Running Gemma on Mac and Windows PCs with Ollama
summary
Demonstration shows Gemma 3 on Ollama running locally and in serverless Google Cloud Run, with Ollama claiming ~150 million model downloads and tens of millions of developers. All four parameter sizes are available: the 1B model runs with 2GB VRAM (very fast on M4 MacBooks), the 4B handles multimodal image and handwriting OCR reliably in the demo, and the 27B runs on medium‑high MacBooks or consumer GPUs; Cloud Run scales 0→hundreds of GPUs with sub‑5s cold GPU start.
AI-written summary · about
chapters (6)
- 0:00Introduction to Ollama
- 1:25Ollama engine internals
- 2:26Gemma 3 demo
- 3:53Multimodal capabilities
- 5:50Model variety and API
- 7:42Cloud Run and summary
related talks
12mDemo: Rapid prototyping with Gemma and Llama.cpp
16mGemma 4: Compacting Intelligence for the Edge
08mDemo: Gemma on-device with MediaPipe and TensorFlow Lite
24mAccelerating AI on Edge
05mBuild your own LLM on Cloud
19mGemma 4 Deep Dive
11mRunning LLMs on your iPhone: 40 tok/s Gemma 4 with MLX
35mLocal LLMs on low end hardware - a practical perspective