LLM Quantization

LLM: Large Language Model

Overview

LLM quantization studies how to represent model weights, activations, and attention key-value caches with fewer bits while preserving useful behavior and making deployment more efficient. Post-training methods such as GPTQ and AWQ use calibration data to round weights, while quantization-aware training, mixed-precision allocation, rotations, and outlier handling aim to reduce errors that accumulate across layers. Work also co-designs number formats and GPU, CPU, and edge-device kernels, evaluating trade-offs with language-model perplexity, zero-shot and reasoning benchmarks, memory, latency, and energy. Four-bit inference is often practical, but performance at very low bit widths remains model- and task-dependent, and an open challenge is ensuring that gains in aggregate accuracy do not hide losses in reasoning, safety, fairness, or uncertainty.

Latest papers

443 papers68 papers in the last four weeks, down 13% on the four weeks before. 0.5% of all new papers.

Apr 5, 2026

RUQuant: Towards Refining Uniform Quantization for Large Language Models

The increasing size and complexity of large language models (LLMs) have raised significant challenges in deployment efficiency, particularly under resource constraints. Post-training quantization (PTQ) has emerged as a practical solution by compressing models without requiring retraining. While existing methods focus on uniform quantization schemes for both weights and activations, they often suffer from substantial accuracy degradation due to the non-uniform nature of activation distributions. In this work, we revisit the activation quantization problem from a theoretical perspective grounded in the Lloyd-Max optimality conditions. We identify the core issue as the non-uniform distribution of activations within the quantization interval, which causes the optimal quantization point under the Lloyd-Max criterion to shift away from the midpoint of the interval. To address this issue, we propose a two-stage orthogonal transformation method, RUQuant. In the first stage, activations are divided into blocks. Each block is mapped to uniformly sampled target vectors using composite orthogonal matrices, which are constructed from Householder reflections and Givens rotations. In the second stage, a global Householder reflection is fine-tuned to further minimize quantization error using Transformer output discrepancies. Empirical results show that our method achieves near-optimal quantization performance without requiring model fine-tuning: RUQuant achieves 99.8% of full-precision accuracy with W6A6 and 97% with W4A4 quantization for a 13B LLM, within approximately one minute. A fine-tuned variant yields even higher accuracy, demonstrating the effectiveness and scalability of our approach.
Apr 4, 2026

Improving Quantized Model Performance in Qualitative Analysis with Multi-Pass Prompt Verification

Quantized Large Language Models (LLMs) are used more often in qualitative analysis because they run fast and need fewer computing resources. This study examines how different lower bits quantization levels (8-bit, 4-bit, 3-bit, and 2-bit) and quantization types affect the performance of LLaMA-3.1 (8B) on qualitative analysis. The study uses expert and non-expert responses from 82 interview transcripts. Low-bit models often produce higher levels of hallucinations and unstable results, especially when reading non-expert language with unclear terms. To improve performance, we propose a quantization-aware multi-pass prompt verification method. This method guides the model through controlled steps that reduce hallucinations. It removes unreliable content and passes the results to the next transcript after verification, improving accuracy. To validate performance, human coders analyzed transcripts using NVivo and BF16 LLaMA. BF16 LLaMA-3.1 produced high-precision output but had semantic drift and hallucination. These errors were corrected manually. The corrected BF16 output and NVivo human coding were combined to create a gold-standard ground truth (GSGT) for thematic extraction and frequency analysis. The results show that 8-bit models stay closest to the GSGT. The 4-bit models lose accuracy but become stable when the proposed method is applied. The 3-bit and 2-bit models drop in performance because of heavy compression, but they improve with the proposed prompt design and verification. The study also finds that models at the same bit level behave differently depending on quantization type. Overall, the method helps low-resource LLMs become more stable, accurate, and suitable for qualitative research at lower cost.
Apr 2, 2026

