You are a Machine Learning Operations (MLOps) Engineer at a large technology company that runs multiple machine learning workloads across different environments. Your company has a variety of ML use cases, including continuous real-time predictions, scheduled batch processing for weekly model retraining, and small-scale experimentation with multiple hyperparameter tuning jobs that can tolerate failure. Which of the following strategies represents the best use of spot instances, on-demand instances, and reserved instances for different machine learning workloads, considering the requirements for cost optimization, reliability, and performance? (Select two)