How might you adapt quantization techniques like QLoRA to optimize physics-specific model inference?

I appreciate the question, but I need to be direct: these preprints don't really address QLoRA or quantization techniques applied to physics-specific model inference.

Here's what's actually in the archive:

What you might expect would be papers showing how to compress physics models using low-rank quantization (like QLoRA does for language models), or applying bit-width reduction to equations-of-motion solvers or field theory simulations. That's not here.

What is here instead falls into three separate categories:

1. Quantum-inspired tensor methods [AG-2024.05-2657] describe QuanTA, which uses quantum circuit structures to improve fine-tuning—but this is about *high-rank* adaptation (opposite of LoRA's low-rank constraint) and targets language models, not physics inference.

2. Physics-domain model training [AG-2026.04-892] shows how to fine-tune small reasoning models on Quantum Field Theory problems, but doesn't discuss quantization or compression techniques during inference.

3. Actual quantum machine learning [AG-2024.05-2251, AG-2024.04-2277, AG-2025.10-1352] explores hybrid quantum-classical approaches and low-parameter quantum circuits for physics tasks—but these are fundamentally different from classical quantization methods like QLoRA.

If you're asking how to adapt QLoRA-style parameter reduction to accelerate classical physics simulations (say, surrogate models for detector response or flow fields), that gap isn't covered here.