README.md

metadata

title: EmbeddingGemma 2
emoji: 🧭
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 6.30.0
app_file: app.py
short_description: Search text and images with one embedding
python_version: '3.12'
startup_duration_timeout: 30m
license: apache-2.0
models:
  - google/embeddinggemma-2

EmbeddingGemma 2

Semantic search across text and images with google/embeddinggemma-2.

Queries use the model's task prefixes (search result, question answering, fact checking, code retrieval, sentence similarity). Documents use title: … | text: …. Images are embedded with no prefix, in the same 768-d space. Vectors can be truncated to 512, 256, or 128 dimensions and are L2-normalized before cosine scoring.

The Space loads the text and vision encoders and skips the audio encoder. Inference runs in float32 on CPU.