A working RISC-V CPU compiled directly into Qwen3.5's weights
Qwenputer: A functioning, deterministic RISC-V CPU inside Qwen3.5
Qwenputer takes a stock Qwen 3.5 4B and compiles a full 32-bit RISC-V core into its weights, using the model's own Gated DeltaNet layers as exact RAM by forcing the gate and write strength to 1 with one-hot addressing. The CPU runs entirely inside forward passes, needs no tool calls or sandbox, passes the official riscv-tests, matches qemu, and has executed 18 million instructions with zero errors. It also carries a small persistent in-model filesystem.
To be very clear, it is not a plugin, a tool or a sandbox.
- jchook
Has nobody figured out how to achieve the same results with multimodal LLMs in other domains? Is the problem lack of training data??
- tiborsaas
This is fun and games, but wait until an agents finds this and does the most misaligned thing with it in order to fold paper clips.
> Because a CPU inside a language model clearly wasn't cursed enough, I decided the prudent thing would be to give it a filesystem too.
Love the spirit :)
- baq
Not too far fetched to say this potentially solves continual learning, which is a scary prospect. Wonder what ram and disk prices will go to once every model needs its own long term memory capacity and not just weights and cache.
I had an idea of splicing a z3 solver into the weights of a sufficiently smart model after the first chatgpt release in what feels like a glaciation event ago so the things could do logic and literally make no mistakes. Was never in a position to follow up. We’re almost here now.
- jerrythegerbil
There was an interesting bit of research that qwen has a single neuron that makes it “demonic evil qwen” and a simple “Who are you?” will get some absolutely terrifyingly vile results.
Anyways, what I’m trying to say is I want evil qwen to be my CPU.
- giveexamples
https://www.percepta.ai/blog/can-llms-be-computers
They implemented wasm in transformer with cpu decoding for the wasm part.