vocabulary: name: UC San Diego TritonAI API Vocabulary description: >- Controlled vocabulary for the UC San Diego TritonAI developer API, an OpenAI-compatible LLM gateway powered by LiteLLM. Terms are derived from the live OpenAPI document at https://tritonai-api.ucsd.edu/openapi.json. version: 1.0.0 terms: - term: model definition: Identifier of a language, embedding, image, or audio model exposed by the gateway. tags: [core, request] - term: messages definition: Ordered list of chat messages forming the conversation passed to chat completions. tags: [chat, request] - term: role definition: The author of a message, one of system, user, assistant, tool, function, or developer. tags: [chat, request] - term: content definition: The body of a message, either a plain string or an array of typed content parts (text, image_url, input_audio). tags: [chat, request] - term: chat completion definition: A model-generated response to a chat conversation returned by POST /v1/chat/completions. tags: [chat, response] - term: completion definition: A text continuation generated for a prompt by POST /v1/completions. tags: [completions, response] - term: embedding definition: A numeric vector representation of input text produced by POST /v1/embeddings. tags: [embeddings, response] - term: input definition: The text or list of texts submitted to the embeddings endpoint. tags: [embeddings, request] - term: image generation definition: Creation of images from a text prompt via POST /v1/images/generations. tags: [images] - term: audio speech definition: Text-to-speech synthesis via POST /audio/speech. tags: [audio] - term: audio transcription definition: Speech-to-text transcription via POST /audio/transcriptions. tags: [audio] - term: usage definition: Token accounting (prompt_tokens, completion_tokens, total_tokens) returned with a response. tags: [response, billing] - term: API key definition: Issued credential passed as a Bearer token to authenticate approved UC San Diego users. tags: [auth, security] - term: gateway definition: The centralized LiteLLM proxy through which all TritonAI model traffic is routed. tags: [infrastructure]