Google EmbeddingGemma 2: free Apache 2.0 multimodal model weights for local use

The offer

Indian developers can download Google 740-million-parameter EmbeddingGemma 2 weights free for text, code, image, audio and video embeddings. Run it with supported libraries on your own hardware; this is a downloadable model, not free hosted inference or a chatbot subscription. The official model card lists Apache 2.0 and requires deployments to follow the Gemma Prohibited Use Policy.

Who can claim it

• Downloadable in India under a worldwide no-charge Apache 2.0 licence.
• Bring suitable hardware and a compatible model runtime; your compute and storage are not included.
• Use it for embeddings, search, classification or retrieval rather than expecting a conversational chatbot.

How to claim it

1. Open Google official EmbeddingGemma 2 model listing on Hugging Face.
2. Read the model card, linked Apache 2.0 licence and Gemma Prohibited Use Policy.
3. Use Files and versions to obtain the weights, or follow the model card sentence-transformers or Transformers quick start.
4. Load google/embeddinggemma-2 in your supported local runtime and follow the documented task-prefix and numerical-precision guidance.

Good to know

There is no free-trial expiry or automatic paid renewal for these downloadable weights. Your hardware, electricity, storage and any cloud compute you choose remain your responsibility.

The official repository is public and ungated, with model.safetensors available. No free hosted inference or training-code release is promised. Apache 2.0 requires licence/notice preservation and notices of modifications, and provides no warranty. The model card also requires the Gemma Prohibited Use Policy. Outputs may carry bias; application safeguards are your responsibility. Use bfloat16 or float32, not float16, as directed by the model card.