Qelm
Year: 2025
Type: article
Abstract: QELM is an end-to-end quantum-hybrid language modeling stack. It introduces a unified tokenizer with “sub-bit” encoding that maps text features into amplitude/phase pairs, and couples this to transformer-style blocks built from quantum layers that run on simulators and real QPUs. The framework supports multiple channels—standard qubit, MBQC/cluster, analog Hamiltonian evolution, QSVT-style polynomial layers, quantum walks, and tensor-network linear maps—with graceful fallbacks. Training uses parameter-shift gradients with a persistent multiprocess executor; search includes Grover-assisted routines. Built-in error mitigation (Pauli twirling, ZNE, dynamic decoupling), a GUI, and reproducibility scripts (perplexity/BLEU) are included. The paper documents design choices, ablations (sub-bit on/off, simulator vs. QPU, mitigation on/off), and hardware runs—positioning QELM as a practical route to quantum-assisted NLP.Code: https://github.com/R-D-BioTech-Alaska/Qelm • Project: https://qelm.org
Author: Carter, Brenton
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7077520867
FWCI:
Cites:
Cited by:
Related to: