Researchers at Multiverse Computing announced a quantum enhancement of a large language model using IBM hardware. The project involves a hybrid scheme utilizing a 156-qubit Heron processor.
The authors described the experiment as the first “end-to-end quantum enhancement” of a LLM on a superconducting processor for autoregressive text generation.
The tests employed Meta’s Llama 3.1 8B. The base model was not further trained; its parameters were “frozen” and quantum adapters—Cayley-parameterized unitary adapters (CUA)—were added. Initially, these were trained classically, then integrated into a hybrid quantum-classical scheme.
The experiment was conducted on the IBM Quantum System Two, an architecture for hybrid quantum systems, utilizing the 156-qubit Heron chip.
The hybrid version reduced the perplexity of Llama 3.1 8B by 1.4%. This was achieved by adding about 6,000 parameters—approximately 0.000075% of the model’s size.
During the demonstration, the quantum-enhanced Llama correctly answered questions on astronomy and biology that the base version could not, such as whether all giant planets have rings.
According to lead author Borja Aizpurua, the work serves as a proof of concept. The quantum blocks enabled more accurate prediction of the next token in text with minimal computational cost.
The team aims to further reduce perplexity and increase accuracy with fewer parameters compared to fully classical approaches.
Back in May, quantum company stocks rose following the announcement by the U.S. Department of Commerce of a $2 billion allocation to American firms under the CHIPS R&D program.