EmbeddingGemma: Powerful and Lightweight Text Representations
Abstract
EmbeddingGemma, a lightweight text embedding model based on Gemma 3, achieves state-of-the-art performance with fewer parameters through encoder-decoder initialization, geometric embedding distillation, and spread-out regularization.
We introduce EmbeddingGemma, a new lightweight, open text embedding model based on the Gemma 3 language model family. Our innovative training recipe strategically captures knowledge from larger models via encoder-decoder initialization and geometric embedding distillation. We improve model robustness and expressiveness with a spread-out regularizer, and ensure generalizability by merging checkpoints from varied, optimized mixtures. Evaluated on the Massive Text Embedding Benchmark (MTEB) across multilingual, English, and code domains, EmbeddingGemma (300M) achieves state-of-the-art results. Notably, it outperforms prior top models, both proprietary and open, with fewer than 500M parameters, and provides performance comparable to models double its size, offering an exceptional performance-to-cost ratio. Remarkably, this lead persists when quantizing model weights or truncating embedding outputs. This makes EmbeddingGemma particularly well-suited for low-latency and high-throughput use cases such as on-device applications. We provide ablation studies exploring our key design choices. We release EmbeddingGemma to the community to promote further research.
Community
We introduce EmbeddingGemma, a new lightweight, open text embedding model based on the Gemma 3 language model family. Our innovative training recipe strategically captures knowledge from larger models via encoder-decoder initialization and geometric embedding distillation. We improve model robustness and expressiveness with a spread-out regularizer, and ensure generalizability by merging checkpoints from varied, optimized mixtures. Evaluated on the Massive Text Embedding Benchmark (MTEB) across multilingual, English, and code domains, EmbeddingGemma (300M) achieves state-of-the-art results. Notably, it outperforms prior top models, both proprietary and open, with fewer than 500M parameters, and provides performance comparable to models double its size, offering an exceptional performance-to-cost ratio. Remarkably, this lead persists when quantizing model weights or truncating embedding outputs. This makes EmbeddingGemma particularly well-suited for low-latency and high-throughput use cases such as on-device applications. We provide ablation studies exploring our key design choices. We release EmbeddingGemma to the community to promote further research.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- Granite Embedding R2 Models (2025)
- Training LLMs to be Better Text Embedders through Bidirectional Reconstruction (2025)
- Conan-Embedding-v2: Training an LLM from Scratch for Text Embeddings (2025)
- Causal2Vec: Improving Decoder-only LLMs as Versatile Embedding Models (2025)
- On The Role of Pretrained Language Models in General-Purpose Text Embeddings: A Survey (2025)
- Context-Adaptive Multi-Prompt Embedding with Large Language Models for Vision-Language Alignment (2025)
- MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interaction (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend
Get this paper in your agent:
hf papers read 2509.20354 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash