Limitations and supported versions

Dataiku makes best effort to ensure that the MLflow import capability is compatible with a wide variety of MLflow models.

However, MLflow imposes extremely few constraints on models, and different MLflow models are allowed to behave in arbitrary non-standard ways and to return completely different kind of data.

It is thus not possible to guarantee unfettered ability to use all features (notably advanced features such as performance evaluation, model comparison or drift analysis) with all types of models.

This page lists known compatibilities and incompatibilities.

Import and scoring

The MLflow model import capability and scoring recipes in “direct output” mode are usually compatible with MLflow models supporting the “pyfunc” variant.

Dataiku does not support R MLflow models nor Spark Mlflow models.

The import of MLflow models in DSS was successfully tested with MLflow versions 2.22.0 and 3.10.1 with Python 3.10. Later versions may also work, but modifications of implementation details of MLflow may cause bugs. Version 3.10.1 with Python 3.10 should generally be preferred.

Warning

MLflow versions prior to 2.0.0 are no longer supported.

The following ML packages are supported in the versions specified in the below table. When using one of these packages, you should create a code environment including:

• dataiku core packages

• scikit-learn==1.0.2

• statsmodels==0.13.5

• numpy<1.27

• the specified version(s) of the ML package(s):

ML package	ML packages versions tested for 2.22.0	ML packages versions tested for 3.10.1.
CatBoost	catboost==1.2.7	catboost==1.2.7
fast.ai 2	fastai==2.7.10	Not supported by MLflow v3
LightGBM	lightgbm>=3.0,<3.1	lightgbm>=4.6,<4.7
ONNX	onnx==1.12.0 onnxruntime==1.12.0	onnx==1.21.0 onnxruntime==1.23.2 onnxscript==0.6.2
PyTorch	torch==1.13.1 torchvision==0.14.1 torchmetrics==0.11.0 pytorch-lightning==1.8.4	torch==2.11.0 torchvision==0.26.0 torchmetrics==1.9.0 pytorch-lightning==2.6.1
TensorFlow 2 / Keras 2	tensorflow==2.13.1 tensorflow-estimator==2.13.0 keras==2.13.1	tensorflow==2.15.1 tensorflow-estimator==2.15.0 keras==2.15.0
XGBoost	xgboost==1.7.6	xgboost==1.7.6

Evaluation

DSS also features the evaluation of regression or classification models on tabular input data. As of April 2026, this feature has been successfully tested in the following circumstances.

Please note that this is not a guarantee that you would necessarily be able to do the same, due to the high variability of models that can be saved (even within a single framework).

In all cases, the prediction_type should be set when importing the model. Please see below supported prediction types for the supported ML packages:

ML package	binary classification	multiclass classification	regression
CatBoost	✓	✓	✓
fast.ai 2 (only for mlflow v2)	✗	✗	✓
LightGBM	✓	✗	✓
ONNX	✓	✓	✓
PyTorch	✓	✓	✓
scikit-learn 1.0.2	✓	✓	✓
TensorFlow 2 / Keras 2	✓	✓	✓
XGBoost	✓	✓	✓