Class SiglipVisionTower

Readonlycfg

ReadonlyembedScale

cacheIdentity

• get cacheIdentity(): string

  Immutable encoder weights and computation settings, hashed once on use.

  Returns string

dispose

• dispose(): void

  Returns void

features

• features(pre: SiglipPreprocessed): MlxArray

  One preprocessed image → language-space soft tokens [1, softTokens, textHidden], pre-divided by embed_scale.

  Parameters

  • pre: SiglipPreprocessed

  Returns MlxArray

preprocess

• preprocess(bytes: Uint8Array): Promise<SiglipPreprocessed>

  Parameters

  • bytes: Uint8Array

  Returns Promise<SiglipPreprocessed>

Staticload

• load(
      modelDir: string,
      cfg: SiglipVisionConfig,
      embedScale: number,
  ): SiglipVisionTower

  Load the vision_tower.* + embed_vision.* tensors from the bf16 sidecar. embedScale = the language model's embed_scale (= sqrt(textHidden)); features come out pre-divided by it.

  Parameters

  • modelDir: string
  • cfg: SiglipVisionConfig
  • embedScale: number

  Returns SiglipVisionTower