• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Changelog
2
3<!--
4
5Changelog follow the https://keepachangelog.com/ standard (at least the headers)
6
7This allow to:
8
9* auto-parsing release notes during the automated releases from github-action:
10  https://github.com/marketplace/actions/pypi-github-auto-release
11* Have clickable headers in the rendered markdown
12
13To release a new version (e.g. from `1.0.0` -> `2.0.0`):
14
15* Create a new `# [2.0.0] - YYYY-MM-DD` header and add the current
16  `[Unreleased]` notes.
17* At the end of the file:
18  * Define the new link url:
19  `[2.0.0]: https://github.com/jax-ml/ml_dtypes/compare/v1.0.0...v2.0.0`
20  * Update the `[Unreleased]` url: `v1.0.0...HEAD` -> `v2.0.0...HEAD`
21
22-->
23
24## [Unreleased]
25
26## [0.5.2] - 2025-01-06
27
28 * Dropped support for Power wheels again. These turned out to cause problems in
29   our release process. We will consider readding these if NumPy ships Power
30   wheels.
31
32## [0.5.1] - 2025-01-06
33
34 * Fixed sign bit handling for float4 and float6 types.
35 * Wheels now support Python 3.13 free-threading.
36 * Wheels now support the Power architecture.
37
38## [0.5.0] - 2024-09-13
39
40* Added new 8-bit float types following IEEE 754 convention:
41  `ml_dtypes.float8_e4m3`, `ml_dtypes.float8_e3m4`
42* Added the 8-bit floating point type `ml_dtypes.float8_e8m0fnu`, which is the
43  OpenCompute MX scale format.
44* Added new 4-bit and 6-bit float types:
45  `ml_dtypes.float4_e2m1fn`, `ml_dtypes.float6_e2m3fn` and `ml_dtypes.float6_e3m2fn`.
46* Fix outputs of float `divmod` and `floor_divide` when denominator is zero.
47
48## [0.4.1] - 2024-09-13
49
50* Updates build requirements to use NumPy 2.0 release
51
52## [0.4.0] - 2024-04-1
53
54* Updates `ml_dtypes` for compatibility with future NumPy 2.0 release.
55* Wheels are built against NumPy 2.0.0rc1.
56
57## [0.4.0b1] - 2024-03-12
58
59* Updates `ml_dtypes` for compatibility with future NumPy 2.0 release.
60* Wheels for the release candidate are built against NumPy 2.0.0b1.
61
62## [0.3.2] - 2024-01-03
63
64* Fixed spurious invalid value warnings when casting between floating point
65  types on Mac ARM.
66* Remove `pybind11` build requirement
67* Update C++ sources for compatibility with NumPy 2.0
68
69## [0.3.1] - 2023-09-22
70
71* Added support for int4 casting to wider integers such as int8
72* Addes support to cast np.float32 and np.float64 into int4
73
74## [0.3.0] - 2023-09-19
75
76* Dropped support for Python 3.8, following [NEP 29].
77* Added support for Python 3.12.
78* Removed deprecated name `ml_dtypes.float8_e4m3b11`;
79  use `ml_dtypes.float8_e4m3b11fnuz` instead.
80
81## [0.2.0] - 2023-06-06
82
83* New features:
84  * added new 4-bit integer types: `ml_dtypes.int4` and `ml_dtypes.uint4`
85
86* Deprecations:
87  * `ml_dtypes.float8_e4m3b11` has been renamed to `ml_dtypes.float8_e4m3b11fnuz` for more
88    consistency with other dtype names. The former name will still be available until
89    version 0.3.0, but will raise a deprecation warning.
90
91## [0.1.0] - 2023-04-11
92
93* Initial release
94
95[Unreleased]: https://github.com/jax-ml/ml_dtypes/compare/v0.5.1...HEAD
96[0.5.2]: https://github.com/jax-ml/ml_dtypes/compare/v0.5.1....v0.5.2
97[0.5.1]: https://github.com/jax-ml/ml_dtypes/compare/v0.5.0....v0.5.1
98[0.5.0]: https://github.com/jax-ml/ml_dtypes/compare/v0.4.0....v0.5.0
99[0.4.1]: https://github.com/jax-ml/ml_dtypes/compare/v0.4.0....v0.4.1
100[0.4.0]: https://github.com/jax-ml/ml_dtypes/compare/v0.4.0b1....v0.4.0
101[0.4.0b1]: https://github.com/jax-ml/ml_dtypes/compare/v0.3.2...v0.4.0b1
102[0.3.2]: https://github.com/jax-ml/ml_dtypes/compare/v0.3.1...v0.3.2
103[0.3.1]: https://github.com/jax-ml/ml_dtypes/compare/v0.3.0...v0.3.1
104[0.3.0]: https://github.com/jax-ml/ml_dtypes/compare/v0.2.0...v0.3.0
105[0.2.0]: https://github.com/jax-ml/ml_dtypes/compare/v0.1.0...v0.2.0
106[0.1.0]: https://github.com/jax-ml/ml_dtypes/releases/tag/v0.1.0
107[NEP 29]: https://numpy.org/neps/nep-0029-deprecation_policy.html
108