Implementation:Mlflow Mlflow Log Artifact

Knowledge Sources	MLflow MLflow API
Domains	ML_Ops, Experiment_Tracking
Last Updated	2026-02-13 20:00 GMT

Overview

Concrete tool for persisting local files, directories, and trained models as artifacts within an MLflow run, provided by the MLflow library.

Description

mlflow.log_artifact uploads a local file or directory to the artifact store associated with the currently active run. It is the general-purpose function for persisting any output file -- plots, data exports, configuration snapshots, or custom serialization formats. For trained models specifically, mlflow.sklearn.log_model (and analogous functions in other flavor modules) provides a higher-level interface that serializes the model, records its dependencies and input/output signature, and optionally registers the model in the MLflow Model Registry.

Together, these two functions cover the full spectrum of artifact persistence needs: log_artifact for arbitrary files and log_model (per-flavor) for self-describing, deployment-ready model packages.

Usage

Use log_artifact for non-model files such as plots, CSV exports, configuration YAML files, or custom binary formats. Use mlflow.sklearn.log_model (or the equivalent for your framework) when logging a trained model that should be loadable via mlflow.sklearn.load_model or servable via mlflow models serve. Both functions require an active run context. Provide an artifact_path to organize artifacts into subdirectories within the run's artifact store.

Code Reference

Source Location

Repository: mlflow
File (log_artifact): mlflow/tracking/fluent.py
Lines (log_artifact): L1500-1533
File (sklearn.log_model): mlflow/sklearn/__init__.py
Lines (sklearn.log_model): L379-501

Signature

# General artifact logging
def log_artifact(
    local_path: str,
    artifact_path: str | None = None,
    run_id: str | None = None,
) -> None: ...

# scikit-learn model logging (flavor-specific wrapper)
def log_model(
    sk_model,
    artifact_path: str | None = None,
    conda_env=None,
    code_paths=None,
    serialization_format=SERIALIZATION_FORMAT_CLOUDPICKLE,
    registered_model_name=None,
    signature: ModelSignature = None,
    input_example: ModelInputExample = None,
    await_registration_for=DEFAULT_AWAIT_MAX_SLEEP_SECONDS,
    pip_requirements=None,
    extra_pip_requirements=None,
    pyfunc_predict_fn="predict",
    metadata=None,
    extra_files=None,
    params: dict[str, Any] | None = None,
    tags: dict[str, Any] | None = None,
    model_type: str | None = None,
    step: int = 0,
    model_id: str | None = None,
    name: str | None = None,
    skops_trusted_types: list[str] | None = None,
    **kwargs,
) -> ModelInfo: ...

Import

import mlflow
import mlflow.sklearn  # for sklearn.log_model

I/O Contract

Inputs (log_artifact)

Name	Type	Required	Description
local_path	str	Yes	Path to the local file or directory to upload as an artifact.
artifact_path	str or None	No	Subdirectory within the run's artifact URI where the artifact will be placed. If None, the artifact is placed at the root.
run_id	str or None	No	If specified, logs the artifact to the given run instead of the active run.

Inputs (sklearn.log_model)

Name	Type	Required	Description
sk_model	sklearn estimator	Yes	The fitted scikit-learn model object to serialize and log.
name	str or None	No	Name for the logged model within the artifact store. Replaces the deprecated `artifact_path`.
signature	ModelSignature or None	No	Input/output schema describing the model's expected data types and shapes.
registered_model_name	str or None	No	If provided, registers the model under this name in the MLflow Model Registry, creating the registered model if it does not exist.
serialization_format	str	No	Serialization format: `"cloudpickle"` (default), `"pickle"`, or `"skops"`.
input_example	ModelInputExample or None	No	Sample input data used for model signature inference and testing.
pip_requirements	list or None	No	Explicit pip requirements for the model environment.
model_type	str or None	No	Model type classification for registry purposes.

Outputs

Name	Type	Description
(log_artifact return)	None	The file is uploaded to the artifact store. No return value.
(sklearn.log_model return)	ModelInfo	Metadata about the logged model, including `model_uri`, `model_id`, and `registered_model_version` (if registered).

Usage Examples

Basic Usage

import tempfile
from pathlib import Path

import mlflow

# Log a text file as an artifact
with mlflow.start_run():
    features = "rooms, zipcode, median_price, school_rating, transport"
    with tempfile.TemporaryDirectory() as tmp_dir:
        path = Path(tmp_dir, "features.txt")
        path.write_text(features)
        mlflow.log_artifact(path)

Logging a scikit-learn Model

import mlflow
import mlflow.sklearn
from mlflow.models import infer_signature
from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier

with mlflow.start_run():
    iris = load_iris()
    model = DecisionTreeClassifier()
    model.fit(iris.data, iris.target)

    signature = infer_signature(iris.data, model.predict(iris.data))
    model_info = mlflow.sklearn.log_model(
        model,
        name="iris_classifier",
        signature=signature,
    )
    print(f"Model URI: {model_info.model_uri}")

Logging with Model Registration

import mlflow
import mlflow.sklearn
from sklearn.ensemble import RandomForestClassifier

with mlflow.start_run():
    model = RandomForestClassifier(n_estimators=100)
    model.fit(X_train, y_train)

    mlflow.sklearn.log_model(
        model,
        name="rf_model",
        registered_model_name="production-classifier",
    )

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment