Foundational Models
Learn about the foundational models in Amazon Bedrock
We'll cover the following
A foundation model (FM) is a generative AI model that has been trained on various parameters and unlabeled data. It can generate an output from a single or multiple input prompts that can be human-language instructions.
Bedrock offers various FMs. After experimenting with different options, we can select the most suitable one based on the use case requirements and tailor it according to our data. An FM can be customized using several techniques, such as RAG, fine-tuning the model, and agents.
How to choose a foundation model
Amazon Bedrock has several model options from multiple providers which may vary based on the region we use the service from. Moreover, the choice of the model would also differ based on the input and output modalities in use. For example, if the input modality is in the form of text and the output is required as an image, then we would need to select a foundation model that processes prompts in text and generates images for the user. Some other factors that could also be considered while selecting a foundation model are inference parameters, streaming support, and model hyperparameters.
Amazon Bedrock does not provide permission to access a model by default. We need to first request access using the Amazon Bedrock console.
Some of the options that this service provides are the following, along with a few examples of their variants based on the modality requirements:
Get hands-on with 1400+ tech skills courses.