What is AI inference?
Inference is running a trained AI model on an input to produce an output. It is the computation that happens when you ask a model a question or generate an image.
Glossary
AI glossary
Definitions, examples and related explanations. Understand the language behind the tools, then follow it into a practical guide.
16 entries
A
C
Checkpoint
A checkpoint is saved model state, usually including weights. In image workflows, the term often refers to the main model file selected by a loader.
ComfyUI
ComfyUI is software for assembling and running AI media workflows through a graph of connected nodes.
Context window
The context window is the amount of token context a model can use for a request, subject to the model and runtime configuration.
D
Diffusion
Diffusion models learn to reverse a noise process and can generate media through repeated refinement steps.
H
Hallucination
In generative AI, hallucination describes fabricated or unsupported content presented as though it were established information.
I
Image-to-video
Image-to-video generates video frames conditioned on one or more images, sometimes with text or other controls.
L
Large language model (LLM)
An LLM is a machine-learning model trained on large amounts of language data to perform tasks such as generating, classifying or transforming text.
LoRA
LoRA, or low-rank adaptation, adapts a model by training comparatively small update matrices while keeping the base weights frozen.
P
Prompt
A prompt is input used to guide a generative model, including instructions and relevant context. Some models also accept images or other media.
Q
Quantisation
Quantisation represents model numbers with lower precision to reduce storage or memory requirements, with possible changes to output quality.
R
Retrieval augmented generation (RAG)
RAG retrieves relevant material from an external source and supplies it to a model as context for generating an answer.
T
Token
A token is a unit of input or output processed by a language model. It can be a word fragment, punctuation or another encoded unit.
Training
Training adjusts model parameters using data and an optimisation process. Fine-tuning continues training from an existing trained model.
V
VRAM
VRAM is memory available to a graphics processor. Local AI uses accelerator memory for model data and working state.
W
Workflow
A workflow is a connected set of processing steps, their settings and inputs. In ComfyUI it is a node graph that can be saved and shared.
No matching entry yet. Try a shorter term or browse another content type.