FastEmbed-rs 🦀

Rust implementation of @qdrant/fastembed

🍕 Features

• Supports synchronous usage. No dependency on Tokio.
• Uses @pykeio/ort for performant ONNX inference.
• Uses @huggingface/tokenizers for fast encodings.
• Supports batch embeddings generation with parallelism using @rayon-rs/rayon.

The default model is Flag Embedding, which is top of the MTEB leaderboard.

🔍 Not looking for Rust?

• Python 🐍: fastembed
• Go 🐳: fastembed-go
• JavaScript 🌐: fastembed-js

🤖 Models

Text Embedding

• BAAI/bge-small-en-v1.5 - Default
• sentence-transformers/all-MiniLM-L6-v2
• mixedbread-ai/mxbai-embed-large-v1

Click to see full List

• BAAI/bge-large-en-v1.5
• BAAI/bge-small-zh-v1.5
• BAAI/bge-base-en-v1.5
• sentence-transformers/all-MiniLM-L12-v2
• sentence-transformers/paraphrase-MiniLM-L12-v2
• sentence-transformers/paraphrase-multilingual-mpnet-base-v2
• nomic-ai/nomic-embed-text-v1
• nomic-ai/nomic-embed-text-v1.5
• intfloat/multilingual-e5-small
• intfloat/multilingual-e5-base
• intfloat/multilingual-e5-large
• Alibaba-NLP/gte-base-en-v1.5
• Alibaba-NLP/gte-large-en-v1.5

Sparse Text Embedding

• prithivida/Splade_PP_en_v1 - Default

Image Embedding

• Qdrant/clip-ViT-B-32-vision - Default
• Qdrant/resnet50-onnx
• Qdrant/Unicom-ViT-B-16
• Qdrant/Unicom-ViT-B-32

Reranking

• BAAI/bge-reranker-base
• BAAI/bge-reranker-v2-m3
• jinaai/jina-reranker-v1-turbo-en
• jinaai/jina-reranker-v2-base-multiligual

🚀 Installation

Run the following command in your project directory:

cargo add fastembed

Or add the following line to your Cargo.toml:

[dependencies]
fastembed = "3"

📖 Usage

Text Embeddings

use fastembed::{TextEmbedding, InitOptions, EmbeddingModel};

// With default InitOptions
let model = TextEmbedding::try_new(Default::default())?;

// With custom InitOptions
let model = TextEmbedding::try_new(
        InitOptions::new(EmbeddingModel::AllMiniLML6V2).with_show_download_progress(true),
    )?;

let documents = vec![
    "passage: Hello, World!",
    "query: Hello, World!",
    "passage: This is an example passage.",
    // You can leave out the prefix but it's recommended
    "fastembed-rs is licensed under Apache  2.0"
    ];

 // Generate embeddings with the default batch size, 256
 let embeddings = model.embed(documents, None)?;

 println!("Embeddings length: {}", embeddings.len()); // -> Embeddings length: 4
 println!("Embedding dimension: {}", embeddings[0].len()); // -> Embedding dimension: 384

Image Embeddings

use fastembed::{ImageEmbedding, ImageInitOptions, ImageEmbeddingModel};

// With default InitOptions
let model = ImageEmbedding::try_new(Default::default())?;

// With custom InitOptions
let model = ImageEmbedding::try_new(
    ImageInitOptions::new(ImageEmbeddingModel::ClipVitB32).with_show_download_progress(true),
)?;

let images = vec!["assets/image_0.png", "assets/image_1.png"];

// Generate embeddings with the default batch size, 256
let embeddings = model.embed(images, None)?;

println!("Embeddings length: {}", embeddings.len()); // -> Embeddings length: 2
println!("Embedding dimension: {}", embeddings[0].len()); // -> Embedding dimension: 512

Candidates Reranking

use fastembed::{TextRerank, RerankInitOptions, RerankerModel};

let model = TextRerank::try_new(
    RerankInitOptions::new(RerankerModel::BGERerankerBase).with_show_download_progress(true),
)?;

let documents = vec![
    "hi",
    "The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear, is a bear species endemic to China.",
    "panda is animal",
    "i dont know",
    "kind of mammal",
];

// Rerank with the default batch size
let results = model.rerank("what is panda?", documents, true, None)?;
println!("Rerank result: {:?}", results);

Alternatively, local model files can be used for inference via the try_new_from_user_defined(...) methods of respective structs.

✊ Support

To support the library, please consider donating to our primary upstream dependency, ort - The Rust wrapper for the ONNX runtime.

⚙️ Under the hood

It's important we justify the "fast" in FastEmbed. FastEmbed is fast because of:

1. Quantized model weights.
2. ONNX Runtime which allows for inference on CPU, GPU, and other dedicated runtimes.
3. No hidden dependencies via Huggingface Transformers.

📄 LICENSE

Apache 2.0 © 2024