Installation

mlbundle requires Python 3.14 or newer. Its only core dependency is Pydantic v2; every model library is an optional extra that is imported lazily, only when a bundle needs it.

These pages document mlbundle 0.2.1.

python -m pip install "mlbundle[sklearn,schema]==0.2.1"

Replace sklearn with the extra for your model library; the extras are listed below.

From a local checkout

To work from source, or before 0.2.1 is available from your package index, install a checkout of the repository:

git clone https://github.com/UlloaSP/mlbundle.git
cd mlbundle
python -m pip install ".[sklearn,schema]"

Inside a uv project, add the checkout as a path dependency instead (adjust the path to where you cloned it):

uv add ../mlbundle --extra sklearn --extra schema

Choose the extras

Extra	Installs	Needed for
sklearn	scikit-learn>=1.6, skops>=0.16.0	Exporting and loading scikit-learn models and pipeline prefixes.
xgboost	xgboost>=3.4.1, scikit-learn>=1.2	XGBClassifier and XGBRegressor.
catboost	catboost>=1.2.10	CatBoost classifiers and regressors. CatBoost brings matplotlib and plotly.
onnx	onnx>=1.23.0, onnxruntime>=1.30.0	Existing ONNX files.
schema	mlschema>=0.3.0,<0.4 on Python 3.14 only	Every export. Also installs pandas.
all	All of the above	Development and mixed workloads.

Pick by what each machine does:

• Training machine: the model extra plus schema, for example mlbundle[xgboost,schema]. Export always infers a schema, even without training data.
• Serving or verifying machine: only the matching model extra, for example mlbundle[xgboost]. Loading validates the schema without pandas or mlschema.
• Metadata only: no extra. inspect and mlbundle inspect never import a model library.

Composite pipelines need both parts: [sklearn,xgboost] or [sklearn,catboost].

Schema inference is Python 3.14 only

The schema extra is limited to Python 3.14 because mlschema supports only that version. Export therefore runs on 3.14. Inspecting and loading validate the schema without mlschema.

Verify the installation

from importlib.metadata import version

import mlbundle

print(version("mlbundle"))  # 0.2.1
print(mlbundle.FORMAT_NAME, mlbundle.FORMAT_VERSION)  # mlbundle 1

The console script is installed with the package:

mlbundle --help

Troubleshooting

Symptom	Cause	Resolution
MissingOptionalDependencyError naming schema	Export runs without pandas or mlschema.	Install mlbundle[schema] on Python 3.14.
MissingOptionalDependencyError naming a library	Load or verify needs the model’s extra.	Install the extra named in the message.
mlbundle: missing optional dependency; install mlbundle[xgboost].	Same, from the CLI.	Install that extra; inspect works without it.
EnvironmentMismatchError on load	A skops load failed and the bundle was written with a different scikit-learn version.	Match the training scikit-learn version.
No matching distribution found	Python is older than 3.14.	Check python --version.