Fast NF4 Dequantization Kernels for Large Language Model Inference

Large language models (LLMs) have grown beyond the memory capacity of single GPU devices, necessitating quantization techniques for practical deployment. While NF4 (4-bit NormalFloat) quantization enables 4×\times memory reduction, inference on current NVIDIA GPUs (e.g., Ampere A100) requires expensive dequantization back to FP16 format, creating a critical performance bottleneck. This paper presents a lightweight shared memory optimization that addresses this gap through principled memory hierarchy exploitation while maintaining full ecosystem compatibility. We compare our technique against the open-source BitsAndBytes implementation, achieving 2.0--2.2×\times kernel speedup across three models (Gemma 27B, Qwen3 32B, and Llama3.3 70B) and up to 1.54×\times end-to-end improvement by leveraging the 12--15×\times latency advantage of shared memory over global memory access. Our optimization reduces instruction counts through simplified indexing logic while using only 64 bytes of shared memory per thread block, demonstrating that lightweight optimizations can deliver substantial performance gains with minimal engineering effort. This work provides a plug-and-play solution for the HuggingFace ecosystem that democratizes access to advanced models on existing GPU infrastructure.
Apr 2, 2026

AdaHOP: Fast and Accurate Low-Precision Training via Outlier-Pattern-Aware Rotation

Hadamard transforms have become a key tool for stabilizing low-precision training, but existing methods apply them uniformly across tensors and computation paths. We show that this one-size-fits-all strategy is inherently limited: Hadamard smoothing reduces quantization error only when its direction is properly aligned with the operand's outlier structure. Through a systematic study of weights, activations, and gradients in LLM training, we identify three stable outlier patterns, Row-wise, Column-wise, and None, and show that each outlier pattern pair in matrix multiplication requires a distinct transform or outlier-handling strategy. We propose AdaHOP, Adaptive Hadamard transform with Outlier-Pattern-aware strategy, which applies Inner Hadamard Transform (IHT) when inner-dimension mixing properly suppresses the operands' outliers, and selectively applies Outlier Extraction (OE) that extracts dominant outlier rows or columns into a high-precision path when it does not. With fused, hardware-aware Triton kernels, AdaHOP enables training from scratch at MXFP4 precision with BF16-level quality, while achieving up to 3.6X memory compression, 1.46X end-to-end training speedup over BF16.
Mar 30, 2026

PolarQuant: Optimal Gaussian Weight Quantization via Hadamard Rotation for LLM Compression

We present PolarQuant, a post-training weight quantization method for large language models (LLMs) that exploits the distributional structure of neural network weights to achieve near-lossless compression. PolarQuant operates in three stages: (1) block-wise normalization to the unit hypersphere, (2) Walsh-Hadamard rotation to transform coordinates into approximately Gaussian random variables, and (3) quantization with centroids matched to the Gaussian distribution. Our ablation reveals that Hadamard rotation alone accounts for 98% of the quality improvement, reducing Qwen3.5-9B perplexity from 6.90 (absmax Q5) to 6.40 (Delta = +0.03 from FP16), making it practically lossless without any calibration data. Furthermore, PolarQuant functions as an effective preprocessing step for downstream INT4 quantizers: PolarQuant Q5 dequantized and re-quantized by torchao INT4 achieves perplexity 6.56 versus 6.68 for direct absmax INT4, while maintaining 43.1 tok/s throughput at 6.5 GB VRAM. Code and models are publicly available.
Mar 30, 2026

Adaptive Block-Scaled Data Types

NVFP4 has grown increasingly popular as a 4-bit format for quantizing large language models due to its hardware support and its ability to retain useful information with relatively few bits per parameter. However, the format is not without limitations: recent work has shown that NVFP4 suffers from its error distribution, resulting in large amounts of quantization error on near-maximal values in each group of 16 values. In this work, we leverage this insight to design new Adaptive Block-Scaled Data Types that can adapt to the distribution of their input values. For four-bit quantization, our proposed IF4 (Int/Float 4) data type selects between FP4 and INT4 representations for each group of 16 values, which are then scaled by an E4M3 scale factor as is done with NVFP4. The selected data type is denoted using the scale factor's sign bit, which is currently unused in NVFP4, and we apply the same insight to design formats for other bit-widths, including IF3 and IF6. When used to quantize language models, we find that IF4 outperforms existing 4-bit block-scaled formats, achieving lower loss during quantized training and achieving higher accuracy on many tasks in post-training quantization. We additionally design and evaluate an IF4 Multiply-Accumulate (MAC) unit to demonstrate that IF4 can be implemented efficiently in next-generation hardware accelerators. Our code is available at https://github.com/mit-han-lab/fouroversix.
Mar 30, 2026

ITQ3_S: High-Fidelity 3-bit LLM Inference via Interleaved Ternary Quantization with Rotation-Domain Smoothing

We present ITQ3_S (Interleaved Ternary Quantization -- Specialized), a novel 3-bit weight quantization format for LLMs integrating TurboQuant (TQ), a rotation-domain strategy based on the Fast Walsh-Hadamard Transform (FWHT). Conventional 3-bit methods suffer precision loss from heavy-tailed weight distributions and inter-channel outliers. ITQ3_S pre-rotates the weight space via FWHT before quantization, spreading outlier energy across the vector and inducing a near-Gaussian distribution amenable to uniform ternary coding. We derive a rigorous dequantization procedure fusing a 256-point Inverse FWHT into the CUDA shared-memory loading stage, ensuring reconstruction error is bounded exclusively by the ternary quantization grid with no additional error from the transform inversion. For any weight vector w∈R256\mathbf{w} \in \mathbb{R}^{256}, the reconstruction satisfies ∥w^−w∥2≤εq\|\hat{\mathbf{w}} - \mathbf{w}\|_2 \leq ε_q, strictly smaller than uniform 3-bit baselines that do not exploit rotation-induced distribution normalization. TurboQuant lacks a native CUDA kernel, precluding direct deployment; naively composing TQ with existing weight quantizers introduces domain mismatch errors that accumulate across layers, degrading quality below standard 3-bit baselines. ITQ3_S resolves this by co-designing the FWHT rotation and quantization kernel as a unified pipeline grounded in the IQ3_S weight format, with the inverse transform fused into the CUDA MMQ kernel. Empirically, on the NVIDIA RTX 5090 (Blackwell), ITQ3_S achieves perplexity competitive with FP16 while delivering throughput exceeding 1.5x that of 4-bit alternatives via optimized DP4A and Tensor Core scheduling. Our results establish ITQ3_S as a practical, mathematically grounded solution for high-fidelity LLM deployment on consumer hardware.
Mar 26, 2026

GlowQ: Group-Shared LOw-Rank Approximation for Quantized LLMs

Quantization techniques such as BitsAndBytes, AWQ, and GPTQ are widely used as a standard method in deploying large language models but often degrades accuracy when using low-bit representations, e.g., 4 bits. Low-rank correction methods (e.g., LQER, QERA, ASER) has been proposed to mitigate this issue, however, they restore all layers and insert error-correction modules into every decoder block, which increases latency and memory overhead. To address this limitation, we propose GlowQ, a group-shared low-rank approximation for quantized LLMs that caches a single shared right factor per input-sharing group and restores only the groups or layers that yield the highest accuracy benefit. GlowQ computes the high-precision projection once per input-sharing group and reuses it across its modules, reducing parameter and memory overhead, and retaining the expressivity of layer-specific corrections. We also propose a selective variant, GlowQ-S, that applies the cached shared module only where it provides the largest benefit. Compared with strong baselines, our approach reduces TTFB by (5.6%) and increases throughput by (9.6%) on average, while reducing perplexity on WikiText-2 by (0.17%) and increasing downstream accuracy by 0.42 percentage points. The selective model GlowQ-S further reduces latency, cutting TTFB by (23.4%) and increasing throughput by (37.4%), while maintaining accuracy within 0.2 percentage points on average. Code is available at https://github.com/ahnselim/GlowQ.
Mar 26, 2026

SliderQuant: Accurate Post-Training Quantization for LLMs

In this paper, we address post-training quantization (PTQ) for large language models (LLMs) from an overlooked perspective: given a pre-trained high-precision LLM, the predominant sequential quantization framework treats different layers equally, but this may be not optimal in challenging bit-width settings. We empirically study the quantization impact of different layers on model accuracy, and observe that: (1) shallow/deep layers are usually more sensitive to quantization than intermediate layers; (2) among shallow/deep layers, the most sensitive one is the first/last layer, which exhibits significantly larger quantization error than others. These empirical observations imply that the quantization design for different layers of LLMs is required on multiple levels instead of a single level shared to all layers. Motivated by this, we propose a new PTQ framework termed Sliding-layer Quantization (SliderQuant) that relies on a simple adaptive sliding quantization concept facilitated by few learnable parameters. The base component of SliderQuant is called inter-layer sliding quantization, which incorporates three types of novel sliding window designs tailored for addressing the varying quantization sensitivity of shallow, intermediate and deep layers. The other component is called intra-layer sliding quantization that leverages an incremental strategy to quantize each window. As a result, SliderQuant has a strong ability to reduce quantization errors across layers. Extensive experiments on basic language generation, zero-shot commonsense reasoning and challenging math and code tasks with various LLMs, including Llama/Llama2/Llama3/Qwen2.5 model families, DeepSeek-R1 distilled models and large MoE models, show that our method outperforms existing PTQ methods (including the latest PTQ methods using rotation transformations) for both weight-only quantization and weight-activation quantization.
Mar 24, 2026

