EmbeddingGemma-300M 🧠✨

Overview 🔊

EmbeddingGemma-300M is a text-to-vector model that transforms any text into 768-dimensional embeddings. This tool is built upon the amazing work of Google DeepMind and their EmbeddingGemma research. We’ve wrapped their embedding-gemma-300m model to work on Replicate! Allowing us to generate high-quality embeddings for search, recommendations, and AI applications!

Support Google DeepMind and learn more about their work through:

• EmbeddingGemma Paper
• Model Card on Hugging Face

Pre-loaded Efficiency 🚀

The EmbeddingGemma-300M comes pre-loaded with base64 output format for immediate use with 3x smaller response sizes compared to arrays.

Getting Different Dimensions 💥

You can customize your EmbeddingGemma-300M’s output using Matryoshka representation learning. Here’s how:

1. Choose your embedding dimension: 128, 256, 512, or 768

2. Set the embedding_dim parameter to your desired size

3. The model automatically truncates the full 768-dimensional embedding to your chosen size

4. Example usage:

5. embedding_dim = 256 → Get 256-dimensional vectors
6. embedding_dim = 512 → Get 512-dimensional vectors
7. embedding_dim = 768 → Get full 768-dimensional vectors (default)

Note: Smaller dimensions are perfect for storage optimization while maintaining strong performance for most tasks.

Code Examples 📚

Python:

import replicate, base64, numpy as np

# Get base64 embedding (default)
b64 = replicate.run("zsxkib/embedding-gemma-300m", input={"text": "Hello world"})

# Decode to numpy array
vec = np.frombuffer(base64.b64decode(b64), dtype=np.float32)
print(vec.shape)  # (768,)

JavaScript:

const b64 = await replicate.run("zsxkib/embedding-gemma-300m", { input: { text: "Hello world" } });
const embedding = new Float32Array(Buffer.from(b64, 'base64').buffer);
console.log(embedding.length);  // 768

Terms of Use 📚

The use of this embedding model for the following purposes is prohibited:

• Generating embeddings for harmful or malicious content.

• Creating systems designed to manipulate or deceive users.

• Building applications that violate user privacy or data protection laws.

• Commercial use without proper licensing from Google.

• Redistributing the model weights without permission.

• Using embeddings to create biased or discriminatory systems.

Disclaimer ‼️

I am not liable for any direct, indirect, consequential, incidental, or special damages arising out of or in any way connected with the use/misuse or inability to use this software.

⭐ Star the repo on GitHub!
🐦 Follow @zsakib_ on X
💻 Check out more projects @zsxkib on GitHub