models

const models: object

Defined in: packages/runntime/src/zoo/models.ts:105

Type Declaration

depthEstimation

readonly depthEstimation: object

depthEstimation.DEPTHART

readonly DEPTHART: object

DepthART relative depth at 448×448, f16 weights.

depthEstimation.DEPTHART.B

readonly B: object = DEPTHART_B

The base size, 24 MB. More detail, slower.

depthEstimation.DEPTHART.B.modelPath

modelPath: string

depthEstimation.DEPTHART.DEFAULT

readonly DEFAULT: object = DEPTHART_S

depthEstimation.DEPTHART.DEFAULT.modelPath

modelPath: string

depthEstimation.DEPTHART.S

readonly S: object = DEPTHART_S

The small size, 13 MB.

depthEstimation.DEPTHART.S.modelPath

modelPath: string

embedding

readonly embedding: object

embedding.ALL_MINILM_L6_V2

readonly ALL_MINILM_L6_V2: object = MINILM

all-MiniLM-L6-v2: English text in 384-dim vectors, 22M parameters.

embedding.ALL_MINILM_L6_V2.DEFAULT

DEFAULT: object & object = f16

Type Declaration

tokenizerPath

tokenizerPath: string

Type Declaration

modelPath

modelPath: string

embedding.ALL_MINILM_L6_V2.F16

F16: object & object

Type Declaration

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f16"

modelPath

modelPath: string

embedding.ALL_MINILM_L6_V2.F32

F32: object & object

Type Declaration

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f32"

modelPath

modelPath: string

embedding.EMBEDDINGGEMMA_2

readonly EMBEDDINGGEMMA_2: object = EMBEDDINGGEMMA_2

Google’s EmbeddingGemma 2: text in 100+ languages and code, images and video in one 768-dim space, inputs up to 2048 tokens. 270M parameters in the text model, 170M in the vision tower.

embedding.EMBEDDINGGEMMA_2.DEFAULT

DEFAULT: object

embedding.EMBEDDINGGEMMA_2.DEFAULT.audioPath

audioPath: string

embedding.EMBEDDINGGEMMA_2.DEFAULT.modelPath

modelPath: string

embedding.EMBEDDINGGEMMA_2.DEFAULT.tokenizerPath

tokenizerPath: string

embedding.EMBEDDINGGEMMA_2.DEFAULT.visionPath

visionPath: string

embedding.EMBEDDINGGEMMA_2.F16

F16: object

embedding.EMBEDDINGGEMMA_2.F16.audioPath

audioPath: string

embedding.EMBEDDINGGEMMA_2.F16.dtype

dtype: "f16"

embedding.EMBEDDINGGEMMA_2.F16.modelPath

modelPath: string

embedding.EMBEDDINGGEMMA_2.F16.tokenizerPath

tokenizerPath: string

embedding.EMBEDDINGGEMMA_2.F16.visionPath

visionPath: string

embedding.EMBEDDINGGEMMA_2.F32

F32: object

embedding.EMBEDDINGGEMMA_2.F32.audioPath

audioPath: string

embedding.EMBEDDINGGEMMA_2.F32.dtype

dtype: "f32"

embedding.EMBEDDINGGEMMA_2.F32.modelPath

modelPath: string

embedding.EMBEDDINGGEMMA_2.F32.tokenizerPath

tokenizerPath: string

embedding.EMBEDDINGGEMMA_2.F32.visionPath

visionPath: string

imageClassification

readonly imageClassification: object

imageClassification.MOBILENETV4

readonly MOBILENETV4: object

MobileNetV4 conv, ImageNet-1k, f16 weights. The model runs in f16 only, so each size has one entry.

imageClassification.MOBILENETV4.DEFAULT

readonly DEFAULT: object = MOBILENETV4_SMALL

imageClassification.MOBILENETV4.DEFAULT.modelPath

modelPath: string

imageClassification.MOBILENETV4.SMALL

readonly SMALL: object = MOBILENETV4_SMALL

The small size, 3.8M parameters, 10 MB.

imageClassification.MOBILENETV4.SMALL.modelPath

modelPath: string

instanceSegmentation

readonly instanceSegmentation: object

instanceSegmentation.YOLO26_SEG

readonly YOLO26_SEG: object

YOLO26 segment, COCO-80 classes with a mask per object, f16 weights.

instanceSegmentation.YOLO26_SEG.DEFAULT

readonly DEFAULT: object = YOLO26_SEG_N

instanceSegmentation.YOLO26_SEG.DEFAULT.modelPath

modelPath: string

instanceSegmentation.YOLO26_SEG.M

readonly M: object = YOLO26_SEG_M

The medium size, 45 MB. More accurate, slower.

instanceSegmentation.YOLO26_SEG.M.modelPath

modelPath: string

instanceSegmentation.YOLO26_SEG.N

readonly N: object = YOLO26_SEG_N

The nano size, 5 MB.

instanceSegmentation.YOLO26_SEG.N.modelPath

modelPath: string

instanceSegmentation.YOLO26_SEG.S

readonly S: object = YOLO26_SEG_S

The small size, 20 MB.

