19:48EmbeddingGemma 2: How natively multimodal embeddings work
From the creator
Learn how EmbeddingGemma 2, Google's open, lightweight multimodal embedding model, maps text, code, images, video, and audio into a single, unified embedding space. The video covers: * How modality-specific encoders and a bidirectional transformer turn multimodal data into a single representation. * The two-stage training recipe: initializing from Gemma 4 and distilling from Gemini Embedding 2, then finetuning with contrastive and spread-out losses. * Efficiency techniques for running in resource-constrained environments. Have questions? Drop them in the comments. 🔗Resources: Read EmbeddingGemma 2 tech report → https://goo.gle/4eAIY7W Download model weights → https://goo.gle/4fbzYpU View docs → https://goo.gle/3WJLEKh Subscribe to Google for Developers → https://goo.gle/developers Speakers: Henrique Schechter Vera, Sahil Dua Products Mentioned: Gemma
Published 2026-11-17 · Watch on YouTube ↗
Availability and playback restrictions are controlled by the original channel.





