Skip to content

MLX compatibility: Constants, Datatypes and Array Attributes #451

New issue
New issue
Closed
Closed
MLX compatibility: Constants, Datatypes and Array Attributes#451

Description

@aaishwarymishra
aaishwarymishra
opened on Jul 18, 2026

Constant

Array API MLX analog status test node result
e mx.e exact array_api_tests/test_constants.py::test_e pass
inf mx.inf exact array_api_tests/test_constants.py::test_inf pass
nan mx.nan exact array_api_tests/test_constants.py::test_nan pass
newaxis mx.newaxis exact array_api_tests/test_constants.py::test_newaxis pass
pi mx.pi exact array_api_tests/test_constants.py::test_pi pass

Activity

  1. aaishwarymishra commented on Jul 18, 2026

    @aaishwarymishra
    aaishwarymishra
    on Jul 18, 2026
    Author

    DataTypes

    Array API MLX analog status test node result
    bool mx.bool exact test_has_names (dtype via _array_module) PASS / defined
    complex128   missing test_has_names missing (stub)
    complex64 mx.complex64 exact test_has_names PASS / defined
    float32 mx.float32 exact test_has_names PASS / defined
    float64 mx.float64 exact test_has_names defined (but unsupported on GPU)
    int16 mx.int16 exact test_has_names PASS / defined
    int32 mx.int32 exact test_has_names PASS / defined
    int64 mx.int64 exact test_has_names PASS / defined
    int8 mx.int8 exact test_has_names PASS / defined
    uint16 mx.uint16 exact test_has_names PASS / defined
    uint32 mx.uint32 exact test_has_names PASS / defined
    uint64 mx.uint64 exact test_has_names PASS / defined
    uint8 mx.uint8 exact test_has_names PASS / defined

    Note: In reporting.py we have

    def to_json_serializable(o):
    if o in dtype_to_name:
        return dtype_to_name[o]

    While running tests all the test failed because this method was being called at each test and the main problem is if the o is of the type None

    MlLX throws an exception for equality like

    if None == mlx.core.float16:
    print("success)

    We either have to modify the method to something like

    def to_json_serializable(o):
    if o is not None and o in dtype_to_name:
        return dtype_to_name[o]

    or try to reach out to mlx team for the fix

  2. aaishwarymishra commented on Jul 18, 2026

    @aaishwarymishra
    aaishwarymishra
    on Jul 18, 2026
    Author

    Array Attributes

    Array API MLX Analog Status Test Node Result Blame / Notes
    T array.T exact array_api_tests/test_array_object.py PASS*
    device missing manual array_api_tests/test_array_object.py FAIL MLX missing device attribute
    dtype array.dtype exact array_api_tests/test_array_object.py PASS*
    mT missing manual array_api_tests/test_array_object.py PASS ml-explore/mlx#4402
    ndim array.ndim exact array_api_tests/test_array_object.py PASS*
    shape array.shape exact array_api_tests/test_array_object.py PASS*
    size array.size exact array_api_tests/test_array_object.py PASS*

    Array Methods and Operators

    Array API MLX Analog Status Test Node Result Blame / Notes
    __abs__ array.__abs__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __add__ array.__add__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __and__ array.__and__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __array_namespace__ array.__array_namespace__ exact array_api_tests/test_array_object.py PASS* presence only
    __bool__ array.__bool__ exact array_api_tests/test_array_object.py PASS
    __complex__ array.__complex__ exact array_api_tests/test_array_object.py PASS
    __dlpack__ array.__dlpack__ exact array_api_tests/test_array_object.py PASS* presence only
    __dlpack_device__ array.__dlpack_device__ exact array_api_tests/test_array_object.py PASS* presence only
    __eq__ array.__eq__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __float__ array.__float__ exact array_api_tests/test_array_object.py NOT RUN float64 skipped
    __floordiv__ array.__floordiv__ exact array_api_tests/test_operators_and_elementwise_functions.py Pass ml-explore/mlx#4450
    __ge__ array.__ge__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __getitem__ array.__getitem__ exact array_api_tests/test_array_object.py FAIL indexing behavior fails
    __gt__ array.__gt__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __index__ array.__index__ exact array_api_tests/test_array_object.py PASS ml-explore/mlx#4388
    __int__ array.__int__ exact array_api_tests/test_array_object.py PASS
    __invert__ array.__invert__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __le__ array.__le__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __lshift__ array.__lshift__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS ml-explore/mlx#4694
    __lt__ array.__lt__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __matmul__ array.__matmul__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __mod__ array.__mod__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __mul__ array.__mul__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __ne__ array.__ne__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __neg__ array.__neg__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __or__ array.__or__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __pos__ array.__pos__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS ml-explore/mlx#4488
    __pow__ array.__pow__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __rshift__ array.__rshift__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS ml-explore/mlx#4694
    __setitem__ array.__setitem__ exact array_api_tests/test_array_object.py PASS
    __sub__ array.__sub__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    __truediv__ array.__truediv__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS ml-explore/mlx#4644
    __xor__ array.__xor__ exact array_api_tests/test_operators_and_elementwise_functions.py PASS
    to_device missing manual array_api_tests/test_array_object.py FAIL MLX is missing to_device()

  3. ev-br commented on Jul 18, 2026

    @ev-br
    ev-br
    on Jul 18, 2026
    Member

    Great. what about functions from the spec?

