AIF-C01 Practice Question 75: Serving Externally Tuned Weights
AWS Certified AI Practitioner (AIF-C01) · Test 8 · question 5 of 10
A company fine-tuned an open-weight Llama model on proprietary data using Amazon SageMaker AI, and the resulting weights now sit in Amazon S3. It wants to call that model exactly the way it calls the rest of its Amazon Bedrock models, the same InvokeModel operation, on demand, with no endpoint to keep running. Which Amazon Bedrock capability makes this possible?
Show answer and explanation
Answer:
Custom Model Import exists for this exact case. AWS describes it as importing *Foundation Models that you have customized in other environments, such as Amazon SageMaker AI*, and gives the same example the stem does: *you might have a model that you have created in Amazon SageMaker AI that has proprietary model weights. You can now import that model into Amazon Bedrock and then use Amazon Bedrock features to make inference calls to the model.* Crucially, *you can use a model that you import with on demand throughput* and call it with the standard InvokeModel or InvokeModelWithResponseStream operations (option A).
- B is wrong because Provisioned Throughput buys dedicated, committed capacity for a model that is *already* in Amazon Bedrock. It is a capacity and pricing choice, and it does nothing to get the company's own weights into the service. - C is wrong because running a Bedrock fine-tuning job would produce a *different* model, derived from a Bedrock base model. The company has already done its training and wants to serve those exact weights, not pay to repeat the work on a different base. - D is wrong because a knowledge base supplies retrieved context to a model at request time. It neither hosts nor serves model weights, so the company's customisation would stay stranded in S3.
Reference: Use Custom model import to import a customized open-source model into Amazon Bedrock