Unsupported or error-prone scenarios
operatorsExplicit and implicit conversion between base types float32
and float16
supports both
implicit
and
explicit
conversion from
,
, and
strings to
, as long as the vector is declared with an
explicit dimension count.
Implicit Conversion is supported only when the target
type is fully declared.
The following examples highlight common errors and limitations when working with half-
precision float
data type in SQL Server.
not
Dimension mismatch
Null handling
currently does
support implicit conversion between
and.
Additionally,
explicit conversion
using
or
is currently
blocked.
The following error is returned:
Output
Conversion between vectors with mismatched dimensions isn’t allowed and raises a dimension
mismatch error.
The following error is returned:
Output
If a vector is declared without a dimension count, assigning a value to it raises an error.
This example works:
Out-of-range values
Mixed base types in functions
However, if the dimension count isn’t specified, it raises an error:
Out-of-range values for
(for example, above 65504.0) raise an error during
assignment.
The following error is returned:
Output
Mixed base types in functions like
aren’t supported and raises a type error.
The following error is returned:
Output
varchar(max)
Unsupported architecture
SIMD overflow
Tools support
Binary transport support for float16 vectors not yet available
isn’t supported on Arm64 architectures, and using it raises a runtime error
The following error is returned:
Output
Single instruction, multiple data (SIMD)-based operations, such as AVX2 or SSE4.2, might
produce overflow errors if values exceed representable ranges.
The behavior depends on the
setting:
results in an error
results in
Management Studio (SSMS) doesn’t currently distinguish between
and
in the UI. Use
to confirm the actual base type used in a schema.
vectors are currently transmitted as
(JSON array) over TDS. Binary
transport support for
isn’t yet available in drivers like ODBC, JDBC, and.NET.
vector
Vector search and vector indexes in the SQL Database Engine
Intelligent applications
７
Note
All limitations that apply to the default
type (with
) also apply to.