  4. aaishwarymishra commented on Jul 18, 2026

    @aaishwarymishra
    aaishwarymishra
    on Jul 18, 2026
    Author

    Great. what about functions from the spec?

    Hi, its still work in progress, I am planning on making sub-issues for them and reference them on the tracking issue, I am still working on this issue investigating test failures etc, once done link this to main tracker issue and continue to do same with methods and functions in the future :)

  5. aaishwarymishra commented on Jul 18, 2026

    @aaishwarymishra
    aaishwarymishra
    on Jul 18, 2026
    Author

    2 unique failures in array_api_tests/test_array_object.py :

    1. mx.complex64 throws in TypeError: float() argument must be a string or a real number, not 'complex' when used complex on them.
    import mlx.core as mx
arr = mx.array(1, dtype=mx.complex64)
arr2 = complex(arr)
    1. No boolean indexing causes few tests to fail
    arr2 = arr[[True,False]] # read operations trows error boolean assignment works

    Other failures were because of missing array attributes like no __index__ and no float64 support on the gpu

  6. aaishwarymishra
    changed the title [-]MLX compatibility: constants[/-] [+]MLX compatibility: constants, Datatypes and Array Attributes[/+] on Jul 18, 2026
  7. aaishwarymishra
    changed the title [-]MLX compatibility: constants, Datatypes and Array Attributes[/-] [+]MLX compatibility: Constants, Datatypes and Array Attributes[/+] on Jul 18, 2026

  8. ev-br commented on Aug 14, 2026

    @ev-br
    ev-br
    on Aug 14, 2026
    Member

    Out of array object attributes,

    • __index__, __pos__ and mT seem to be small additions likely to be accepted upstream;
    • indexing could be nice investigate in detail: Boolean indexing is unlikely to land upstream, what else is failing?
    • casting test failure : is it just the lack of float64? I

  9. aaishwarymishra commented on Aug 25, 2026

    @aaishwarymishra
    aaishwarymishra
    on Aug 25, 2026
    Author

    Hi, the .T follows the numpy convention of reversing the axes and I dont think they will be open for restricting it to 2D arrays as it will be back breaking like numpy

  10. aaishwarymishra commented on Oct 5, 2026

    @aaishwarymishra
    aaishwarymishra
    on Oct 5, 2026
    Author

    Hi, I updated the table with new pr's that got merged, I noticed that recently the __true_divide or divide in general started failing, maybe I had missed it before so I marked it as failing and will work on new fixes soon :)

    edit: it fails for the numerically large complex 64 numbers

  11. aaishwarymishra commented on Oct 29, 2026

    @aaishwarymishra
    aaishwarymishra
    on Oct 29, 2026
    Author

    @ev-br do you think this issue can be closed? I have fixed all the fixable items I could.

  12. ev-br commented on Oct 29, 2026

    @ev-br
    ev-br
    on Oct 29, 2026
    Member

    Thanks for the great effort @aaishwarymishra ! Agreed this can be closed as completed. WDYT about other MLX-related issues here, are they still useful/actionable?

  13. ev-br
    closed this as completedon Oct 29, 2026

  14. aaishwarymishra commented on Oct 29, 2026

    @aaishwarymishra
    aaishwarymishra
    on Oct 29, 2026
    Author

    it's up to you, maybe we can move the to MLX themselves what you suggest? I do update the issues regularly after fixing an issue here.

  15. ev-br commented on Oct 30, 2026

    @ev-br
    ev-br
    on Oct 30, 2026
    Member

    If MLX maintainers are open to having open issues in the MLX tracker, then yes that'd be ideal. No idea if they are.
    In this repository, I think six open issues are a bit much; maybe prune them? However, if you keep working on MLX support and some of issues here are useful to you, then great.
    The canonical representation of the MLX support of Array API is I think upstream: https://github.com/ml-explore/mlx/tree/main/python/tests/xp-tests

  16. aaishwarymishra commented on Oct 30, 2026

    @aaishwarymishra
    aaishwarymishra
    on Oct 30, 2026
    Author

    20. If MLX maintainers are open to having open issues in the MLX tracker, then yes that'd be ideal. No idea if they are.
    In this repository, I think six open issues are a bit much; maybe prune them? However, if you keep working on MLX support and some of issues here are useful to you, then great.
    The canonical representation of the MLX support of Array API is I think upstream: https://github.com/ml-explore/mlx/tree/main/python/tests/xp-tests

    Hi, @zcbenz do you think we can port these issues to MLX?

  17. zcbenz commented on Oct 31, 2026

    @zcbenz
    zcbenz
    on Oct 31, 2026

    I'm good having one master issue in MLX. For tracking todo work I think the xfail/skips files usually do a better job than having multiple issues.

  18. aaishwarymishra commented on Nov 3, 2026

    @aaishwarymishra
    aaishwarymishra
    on Nov 3, 2026
    Author

    Sure I will make an PR on MLX :)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions