#litert

Three chips can run the same LLM, but it takes two files - one shared by CPU and GPU, one just for the NPU. Understanding why took me into graphs, subgraphs, and the difference between just-in-time and ahead-of-time compilation - the layer LiteRT calls delegates.

A technical walkthrough of “Which Celebrity Vegetable Are You?” — private WebAI inference with...

Running Gemma 4 On-Device on Android: Building a Private AI Assistant with LiteRT I built...

The Real Problem: On-Device AI Fragmentation and Bottlenecks For years, the promise of...

The landscape of client-side machine learning is shifting. For years, deploying high-performance...