Tasks and inputs
EmbeddingGemma 2 is listed for Embeddings, Image feature extraction. It accepts text, image, audio, video and produces embedding. Use the publisher’s input format and preprocessing for your chosen deployment.
EmbeddingGemma 2 creates shared embeddings for text, code, images, video and audio for multimodal retrieval.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
EmbeddingGemma 2 creates shared embeddings for text, code, images, video and audio for multimodal retrieval.
Google’s model card describes a modular 740M-parameter system whose text, vision and audio components can be selected for the intended workload. It produces 768-dimensional embeddings, with Matryoshka options for shorter vectors. The release uses Apache 2.0 and is intended for similarity, search and retrieval across supported modalities.
Use a small collection containing the material your users will search: passages, diagrams, short clips or audio segments. Label relevant results for a set of queries, including paraphrases and cases where no useful match exists. Measure retrieval quality before adding a language model to explain the results. A persuasive generated answer does not reveal whether the retriever found the correct evidence.
Apply the documented preprocessing and query or document instructions consistently. Choose vector dimension before building the index, and keep the same model, dimension and normalization settings on both sides of retrieval. Switching these choices usually requires rebuilding stored vectors. Evaluate the memory and search-time savings of shorter vectors against lost relevant results. Modality support describes what can be embedded; it does not guarantee that every cross-modal query will retrieve the intended item. Keep ordinary chat-output limits and token prices unknown unless a specific managed service documents them.
Compare the model on the work you need to complete.
EmbeddingGemma 2 is listed for Embeddings, Image feature extraction. It accepts text, image, audio, video and produces embedding. Use the publisher’s input format and preprocessing for your chosen deployment.
Evaluate retrieval over a labelled query set. Keep the encoder, vector dimensions and index settings consistent.
The weights are downloadable under Apache 2.0. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Tencent
WeMM Embedding 2B encodes text, images, video and visual documents for multimodal similarity search.
Tencent
WeMM Embedding 4B encodes text, images, video and visual documents for multimodal similarity search.
Tencent
WeMM Embedding 9B encodes text, images, video and visual documents for multimodal similarity search.