Added onnx 1.12.0 - #8753
Conversation
| # scikit-learn 1.9.0 | ||
| # torch 2.13.0 | ||
| # onnxruntime 1.26.0 | ||
| # ml_dtypes 0.5.4 |
There was a problem hiding this comment.
Need to built with
ml-dtypes 0.6.0
xgboost-cpu 3.4.0
pyarrow 25.0.0
lightgbm 4.7.0
if the packages are not present in the devfirst index. Built it within the script. You can use the pr's already raised for the updated versions.
| # --------------------------------------------------------------------------- | ||
| export PKG_CONFIG_PATH="/usr/lib64/pkgconfig:/usr/share/pkgconfig:${PKG_CONFIG_PATH:-}" | ||
|
|
||
| git clone https://github.com/numpy/numpy numpy_src |
There was a problem hiding this comment.
No need to build numpy from source and then install as a wheel. Just include pip install numpy=2.0.5 it will fetch the tar and then build and install the package.
| "scikit-learn==1.9.0" | ||
|
|
||
| # torch 2.13.0 + onnxscript (required by torch.onnx.export) — IBM wheels | ||
| python3.14 -m pip install \ |
There was a problem hiding this comment.
Instead if installing required dependencies one by one. Install the required dependent packages from devfirst url in a single line of code.
The old buildscript is of 79 lines and this one is 379.
Don't use Bob blindly lets make the buildscript crisp and perfect.
| cd "${CURRENT_DIR}" | ||
|
|
||
| # protobuf Python runtime (py3-none-any; used by the installed onnx package) | ||
| python3.14 -m pip install "protobuf>=4.25.1" |
There was a problem hiding this comment.
RT26.1 requires protobuf =6.33.6
3 participants
Checklist
set -eoption enabled and observe success ?