Google claims EmbeddingGemma 2 outperforms rival embedding models twice its size
decoder ·
Google has released EmbeddingGemma 2, an open-source model designed to convert text, images, video, audio, and code into numerical vectors for easier comparison. With 740 million parameters, it is described as the most compact model of its kind while outperforming competitors that are twice its size on multimodal benchmarks. The model achieved a score of 78.68 on the Massive Text Embedding Benchmark for Code, marking a significant improvement over its predecessor. It operates locally without requiring an API key, utilizing WebGPU in the browser with query times between 20 and 70 milliseconds. The system requires only about 191 MB of RAM and can reduce local vector database storage by up to six times. For simpler text-only tasks, a smaller version with 270 million parameters is available. When paired with models like Gemma 4, it enables offline Retrieval-Augmented Generation applications without sending data to external servers. Weights and documentation are currently available on Hugging Face and Kaggle.
decoder