Options to configure the MediaPipe Universal Embedder Task.
Signature:
export declare interface UniversalEmbedderOptions extends TaskRunnerOptions
Extends: TaskRunnerOptions
Properties
| Property | Type | Description |
|---|---|---|
| activationDataType | 'FLOAT32' | 'FLOAT16' | 'INT8' | undefined | Optional activation precision override for inference execution. |
| l2Normalize | boolean | undefined | Whether to L2-normalize the output embedding vectors. Defaults to true. |
| maxInputLength | number | undefined | Maximum sequence length in tokens for text and multimodal inputs. Defaults to 512. |
| visionTokensPerImage | 64 | 256 | undefined | Number of vision tokens allocated per input image (64 or 256). Defaults to 64. |
UniversalEmbedderOptions.activationDataType
Optional activation precision override for inference execution.
Signature:
activationDataType?: 'FLOAT32' | 'FLOAT16' | 'INT8';
UniversalEmbedderOptions.l2Normalize
Whether to L2-normalize the output embedding vectors. Defaults to true.
Signature:
l2Normalize?: boolean;
UniversalEmbedderOptions.maxInputLength
Maximum sequence length in tokens for text and multimodal inputs. Defaults to 512.
Signature:
maxInputLength?: number;
UniversalEmbedderOptions.visionTokensPerImage
Number of vision tokens allocated per input image (64 or 256). Defaults to 64.
Signature:
visionTokensPerImage?: 64 | 256;