Snowflake ML Python

snowflake-ml-python is the Python library for AI and machine learning on Snowflake. Use it to build and manage features with the Snowflake Feature Store, train with the open-source frameworks you already use, and manage, serve, and monitor models on Snowflake compute.

Train machine learning models in the Snowflake Container Runtime, a managed environment for notebooks and batch workloads on CPU or GPU, with common ML packages already available. Work in Snowflake Notebooks, or send the same workloads to the Container Runtime from your own IDE.

See the Snowflake ML developer guide for the full platform.

APIs

• Model Registry (snowflake.ml.registry). Log and manage versions of models trained in Snowflake or elsewhere, including scikit-learn, XGBoost, LightGBM, PyTorch, TensorFlow, Keras, Hugging Face, and MLflow models, as well as your own Python models. Run inference in a warehouse, or deploy the model to Snowpark Container Services.

• Feature Store (snowflake.ml.feature_store). Define, discover, and maintain features derived from your data. Pipelines refresh from batch and streaming sources, and the same definitions serve both training and online inference.

• ML Jobs (snowflake.ml.jobs). Run training and other Python ML workloads on Container Runtime from VS Code, Jupyter, or a pipeline. Submit a file, a directory, or a function.

• Experiment tracking (snowflake.ml.experiment). Log parameters, metrics, and models while you train, then compare runs and choose what to deploy.

• Datasets (snowflake.ml.dataset). Keep immutable, versioned snapshots of training data, and read them from Snowpark, PyTorch, or TensorFlow.

• ML Observability (snowflake.ml.monitoring). Monitor performance and drift for models in the registry, and explain model predictions.

• ML Lineage (snowflake.ml.lineage). Trace source data through features, datasets, and models.

• Modeling and preprocessing (snowflake.ml.modeling). snowflake.ml.modeling.preprocessing runs transformations such as scaling and encoding in a Snowflake warehouse. The package also includes scikit-learn, XGBoost, and LightGBM estimators.

To load Snowflake tables and stages into pandas, PyTorch, or TensorFlow while training in Container Runtime, use the Data Connector (snowflake.ml.data).

Getting started

If you don't have a Snowflake account yet, you can sign up for a 30-day free trial.

Installation

Python 3.10 through 3.14 are supported.

snowflake-ml-python is preinstalled in Container Runtime. The version in the image depends on the runtime you select. Each runtime's package list is in the Container Runtime release notes.

In warehouse notebooks and worksheets, select snowflake-ml-python from the Packages menu. The package and its dependencies must be allowed by your organization's package policy. See Python APIs for Snowflake ML.

Locally, install from PyPI:

python -m pip install snowflake-ml-python

Some libraries are optional extras rather than part of the base install. For example:

python -m pip install "snowflake-ml-python[scikit-learn]"
python -m pip install "snowflake-ml-python[feature_store]"  # online feature reads
python -m pip install "snowflake-ml-python[all]"

The Snowflake Anaconda channel is the preferred conda source. Packages there are validated with Snowflake ML:

conda create --name snowflake-ml
conda activate snowflake-ml
conda install \
  --override-channels \
  --channel https://repo.anaconda.com/pkgs/snowflake/ \
  snowflake-ml-python

The package is also on conda-forge:

conda install \
  --override-channels \
  --channel https://conda.anaconda.org/conda-forge/ \
  snowflake-ml-python

Conda does not support pip extras. Install optional libraries such as scikit-learn, xgboost, or httpx (online Feature Store reads) as separate conda packages.

Verifying a release

Release artifacts on PyPI are signed with cosign. For a given version, download the sdist or wheel from PyPI and the .pub and .sig files from the matching GitHub release.

VERSION=2.5.0  # use the version you downloaded

cosign verify-blob \
  "snowflake_ml_python-${VERSION}.tar.gz" \
  --key "snowflake-ml-python-${VERSION}.pub" \
  --signature "resources.linux.snowflake_ml_python-${VERSION}.tar.gz.sig"

cosign verify-blob \
  "snowflake_ml_python-${VERSION}-py3-none-any.whl" \
  --key "snowflake-ml-python-${VERSION}.pub" \
  --signature "resources.linux.snowflake_ml_python-${VERSION}-py3-none-any.whl.sig"