instanceSegmentation.YOLO26_SEG.S.modelPath

modelPath: string

keypointDetection

readonly keypointDetection: object

keypointDetection.YOLO26_POSE

readonly YOLO26_POSE: object

YOLO26 pose, people with 17 COCO landmarks, f16 weights.

keypointDetection.YOLO26_POSE.DEFAULT

readonly DEFAULT: object = YOLO26_POSE_N

keypointDetection.YOLO26_POSE.DEFAULT.modelPath

modelPath: string

keypointDetection.YOLO26_POSE.M

readonly M: object = YOLO26_POSE_M

The medium size, 42 MB. More accurate, slower.

keypointDetection.YOLO26_POSE.M.modelPath

modelPath: string

keypointDetection.YOLO26_POSE.N

readonly N: object = YOLO26_POSE_N

The nano size, 6 MB.

keypointDetection.YOLO26_POSE.N.modelPath

modelPath: string

keypointDetection.YOLO26_POSE.S

readonly S: object = YOLO26_POSE_S

The small size, 20 MB.

keypointDetection.YOLO26_POSE.S.modelPath

modelPath: string

objectDetection

readonly objectDetection: object

objectDetection.YOLO26

readonly YOLO26: object

YOLO26 detect, COCO-80 classes, f16 weights.

objectDetection.YOLO26.DEFAULT

readonly DEFAULT: object = YOLO26_N

objectDetection.YOLO26.DEFAULT.modelPath

modelPath: string

objectDetection.YOLO26.M

readonly M: object = YOLO26_M

The medium size, 39 MB. More accurate, slower.

objectDetection.YOLO26.M.modelPath

modelPath: string

objectDetection.YOLO26.N

readonly N: object = YOLO26_N

The nano size, 5 MB.

objectDetection.YOLO26.N.modelPath

modelPath: string

objectDetection.YOLO26.S

readonly S: object = YOLO26_S

The small size, 18 MB.

objectDetection.YOLO26.S.modelPath

modelPath: string

privacyFilter

readonly privacyFilter: object

privacyFilter.PRIVACY_FILTER

readonly PRIVACY_FILTER: object

OpenAI’s privacy-filter: 1.5B parameters, 50M active per token, int8 weights in groups of 640. The one hosted format.

privacyFilter.PRIVACY_FILTER.DEFAULT

readonly DEFAULT: object = PRIVACY_FILTER_INT8

privacyFilter.PRIVACY_FILTER.DEFAULT.modelPath

modelPath: string

speechToText

readonly speechToText: object

speechToText.MOONSHINE

readonly MOONSHINE: object

English.

speechToText.MOONSHINE.BASE

readonly BASE: object = MOONSHINE_BASE

61M parameters. More accurate, slower.

speechToText.MOONSHINE.BASE.DEFAULT

DEFAULT: object & object = f16

Type Declaration

arch

arch: "moonshine" = 'moonshine'

tokenizerPath

tokenizerPath: string

Type Declaration

modelPath

modelPath: string

speechToText.MOONSHINE.BASE.F16

F16: object & object

Type Declaration

arch

arch: "moonshine" = 'moonshine'

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f16"

modelPath

modelPath: string

speechToText.MOONSHINE.BASE.F32

F32: object & object

Type Declaration

arch

arch: "moonshine" = 'moonshine'

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f32"

modelPath

modelPath: string

speechToText.MOONSHINE.TINY

readonly TINY: object = MOONSHINE_TINY

27M parameters.

speechToText.MOONSHINE.TINY.DEFAULT

DEFAULT: object & object = f16

Type Declaration

arch

arch: "moonshine" = 'moonshine'

tokenizerPath

tokenizerPath: string

Type Declaration

modelPath

modelPath: string

speechToText.MOONSHINE.TINY.F16

F16: object & object

Type Declaration

arch

arch: "moonshine" = 'moonshine'

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f16"

modelPath

modelPath: string

speechToText.MOONSHINE.TINY.F32

F32: object & object

Type Declaration

arch

arch: "moonshine" = 'moonshine'

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f32"

modelPath

modelPath: string

speechToText.MOONSHINE_STREAMING

readonly MOONSHINE_STREAMING: object

English, the streaming family.

speechToText.MOONSHINE_STREAMING.TINY

readonly TINY: object = MOONSHINE_STREAMING_TINY

34M parameters.

speechToText.MOONSHINE_STREAMING.TINY.DEFAULT

DEFAULT: object & object = f16

Type Declaration

arch

arch: "moonshine-streaming" = 'moonshine-streaming'

tokenizerPath

tokenizerPath: string

Type Declaration

modelPath

modelPath: string

speechToText.MOONSHINE_STREAMING.TINY.F16

F16: object & object

Type Declaration

arch

arch: "moonshine-streaming" = 'moonshine-streaming'

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f16"

modelPath

modelPath: string

speechToText.MOONSHINE_STREAMING.TINY.F32

F32: object & object

Type Declaration

arch

arch: "moonshine-streaming" = 'moonshine-streaming'

tokenizerPath

tokenizerPath: string

Type Declaration

dtype

dtype: "f32"

modelPath

modelPath: string