DevEx Engineer @ Google DeepMind Working on Gemma open models | Bring frontier intelligence to the edge! | BSky: https://xitter.twitterapi.xyz/t.co/JPipCfv3P5 | LI: ianballantyne
UK
Joined May 2011
- Tweets2.4K
- Following956
- Followers967
- Likes1.6K
We took "Reachy on Wheels" a @googlegemma demo to #pytorchconf. It uses Gemma 4 E4B as a local voice agent and EmbeddingGemma 2 as a "Gemma-based decision model" to drive around. All of this running on an @arduino VentunoQ! Check out the video to understand how it works ⚙️
Looking forward to Open Source AI Awards Night, tonight in San Francisco. The community is coming together to celebrate the amazing contributions and very proud that Google DeepMind is supporting it 🙌♥️🫂 share.google/LXQE44SbUztpqDg…
1
128
#PyTorchConf + Open Source AI Week are calling. Gemma open models answers the call. It's a packed week in SF / San Jose and I'm ready for it 🫡
💪 Arm Create: Gemma on-device session (Oct 18): luma.com/w9n0o36l
🎤 Open Weight Debate Night (Oct 19): luma.com/592fcwx2
🔦 PyTorchConf: Talks, Posters, Demos (Oct 20 - 21st): pytorch.org/event/pytorch-co…
🏆 Open Source AI Awards Night (Oct 23): luma.com/7ebvz0oq
🛠️ The Google Gemma Hardware Hackathon w/ Android, Arduino, Coral, LiteRT, Nvidia, Pixel, Qualcomm, Raspberry Pi, Synaptics (Oct 24): agihouse.org/events/the-goog…
In town? Come say hi 👋 and join the open source AI community!
🤖 Made with AI
Thanks to @open_intel_fdn @matthew_d_white @agihouse_org @googlegemma @GoogleDeepMind all our partners and more for enabling this incredible lineup! 🙏
1
72
Easier to navigate, more interactive, friendly to agents. These are all things I'd use to describe gemma.ai our new destination for everything Gemma!
Super proud of the Gemma DevX team for this one. Congrats @MaartenGr and Ju-yeong for your awesome work👏
Introducing gemma.ai
The new hub for Gemma, including
📚Inference and training guides
🧠 Deep visual architectural guides
🖼️ Online demos, even running models in the browser
And a lot more to come, check it out!
1
14
449
Imagine finding a photo of your cat by searching with their "Meow" 🙀
EmbeddingGemma 2 is a (very) small (740M), multimodal (text, code, images, video and audio) model you can use to search... pretty much anything. A photo, a sound, a line of code.
Btw: runs on phones / IoT 📱
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
It was so awesome to help kick off @hacktoberfest with @SwiftAlphaOne and talk all things @googlegemma and running open models ❤️. The opportunity in this space is really wide open and I'm super excited to see what people build 🙌 #hacktoberfest2026 #googlegemma
Hacktoberfest 2026 is officially live!
This year, we're focused on open-weight model and open-source AI. We've got over 300+ events, livestreams, challenges, and free swag for participants all month long.
Learn more about our vision and how the celebration has evolved: blog.mlh.com/everything-you-…
I use Colab daily and I can't believe I missed this announcement 😞. This alone is worth getting a Google AI plan for. An A100 with 80GB of RAM is enough to run a Gemma 4 31B in bf16 precision 🖐️🎤
Wow this is amazing 🔥
Google is now including Colab inside their Gemini plans... so you have access to premium GPUs!!
You can basically:
- Run any open source model
- Even image/audio/video gen
- Fine-tune a lot of OSS models
Everything on an expensive Nvidia GPU without having to rent it or own it. So many possibilities.
Time to build!
Me walking through town: There's a distinct difference in listening to music from the 80s vs music inspired by the 80s. The former is the nostalgic joy of something that has lasted the test of time, the later is an excitement about what is possible today, driven through nostalgia. I regularly flip between both perspectives.
3
2
419
Yes! Yes! Yes! Gemma models unlocked in Antigravity SDK 🔓 Got some interesting uses to try for sure 😎
Well that escalated quickly 👀
Amazing work by @mmastrac on the implementation and @dylayed on making it super accessible.
We live in an incredible age of rapid exploration!
What @mmastrac done to build out DiffusionGemma-Jev (djev) is amazing!
You can now deploy djev to Google Cloud Run and have your own Jev API-compatible endpoint to play with in just one gcloud commands:
github.com/taeold/djev-run
Costs ~$3/hr while active ($0 when idle)
Super fun way to get a glimpse of the future without spending $$$$ on a local Blackwell setup 😅
I use Gemini Live almost daily to brainstorm, make quick decisions, answer questions. The UX is what I love most about voice AI, especially when it gets me exactly what I'm after... i.e. A bacon roll 🥓
👇 Check out Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking to build your own experience.
➡️ Blog: blog.google/innovation-and-a…
➡️ Live API: ai.google.dev/gemini-api/doc…
1
1
428
As someone that now spends more time talking to agents then typing to them I say "Gemini, what can we do today?"
Say hello (literally) to Gemini 3.8 Live and 3.8 Live Extended Thinking, our new SOTA live audio models, available with frontier price + performance.
3.8 Live supports 97 languages (can seamlessly switch), async tool calls, and more!
1
5
538
SPAAAACEEE. Very cool to see Gemma models providing an important role in places where connectivity can't be relied upon. Can't wait to see what they do with the vision capabilities 🛰️
Running AI in space requires strict power and bandwidth management.
Satlyt is deploying Gemma directly on satellites to analyze telemetry and diagnose faults locally. Running Gemma 3 1B onboard cuts diagnostic data payloads by over 64%, turning raw system logs into compact summaries before they are sent back to Earth.
Next launch target: Gemma 4 E2B on NVIDIA Jetson Orin Nano to provide further reasoning and vision capabilities at the edge… of space.
1
8
535
The flywheel is spinning up. Very excited to see the improvements building momentum. 3.7 was a leap while maintaining speed (now my daily driver). Congrats to Gemini team on the pace of improvements 🚅
Two new Gemini models are here to help scale your AI agents and secure code:
🔘 3.8 Flash: our most intelligent model yet with significant gains from 3.7 Flash across software engineering, agentic tasks, and multi-step reasoning.
🔘 3.8 Flash Cyber: our most capable cybersecurity model with frontier-level vulnerability detection and automated patching.
1
10
385
Very proud of the whole Gemma team, past and present on this achievement, but mostly YOU THE COMMUNITY ♥️♥️♥️
Tonight we’re gathering in SF to celebrate 1 BILLION Gemma downloads!
While we can only fit a few of you in the room, we’ll be raising a glass to the millions of developers worldwide who are driving the Gemmaverse forward. 🥂
To celebrate, we rounded up some of the most out of this world ways you're using Gemma. From space operations to breakthroughs in medicine, plus a new GitHub repo to help you build!
Check it out:
github.com/google-gemma/awes…
"Dog" (via @GoogleAIStudio). That's it. That's all I asked it to build. Now I've got a Dogagotchi 🐶. Well played Gemini 3.7 Flash, well played👏
1
9
321
What's the shortest prompt you've tried @DynamicWebPaige @patloeber @thorwebdev @fofrAI @osanseviero @_philschmid
3
73
Gemini 3.7 Flash surprised me and I use it daily 😲 Wanted to see how it handled simple a prompt. "Frogger in three.js" gave me this in @GoogleAIStudio. Task comprehension and breakdown is what's nuts. Added SFX too. Gunna try more "minimal" prompts... for science of course 🧪
1
3
254
