🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Releasing Gemma 3n, our new open-weight model processing audio, images and text (with improved multilingual capabilities), optimized for on-device usage with MatFormer architecture (enabling adaptive compute) and reaching 1283 on Chatbot Arena. Read more: developers.googleblog.com/en/introducing….

Announcing Gemma 3n preview 🎙️ Multimodal: audio, video, text 🧠New architecture with 4B and 2B effective params 🤯LMArena score of 1283 👀Available in AI Studio More info soon! developers.googleblog.com/en/introducing…

Beyond thrilled to share what we've been cooking the past few months! 🥳 Veo can now generate sounds and dialog! We are the state-of-the-art model on audio-video generation: deepmind.google/models/veo/eva… Try out Veo here: labs.google/flow/about

This is something I've been working on with some amazing collaborators for a while. Model-software-hardware co-design. Making things run fast on real devices. A lot of learning. And happy to share this with the open-source community and beyond. developers.googleblog.com/en/introducing…

As you probably already heard, Gemma 3n can use different amounts of parameters during inference (from 2B to 5B) depending on the device it runs on. This is thanks to the Matryoshka architecture proposed in the MatFormer paper. MatFormer is based on "nested" feedforward network

This is my favorite demo of Gemma 3n -- multimodal live video understanding and intelligence, locally on your phone 🤯! This was only possible with the peak of foundation models at I/O last year -- the Astra demo -- the progress of small models is incredible

Gemma 3n is out! 🚀🚀🚀 The frontier models from a year ago can now run locally on a phone! Lots of innovations (e.g. matformers, mix’n’match, per layer embeddings) to make this model mobile first. And we finally have audio/video as an input for Gemma models! 1/2

Our post-training team (once again) went beyond to post-craft this fantastic performance-for-size model: Johan Ferret Alexandre Ramé Sarah Perrin Geoffrey Cideron Nino Vieillard Sabela Angéline Pouget V Carbune L Rouillard L Hussenot Gaël Liu Danila Sinopalnikov Olivier Bachem 2/2

MatFormers are very powerful alternatives to transformers. Similar to a regular transformer, but after training, you can split up the model to any size you like and get very strong performance that scales just like a regular transformer. So train once, get models of all sizes!

There was a lot of announcements at IO, easy to overlook the new 2.5 Flash. It's pushing new boundaries in capability vs speed!

Introducing Gemma 3n, our multimodal model built for mobile on-device AI. 🤳 It runs with a smaller memory footprint, cutting down RAM usage by nearly 3x – enabling more complex applications right on your phone, or for livestreaming from the cloud. Now available in early

What can you do with Gemma 3n? 🛠️Generate smart text from audio, images, video, and text 🛠️Create live, interactive apps that react to what users see and hear 🛠️Build advanced audio apps for real-time speech, translation, and voice commands

Gemma 3n was built to be fast and efficient. 🏃 Engineered to run quickly and locally on-device – ensuring reliability, even without the internet. Think up to 1.5x faster response times on mobile! Preview Gemma 3n now on Google AI Studio. → goo.gle/4jrSOZq

The Gemma team keeps shipping. In 6 months: - PaliGemma 2 - PaliGemma 2 Mix - Gemma 3 - ShieldGemma 2 - TxGemma - Gemma 3 QAT - Gemma 3n Preview - MedGemma Early - DolphinGemma - SignGemma And so much more to come! 🚀

Introducing Matryoshka Teaching Assistant aka matTA framework! This allows learning a "Matformed TA-student" pair with distillation from a much more relatable TA, and provides elasticity/adaptivity/flexibility of Matryoshka models. Key benefits: a. more accurate, servable

Continuing impact from Matryoshka Models, so proud of our team's contributions!

Super blessed to work with such visionary and kind leader 🙏

We are hiring Research Scientists for our Machine Learning and Optimization team at Google DeepMind Bangalore. If you're passionate about cutting-edge AI research and building efficient, elastic, customized, and safe LLMs, we'd love to hear from you. We are looking for

For strong researchers, it doesn't get any better. You get to work with one of the best ML research teams in the world and contribute to Google's frontier models impacting billions of people!

I recently left FAIR and joined Google DeepMind. I'm deeply grateful to the FAIR leadership, mentors, collaborators & friends who believed in me, encouraged me to aim higher & celebrated my wins. Grateful for the journey so far & excited to help advance AI research at DeepMind.