//clika-runtime/io.clika.modelverse/Capabilities

Capabilities

[common]
data class Capabilities(val inputModalities: List<String>, val outputModalities: List<String>, val hasChatTemplate: Boolean, val contextLength: Int, val maxImagesPerRequest: Int, val thinking: ThinkingControl, val defaults: GenerationConfig, val family: String, val modelType: String, val device: String, val voices: List<String> = emptyList(), val languages: List<String> = emptyList(), val needsReference: Boolean = false, val speakKnobs: List<SpeakKnob> = emptyList(), val visionTasks: List<String> = emptyList(), val labels: List<String> = emptyList(), val openVocabulary: Boolean = false, val embeddingDim: Int = 0, val defaultImageWidth: Int = 0, val defaultImageHeight: Int = 0, val defaultFrames: Int = 0, val defaultFps: Double = 0.0, val metricDepth: Boolean = false, val supportsRepetitionKnobs: Boolean = false, val supportsTools: Boolean = false, val toolCallFormat: String = "", val templateRendersTools: Boolean = false)

What a loaded model can take and give, read from the checkpoint's documents and the family's registration when the model opens. A consumer greys out the controls a model cannot serve from this and nothing else.

Constructors

Capabilities

[common] constructor(inputModalities: List<String>, outputModalities: List<String>, hasChatTemplate: Boolean, contextLength: Int, maxImagesPerRequest: Int, thinking: ThinkingControl, defaults: GenerationConfig, family: String, modelType: String, device: String, voices: List<String> = emptyList(), languages: List<String> = emptyList(), needsReference: Boolean = false, speakKnobs: List<SpeakKnob> = emptyList(), visionTasks: List<String> = emptyList(), labels: List<String> = emptyList(), openVocabulary: Boolean = false, embeddingDim: Int = 0, defaultImageWidth: Int = 0, defaultImageHeight: Int = 0, defaultFrames: Int = 0, defaultFps: Double = 0.0, metricDepth: Boolean = false, supportsRepetitionKnobs: Boolean = false, supportsTools: Boolean = false, toolCallFormat: String = "", templateRendersTools: Boolean = false)

Properties

Name	Summary
contextLength	[common] val contextLength: Int The context window the model loaded with, in tokens.
defaultFps	[common] val defaultFps: Double = 0.0
defaultFrames	[common] val defaultFrames: Int = 0 The frame count and the frame rate a video-generation request of zeros takes (a video-generation model; 0 otherwise).
defaultImageHeight	[common] val defaultImageHeight: Int = 0
defaultImageWidth	[common] val defaultImageWidth: Int = 0 The picture size an image-generation request of 0 by 0 takes (an image-generation model; 0 otherwise).
defaults	[common] val defaults: GenerationConfig The sampling defaults the checkpoint ships (generation_config.json); a zero temperature decodes greedily.
device	[common] val device: String The device the weights landed on, as the compute API's name (CPU; an accelerator among several carries its ordinal, CUDA:1).
embeddingDim	[common] val embeddingDim: Int = 0 The width of the vectors an embedding model answers; 0 for every other model.
family	[common] val family: String The model family as the library names it (qwen, llama).
hasChatTemplate	[common] val hasChatTemplate: Boolean True when the chat route renders a template: the checkpoint's own, or its family's.
inputModalities	[common] val inputModalities: List<String> The modalities the model's requests may carry, in the library's own spellings: text, audio, vision, video, bounding_boxes, embedding, depth, segmentation, ranking, keypoints. A model whose inputs carry vision takes images in its turns (ChatMessage.images); maxImagesPerRequest says how many one request may carry.
labels	[common] val labels: List<String> A detector's own label table, in id order; empty for an open-vocabulary detector (its labels are the request's) and for every other model.
languages	[common] val languages: List<String> The language codes a speech model declares; empty when it declares none.
maxImagesPerRequest	[common] val maxImagesPerRequest: Int The most images one request may carry across all of its turns: the model's image slots for a model that takes images, 0 for one that does not. A request past it is refused before any decode.
metricDepth	[common] val metricDepth: Boolean = false True when a depth model's values are meters; false for relative depth (larger is closer).
modelType	[common] val modelType: String The checkpoint's own type (qwen3, llama).
needsReference	[common] val needsReference: Boolean = false True when a text-to-speech model conditions every utterance on a reference voice (TtsModel.speak with a VoiceReference); a request without one is refused and ends in SpeakListener.onError. False for a model that speaks its own voice, and for every model that is not a text-to-speech model.
openVocabulary	[common] val openVocabulary: Boolean = false True for a detector that looks for the phrases a request names (DetectOptions.labels).
outputModalities	[common] val outputModalities: List<String> The modalities a reply carries, in the same spellings (text).
speakKnobs	[common] val speakKnobs: List<SpeakKnob> The synthesis knobs a text-to-speech model reads, each with the range the binding accepts and the value that keeps the family's own default; empty for every other model.
supportsRepetitionKnobs	[common] val supportsRepetitionKnobs: Boolean = false True when the library carries the repetition knobs (GenerationConfig.repetitionPenalty, GenerationConfig.noRepeatNgramSize); false refuses any value but their defaults.
supportsTools	[common] val supportsTools: Boolean = false True when the model takes tool definitions and its replies' calls are read: it has a chat template whose call syntax a known format reads.
templateRendersTools	[common] val templateRendersTools: Boolean = false True when the chat template renders the offered tools itself (a tools block in the prompt).
thinking	[common] val thinking: ThinkingControl The reasoning controls the model admits.
toolCallFormat	[common] val toolCallFormat: String The call format stamped at the load: the first of the family's formats the chat template writes (hermes, mistral, llama3_json, pythonic, qwen_xml), else generic, the fallback reader, a model without a template included; whether the model takes tools is supportsTools.
visionTasks	[common] val visionTasks: List<String> The tasks a vision model serves, as the VisionTask constants spell them (VisionTask.DETECT, VisionTask.DEPTH, VisionTask.OCR, VisionTask.EMBED, VisionTask.CLASSIFY); a VisionModel.detect on a model whose list lacks the task ends in VisionListener.onError. Empty for every other model.
voices	[common] val voices: List<String> The preset voices a text-to-speech model names; empty for a model that clones the reference voice it is given, and for every model that is not a speech model.