TLMs: Tiny LLMs and Agents on Edge Devices with @cormacb Function Gemma ships at 270 million parameters and runs nearly 2,000 tokens per second pre…

Google's Function Gemma (270M parameters) achieves 2,000 tokens/second on Pixel 7, reaches 46% accuracy on app intents out-of-box and 90% after fine-tuning on synthetic data, enabling on-device LLM agents.