embeddinggemma

Upstream reference: EmbeddingGemma

Publisher: Google DeepMind

Google describes EmbeddingGemma as a compact model that represents text as numerical vectors, intended for search, retrieval and related text-analysis tasks rather than conversational text output.

Publisher-described context, separate from Sigma5C observations and recorded catalog statistics.

Published upstream specifications

Published size

300 million parameters Source

Output

Text embedding vectors Source

Native output dimensions

768 Source

Documented reduced dimensions

512, 256 or 128 with Matryoshka truncation and renormalization Source

Publisher-described uses

• Semantic similarity search, classification and clustering. Source

Reference limitations

• An embedding vector is not a generated answer or a conversational response. Source

A recorded tag and an upstream card association do not verify the deployed checkpoint, quantization, configured context or current hosted availability.

Publisher sources

• EmbeddingGemma official documentation — Google DeepMind. Checked 2026-11-02.

Where embeddinggemma comes from

Named in 1 observation page

• Large-paper embedding observation · V1