Generative AI platform
Amazon Bedrock Custom Model Import
Imports supported custom foundation models into Bedrock for managed inference.
Key points
- Allows supported externally trained or customized models to be brought into Bedrock.
- Provides managed Bedrock inference access for imported models instead of self-hosting them.
- Keeps the model within Bedrock operational controls such as IAM and API-based invocation.
- Targets cases where an organization already has a model it wants to serve through Bedrock.
- Differs from tuning a Bedrock base model because the model artifact is imported.
When to use it
- Choose Custom Model Import when a supported custom foundation model needs managed inference in Bedrock.
- Use it when teams want Bedrock APIs and operations for a model trained outside Bedrock.
- Use it to standardize generative AI serving without rebuilding the model in SageMaker.
Exam tips
- Import means bringing an existing supported model into Bedrock; customization means adapting a Bedrock-supported model.
- For building and training arbitrary ML models end to end, SageMaker is usually the better answer.
- Imported models still need normal security controls such as IAM, encryption, and access governance.
- Do not assume every open model can be imported; AWS supports specific model architectures and formats.