Skip to main content

Glossary

This glossary provides definitions and explanations for key terms related to artificial intelligence, large language models, and knowledge graphs. Use this as a reference to better understand the concepts discussed throughout our documentation.

Foundation Concepts

Foundation Model

Type of large-scale machine learning model that is pretrained on extensive datasets and can be adapted for a wide variety of tasks across different domains.

LLM (Large Language Model)

Advanced machine learning models capable of generating human-like text by processing large datasets.

Model Parameters

The adjustable components within an LLM that influence its ability to process input data and generate output. They are essentially the "settings" that the model fine-tunes based on the training data it receives.

Model Family

Describes the architecture type of the foundation model.

Alignment

Ensuring that artificial intelligence systems operate in accordance with human intentions, values, and goals.

Learn more about SAP AI Ethics

Text Processing and Tokens

Token

The basic unit of text that the model processes. A token can be a word, part of a word, or even a character.

Learn more about tokens

Tokenization

The method of breaking down a sequence of text into discrete components (tokens), which can be words, subwords, characters, or symbols.

Learn more about tokenization

Context Window

The length of the prompt that is supported by each LLM model, usually related to Model family.

LLM Configuration Parameters

Temperature

What sampling temperature to use, between 0 and 2. Higher values will make the output more random, while lower values will make it more focused and deterministic (meaning it will choose the most probable next token based on its training).

Learn more about temperature (Tutorial:Querying the LLMs)

Frequency Penalty

Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.

Learn more about frequency penalty (Tutorial:Querying the LLMs)

Presence Penalty

Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.

Learn more about presence penalty (Tutorial:Querying the LLMs)

Max Tokens

The maximum number of tokens allowed for the generated answer.

Learn more about max tokens (Tutorial:Querying the LLMs)

Stop Sequence

A determined token that stops response generation when encountered.

Infrastructure and Deployment

Endpoint Configuration

Defining the URL, HTTP methods, and security protocols.

Inference Endpoint

Key concepts such as API management, secure data transmission, load balancing, and session management. It acts as the interface between SAP AI models and client applications, ensuring that all interactions are optimized for performance and security.

Request Handling

Routing, data validation, and response generation.

Scalability and Monitoring

Ensuring the endpoint can handle varying loads and providing insights into its performance.

Advanced AI Techniques

Prompt

A natural language instruction or query given to a generative AI model to elicit a response.

Grounding

A technique to provide additional context (closed source information/ Subject matter expertise / Domain specific knowledge) to Large Language models to improve the quality of the generated output.

Vector Embeddings

Vector embeddings are vectors generated by neural networks to map objects like text or images into a high-dimensional space, preserving semantic similarity.

RAG

RAG (Retrieval-Augmented Generation) is an AI technique that combines information retrieval with text generation to produce more accurate and contextually relevant responses.

CAG

[Content to be added]

VectorRAG

This is a method that extends the traditional Retrieval Augmented Generation model by using vector representations for information retrieval. This technology uses dense vectors to estimate the semantic similarity between a query and potential candidates from a database, allowing you to:

  • More accurately find relevant data based on context, rather than just an exact match of keywords
  • Efficiently process large amounts of unstructured data
  • Integrate with deep learning models to improve search results and answer generation

GraphRAG

Is a structured, hierarchical approach to Retrieval Augmented Generation (RAG), as opposed to naive semantic search approaches using chunks of plain text. The GraphRAG process involves extracting a knowledge graph from raw text, constructing a domain entities hierarchy, creating summaries for these entities, and then using these KG to perform RAG-based tasks.

Learn more about GraphRAG

Knowledge Graph Concepts

LPG (Labeled Property Graphs)

Labeled Property Graphs have been first developed in Sweden in the context of an enterprise content management. The focus was ease of storage and quick traversals.

Learn more about LPG

RDF (Resource Description Framework)

RDF was inspired by the vision of the Semantic Web, in which autonomous agents retrieve and process information on behalf of human users. The original motivation for RDF was to become a means of representing knowledge, i.e., expressing meaning through ontologies to enable agents to perform automated reasoning.

Learn more about RDF (W3C)
Learn more about RDF (Wikipedia)

Ontology

A formal representation of a set of concepts within a domain and the relationships between those concepts.

SPARQL

SPARQL Protocol and RDF Query Language, is a powerful tool designed for querying and manipulating RDF (Resource Description Framework) data. It allows users to perform complex queries over diverse data sources that are formatted in RDF, supporting operations like querying required and optional graph patterns, along with their conjunctions and disjunctions.

Learn more about SPARQL

Graph Workspace

A user interface in SAP HANA Graph that allows for the visualization and manipulation of graph data.