LFM2.5-350M on Apple Silicon

Why activation quantization is harder than weight quantization (and what to do about it).

How to improve acceptance rate in speculative decoding?

Exploring Metal Performance Primitives on any M-series chip

Mirai raises $10M to build the on-device AI capability layer.

Mirai iOS SDK Public Preview

Part 4: Brief history of Apple ML Stack

Part 3: iPhone Hardware and How It Powers On-Device AI

Part 2: How to Understand On-Device AI

Part 1: Introduction to Deploying LLMs on Mobile