Train a classifier on requests with verified category labels and save the trained model. In operation, keep that model fixed and return categories for new requests.
No process retrains the model for each new input. Compare classifying one request at a time with processing many requests together overnight.
Choose TWO appropriate explanations of this flow.
Distinguish obtaining a model from data during training from applying a saved model during inference. New inputs, many inputs, or added reference documents alone do not mean a fixed model has been retrained.
Detailed explanation
The scenario fixes the model during inference. Supplying an input and obtaining output does not itself perform retraining or a model update.
Batch inference can apply a trained model to many inputs. Distinguish the purpose and whether a model is being learned rather than relying on record count or time of day.
Adding inference-time information differs from updating learned parameters. Using reference information does not by itself establish that retraining occurred.
This stage obtains the model from data. It differs from applying a saved model to new inputs.
This uses a model that has already learned. Processing many records does not by itself turn it into training.
Check official documentation
Compare the following points with the official documentation.
Review SageMaker AI training, inference, and batch-inference documentation and the AWS RAG explanation. Diagram invented training data, a saved classifier, new inputs, outputs, and whether an update process exists. Do not train models or call APIs.
Official references
- Official documentation 1
- Official documentation 2
- Official documentation 3
- Official documentation 4
- Official documentation 5
What to confirm
Distinguish training, single-request inference, and batch inference. Do not infer retraining of fixed model weights merely from new inputs or reference documents.
Key points
- Learning versus use
- Fixed model
- Batch inference
- Input versus weight updates
Notes
- Environment: Current AIF-C01 task and related AWS primary-source review using invented examples; no model training or inference is performed.
- Review SageMaker AI training, inference, and batch-inference documentation and the AWS RAG explanation. Diagram invented training data, a saved classifier, new inputs, outputs, and whether an update process exists. Do not train models or call APIs.
- Distinguish training, single-request inference, and batch inference. Do not infer retraining of fixed model weights merely from new inputs or reference documents.
Foundation review
Separate obtaining a model from using it
Training obtains the model; inference applies it. Some operational systems may include retraining, but retain this scenario condition that the model is fixed during inference.
Distinguish volume from purpose
Processing many inputs together using a saved model is still batch inference. Real-time or overnight execution does not decide whether the role is learning or applying a model.
Separate input information from weight updates
Reference documents can supply new information at inference time without retraining. Updating reference material or input context differs from updating learned model parameters.