Interface: HuggingFaceConfig

Defined in: vectorizers/huggingface-vectorizer.ts:7

Configuration options for HuggingFaceVectorizer

Properties

device?

optional device?: "cpu" | "webgpu"

Defined in: vectorizers/huggingface-vectorizer.ts:23

Device to run inference on.

• 'cpu': Run on CPU (default)
• 'webgpu': Run on GPU (if available)

---

dtype?

optional dtype?: "fp32" | "fp16" | "q8"

Defined in: vectorizers/huggingface-vectorizer.ts:31

Data type for model weights.

• 'fp32': 32-bit floating point (default, best quality)
• 'fp16': 16-bit floating point (faster, less memory)
• 'q8': 8-bit quantized (fastest, smallest)

---

model

model: string

Defined in: vectorizers/huggingface-vectorizer.ts:16

The Hugging Face model to use for embeddings.

Examples:

• 'Xenova/all-MiniLM-L6-v2' (384 dimensions, fast)
• 'Xenova/all-mpnet-base-v2' (768 dimensions, better quality)
• 'sentence-transformers/all-MiniLM-L6-v2'

---

normalize?

optional normalize?: boolean

Defined in: vectorizers/huggingface-vectorizer.ts:44

Whether to normalize embeddings to unit length (L2 norm = 1). Default: true

---

pooling?

optional pooling?: "mean" | "cls"

Defined in: vectorizers/huggingface-vectorizer.ts:38

Pooling strategy for combining token embeddings.

• 'mean': Average all token embeddings (default)
• 'cls': Use [CLS] token embedding