ML Systems Infrastructure
Building robust ML systems.
Compute
GPUs: training deep learning. TPUs: Google Cloud. Edge: specialized hardware.
Storage
Object storage: S3, GCS. Feature store. Model registry.
Orchestration
Airflow, Prefect, Dagster. Kubeflow for ML pipelines. Kubernetes for scale.
Key Takeaways
- GPU/TPU for training
- Feature stores for consistent features
- Orchestration for pipelines