QuanLLM-v2.0-qm, QuanLLM Harness, and QMStudy: An Integrated System for Verifiable Quantum-Mechanics Learning
Chengtian Liang, Fan Yang, Jizhong Shen, Xuanting Hong, Qixuan Zhang, Tianrun Li, Xin Shao, Renxi Lu, Zihan Shen, Zhenghua Li, Yike Qian, Yixuan Lin
Source record
Source: Crossref
Published: Oct 10, 2026
DOI: 10.20944/preprints202610.0829.v1
Open original source ↗Source abstract
We present an integrated quantum-mechanics learning system comprising QuanLLM-v2.0-qm, the QuanLLM Harness verification framework, and the QMStudy interactive learning platform. QuanLLM-v2.0-qm adapts Qwen3-14B through rank-8 LoRA fine-tuning over three epochs, with a random ten-percent validation split, and supports explanatory reasoning and structured generation through an OpenAI-compatible interface. The Harness transforms generated answers into auditable candidates using isolated solvers, claim extraction, semantically reviewed tool calls, deterministic evidence, independent semantic checks, issue adjudication, and bounded recovery. Its v0.1.5 release separates domain-neutral orchestration from an explicitly activated quantum-mechanics teaching plugin. QMStudy combines a sixteen-lesson digital textbook, interactive numerical experiments, contextual AI tutoring, generated assessments, and a rule-driven learner-state engine. Assessment records connect knowledge-node performance to error tracking, review tasks, and learning reports, while streamed events expose longer verification runs. We describe both the independently released Harness and the locally adapted v0.1.3 implementation embedded in QMStudy v2.8. The Harness release record reports successful independent deterministic checks of delivered answers in ten reasoning-and-recovery reproductions, including three runs retaining degraded status after timeouts. The system establishes a modular path from domain-model adaptation to computationally checked explanations and persistent learning feedback. The visual summary below connects model generation, answer verification, and learner-state estimation.
Evidence graph
No public relationships recorded yet.
Integrity note: This page is a factual metadata record created by deterministic ingestion. It is not a claim that the work moves a mathematical frontier or has been independently verified.