MLOps is a collaborative approach that unifies data engineering, ML deployment, and DevOps. While DevOps focuses on the software development life cycle (SDLC), MLOps tailors these best practices to the Machine Learning lifecycle.

MLOps Core Tenets

1. Collaborative Practices: Emphasizes integration among data scientists, machine learning engineers, and IT operations.

2. Reproducibility: Consistently captures all data, code, and models associated with each ML iteration.

3. Continuous Integration & Continuous Deployment (CI/CD): Automates testing, building, and deploying of ML models.

4. Monitoring & Governance: Ensures deployed models are both accurate and ethical, requiring regular performance monitoring and compliance checks.

5. Scalability: Designed to sustain the increasing demand for ML deployment across an organization or beyond.

6. Version Control: Tracks all steps in the ML pipeline, including data versions, model versions, and experimentation details.

7. Security: Adheres to industry security standards, ensuring sensitive data is protected.

8. Resource Management: Handles computational resources efficiently, considering factors such as GPU usage and data storage.

Key Components in MLOps

1. Data Versioning: Tracks data changes over time, crucial for model reproducibility.
2. Feature Store: A central repository for machine learning features, facilitating feature sharing and reuse.
3. Model Registry: Manages model versions, associated metadata, and deployment details.
4. Experiment Tracking: Records experiments, including code, data, and hyperparameters, allowing teams to reproduce and compare results.
5. Deployment Strategies: Considers whether to deploy models in batch or real-time mode, and the environment, such as cloud or on-premises.

Processes in MLOps

1. Model Development: The iterative process of training and evaluating machine learning models.
2. Model Deployment & Monitoring: The staged deployment of models into production systems, followed by continuous monitoring.
3. Feedback Loops: The process of collecting real-world data on model predictions, assessing model performance, and using this feedback to improve model quality.
4. Model Retraining: The automated process of retraining models periodically using the latest data.

Tools & Frameworks in MLOps

• Version Control Systems: Git, Mercurial
• Continuous Integration / Continuous Deployment: Jenkins, GitLab CI/CD, Travis CI
• Containerization & Orchestration: Docker, Kubernetes
• Data Versioning: DVC, Pachyderm
• Feature Store: Hopsworks, Tecton
• Model Registry: MLflow, DVC, Seldon
• Experiment Tracking: MLflow, Neptune, Weights & Biases

Key Differences between DevOps and MLOps

1. Data-Centricity: MLOps puts data at the core of the ML lifecycle, focusing on data versioning, feature engineering, and data quality.

2. Dependency Management: While both involve managing dependencies, the nature of dependencies is different. DevOps focuses on code dependencies, while MLOps looks at data and model dependencies.

3. Testing Strategies: MLOps requires specialized model evaluation and testing methods, including methods like back-testing for certain ML applications.

4. Deployment Granularity: DevOps typically operates on a code-level granularity, whereas MLOps may involve feature-level, model-level, or even ensemble-level deployments.