Term Definitions

Agent

In the context of LLM, an agent typically refers to a program or system that utilizes the LLM to perform specific tasks or interact with users or other agents....

An API, or Application Programming Interface, is a set of protocols and tools for building and integrating application software. It allows different software applications to communicate with each other, facilitating the exchange of data and functionality. In the context of LLM we may sometimes use a commercial LLM via an...

Architecture

The architecture of a large language model (LLM) consists of a deep learning network built upon the transformer architecture, which utilizes attention mechanisms to weigh the influence of different input words on each output word. This setup allows LLMs to generate coherent and contextually relevant text based on the input...

Embedding

An embedding is a mathematical representation of objects, such as words or text, in a continuous vector space, capturing their semantic relationships for use in machine learning tasks like natural language processing....

Fine-tune

To fine-tune a model such as an LLM means to take an LLM that was already trained and to train it a bit further on specific data....

Foundation model

A foundation model LLM is a versatile AI model pre-trained on vast data, designed to be adaptable for various tasks without needing retraining for each new application....

Hallucination

In the context of LLMs, hallucination refers to a phenomenon where the model generates text that is incorrect, nonsensical, or not real....

Inference

To use a model for inference means to use it for production (in contrast to training or learning). This is the common use of an LLM....

Instruction-tuned model

An instruction-tuned model is a form of AI that has been specifically trained to understand and execute textual instructions with high accuracy. By undergoing instruction tuning, these models can adapt to a wide array of tasks directly from user commands, enhancing their versatility and applicability across various domains without the...

LLM

A large language model (LLM) is a type of artificial intelligence (AI) algorithm that uses deep learning techniques and massively large data sets to understand, summarize, generate and predict new content....

Model parameters

The model parameters of a large language model (LLM) are the weights, biases, and other learned parameters that are optimized during a learning process....

Multimodal model

A multimodal language model or multimodal model is capable of processing and generating text, images, and other modalities simultaneously. It integrates multiple sources of information, such as text and visual data, to enhance understanding and generation of content....

Prompt

In the context of LLM, a prompt is a specific input or instruction provided to the model to generate a desired output, such as text or code....

RAG

Retrieval augmented generation, or RAG, is an architectural approach that can improve the efficacy of large language model (LLM) applications by leveraging custom data....

Train

To train a model such as an LLM means to go through a process of learning the model parameters based on training data....

Training data

The training data of an LLM is the data used to learning the model parameters of the LLM. The data itself is not part of the LLM. Some data can include large bodies of text used for foundation models and other data is part of instruction tuning the model....

Vector database

A vector database is a specialized database designed for storing and efficiently querying vector data (or embeddings), often used for tasks like similarity search and recommendation systems....