EmbeddingGemma 2 unifies text, image, audio and video embeddings
Original title: EmbeddingGemma 2: Unified and Efficient Multimodal Representations
The open model claims a new Pareto frontier while supporting quantization and selective encoder loading.
Researchers present EmbeddingGemma 2, an open multimodal embedding model that places text, images, audio, video, and combinations in one space. They describe simplified per-layer embeddings and a multi-stage curriculum across modalities and objectives. On embedding benchmarks, they say it beats specialist models and models two to twelve times larger. The release supports selective modality encoders, adjustable vision token budgets, truncatable dimensions, and quantized weights.
Unconfirmed: Benchmark superiority is the authors' claim.
- It natively embeds mixed modalities in one space.
- Authors report gains over models 2x to 12x larger.
- Ablations of key design decisions are included.
- Weights are released for further research.
From the original article
1 min read at huggingface.co
We present EmbeddingGemma 2, an open multimodal embedding model which natively represents text, images, audio, video, and combinations thereof in a unified space. Our approach introduces improvements such as simplified per-layer embeddings and a multi-stage curriculum with modality and objective progression which ensure strong quality across unimodal, cross-modal, and mixed-modal evaluations.
By Henrique Schechter Vera, Sahil Dua, Jovan Andonov · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.