APreQEL: Adaptive Mixed Precision Quantization For Edge LLMs

Today, large language models have demonstrated their strengths in various tasks ranging from reasoning, code generation, and complex problem solving. However, this advancement comes with a high computational cost and memory requirements, making it challenging to deploy these models on edge devices to ensure real-time responses and data privacy. Quantization is one common approach to reducing memory use, but most methods apply it uniformly across all layers. This does not account for the fact that different layers may respond differently to reduced precision. Importantly, memory consumption and computational throughput are not necessarily aligned, further complicating deployment decisions. This paper proposes an adaptive mixed precision quantization mechanism that balances memory, latency, and accuracy in edge deployment under user-defined priorities. This is achieved by analyzing the layer-wise contribution and by inferring how different quantization types behave across the target hardware platform in order to assign the most suitable quantization type to each layer. This integration ensures that layer importance and the overall performance trade-offs are jointly respected in this design. Our work unlocks new configuration designs that uniform quantization cannot achieve, expanding the solution space to efficiently deploy the LLMs on resource-constrained devices.
Mar 23, 2026

FAAR: Format-Aware Adaptive Rounding for NVFP4

Deploying large language models (LLMs) on edge devices requires extremely low-bit quantization. Ultra-low precision formats such as NVFP4 offer a promising solution for reducing memory footprint and accelerating computation. However, existing quantization methods typically rely on conventional rounding strategies and fail to account for the non-uniformity of the NVFP4 numerical grid, resulting in suboptimal rounding decisions and amplified quantization errors. To address this, we propose Format-Aware Adaptive Rounding (FAAR), a learnable rounding strategy tailored for the NVFP4 format. Unlike conventional quantization paradigms, FAAR explicitly incorporates the non-uniform NVFP4 grid into the optimization process. By adaptively adjusting rounding decisions guided by loss gradients, our method effectively approximates the theoretically optimal quantization. To complement FAAR, we introduce a 2-stages Format Alignment (2FA) fine-tuning scheme that aligns LLM parameters layer-by-layer to the NVFP4 numerical space, further narrowing the performance gap. Remarkably, this learnable optimization incurs a minimal training overhead of only 4 GPU hours on Llama3-1B. Extensive experiments demonstrate the effectiveness of our approach. Compared with Round-to-Nearest (RTN), our method reduces perplexity on WikiText-2 from 14.28 to 12.60 on Llama3-1B and from 23.06 to 21.27 on Qwen3-1.7B. Additionally, our method consistently outperforms state-of-the-art approaches across various zero-shot downstream tasks.
Mar 20, 2026

DAQ: Delta-Aware Quantization for Post-Training LLM Weight Compression

We introduce Delta-Aware Quantization (DAQ), a data-free post-training quantization framework that preserves the knowledge acquired during post-training. Standard quantization objectives minimize reconstruction error but are agnostic to the base model, allowing quantization noise to disproportionately corrupt the small-magnitude parameter deltas (ΔWΔW) that encode post-training behavior -- an effect we analyze through the lens of quantization as implicit regularization. DAQ replaces reconstruction-based objectives with two delta-aware metrics -- Sign Preservation Rate and Cosine Similarity -- that directly optimize for directional fidelity of ΔWΔW, requiring only the base and post-trained weight matrices. In a pilot FP8 study, DAQ recovers style-specific capabilities lost under standard quantization while maintaining general performance.
Mar 18, 2026

Weight Clustering Reveals the Importance of Relative Rank in Large Language Models

