Documentation
Open Cloudsmith

Integrations

Integrating MLflow

How to integrate MLflow with Cloudsmith

The MLflow Cloudsmith plugin lets you use Cloudsmith as an artifact store for MLflow runs. Artifacts are stored as Cloudsmith RAW packages, with tagging that preserves MLflow experiment and run context.

For full details and examples, see the MLflow Cloudsmith Plugin repository.

Using Cloudsmith as the MLflow artifact store

Install the MLflow Cloudsmith plugin via pip:

bash
pip install mlflow-cloudsmith

Configure MLflow to use a Cloudsmith repository as the artifact destination by setting the MLflow artifact URI and your Cloudsmith API key:

python
import os
import mlflow

os.environ["CLOUDSMITH_API_KEY"] = "<your-api-key>"
os.environ["MLFLOW_ARTIFACT_URI"] = "cloudsmith://<workspace>/<repo>"

with mlflow.start_run():
    mlflow.log_param("learning_rate", 0.01)
    mlflow.log_metric("accuracy", 0.95)
    mlflow.log_artifact("model.pkl")

Cloudsmith will store each logged artifact as a RAW package in the specified repository, tagged for easy filtering (e.g. mlflow, experiment-<id>, run-<id>, path-<artifact_path>).

Direct repository usage

You can also work with Cloudsmith artifact storage directly using the plugin's repository class:

python
from plugin.cloudsmith_repository import CloudsmithArtifactRepository

repo = CloudsmithArtifactRepository("cloudsmith://<workspace>/<repo>")
repo.log_artifact("model.pkl", "models/production")

for info in repo.list_artifacts("models"):
    print(info.path, info.file_size, info.is_dir)

URI format

Cloudsmith URIs for MLflow artifacts use the following format:

text
cloudsmith://<workspace>/<repository>[/<path>]

Examples:

  • cloudsmith://my-workspace/ml-artifacts
  • cloudsmith://my-workspace/ml-artifacts/experiments