mlops-radar

An MLOps radar

This is an MLOps Radar. An ML project can be in two phases, either (A) exploration or (B) productionisation & iteration.

Exploration phase

Notebooks + compute

  1. Jupyter on local machine
  2. Jupyter on K8s multi-user
  3. Deepnote
  4. Google colab
  5. SageMaker Studio Lab
  6. GitHub Codespaces
  7. Databricks community
  8. Kubeflow Notebooks
  9. Vertex AI Workbench

Exploration tools

  1. Einblick

Experimentation Tracking

  1. Weights & Biases
  2. MLflow
  3. Tensorboard
  4. Comet.ml
  5. Neptune.ai
  6. DVC
  7. SageMaker experiments
  8. Aim

Productionisation & Iteration phase

Pipelining

  1. SageMaker pipelines
  2. AWS Step Functions
  3. Metaflow
  4. Kubeflow pipelines
  5. Valohai
  6. Airflow
  7. TFX
  8. Bodywork
  9. ZenML

Feature Stores

  1. SageMaker Feature Store
  2. Tecton
  3. Databricks Feature Store
  4. Feast

Train/Tune

  1. SageMaker
  2. AzureML
  3. Vertex AI
  4. Vertex AI Vizier
  5. Anyscale

Model Registry

  1. SageMaker
  2. AzureML
  3. Vertex AI

Inference

  1. Kubernetes
  2. TensorFlow serving
  3. TorchServe
  4. SageMaker
  5. AzureML
  6. Seldon Core
  7. BentoML
  8. Anyscale

Observability

  1. Superwise.ai
  2. SageMaker model monitoring
  3. SageMaker clarify
  4. AzureML interpret
  5. AzureML studio
  6. Alibi Detect
  7. Alibi Explain
This site is open source. Improve this page.