Large Language Models (LLMs) face deployment challenges due to high computational costs, and while Post-Training Quantization (PTQ) offers a solution, existing rotation-based methods struggle at very low bit-widths like 2-bit.We introduce a novel, training-free approach to construct an improved rotation matrix, addressing the limitations of current methods.The key contributions include leveraging the Walsh-Hadamard transform with sequency ordering, which clusters similar frequency components to reduce quantization error compared to standard Hadamard matrices, significantly improving performance.Furthermore, we propose a Grouped Sequencyarranged Rotation (GSR) using block-diagonal matrices with smaller Walsh blocks, effectively isolating outlier impacts and achieving performance comparable to optimization-based methods without requiring any training.Our method demonstrates robust performance on reasoning tasks and Perplexity (PPL) score on WikiText-2.Our method also enhances results even when applied over existing learned rotation techniques.
Large Language Models (LLMs) face deployment challenges due to high computational costs, and while Post-Training Quantization (PTQ) offers a solution, existing rotation-based methods struggle at very low bit-widths like 2-bit.We introduce a novel, training-free approach to construct an improved rotation matrix, addressing the limitations of current methods.The key contributions include leveraging the Walsh-Hadamard transform with sequency ordering, which clusters similar frequency components to reduce quantization error compared to standard Hadamard matrices, significantly improving performance.Furthermore, we propose a Grouped Sequencyarranged Rotation (GSR) using block-diagonal matrices with smaller Walsh blocks, effectively isolating outlier impacts and achieving performance comparable to optimization-based methods without requiring any training.Our method demonstrates robust performance on reasoning tasks and Perplexity (PPL) score on WikiText-2.Our method also enhances results even when applied over existing learned rotation techniques.
Rotating the activation and weight matrices to reduce the influence of outliers in large language models (LLMs) has recently attracted significant attention, particularly in the context of model quantization. Prior studies have shown that in low-precision quantization scenarios, such as 4-bit weights and 4-bit activations (W4A4), randomized Hadamard transforms can achieve significantly higher accuracy than randomized orthogonal transforms. Notably, the reason behind this phenomenon remains unknown. In this paper, we find that these transformations show substantial improvement in eliminating outliers for common tokens and achieve similar quantization error. The primary reason for the accuracy difference lies in the fact that randomized Hadamard transforms can slightly reduce the quantization error for tokens with massive activations while randomized orthogonal transforms increase the quantization error. Due to the extreme rarity of these tokens and their critical impact on model accuracy, we consider this a long-tail optimization problem, and therefore construct a simple yet effective method: a weighted loss function. Additionally, we propose an optimization strategy for the rotation matrix that involves alternating optimization of quantization parameters while employing orthogonal Procrustes transforms to refine the rotation matrix. This makes the distribution of the rotated activation values more conducive to quantization, especially for tokens with massive activations. Our method enhances the Rotated LLMs by achieving dual free, Outlier-Free and Massive Activation-Free, dubbed as DFRot. Extensive experiments demonstrate the effectiveness and efficiency of DFRot. By tuning the rotation matrix using just a single sample, DFRot achieves a perplexity improvement of 0.98 and 0.95 on W4A4KV4 and W4A4KV16, respectively, for LLaMA3-70B, a model known for its quantization challenges.
Recent post-training quantization (PTQ) methods have adopted block rotations to diffuse outliers prior to rounding. While this reduces the overhead of online full-vector rotations, the effect of block structure on outlier suppression remains poorly understood. To fill this gap, we present the first systematic, non-asymptotic analysis of outlier suppression for block Hadamard rotations. Our analysis reveals that outlier suppression is fundamentally limited by the geometry of the input vector. In particular, in the deterministic worst case, post-rotation outliers are minimized when the pre-rotation ℓ1 norm mass is evenly distributed across blocks. Guided by these insights, we introduce PeRQ (Permute, Rotate, then Quantize), a PTQ framework that redistributes activation mass via permutations prior to rotation. We propose a greedy mass diffusion algorithm to calibrate permutations by equalizing the expected blockwise ℓ1 norms. To avoid adding inference overhead, we identify permutation-equivariant regions in transformer architectures to merge these permutations into model weights before deployment. Experiments show that PeRQ consistently improves accuracy across all block sizes, recovering up to 90% of the full-vector rotation perplexity when quantizing Llama3 1B to INT4 with block size 16, compared to 46% without permutations.
