AI model gets a 32-bit RISC-V CPU compiled directly into its weights, then runs DOOM on it
Qwenputer is an experimental blend of the stock Qwen 3.5 4B LLM with a 32-bit RISC-V CPU compiled and inserted directly into its weights.
AI engineer Divyansh Agrawal unveiled a groundbreaking blend of AI and computer architecture called the Qwenputer. This experimental system integrates a 4B Qwen AI language model with a 32-bit RISC-V CPU directly embedded within its weight structure. Agrawal demonstrated that the Qwenputer can execute arithmetic operations, code written in C, and even the game DOOM, albeit at a slow pace.
The RISC-V CPU is compiled into the model's weights, eliminating the need for external runtimes or tool use. Agrawal utilized Qwen's Gated DeltaNet (GND) state as RAM, a unique approach that he acknowledges isn't novel but is a first implementation of this idea. This CPU-based implementation is 11 transformer layers ahead of Qwen's own architecture, with Agrawal boasting that it passes official riscv-tests, matches qemu, and has run 18 million instructions without any errors.
The project also introduces a filesystem storage system using DeltaNet state, capable of holding up to 512 files. While the Qwenputer effectively improves computational accuracy within LLMs, it has doubled the size of the original 4B Qwen model to nearly 19GB. Agrawal anticipates that more streamlined coding and speed improvements will be forthcoming.
The project, marked as experimental, showcases Agrawal's exploratory approach rather than solving a specific problem. For those interested in exploring this innovative project, the MIT-licensed code is available on GitHub and the model can be accessed directly from Hugging Face.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
