llm-architecture.

A RISC-V CPU Compiled Into Qwen 3.5 Weights

Qwenputer embeds a working 32-bit CPU in transformer weights using GDN layers as RAM. No external runtime needed. What this means for AI systems.