Get Started with JFrog ML

Deploy your first model on JFrog ML in minutes!

Building Your First Model

It's time to build your first ML model on JFrog ML.🚀

A model build is a trained, serialized and tested model instance, bundled with necessary dependencies that can be later deployed to production.

JFrog ML builds models on a scalable infrastructure that allows you to customize resources, whether using a pre-trained model or running live training of any size.

Creating a Model

To begin, you need to create a new model and a new project on JFrog ML. Projects enable you to group and organize your models in a single location.

Creating models and projects can be done either through the user interface or by using the FrogML CLI.

This example tutorial, shows how to create a model using the FrogML CLI.

Viewing Build Logs

To monitor the progress of your build on JFrog ML, view logs via the CLI or UI.

Logs provide real-time updates on the process, including any errors or warnings or debug information.

CLI: Run frogml models builds logs -b <build-id> --follow and replace build-id with your build id.

App: To view build logs via the UI, follow the links provided in the terminal when building a model.

frogml models build ./credit_risk --model-id “credit_risk” --name "credit_risk_build_v1"
✅ Fetching model code (0:00:00.30)
✅ Registering frogml build - 100% (0:00:06.01)

Build ID ed761a55-72ff-4136-9484-f6a9d667e000 triggered remotely

########### Follow build logs in the CLI
frogml models builds logs -b ed761a55-72ff-4136-9484-f6a9d667e000 --follow

########### Follow build logs in the platform
https://mydemo.jfrog.io/ui/ml/models/credit_risk/build/ed761a55-72ff-4136-9484-f6a9d667e000

An example of the build logs page on FrogML

Deploying Your Model

After a successful build, our model can be deployed as a real-time inference endpoint on JFrog ML, ready to handle predictions. Copy the build ID from the previous build step and replace it with YOUR_BUILD_ID

frogml models deploy realtime --model-id credit_risk --build-id {YOUR_BUILD_ID}

After running the deployment command, you can expect to see the following output:

╒═══════════════╤══════════════════════════════════════╕
│ Environment   │ jfrog_demo                            │
├───────────────┼──────────────────────────────────────┤
│ Model ID      │ credit_risk                          │
├───────────────┼──────────────────────────────────────┤
│ Build ID      │ f42af8a7-2942-459f-b768-a981b7098cb7 │
├───────────────┼──────────────────────────────────────┤
│ Deployment ID │ e2d9a66f-26be-4201-b1ea-f1ccf312d5d4 │
╘═══════════════╧══════════════════════════════════════╛

Deployment initiated successfully, Use --sync to wait for deployment to be ready.

Testing Your Model

After a successful model deployment, you can test your live inference endpoint.

The Frogml Python SDK includes a real-time client module which you have to separately install:

pip install frogml-inference

You can use it to run inference and predictions using your deployed real-time model:

from frogml_inference.realtime_client import RealTimeClient

FROGML_MODEL_ID = 'credit_risk'

if __name__ == '__main__':
  feature_vector = [
    {
      "Purpose": "male",
      "Age_cat": "male",
      "Housing": "male",
      "Age": 3,
      "UserId": "male",
      "Job": 2,
      "Saving accounts": "male",
      "Checking account": "male",
      "Duration": 4,
      "Sex": "male",
      "Credit amount": 54.2
    }]
  
  client = RealTimeClient(model_id=FROGML_MODEL_ID)
  response = client.predict(feature_vector)
  print(response)

Once you begin making predictions using the deployed model, you will be able to view relevant metrics in the Health dashboard on the Model Overview tab.

Querying Model Predictions

Querying model predictions is an essential step in the machine learning development process. With JFrog ML, it's easy to query your model's predictions and view relevant metrics.

^{The Analytics tab on the model page}