gitlab-mlops 0.3.0

No project description provided

GitLab MLOps Python Client

A Python client for integrating GitLab's MLOps features, specifically designed to leverage GitLab's experiment tracking and model registry capabilities. This client simplifies the management of machine learning workflows by combining MLflow's flexibility with GitLab's powerful MLOps tools.

Features

• GitLab Experiment Tracking: Easily track machine learning experiments within GitLab.
• Model Registry Integration: Register and manage models in GitLab's model registry.
• Experiment Management: Create and manage experiments directly from the client.
• Run Tracking: Initiate and monitor training runs with ease.
• Model Lifecycle Management: Promote runs to model versions effortlessly.

Feedback

The MLOPs Python client is currently in Beta. We welcome feedback and feature requests either in this repository's issue tracker or on our feedback issue

Installation

Install the client using pip:

pip install gitlab-mlops

Alternatively, to install the latest version from the source, use Poetry:

# Clone the repository
git clone https://gitlab.com/gitlab-org/modelops/mlops/gitlab-mlops.git
cd gitlab-mlops

# Install dependencies
poetry install

Usage

Configuration

Configure the client using environment variables or initialization arguments.

Using Environment Variables:

export MLFLOW_TRACKING_URI="https://gitlab.com/api/v4/projects/<project_id>/mlflow"
export MLFLOW_TRACKING_TOKEN="your_access_token"

from gitlab_mlops import Client

client = Client()

Using Initialization Arguments:

from gitlab_mlops import Client

client = Client(
    tracking_uri="https://gitlab.com/api/v4/projects/<project_id>/mlflow",
    gitlab_token="your_access_token"
)

Create a New Experiment

from gitlab_mlops import Client

client = Client()
experiment = client.create_experiment("Experiment Name")

Create a New Run

from gitlab_mlops import Client

client = Client()
experiment = client.get_experiment(name="Experiment Name")
run = experiment.create_run()

Log Parameters, Metrics, and Model artifacts

from gitlab_mlops import Client

client = Client()

run = client.get_run(run_id="19e547d3-8de7-42c3-9575-fae1142dec69")

run.log_param(key="param_name", value="value")
run.log_metric(key="metric_name", value=0.95)
run.log_artifact(
    local_path="<path to the model file or directory>",
    artifact_path="<optional relative path to log the artifact(s) at>"
)

Create a New Model

from gitlab_mlops import Client

client = Client()
model = client.create_model(name="Model Name", description="Model Description")

Create a New Model Version

from gitlab_mlops import Client

client = Client()

model = client.get_model(name="Model Name")

model_version = model.create_version(description="New version", version="1.0.0")

model_version.log_param(key="param_name", value="value")
model_version.log_text("How to run this model [...]", "README.md")

Importing existing models

To import existing models from various external sources, you can use the export_to_gitlab method. This method allows you to import models from the following sources:

• Google Cloud Platform - Vertex AI

Importing from Google Cloud Platform - Vertex AI

from gitlab_mlops import Client
from gitlab_mlops.providers.gcp.vertex import VertexAIClient

client = Client()
vertex = VertexAIClient(
    project_id="your_project_id",
    location="us-central1",
    client=client
)

resource_name = "projects/your_project_id/locations/us-central1/models/your_model_id"

vertex.export_model_to_gitlab(
    model_id=resource_name
)