Researcher divagr18 has embedded an NPU capable of running Andrej Karpathy's stories260k LLM within the weights of Qwen 3.5-4B, following their prior work implementing a RISC-V CPU in the model. The stories260k model was compiled as a C program to execute on this internal neural processing unit, demonstrating nested model execution. Code and weights are available on GitHub and Hugging Face under the QwenPU project.
Read original
reddit/r/LocalLLaMA