Large language models (LLMs) contain billions of parameters, but many exact weight values are not essential. We apply scalar KK-means clustering to pretrained models, replacing each weight matrix with only 16--64 shared values. For Llama~3.1-8B-Instruct and SmolLM2-135M, this preserves strong accuracy without retraining, providing a simple way to compress LLMs on disk. Fine-tuning only the cluster centroids further recovers 30--40% of the remaining accuracy loss. Using clustering as an analysis tool, we then keep the cluster assignments fixed and perturb only the shared values, to study which of their properties matter most. We find that preserving their relative rank (which connections are stronger or weaker than others) is more important than preserving their exact magnitudes. Breaking this ordering greatly increases perplexity, while changing the values but keeping the ordering causes little loss in many middle and late layers. However, rank preservation alone is not sufficient. Model performance also depends on maintaining the variance of the reconstructed weights, while early layers remain more sensitive; perturbing the cluster assignments themselves shows that their spatial geometry is also load-bearing. We observe the same behavior across four LLMs from two architecture families. These results suggest that the ordering of weights, together with their overall variance, plays a central role in LLM compression and robustness.
Mar 18, 2026

RAMP: Reinforcement Adaptive Mixed Precision Quantization for Efficient On Device LLM Inference

Post training quantization is essential for deploying large language models (LLMs) on resource constrained hardware, yet state of the art methods enforce uniform bit widths across layers, yielding suboptimal accuracy efficiency trade offs. We present RAMP (Reinforcement Adaptive Mixed Precision), an off policy Soft Actor Critic framework that learns per layer bit width assignments to minimize perplexity under a global bit budget. The policy conditions on an 11 dimensional embedding of activation statistics, weight properties, and structural descriptors, enabling zero shot transfer across model families and scales. To enable stable sub 4 bit quantization, we introduce Scale Folding, a preconditioning technique that migrates activation outliers into weights via per channel scaling and normalization layer compensation. A quality prioritized reward with asymmetric penalties and budget cliffs drives rapid convergence. On Llama 2 7B, RAMP achieves 5.54 perplexity at 3.68GB (3.65 effective bits), outperforming uniform 4 bit AWQ (5.60 at 3.90 GB) and GPTQ by 6% in size and 1% to3% in quality. Critically, a policy trained only on Llama 2 7B generalizes zero shot to Llama 2 13B and Mistral 7B, often surpassing target specific training, supporting the hypothesis that quantization sensitivity is primarily architectural. The HALO pipeline exports allocations to GGUF format for kernel free inference on CPUs, GPUs, and edge devices, retaining 99.5% of FP16 commonsense reasoning performance.
Mar 17, 2026

Understanding Quantization of Optimizer States in LLM Pre-training: Dynamics of State Staleness and Effectiveness of State Resets

Quantizing optimizer states is becoming an important ingredient of memory-efficient large-scale pre-training, but the resulting optimizer dynamics remain only partially understood. We study low-precision exponential moving average (EMA) optimizer states and show how quantization can cause many nominal updates to round back to the same stored value, making the state effectively stale and slowing adaptation beyond what the nominal decay would suggest. We then develop a simple predictive model of stalling that estimates one-step stalling probabilities and characterizes how stalling builds up over time after the initialization. This perspective provides a mechanistic explanation for why optimizer-state resets help in low precision: once a quantized EMA becomes effectively stale, resetting it can temporarily restore responsiveness. Motivated by this picture, we derive a simple theory-guided method for choosing useful reset periods, showing that in low precision the key question is not only whether resets help, but when they should be applied. Experiments in controlled simulations and LLM pre-training show that suitable reset schedules recover the performance lost to low-precision state storage while substantially reducing optimizer-state memory.
Mar 17, 2026

BATQuant: Outlier-resilient MXFP4 Quantization via Learnable Block-wise Optimization

