Public large language models (LLMs) are typically safety-aligned during pretraining, but task-specific fine-tuning often erodes this alignment, introducing significant safety risks. Existing defenses are largely coupled with the fine-tuning process, incurring high computational overhead and complex workflows. To address these challenges, we propose Q-realign, a post-hoc defense based on quantization. Departing from the conventional view of quantization as a safety-eroding compromise, we reframe it as a dual-purpose mechanism for both compression and safety recovery. This allows Q-realign to decouple safety alignment from fine-tuning and naturally piggyback onto modern deployment pipelines. Experiments across multiple models and datasets demonstrate that our method substantially reduces unsafe behaviors while preserving task performance, significantly improving memory and GPU efficiency. Notably, our approach can recover the safety alignment of a fine-tuned 7B LLM on a single RTX 4090 within 40 minutes. Overall, our work provides a practical, turnkey solution for safety-aware deployment. Our code is released at https://anonymous.4open.science/r/Q-realign-copy-0D71.

Quantization reduces the deployment cost of large language models (LLMs). However, post-training quantization (PTQ) cannot fine-tune model parameters and often degrades accuracy in low-bit settings, while quantization-aware training (QAT) requires backpropagation and incurs high memory overhead. We propose ZeroQAT, a zeroth-order QAT framework supporting both weight and activation quantization. By using forward-only gradient estimation, ZeroQAT removes backpropagation and substantially reduces memory and computation while retaining end-to-end optimization benefits. We further introduce a lightweight variant of ZeroQAT for quantized fine-tuning, which freezes and pre-quantizes most parameters to further cut memory usage. Experiments show that ZeroQAT consistently outperforms representative PTQ and QAT baselines while requiring significantly less memory. For example, ZeroQAT enables fine-tuning of a 13B model at extremely low bit-widths (e.g., 2-4 bits) on a single 8GB GPU, and even allows fine-tuning a 6.7B model on a OnePlus 12 smartphone, demonstrating its practicality for end-to-end QAT on resource-limited edge devices.

Students’ wrong answers in science assessments can reveal plausible misconceptions, yet standard LLM benchmarks usually evaluate only answer correctness. We introduce MisconceptionBench, a benchmark for evaluating whether LLMs can reason about misconception-linked distractors and empirical student response patterns. It contains 761 K-12 science items, 1,474 expert-annotated distractors, and response distributions from more than 150,000 students. We evaluate eight LLMs on three tasks: misconception matching, misconception generation, and student response distribution prediction. In matching, text-similarity baselines remain competitive in two- and three-candidate settings; clear LLM gains emerge only with four candidates, where GPT-4o and Llama-3.3-70B significantly outperform TF-IDF and MiniLM. In generation, the strongest models are tightly clustered, with Hit@1 of 38.5-39.6%, while retrieval-only baselines recover many targets without an LLM. Response distribution prediction remains difficult: no model exceeds F1=0.21 for identifying grade-band differences. Overall, misconception-centered reasoning requires evaluation beyond answer accuracy.