NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #583 most downloaded on PyPI
ml_dtypes is a stand-alone implementation of several NumPy dtype extensions used in machine learning.
Latest stable
0.6.0
Last release 3 months ago
13 Aug 2026
Ships fairly regularly
a new release about every 4 months
Most releases are documented
notes for 12 of 16 stable releases
Nothing withdrawn
no release was ever pulled
4 years old
19 releases · first in 2023
Drag across the ruler to compare, or tap a release to open it.
Added new 16-bit complex types: ml_dtypes.complex32 (based on float16) and ml_dtypes.bcomplex32 (based on bfloat16) (#351).
ml_dtypes.complex32 (based on float16) and
ml_dtypes.bcomplex32 (based on bfloat16) (#351).ml_dtypes.int1 and ml_dtypes.uint1.__format__ method to custom float, complex, and integer types
(#341). Previously,
formatting custom scalars (e.g. in f-strings) fell back to string formatting,
which could truncate exponents or fail on numeric format specifiers.ml_dtypes.finfo and ml_dtypes.iinfo now allow passing an array object to
their constructor (#350).==, !=) when comparing custom dtypes against
incompatible types like strings or None.scikit-build-core and CMake (#361).Your coding agent can read these notes before it upgrades. Set up the MCP server →
We now register casts from int2 and int4 to all of the custom float types, except float6_e2m3fn and float8_e8m0fnu.
NPY_NEEDS_PYAPI was removed from the dtype flags. This should improve the speed of array operations, but it does mean that values pickled using previo
NPY_NEEDS_PYAPI was removed from the dtype flags. This should improve the speed of array operations, but it does mean that values pickled using previous versions of ml_dtypes are incompatible with the current release and should be regenerated with the current release.Fixed sign bit handling for float4 and float6 types.
Added new 8-bit float types following IEEE 754 convention: ml_dtypes.float8_e4m3, ml_dtypes.float8_e3m4
ml_dtypes.float8_e4m3, ml_dtypes.float8_e3m4ml_dtypes.float8_e8m0fnu, which is the
OpenCompute MX scale format.ml_dtypes.float4_e2m1fn, ml_dtypes.float6_e2m3fn and ml_dtypes.float6_e3m2fn.divmod and floor_divide when denominator is zero.Updates build requirements to use NumPy 2.0 release
Updates ml_dtypes for compatibility with the upcoming NumPy 2.0 release.
ml_dtypes for compatibility with the upcoming NumPy 2.0 release.ml_dtypes for compatibility with future NumPy 2.0 release.Updates ml_dtypes for compatibility with future NumPy 2.0 release.
ml_dtypes for compatibility with future NumPy 2.0 release.Fixed spurious invalid value warnings when casting between floating point types on Mac ARM.
pybind11 build requirementAdded support for int4 casting to wider integers such as int8
int4 casting to wider integers such as int8np.float32 and np.float64 into int4Removed deprecated name ml_dtypes.float8_e4m3b11; use ml_dtypes.float8_e4m3b11fnuz instead.
ml_dtypes.float8_e4m3b11; use ml_dtypes.float8_e4m3b11fnuz instead.The former name will still be available until version 0.3.0, but will raise a deprecation warning.
Version 0.2.0 release
New features:
ml_dtypes.int4 and ml_dtypes.uint4Deprecations:
ml_dtypes.float8_e4m3b11 has been renamed to ml_dtypes.float8_e4m3b11fnuz for more
consistency with other dtype names. The former name will still be available until
version 0.3.0, but will raise a deprecation warning.…[NEP 29]: https://numpy.org/neps/nep-0029-deprecation_policy.html
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Test release for testing auto-publish to PyPI
Test release for testing auto-publish to PyPI
Nothing published for this version
One column per quarter.