AttributeError: module 'ml_dtypes' has no attribute 'float4_e2m1fn' with v1.19.0

[nishantkumar1292]

Bug Report

Is the issue related to model conversion?

No. The error is from the onnx package itself.

Describe the bug

Getting the below error:

from onnxmltools.convert import convert_lightgbm as convert_lightgbm_booster
  File "/home/runner/work/credit-model-training/credit-model-training/.venv/lib/python3.13/site-packages/onnxmltools/__init__.py", line 16, in <module>
    from .convert import convert_coreml
  File "/home/runner/work/credit-model-training/credit-model-training/.venv/lib/python3.13/site-packages/onnxmltools/convert/__init__.py", line 3, in <module>
    from .main import convert_coreml
  File "/home/runner/work/credit-model-training/credit-model-training/.venv/lib/python3.13/site-packages/onnxmltools/convert/main.py", line 5, in <module>
    import onnx
  File "/home/runner/work/credit-model-training/credit-model-training/.venv/lib/python3.13/site-packages/onnx/__init__.py", line 131, in <module>
    from onnx import (
    ...<12 lines>...
    )
  File "/home/runner/work/credit-model-training/credit-model-training/.venv/lib/python3.13/site-packages/onnx/compose.py", line 8, in <module>
    from onnx import (
    ...<7 lines>...
    )
  File "/home/runner/work/credit-model-training/credit-model-training/.venv/lib/python3.13/site-packages/onnx/helper.py", line 21, in <module>
    from onnx import _mapping, defs, subbyte
  File "/home/runner/work/credit-model-training/credit-model-training/.venv/lib/python3.13/site-packages/onnx/_mapping.py", line 104, in <module>
    np.dtype(ml_dtypes.float4_e2m1fn),
             ^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: module 'ml_dtypes' has no attribute 'float4_e2m1fn'. Did you mean: 'float8_e4m3fn'?

System information

• OS Platform and Distribution (e.g. Linux Ubuntu 20.04): Ubuntu 24.04.2
• ONNX version (e.g. 1.13): 1.19.0 (latest)
• Python version: 3.13

Reproduction instructions

Version v1.19.0 has started depending on ml-dtypes, and this is where the bug started in the workflow. With the earlier version, it was running fine.

Run these in a fresh env

pip install onnx==1.19.0
pip install onnxmltools==1.14.0

from onnxmltools.convert import convert_lightgbm as convert_lightgbm_booster

### Expected behavior
It should just work..

### Notes
<!-- Any additional information -->

[AntoniosBarotsis]

Same here, downgrading to 1.18 works for me.

[justinchuby]

What is the version of ml_dtypes on your system? It should be higher than 0.5.0: https://github.com/jax-ml/ml_dtypes/releases/tag/v0.5.0

[nishantkumar1292]

What is the version of ml_dtypes on your system? It should be higher than 0.5.0: https://github.com/jax-ml/ml_dtypes/releases/tag/v0.5.0

For me, ml-dtypes==0.4.1. If this should be >0.5.0, why is there no version constraint in the requirements.txt file?

EDIT: Oh..seems you added it here: #7254. What is the plan for merging it?

[justinchuby]

It will have to be in the next release I’m afraid. For now please upgrade ml_dtypes manually. Thanks for reporting the issue!

[lxe]

and ml-dtypes>=0.5.0 depends on numpy{python_full_version >= '3.13'}>=2.1.0, we can conclude that ml-dtypes>=0.5.0 depends on numpy>=2.1.0.
      And because only the following versions of ml-dtypes are available:
          ml-dtypes<=0.5.0
          ml-dtypes==0.5.1
          ml-dtypes==0.5.3
      we can conclude that ml-dtypes>=0.5.0 depends on numpy>=2.1.0.
      And because your project depends on ml-dtypes>=0.5.0 and numpy<2, we can conclude that your project's requirements are unsatisfiable.

We need #7231 as well

[andife]

and ml-dtypes>=0.5.0 depends on numpy{python_full_version >= '3.13'}>=2.1.0, we can conclude that ml-dtypes>=0.5.0 depends on numpy>=2.1.0.
      And because only the following versions of ml-dtypes are available:
          ml-dtypes<=0.5.0
          ml-dtypes==0.5.1
          ml-dtypes==0.5.3
      we can conclude that ml-dtypes>=0.5.0 depends on numpy>=2.1.0.
      And because your project depends on ml-dtypes>=0.5.0 and numpy<2, we can conclude that your project's requirements are unsatisfiable.

We need #7231 as well

I don't really understand why this should be a solution, as the proposed PR increases the maximum required numpy version.

https://github.com/jax-ml/ml_dtypes/blob/main/pyproject.toml

dependencies = [
    # Ensure numpy release supports Python version.
    "numpy>=1.21",
    "numpy>=1.21.2; python_version>='3.10'",
    "numpy>=1.23.3; python_version>='3.11'",
    "numpy>=1.26.0; python_version>='3.12'",
    "numpy>=2.1.0; python_version>='3.13'",
]

Support for python 3.13 was added at NumPy 2.1.0 (https://github.com/numpy/numpy/blob/f579bee88c73db688f6c1b6452d7e5abb8a2ccc9/doc/source/release/2.1.0-notes.rst#L16)

[justinchuby]

your project depends on ml-dtypes>=0.5.0 and numpy<2

I don't understand this line. There isn't numpy<2 specified anywhere for onnx.