onnxruntime 1.24 has implicit ml_dtypes dependency

[sirhcm]

Describe the issue

#25678 introduced a dependency on ml_dtypes, but this dependency is not specified in requirements.txt, and as a result is not present when onnx earlier than 1.19 is installed.

To reproduce

$ python3 -m pip install onnx==1.18 onnxruntime
...
$ python3 -c "import onnxruntime.quantization"
...
ModuleNotFoundError: No module named 'ml_dtypes'

Urgency

No response

Platform

Linux

OS Version

Ubuntu 24.04

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.24.1

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response

[tianleiwu]

Thanks for reporting. We will add such optional dependency to extras_require in setup.py.

[chaya2350]

I re-tested this in a completely clean environment and can confirm it still reproduces on 1.24.2.

Related issue: #27257

Repro (Linux, Python 3.12):

python -m venv /tmp/ort-issue-27257-repro-fresh
source /tmp/ort-issue-27257-repro-fresh/bin/activate
python -m pip install --no-cache-dir onnx==1.18.0 onnxruntime==1.24.2
python -c "import onnxruntime.quantization"

Actual result:

ModuleNotFoundError: No module named 'ml_dtypes'

Workaround verified:

python -m pip install ml_dtypes
python -c "import onnxruntime.quantization"

After installing ml_dtypes, the import succeeds immediately.

So this appears to still be reproducible in 1.24.2 (not only 1.24.1).

[microsoft-github-policy-service]

Applying stale label due to no activity in 30 days. Please react to the issue by assigning it or labeling it as 'contributions welcome', 'documentation', 'feature request', 'regression', or 'no stale'.