cpu-embeddings

Embeddings generated on CPU for lightweight classification tasks using the Xenova Transformers library.

Features

• CPU-based: Runs entirely on CPU without requiring GPU acceleration.
• Embedded Model: Includes a pre-trained quantized ONNX model (all-MiniLM-L6-v2) bundled with the package for convenience.
• Lightweight: Optimized for small-scale classification tasks.
• TypeScript Support: Fully typed for better development experience.

Installation

npm install cpu-embeddings

API Reference

EmbeddingsOptions

Interface for embedding options.

• modelName?: string - Model name (default: "Xenova/all-MiniLM-L6-v2")
• modelPath?: string - Local model path (default: "models")
• numThreads?: number - Number of threads for ONNX (default: 1)

embeddings(text: string | string[], opts?: EmbeddingsOptions): Promise<number[]>

Generates embeddings for the given text(s).

• Parameters:
  • text: A single string or an array of strings to embed
  • opts: Optional embedding options
• Returns: A promise that resolves to an array of numbers representing the embeddings. For a single string, returns 384 numbers. For an array of n strings, returns 384 * n numbers.

Implementation Details

Uses the Xenova Transformers library to run transformer models in Node.js without Python. Runs on CPU with ONNX Runtime WebAssembly, using a quantized all-MiniLM-L6-v2 model with mean pooling and L2 normalization. Model files are bundled for offline use.

Useful Commands

Development

# Install dependencies
npm install

# Run tests
npm test

# Build the library
npm run build

# Format code
npm run format

# Start development server
npm run dev

Release

# Patch release
npm run release:patch

# Minor release
npm run release:minor

# Major release
npm run release:major

License

MIT