Interface: TextEmbeddingsModel

Defined in: types/textEmbeddings.ts:60

A text embeddings model config. Two optional flags drive forward: prompts makes a role argument required, and multiVector makes it return a per-token EmbeddingResult instead of a pooled Float32Array.

Properties

modelName

modelName: TextEmbeddingsModelName

Defined in: types/textEmbeddings.ts:61

---

modelSource

modelSource: ResourceSource

Defined in: types/textEmbeddings.ts:62

---

multiVector?

optional multiVector: boolean

Defined in: types/textEmbeddings.ts:65

---

prompts?

optional prompts: EmbeddingPrompts

Defined in: types/textEmbeddings.ts:64

---

skipListIds?

optional skipListIds: number[]

Defined in: types/textEmbeddings.ts:72

Document token ids to exclude from late-interaction scoring (e.g. ColBERT's punctuation skipList). Derived from the model's training config, so it's shipped here rather than reconstructed by the consumer, who passes it to their own MaxSim scoring.

---

tokenizerSource

tokenizerSource: ResourceSource

Defined in: types/textEmbeddings.ts:63