How fast is #gemma3 270M out of the box on a @raspberrypi.com 5? About 30 tok/s on CPU for the Q4_0 model using #ollama. Tried using #llamacpp, got about 32 tok/s. 🤯 Extremely promising for edge compute with no accelerator! Model is only ~250Mb on disk. Can't wait to fine-tune for my IoT projects 👨‍💻

Great news. We have a new position on the team for a Developer Advocate at the intersection of Web, Chrome, ChromeOS and Android. The position is in Waterloo - Canada www.google.com/about/career...

Understanding an LLM's ability to perform real-world tasks is going to be a much better method of "usefulness". I applaud efforts that attempt to quantify this.

It's been a hot minute since I wrote something on @hacksterio.bsky.social, but today I put together a quick pro-tip article on connecting an #esp32 #esp8266 to the #Gemini API. www.hackster.io/PaulTR/conne...

The price <> capability relationship of Gemini 2.0 models is really exciting right now. Using aistudio.google.com to test the models, then turn those experiments into production apps is much more feasible from a cost basis. ✨💖

Just saw a comment saying that model comparisons should only be objective and based on research papers. Completely disagree. Have you ever made a decision based on how you feel? Yeah, thought so.

So today's the day I move to Google DeepMind. It's still mind bogging that I'm going to work with the amazing people there to bring life-changing AI research to developers and users. I'm beyond excited to help carve out a future that can really benefit society! 🤩 Let's go! ✨🚀❤️ deepmind.google

Using LoRA to customize a model like Gemma 2 is a brilliant strategy for model deployment! You get a range of specialist behaviours, while only needing to deploy a single model + N sets of (much smaller) LoRA weights. 🚢=💎+🏋️

STOP THE PRESS!! It has come to my attention that (some) Costa Express machines at petrol stations now sell Tea. Non-UK folks: This news is as big as a nation collectively winning the superbowl.

👩‍💻 Did you know that you can use Google AI Studio's Gemini 2.0 Flash and Gemini 2.0 Thinking Mode models directly in Cursor? To customize, all you have to do is modify the selected models in your settings: ⚙️ Settings ➡️ Models ➡️ Model Names

Wrote up my initial impressions of the new Google Gemini 2.0 Flash model - it's really good, and the streaming mode (where you can stream video and audio to it and get audio streamed right back) is pure science-fiction simonwillison.net/2024/Dec/11/...

🎅 Love that developers are already starting to build with Gemini 2.0 Flash! This project leverages AI + web automation to create an agent capable of navigating and interacting w Instacart The agent is designed to help users efficiently order Christmas-themed groceries. 🔗 github.com/peytoncasper...