cs.LGNov 17, 2026

BQ-LoRA: Binary-Quantized Low-Rank Adapters as Implicit Regularizers for Parameter-Efficient Fine-Tuning

Authors: Juyoung Park, Heejun Lee

Organizations: Independent Researcher · KAIST, DeepAuto.ai

Abstract

Parameter-efficient fine-tuning of quantized large language models is dominated by QLoRA, which combines 4-bit weight quantization with FP16 low-rank adapters. We ask whether the FP16 adapter precision is necessary. Replacing it with a single binary bit per adapter parameter, combined with a learned per-layer scale ss and a Scaled Straight-Through Estimator, reduces adapter storage by 13-16x. We introduce two variants -- NormScale, which binarizes under a norm-preserving scale with a clipped STE, and LearnThresh, which makes the binarization threshold learnable -- unified under a framework we call BQ-LoRA. Our main theoretical contribution is a tight characterization of the BQ-LoRA hypothesis class: it has bounded entrywise norm, a typical-set spectral norm bound that is Θ(r)Θ(r) times sharper than the worst case, a full-rank, near-isotropic update, and consequently a Rademacher complexity that yields an explicit O~(smax⁡αdout/(rn))\tilde{O}(s_{\max}α\sqrt{d_{out}/(rn)}) generalization bound. FP16 LoRA has no comparable a priori norm or isotropy control. On Qwen2.5 (1.5B-72B), BQ-LoRA exceeds QLoRA's benchmark averages at every scale by 0.8-3.5 points; on LLaMA-2 (7B/70B) it remains competitive at the same storage reduction but does not uniformly lead the strongest FP16-adapter baselines (LoftQ at 7B, QLoRA at 70B). On Qwen2.5-7B the binary adapter preserves the 4-bit base WikiText-2 PPL within 0.06 (9.86 vs. 9.80) while QLoRA worsens it by 0.59. The reduction applies to inference-time adapter storage and serving; training memory matches QLoRA and step time rises by under 4%. We translate the storage saving into a deployment study, BQ-SGMV, by extending Punica's SGMV kernel with a binary inner-product primitive: the same A100-80GB holds 15.6x more adapters in HBM, and our research prototype eliminates cache misses under a skewed synthetic workload.

Figures & tables

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

May 22, 2026

Signs Beat Floats: Low-Rank Double-Binary Adaptation for On-Device Fine-Tuning

On-device adaptation of large language models commonly keeps a quantized base model frozen while training and deploying a small, task-specific LoRA adapter. In the unmerged adapter-mode setting, however, the adapter is more than a compact storage module; it introduces an additional dense floating-point branch, maintains a trainable state for local updates, and acts as a unit of communication and hot-swapping.We introduce LoRDBA, a LoRA-compatible adapter that replaces both low-rank factors with binary sign carriers while representing magnitudes through lightweight, channel-wise scales, converting the dense adapter branch into two sign-accumulation matrix multiplications interleaved with channel-wise scaling. A finite-sample analysis shows that reconstruction quality is governed by the residual-to-magnitude ratio of the original LoRA factors. In adapter-mode experiments, LoRDBA outperforms low-bit baselines at matched model sizes while matching fp16 LoRA quality in selected regimes. The unmerged adapter incurs at most 8% prefill latency overhead at matched rank r=16 despite an over 10x reduction in adapter footprint, with moderate training memory overhead of approximately 1.6x that of fp16 LoRA.
May 30, 2026

ProjQ: Project-and-Quantize for Adapter-Aware LLM Compression

Post-Training Quantization (PTQ) and Low-Rank Adaptation (LoRA) constitute the standard pipeline for efficient Large Language Model (LLM) deployment. However, applying them sequentially poses a problem: PTQ often leaves behind random noise that is spread out (across the model's weights) in a way LoRA can't easily fix, meaning that LoRA ends up wasting its limited capacity trying to fix uncorrectable noise instead of improving task performance. In this paper, we propose ProjQ, a novel framework for constraining quantization noise to the low-rank manifold via orthogonal subspace projection. We derive an efficient alternating algorithm that shapes the quantization noise into a low-rank structure, effectively offloading dominant error components to the subsequent adapter while minimizing the residual error in the orthogonal "uncorrectable" subspace. Our theoretical analysis demonstrates that ProjQ preserves strictly greater model plasticity for downstream tasks compared to standard PTQ. Extensive experiments on LLaMA-2, Qwen2.5 and Qwen3 confirm that ProjQ consistently outperforms existing methods in both quantization error compensation and downstream task fine-tuning, achieving up to 2×2\times lower evaluation loss for compensation and matching the performance of standard 4-bit baselines on language modeling tasks with only 3 bits. The code is available on https://github.com/yy9301/ProjQ .
Aug 24, 2026

AQLoRA: A Zero-Search Recipe for Fast Quantized LoRA Fine-Tuning

Quantized fine-tuning (QLoRA) saves memory but not time. It dequantizes every 4-bit weight on the fly, so it trains more slowly than fp16 LoRA. We present AQLoRA (Adaptive-Quantization LoRA), a recipe that buys part of that time back. One CPU pass ranks layers by NF4 reconstruction error, with no search and no calibration data, and keeps a fixed fraction of them in fp16. Those layers skip dequantization, which is where the speed comes from. A quality setting adapts every layer. A speed setting adapts only the top blocks, so the backward pass stops early. The selection matches the accuracy of Unsloth's hand-curated dynamic-4bit list. We evaluate on Commonsense-170K across six models and four architecture families, from 1.4B to 14B. The speed setting trains 11.3 +/- 2.3% faster than well-tuned QLoRA and gives up 0.1 to 1.5 accuracy points. It was faster in all thirteen independent timing sessions, at worst by 7%. The quality setting trains 4.8 +/- 2.0% faster, matches QLoRA's accuracy on every model, and is within a point of fp16 LoRA. The memory premium depends on the architecture: 0.06 to 0.8 GiB on the 1-4B models, 0.5 GiB at 7B, 1.9 GiB at 14B. These error bars are measured between independent sessions, not within one. Timing on shared hardware needs three rules: fix the measurement duration, not the step count; measure the noise floor from a duplicated arm; and repeat whole sessions, since on one model three sessions that each passed the floor test spread over 8.2 points. We validate the recipe with controls and report the two that failed. Choosing adapter layers by weight density is no better than random. Choosing protected layers by quantization error is not either, on five models at three budgets with memory-matched controls and downstream on two. The number of protected layers sets the speed; which layers they are does not measurably change accuracy.