A team fine-tuned a Mistral checkpoint with its own tooling outside AWS and wants to serve it on Bedrock through the standard InvokeModel call. Which Bedrock feature handles this?
Show answer and explanation
Answer: D
Custom Model Import lets a team upload model weights from Amazon S3 for supported architectures such as Llama, Mistral, and Flan, then invoke the imported model through the Bedrock runtime APIs. It is designed for checkpoints trained outside Bedrock. Provisioned Throughput reserves capacity for models Bedrock already hosts, Marketplace lists publisher models for subscription, and Model Evaluation scores outputs. None of these deploy external weights.
Why the other options are wrong
- A. Provisioned Throughput reserves capacity for a model that Bedrock already hosts. It does not load new weights from outside AWS.
- B. Marketplace lists models from publishers for subscription. It does not accept an externally fine-tuned checkpoint from the team's own storage.
- C. Model Evaluation runs benchmark or judge-based comparisons on outputs. It does not serve a model for production calls.