Applications of Foundation Models
Model Selection and Cost
Model selection balances quality, cost, latency, throughput, modality, context length, and operational needs. Larger models are not always the best answer when a smaller or provisioned option satisfies the task.
Concepts
- Smaller or lower-latency models may be better for high-volume simple tasks.
- Larger models may improve quality on complex reasoning or generation tasks but usually cost more and add latency.
- Provisioned Throughput can support predictable Bedrock throughput needs; on-demand invocation fits variable or experimental use.
Exam tips
- Choose smaller or lower-latency models for high-volume simple tasks when quality is sufficient.
- Provisioned Throughput helps when predictable Bedrock capacity is required.
- On-demand invocation fits variable, early-stage, or experimental workloads.