Add MLX tensor adapter with dtype conversion and comprehensive tests

[Copilot]

Implements tensor adapter for MLX arrays following the established TorchTensor pattern.

Implementation

• MlxArray class: Wraps MLX arrays to implement TensorProtocol

  • numpy(): Converts via np.from_dlpack() for efficient zero-copy transfer
  • tobytes(), tofile(): Standard serialization methods
  • __array__(): NumPy protocol support
  • Special handling for bfloat16 dtype

• Dtype conversion: from_mlx_dtype() / to_mlx_dtype()

  • Supports 13 dtypes: bool_, uint8-64, int8-64, float16, float32, bfloat16, complex64
  • Lazy imports to avoid hard MLX dependency

• Test infrastructure: Enhanced skip_if_no() to handle import failures and platform checks

  • Tests automatically skip on Windows where MLX is not supported
  • Improved error handling for missing parent modules

• Platform-specific installation: MLX only installed on non-Windows platforms via mlx; sys_platform != "win32" in requirements-dev.txt

Usage

import mlx.core as mx
import onnx_ir as ir

mlx_array = mx.array([1, 2, 3], dtype=mx.float32)
ir_tensor = ir.tensor_adapters.MlxArray(mlx_array)

# Use in graph as any other tensor
attr = ir.AttrTensor("x", ir_tensor)
proto = ir.serde.serialize_tensor(ir_tensor)

66 new tests covering all dtypes and operations. All existing tests pass. Tests gracefully skip on platforms without MLX support.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Create tensor adaptor for mlx</issue_title>
<issue_description>Create a tensor adapter for mlx arrays in src/onnx_ir/tensor_adapters.py. Implement all the methods and follow existing examples. Create proper tests.</issue_description>

<agent_instructions>You can find mlx documentation at https://ml-explore.github.io/mlx/build/html/usage/quick_start.html</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes Create tensor adaptor for mlx #213

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[codecov]

❌ 4 Tests Failed:

Tests completed	Failed	Passed	Skipped
5921	4	5917	0

View the top 3 failed test(s) by shortest run time

src.onnx_ir.tensor_adapters_test.MlxArrayTest::test_numpy_returns_correct_dtype_11_mlx_core_bfloat16

Stack Traces | 0.001s run time

..../test/lib/python3.10.../site-packages/parameterized/parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
src/onnx_ir/tensor_adapters_test.py:255: in test_numpy_returns_correct_dtype
    self.assertEqual(tensor.numpy().dtype, np_dtype)
src/onnx_ir/tensor_adapters.py:303: in numpy
    return np.asarray(self.raw)
E   RuntimeError: Item size 2 for PEP 3118 buffer format string B does not match the dtype B item size 1.

src.onnx_ir.tensor_adapters_test.MlxArrayTest::test_tobytes_11_mlx_core_bfloat16

Stack Traces | 0.001s run time

..../test/lib/python3.10.../site-packages/parameterized/parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
src/onnx_ir/tensor_adapters_test.py:284: in test_tobytes
    self.assertEqual(tensor.tobytes(), tensor.numpy().tobytes())
src/onnx_ir/tensor_adapters.py:313: in tobytes
    return self.numpy().tobytes()
src/onnx_ir/tensor_adapters.py:303: in numpy
    return np.asarray(self.raw)
E   RuntimeError: Item size 2 for PEP 3118 buffer format string B does not match the dtype B item size 1.

src.onnx_ir.tensor_adapters_test.MlxArrayTest::test_tofile_11_mlx_core_bfloat16

Stack Traces | 0.001s run time

..../test/lib/python3.10.../site-packages/parameterized/parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
src/onnx_ir/tensor_adapters_test.py:329: in test_tofile
    tensor.tofile(temp_file)
src/onnx_ir/tensor_adapters.py:317: in tofile
    return self.numpy().tofile(file)
src/onnx_ir/tensor_adapters.py:303: in numpy
    return np.asarray(self.raw)
E   RuntimeError: Item size 2 for PEP 3118 buffer format string B does not match the dtype B item size 1.

src.onnx_ir.tensor_adapters_test.MlxArrayTest::test_tofile_method_exists_and_works

Stack Traces | 0.001s run time

src/onnx_ir/tensor_adapters_test.py:295: in test_tofile_method_exists_and_works
    tensor.tofile(buffer)
src/onnx_ir/tensor_adapters.py:317: in tofile
    return self.numpy().tofile(file)
E   io.UnsupportedOperation: fileno

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[justinchuby]

@copilot install mlx only when platform is not Windows. Skip the tests on Windows because mlx is not available.

[Copilot]

@copilot install mlx only when platform is not Windows. Skip the tests on Windows because mlx is not available.

MLX is now installed only on non-Windows platforms via mlx; sys_platform != "win32" in requirements-dev.txt. The skip_if_no decorator checks for Windows and skips MLX tests with message "mlx is not available on Windows". Also improved error handling for missing parent modules. (commit 94e1ef0)