Microscaling floating-point (MXFP) formats have emerged as a promising standard for deploying Multi-modal Large Language Models (MLLMs) and Large Language Models (LLMs) on modern accelerator architectures. However, existing Post-Training Quantization (PTQ) methods, particularly rotation-based techniques designed for integer formats, suffer from severe performance collapse when applied to MXFP4. Recent studies attribute this failure to a fundamental format mismatch: global orthogonal rotations inadvertently transfer outlier energy across quantization blocks, inducing new outliers that disrupt local block-wise scaling, while often creating bimodal activation distributions that underutilize the limited quantization range. To address these issues, we propose BATQuant (Block-wise Affine Transformation), which restricts transformations to align with MXFP granularity to prevent cross-block outlier propagation, while relaxing orthogonality constraints to optimize distribution shaping. To ensure parameter efficiency, we introduce Global and Private Kronecker (GPK) decomposition to effectively reduces storage and runtime overhead and incorporate Block-wise Learnable Clipping to suppress residual outliers. Extensive experiments on both MLLMs and LLMs demonstrate that BATQuant establishes new state-of-the-art results under aggressive W4A4KV16 configurations, recovering up to 96.43% of full-precision performance on multimodal benchmarks and clearly outperforming existing methods across diverse tasks.
Mar 17, 2026

Frequency Matters: Fast Model-Agnostic Data Curation for Pruning and Quantization

Post-training model compression is essential for enhancing the portability of Large Language Models (LLMs) while preserving their performance. While several compression approaches have been proposed, less emphasis has been placed on selecting the most suitable set of data (the so-called calibration data) for finding the compressed model configuration. The choice of calibration data is a critical step in preserving model capabilities both intra- and inter-tasks. In this work, we address the challenge of identifying high-performance calibration sets for both pruning and quantization by analyzing intrinsic data properties rather than model-specific signals. We introduce ZipCal, a model-agnostic data curation strategy that maximizes lexical diversity based on Zipfian power laws. Experiments demonstrate that our method outperforms standard uniform random sampling across various pruning benchmarks. Notably, it also performs on par, in terms of downstream performance, with a state-of-the-art method that relies on model perplexity. The latter becomes prohibitively expensive for large-scale models and datasets, while ZipCal is on average ∼\sim240×\times faster due to its tractable linear complexity. We make the code and the experiments available at https://github.com/FrancescoMonaco/ZipCal.
Mar 12, 2026

Adapting Methods for Domain-Specific Japanese Small LMs: Scale, Architecture, and Quantization

This paper presents a systematic methodology for building domain-specific Japanese small language models using QLoRA fine-tuning. We address three core questions: optimal training scale, base-model selection, and architecture-aware quantization. Stage 1 (Training scale): Scale-learning experiments (1k--5k samples) identify n=4,000 as optimal, where test-set NLL reaches minimum (1.127) before overfitting at 5k samples. Stage 2 (Compare finetuned SLMs): Comparing four Japanese LLMs shows that Llama-3 models with Japanese continual pre-training (Swallow-8B, ELYZA-JP-8B) outperform multilingual models (Qwen2.5-7B). Stage 3 (Quantization): Llama-3 architectures improve under Q4_K_M quantization, while GQA architectures degrade severely (Qwen2.5: -0.280 points). Production recommendation: Swallow-8B Q4_K_M achieves 2.830/3 score, 8.9 s/question, 4.9 GB size. The methodology generalizes to low-resource technical domains and provides actionable guidance for compact Japanese specialist LMs on consumer hardware.
Mar 11, 2026

Leech Lattice Vector Quantization for Efficient LLM Compression

Scalar quantization of large language models (LLMs) is fundamentally limited by information-theoretic bounds. While vector quantization (VQ) overcomes these limits by encoding blocks of parameters jointly, practical implementations must avoid the need for expensive lookup mechanisms or other explicit codebook storage. Lattice approaches address this through highly structured and dense packing. This paper explores the Leech lattice, which, with its optimal sphere packing and kissing configurations at 24 dimensions, is the highest dimensional lattice known with such optimal properties. To make the Leech lattice usable for LLM quantization, we extend an existing search algorithm based on the extended Golay code construction, to i) support indexing, enabling conversion to and from bitstrings without materializing the codebook, ii) allow angular search over union of Leech lattice shells, iii) propose fully-parallelisable dequantization kernel. Lastly, we provide a geometric reinterpretation of combining shape--gain quantization with GPTQ-style Hessian corrections: the standard scale-correction step of shape--gain acts as a retraction onto a product of spheres, yielding a Spherical GPTQ primarily acting on directions. We find that low-angular-distortion LLVQ reduces sensitivity to Hadamard/rotation preprocessing, and enables a strong Hadamard-free PTQ in practice. LLVQ delivers state-of-the-art LLM quantization performance, outperforming recent methods such as Quip#, QTIP, and PVQ. The results highlight the effectiveness of high-dimensional lattices for scalable, theoretically grounded model compression.
Mar 11, 2026

