Skip to content
ruNNtime

SpeechToTextModel

Defined in: packages/runntime/src/zoo/tasks/audio/speechToText.ts:39

What to load. Every field is optional; a missing one comes from the family’s default entry in models.speechToText.

Properties

Section titled “Properties”

arch?

Section titled “arch?”

readonly optional arch?: SpeechToTextArch

Defined in: packages/runntime/src/zoo/tasks/audio/speechToText.ts:41

The family the weights belong to. Default moonshine.

dtype?

Section titled “dtype?”

readonly optional dtype?: "f32" | "f16"

Defined in: packages/runntime/src/zoo/tasks/audio/speechToText.ts:50

Precision on the GPU. f16 runs the whole model in half precision, weights and activations, for half the memory traffic; the text stays the same. Needs a device with shader-f16. Default: f16 where the device has it, f32 elsewhere. Either loads any weight file.

modelPath?

Section titled “modelPath?”

readonly optional modelPath?: ModelPath

Defined in: packages/runntime/src/zoo/tasks/audio/speechToText.ts:43

The safetensors checkpoint.

tokenizerPath?

Section titled “tokenizerPath?”

readonly optional tokenizerPath?: string

Defined in: packages/runntime/src/zoo/tasks/audio/speechToText.ts:45

The Hugging Face tokenizer.json.