README.md

---
title: EmbeddingGemma 2 Search Lab
emoji: 🔎
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.29.1
app_file: app.py
short_description: Explore task-aware embeddings and semantic search
python_version: "3.12"
startup_duration_timeout: 30m
---

# EmbeddingGemma 2 Search Lab

Try Google's multilingual text embedding model in a small semantic-search workbench. Edit the document collection, enter a query, choose a retrieval task and embedding size, then compare the ranked results and cosine scores.

The app uses the model's asymmetric task instructions for queries and documents, and re-normalizes vectors after Matryoshka truncation. It loads only the text encoders to keep the demo lightweight. The underlying model also supports image, audio, and video embeddings.

## Model

[`google/embeddinggemma-2`](https://huggingface.co/google/embeddinggemma-2) · Apache 2.0 · 768-dimensional vectors · 128/256/512/768 dimensions supported.

## Run locally

```bash
pip install -r requirements.txt
python app.py
```

The model files are downloaded from Hugging Face on first startup. Access to the model may require accepting its terms on the model page.