AIF-C01 ยท ML lifecycle, MLOps and metrics
Model sources and ways to use a model
Open source pre-trained models versus custom training, and consuming a model through a managed API versus a self-hosted endpoint.
Key points
- Pre-trained (often open source) model: reuse or fine-tune it; cheaper and faster.
- Custom model from scratch: when you need full control and have the data.
- Managed API such as Amazon Bedrock: no servers to run, pay per use.
- SageMaker AI endpoint: host the model yourself; more control over instance, container and scaling, more to operate.
Source: https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html
Practice questions
Which service gives API access to foundation models without you managing any model-serving infrastructure?
- Amazon Bedrock
- A self-managed Amazon EC2 GPU fleet
- A SageMaker AI endpoint you size yourself
- Amazon S3
Show answer
Answer: A. Amazon Bedrock
Amazon Bedrock is a fully managed service exposing foundation models through an API. A SageMaker AI endpoint and EC2 both require you to choose and operate the compute.
Which are advantages of starting from a pre-trained model instead of training from scratch?
- Less training data is typically needed
- Shorter time to a working prototype
- Complete control over every training example
- Guaranteed higher accuracy on any domain
- No licence terms to check
Show answer
Answer: A. Less training data is typically needed, B. Shorter time to a working prototype
Pre-trained models usually need far less data, time and compute. Full control over training data and guaranteed best accuracy on every niche task are not guaranteed benefits.
An open source pre-trained model can always be used commercially without reading its licence.
Show answer
Answer: False
False. Open source models ship under different licences, some with commercial or usage restrictions, so licence terms must be checked.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
