Key terminology

Foundation model (FM) – An AI model with a large number of parameters and trained on a massive amount of diverse data. A foundation model can generate a variety of responses for a wide range of use cases. Foundation models can generate text or image, and can also convert input into embeddings. Before you can use an Amazon Bedrock foundation model, you must request access. For more information about foundation models, see Supported foundation models in Amazon Bedrock.

Base model – A foundation model that is packaged by a provider and ready to use. Amazon Bedrock offers a variety of industry-leading foundation models from leading providers. For more information, see Supported foundation models in Amazon Bedrock.

Model inference – The process of a foundation model generating an output (response) from a given input (prompt). For more information, see Submit prompts and generate responses with model inference.

Prompt – An input provided to a model to guide it to generate an appropriate response or output for the input. For example, a text prompt can consist of a single line for the model to respond to, or it can detail instructions or a task for the model to perform. The prompt can contain the context of the task, examples of outputs, or text for a model to use in its response. Prompts can be used to carry out tasks such as classification, question answering, code generation, creative writing, and more. For more information, see Prompt engineering concepts.

Token – A sequence of characters that a model can interpret or predict as a single unit of meaning. For example, with text models, a token could correspond not just to a word, but also to a part of a word with grammatical meaning (such as "-ed"), a punctuation mark (such as "?"), or a common phrase (such as "a lot").

Model parameters – Values that define a model and its behavior in interpreting input and generating responses. Model parameters are controlled and updated by providers. You can also update model parameters to create a new model through the process of model customization.

Inference parameters – Values that can be adjusted during model inference to influence a response. Inference parameters can affect how varied responses are and can also limit the length of a response or the occurrence of specified sequences. For more information and definitions of specific inference parameters, see Influence response generation with inference parameters.

Playground – A user-friendly graphical interface in the AWS Management Console in which you can experiment with running model inference to familiarize yourself with Amazon Bedrock. Use the playground to test out the effects of different models, configurations, and inference parameters on the responses generated for different prompts that you enter. For more information, see Generate responses in a visual interface using playgrounds.

Embedding – The process of condensing information by transforming input into a vector of numerical values, known as the embeddings, in order to compare the similarity between different objects by using a shared numerical representation. For example, sentences can be compared to determine the similarity in meaning, images can be compared to determine visual similarity, or text and image can be compared to see if they're relevant to each other. You can also combine text and image inputs into an averaged embeddings vector if it's relevant to your use case. For more information, see Submit prompts and generate responses with model inference and Retrieve data and generate AI responses with knowledge bases.

Orchestration – The process of coordinating between foundation models and enterprise data and applications in order to carry out a task. For more information, see Automate tasks in your application using conversational agents.

Agent – An application that carry out orchestrations through cyclically interpreting inputs and producing outputs by using a foundation model. An agent can be used to carry out customer requests. For more information, see Automate tasks in your application using conversational agents.

Retrieval augmented generation (RAG) – The process of querying and retrieving information from a data source in order to augment a generated response to a prompt. For more information, see Retrieve data and generate AI responses with knowledge bases.

Model customization – The process of using training data to adjust the model parameter values in a base model in order to create a custom model. Examples of model customization include Fine-tuning, which uses labeled data (inputs and corresponding outputs), and Continued Pre-training, which uses unlabeled data (inputs only) to adjust model parameters. For more information about model customization techniques available in Amazon Bedrock, see Customize your model to improve its performance for your use case.

Hyperparameters – Values that can be adjusted for model customization to control the training process and, consequently, the output custom model. For more information and definitions of specific hyperparameters, see Custom model hyperparameters.

Model evaluation – The process of evaluating and comparing model outputs in order to determine the model that is best suited for a use case. For more information, see Choose the best performing model using Amazon Bedrock evaluations.

Provisioned Throughput – A level of throughput that you purchase for a base or custom model in order to increase the amount and/or rate of tokens processed during model inference. When you purchase Provisioned Throughput for a model, a provisioned model is created that can be used to carry out model inference. For more information, see Increase model invocation capacity with Provisioned Throughput in Amazon Bedrock.