CLOVA Studio Glossary

Available in Classic and VPC

Familiarize yourself with these key terms to get the most out of CLOVA Studio.

Completion

The model's expected output (string) in response to a user query within a conversation.

Datasets

Training and validation data used for tuning.

Domain

A category used by the router to classify user requests.

Router

A component that classifies and filters user requests.

LangChain

An open-source framework that helps you build applications easily by connecting language models and tools.

Model

A language model or vision-language model used to generate completions.

Service app

Required to apply CLOVA Studio APIs to real services. Each service app has its own usage limit.

Safety filter

A filter that detects unethical or sensitive content within a specific domain in the router.

Skill

A tool that selects the API for answering user input, generates required parameters, calls the API, and returns a natural-language response based on observations.

Skillset

A group of skills. Skills in the same group belong to the same skillset.

System prompt

Initial instructions that define the role or behavior of the model.

Thinking

A feature that allows the model to analyze queries and generate intermediate reasoning tokens before producing the final answer.

Vision-language model

A model that can understand both natural language and images.

Usage control policy

A policy that limits API request counts and tokens to maintain service availability and stability.

Explorer

A workspace where you can use various APIs needed for AI-powered services, such as chunking, sliding windows, embeddings, and summarization.

Embedding

An API that converts text into numerical vectors for RAG implementations and document indexing.

Content filter

A filter that detects topics that must be exceptionally blocked within a specific domain in the router.

Query

A question or request entered by the user.

Token

A subword unit used for natural language processing.

Tuning

A method for creating a customized model by training on user datasets.

Parameter

A configuration value that affects model output.

Playground

A workspace where you can test and experiment with prompts.

Task(s)

The base unit for tuning.

Probabilistic language model

A model that predicts the next word based on probability. CLOVA Studio uses HyperCLOVA X models.

API key

An authentication key used to verify access when calling CLOVA Studio APIs.

Chat Completions

An API that provides conversational outputs using HyperCLOVA X models.

Context length

The maximum number of input and output tokens the model can process in a single request.

Function calling

A feature that allows the model to call external functions or APIs to retrieve information or perform operations.

RAG (Retrieval Augmented Generation)

A method that reduces hallucinations by retrieving information from a reliable external knowledge base before generating a response.

Instruction dataset

A conversational dataset format used to tune HyperCLOVA X models.

Max tokens

A parameter that sets the maximum number of output tokens for the model.

QPM (Queries per Minute)

The number of queries allowed per minute. Exceeding this limit results in rate limiting.

Structured Outputs

A feature that generates structured data based on a user-defined JSON Schema.

Temperature

A value that adjusts the diversity of generated text by changing the weighting of the probability distribution.

Top K

A criterion for selecting the next token from the K most probable candidates.

Top P

A criterion that selects tokens within a specified cumulative probability threshold and removes tokens outside that range.

TPM (Tokens per Minute)

The number of tokens allowed per minute. Exceeding this limit results in rate limiting.