Stanford, CA
Joined December 2009
- Tweets234
- Following355
- Followers393
- Likes728
Pinned Tweet
Bucket list item checked ✅ Recording a cool demo video tweeted by @Google!
The incredibly powerful Gemma 4 12B is now fully rolled out globally in Eloquent. Download ai.google.dev/edge/eloquent/ and experience a powerhouse AI dictation and editing assistant that is fully local, private, subscription-free, no cap! 🧢
Our new unified architecture allows Gemma 4 12B to process multimodal inputs natively. Here's how ⬇️
Traditional models rely on separate encoders for images and audio. This adds latency and increases memory usage. So we streamlined this:
👁️ Vision: We took a novel approach to replace the encoder with a lightweight embedding module, letting the LLM backbone take over visual processing.
🎙️ Audio: We removed the encoder entirely, projecting the raw audio signal directly into the same space as text tokens.
Meet Google AI Edge Foresight, our newest Labs app (developers.google.com/edge/f…), offering an ambient note-taking meeting companion and personal knowledge assistant for Mac. Your audio, transcripts and sensitive files in your personal knowledge base never leave your device courtesy of local processing with EmbeddingGemma2 & Gemma4 suite of models.
- Live Notes Enhancement: Jot down quick shorthand notes during a meeting and Foresight enriches it in real time using the live meeting conversation.
- Ambient Audio Query Detection: Detects questions spoken aloud mid-meeting and surfaces answers live.
- Blazing-Fast Grounded Responses: Point it at your local project folders, docs, transcripts or Drive account for instant cross-modal retrieval and Q&A.
3/ (On-device Jev) MediaPipe Decision API (developers.google.com/edge/m…)
EG2 is a surprisingly good decision engine! With the MP Decision API, you can define strongly typed primitives (Choice, Boolean, Score) and run zero-shot classification and routing in the order of milliseconds!
2/ On-device MediaPipe Tasks so you can deploy EmbeddingGemma2 in your apps in a few lines of code across desktop, web and mobile! developers.google.com/edge/m…
* UniversalEmbedder: Vectorize text, images, audio, or interleaved multimodal content in one unified API.
* SemanticRetriever: An edge-optimized embedded vector search engine that pairs directly with the UniversalEmbedder for fast ANN searches in milliseconds.
* Live Notes Enhancement: Jot down quick shorthand notes during a meeting and Foresight enriches it in real time using the live meeting conversation.
* Ambient Audio Query Detection: Detects questions spoken aloud mid-meeting and surfaces answers live.
* Blazing-Fast Grounded Responses: Point it at your local project folders, docs, transcripts or Drive account for instant cross-modal retrieval and Q&A.
1/ Meet Foresight, our newest Labs app (developers.google.com/edge/f…), an on-device ambient meeting companion and personal knowledge assistant for Mac. Your audio and sensitive files in your personal knowledge base never leave your device courtesy of local processing with EmbeddingGemma2 & Gemma4 suite of models.
Super jazzed for the EmbeddingGemma2 launch today! It crushes the benchmarks but the real story is how much new product surface area it unlocks on the edge! As seen in the video, there's a lot to unwrap. Here is everything we built with EG2 launching today 🧵
Introducing EmbeddingGemma 2! 🚀
Our lightweight, multimodal embedding model maps text, code, images, video, and audio into a single, unified embedding space. Optimized for on-device use cases, it features:
- 740M parameter form factor with modular encoders
- Flexible dimension sizes (768dim-128dim) via Matryoshka Representation Learning (MRL)
- 8K context window (4x larger than text-only EmbeddingGemma)
- A commercially permissive Apache 2.0 license