Post-training quantization (PTQ) is widely used to compress large language models without retraining. However, many existing weight-only methods rely on heuristic objectives and greedy rounding strategies, leading to noticeable degradation under low-bit quantization. We propose OJBKQ, a layer-wise PTQ method that formulates weight quantization as a joint optimization problem over activations and weights, yielding a multiple-right-hand-side box-constrained integer least squares (BILS) problem per layer. To solve this NP-hard problem, we extend Klein’s randomized variant of Babai’s nearest-plane algorithm to handle box constraints and run the extended algorithm K times to generate K box-constrained Klein points. The minimum-residual candidate among these points and the box-constrained Babai point is selected as a suboptimal solution. Experiments on large language models show that OJBKQ achieves lower perplexity at 3–4 bits than existing PTQ methods, while maintaining comparable computational cost.
The growing need for privacy protection, instant response, and offline access has pushed large language model inference from the cloud to local edge devices. However, decoder-only LLMs remain hard to run on smartphones and embedded systems, as their performance is limited not just by parameter count, but also by memory bandwidth, hardware operator compatibility, and the rising cost of long-context generation. Post-training quantization (PTQ) is the most viable approach today, as it shrinks model size and speeds up inference without full retraining. Yet most PTQ methods are tested on cloud GPUs, and their claimed improvements rarely hold up on real edge hardware. This survey reviews on-device PTQ for decoder-only LLMs from a hardware-focused angle, covering static and adaptive schemes, key hardware bottlenecks, evaluation flaws, and KV cache scaling issues. It concludes that model size alone is a poor measure of edge efficiency, and future work must better align quantization algorithms, runtime backends, and on-device evaluation standards.
The best two-bit weight quantizers for large language models, such as QTIP and Proteus, rotate each weight matrix by a random orthogonal transform, which must be undone at every decoding step, then encode it with a trellis or lattice code under a Euclidean search; the layer Hessian enters only through error feedback between coding blocks. We show that this leaves part of the Hessian unused. Error feedback turns the loss into a weighted sum of per-coordinate rounding errors whose weights, the diagonal of the Hessian's LDL factorization, existing quantizers compute but never read. We put these weights into the Viterbi branch metric, so the search follows the curvature within each coding block. This also explains the rotation: it removes this within-block variation, so weighting in the native basis and rotating are substitutes. On three models the weighted native search matches a full-dimension randomized Hadamard to within about one point of downstream accuracy, and weighting after the rotation gains little. Around this search we build CurveTQ, a trellis codec with no rotation, which handles the weights' amplitude and marginal shape with a factored scale field and a closed-form quantile table, and stores a start state per coding block so the trellis can adapt to the residual that error feedback carries into it. At two bits CurveTQ is 1-3 points higher in mean downstream accuracy than QTIP and Proteus on three 4-8B Instruct models, even after both are given our start state, which alone lifts either baseline by 1-3 points. It also leads on a 35B mixture of experts, to our knowledge the first trellis-coded result on such a model. With no rotation to undo, our decoder is the fastest of the three at all tested batch sizes and bit widths.
Fine-tuning Low-bit models aims to adapt a quantized model while keeping the final deployed checkpoint in the same low-bit form. This setting is practically important as it reduces memory and inference cost for storage and deployment. Under this constraint, adaptation becomes an optimization problem over quantization codes and scales. Existing continuous low-bit training is efficient, but it can be distorted by straight through estimation error or by post-quantize gap; discrete search is deployment-faithful, but it is often too inefficient under a finite training budget. We propose code surrogate gradient as the first order signal in deployable code space to acceleate optimization, and performing guided search to preserve deployment faithfulness. Experiments across arithmetic reasoning, instruction following, and structured language understanding show that GradCodes consistently improves fine-tuning low-bit models across different quantization datatypes. Code is provided at https://github.com/ovo67/GradCodes.