Vector Data Types

This topic describes the vector data types.

Data Types

Snowflake supports a single vector data type, VECTOR.

Note

The VECTOR data type is only supported in SQL, the Python connector and the Snowpark Python library. No other languages are supported.

VECTOR

With the VECTOR data type, Snowflake encodes and processes vectors efficiently. This data type supports semantic vector search and retrieval applications, such as RAG-based applications, and common operations on vectors in vector-processing applications.

To specify a VECTOR type, use the following syntax:

VECTOR( <type>, <dimension> )
Copy

Where:

  • type is the Snowflake data type of the elements, which can be 32-bit integers or 32-bit floating-point numbers.

    You can specify one of the following types:

    • INT

    • FLOAT

  • dimension is the dimension (length) of the vector. This must be a positive integer value with a maximum value of 4096.

Note

Direct vector comparisons (e.g. v1 < v2) are byte-wise lexicographic and, while deterministic, won’t produce results that you’d expect from number comparisons. So while you can use VECTOR columns in ORDER BY clauses, for vector comparisons, use the vector similarity functions provided.

The following are examples of valid definitions of vectors:

  • Define a vector of 256 32-bit floating-point values:

    VECTOR(FLOAT, 256)
    Copy

  • Define a vector of 16 32-bit integer values:

    VECTOR(INT, 16)
    Copy

The following are examples of invalid definitions of vectors:

  • A vector definition using an invalid value type:

    VECTOR(STRING, 256)
    Copy

  • A vector definition using an invalid vector size:

    VECTOR(INT, -1)
    Copy

Vector conversion

This section describes how to convert to and from a VECTOR value. For details on casting, see Data Type Conversion.

Converting a value to a VECTOR value

VECTOR values can be explicitly and implicitly cast from the following types:

Converting a value from a VECTOR value

VECTOR values can be explicitly and implicitly cast to the following types:

VECTOR values are implicitly cast to the same types automatically by Snowflake.

Loading and unloading vector data

Directly loading and unloading a VECTOR column is not supported. For VECTOR columns, you must load and unload data as an ARRAY and then cast it to a VECTOR when you use it. To learn how to load and unload ARRAY data types, see Introduction to Loading Semi-structured Data. A common use case for vectors is to generate a vector embedding.

The following example shows how to unload a table with a VECTOR column to an internal stage named mystage:

CREATE TABLE mytable (a VECTOR(float, 3), b VECTOR(float, 3));
INSERT INTO mytable SELECT [1.1,2.2,3]::VECTOR(FLOAT,3), [1,1,1]::VECTOR(FLOAT,3);
INSERT INTO mytable SELECT [1,2.2,3]::VECTOR(FLOAT,3), [4,6,8]::VECTOR(FLOAT,3);

COPY INTO @mystage/unload/
  FROM (SELECT TO_ARRAY(a), TO_ARRAY(b) FROM mytable);
Copy

The following example shows how to load a table from a stage and then cast the ARRAY columns as VECTOR columns:

CREATE OR REPLACE TABLE arraytable (a ARRAY, b ARRAY);

COPY INTO arraytable
  FROM @mystage/unload/mydata.csv.gz;

SELECT a::VECTOR(FLOAT, 3), b::VECTOR(FLOAT, 3)
  FROM arraytable;
Copy

Examples

Construct a VECTOR by casting a constant ARRAY:

SELECT [1, 2, 3]::VECTOR(FLOAT, 3) as vec;
Copy

Add a column with the VECTOR data type:

ALTER TABLE issues ADD COLUMN issue_vec VECTOR(FLOAT, 768);

UPDATE TABLE issues
  SET issue_vec = SNOWFLAKE.CORTEX.EMBED_TEXT_768('e5-base-v2', issue_text);
Copy

Limitations

  • There is limited language support for the VECTOR data type. Languages not represented in this table are not supported.

    Snowflake feature

    Python

    SQL

    UDFs

    ✔

    ✔

    UDTFs

    ✔

    ✔

    Drivers/Connectors

    ✔

    ✔

    Snowpark API

    ✔

  • Vectors are not supported in VARIANT columns.

  • Arithmetic operations on vectors are not supported other than through user-defined functions (UDFs).

  • Server-side binding is not supported. This means that when writing to a VECTOR column through a Snowflake driver, you must cast the VECTOR values in the query before running the query.

  • Vectors are allowed in hybrid tables but not as primary and foreign keys.

  • The VECTOR data type is not supported for use with the following Snowflake features:

Language: English