Multiverse Computing ran a pretrained Llama model through quantum circuit blocks on IBM’s 156-qubit machine and cut its perplexity while adding almost no parameters — the first end-to-end “quantum enhancement” of a production LLM.