector_l2_ops);
SELECT COUNT(1) FROM (SELECT 1 FROM t ORDER BY val <-> '[0.5,0.5,0.5]' limit 100) t2;
DROP INDEX hnsw_l2_index;
Multiply types of indices are accepted:
SET vectors.pgvector_compat=on;
-- [hnsw + vector_l2_ops] index with default options
CREATE INDEX hnsw_l2_index ON t USING hnsw (val vector_l2_ops);
-- [hnsw + vector_cosine_ops] index with single ef_construction option
CREATE INDEX hnsw_cosine_index ON t USING hnsw (val vector_cosine_ops) WITH (ef_construction = 80);
-- anonymous [hnsw + vector_ip_ops] with all options
CREATE INDEX ON t USING hnsw (val vector_ip_ops) WITH (ef_construction = 80, m = 12);
-- [ivfflat + vector_l2_ops] index with default options
CREATE INDEX ivfflat_l2_index ON t USING ivfflat (val vector_l2_ops);
-- [ivfflat + vector_ip_ops] index with all options
CREATE INDEX ivfflat_ip_index ON t USING ivfflat (val vector_cosine_ops) WITH (nlist = 80);
-- anonymous [ivf + vector_ip_ops] with all options
CREATE INDEX ON t USING ivfflat (val vector_ip_ops) WITH (lists = 80)
Limitation
For compatibility, we strive to maintain a consistent user experience, but there are still some limitations in two aspects:
- Some features of pgvector.rs are not available in `compatibility mode'.
- Some features of pgvector are different in `compatibility mode
Unavailable features of pgvecto.rs
When pgvecto.rs is running in compatibility mode
, some features of pgvecto.rs are not available:
flat
index- Quantization, including
scalar quantization
andproduct quantization
- prefilter and vbase
And for index ivfflat
and hnsw
only the following options are available. Their default value is different from pgvecto.rs original, which keeps the same from pgvector
.
Options for `ivfflat
Key | Type | Default | Description |
---|---|---|---|
lists | integer | 100 | Number of cluster units. |
Options for hnsw
.
key | type | default | description |
---|---|---|---|
m | integer | 16 | Maximum degree of the node. |
ef_construction | integer | 64 | Search extent in construction. |
Difference from pgvector
Although in `compatibility mode', there are still differences from pgvector.
Known problems are not limited to:
- Create vector indexes is asynchronous at pgvecto.rs, instead of synchronous at pgvector
- Query Options ivfflat.probes need to be
vectors.ivf_nprobe
, and hnsw.ef_search need to bevectors.hnsw_ef_search
- Vector Functions and Aggregate Functions are not supported