Class: HuggingFaceEmbedding
Uses feature extraction from '@xenova/transformers' to generate embeddings. Per default the model XENOVA_ALL_MINILM_L6_V2 is used.
Can be changed by setting the modelType
parameter in the constructor, e.g.:
new HuggingFaceEmbedding({
modelType: HuggingFaceEmbeddingModelType.XENOVA_ALL_MPNET_BASE_V2,
});
Hierarchy
-
↳
HuggingFaceEmbedding
Constructors
constructor
• new HuggingFaceEmbedding(init?
): HuggingFaceEmbedding
Parameters
Name | Type |
---|---|
init? | Partial <HuggingFaceEmbedding > |
Returns
Overrides
Defined in
packages/core/src/embeddings/HuggingFaceEmbedding.ts:27
Properties
embedBatchSize
• embedBatchSize: number
= DEFAULT_EMBED_BATCH_SIZE
Inherited from
Defined in
packages/core/src/embeddings/types.ts:9
extractor
• Private
extractor: any
Defined in
packages/core/src/embeddings/HuggingFaceEmbedding.ts:25
modelType
• modelType: string
= HuggingFaceEmbeddingModelType.XENOVA_ALL_MINILM_L6_V2
Defined in
packages/core/src/embeddings/HuggingFaceEmbedding.ts:22
quantized
• quantized: boolean
= true
Defined in
packages/core/src/embeddings/HuggingFaceEmbedding.ts:23
Methods
getExtractor
▸ getExtractor(): Promise
<any
>
Returns
Promise
<any
>
Defined in
packages/core/src/embeddings/HuggingFaceEmbedding.ts:32
getQueryEmbedding
▸ getQueryEmbedding(query
): Promise
<number
[]>
Parameters
Name | Type |
---|---|
query | string |
Returns
Promise
<number
[]>
Overrides
BaseEmbedding.getQueryEmbedding