The Curse and Blessing of Mean Bias in FP4-Quantized LLM Training

FP4 training promises substantial memory and compute savings for large language models, but remains fragile because blockwise quantization is dictated by extreme activation magnitudes, which inflate dynamic range and compress long-tail signals. We identify a counterintuitive source of this failure: dominant activation outliers are not merely arbitrary sparse events, but are largely induced by a coherent rank-one mean bias, whose direction aligns with the leading anisotropic spectral component. This mean component strengthens during training, is amplified and reshaped by attention and FFN operators, and increasingly dominates top activation magnitudes. Crucially, this discovery reveals that a seemingly complex outlier-suppression problem admits a truly simple solution: isolate the coherent mean before quantization. We therefore propose Averis, a mean-residual splitting quantization method that separates the mean component using only reductions and elementwise subtractions before FP4 quantization. Across Qwen3 0.6B Dense trained on 100B tokens and Qwen3 7B A1.5B MoE trained on 50B tokens, Averis enables robust W4A4G4 FP4 training, reducing BF16 loss gaps to 1.19%/0.81% versus 2.05%/1.10% for NVIDIA's recently released Hadamard-based outlier-smoothing method, while limiting downstream gaps to 0.89/0.71 points. With only 2.20% end-to-end overhead over vanilla NVFP4, about 30% of NVIDIA's Hadamard-based design, Averis provides a hardware-efficient path to stable low-bit LLM training. Complementary to Hadamard, Averis further reduces the Qwen3-0.6B loss and downstream gaps to 0.94% and 0.73 points when combined. Code is available at: https://anonymous.4open.science/r/averis-504D.
Mar 11, 2026

TTQ: Activation-Aware Test-Time Quantization to Accelerate LLM Inference On The Fly

To tackle the huge computational demand of large foundation models, activation-aware compression techniques without retraining have been introduced. However, since these methods highly rely on calibration data, domain shift issues may arise for unseen downstream tasks. We propose a test-time quantization (TTQ) framework which compresses large models on the fly at inference time to resolve this issue. With an efficient online calibration, instant activation-aware quantization can adapt every prompt regardless of the downstream tasks, yet achieving inference speedup. Several experiments demonstrate that TTQ can improve the quantization performance over state-of-the-art baselines.
Mar 9, 2026

SERQ: Saliency-Aware Low-Rank Error Reconstruction for LLM Quantization

Post-training quantization (PTQ) has emerged as a prevailing technique for deploying large language models (LLMs) efficiently in terms of both memory and computation, across edge devices and server platforms. Existing PTQ methods primarily aim to reduce precision in weights and activations by mitigating quantization errors caused by channel-wise outlier activations (e.g., pre-quantization scaling, online transformations, or low-rank error reconstruction). Among these approaches, error reconstruction with low-rank adaptation (LoRA) has proven particularly effective, as it introduces a lightweight auxiliary computation path without requiring heavy optimization or additional online layers. However, prior studies reveal severe accuracy degradation under W4A4 settings, and conventional low-rank adaptations rely on two sequential factors, necessitating intermediate quantization during inference and thereby limiting low-precision efficiency. In this work, we propose SERQ, a saliency-aware error reconstruction method for low-bit LLM inference that employs a single low-rank compensation matrix. SERQ preserves efficient 4-bit matrix multiplication in linear layers by jointly mitigating quantization errors arising from both activation and weight saliency through three stages: (1) static activation flattening, (2) saliency-aware error reconstruction, and (3) offline weight permutation. The method incurs additional computation only for low-rank error reconstruction via a single decomposition, while all other operations are performed offline, thereby keeping latency overhead minimal. Empirically, SERQ outperforms prior error reconstruction methods under both W4A8 and W4A4 settings, and achieves higher accuracy than state-of-the-art rotation-based W4A4 approaches, while substantially reducing calibration complexity. Code is available at https://github.com/acalabys/SERQ
Mar 5, 2026

Diagnosing FP4 inference: a layer-wise and block-wise sensitivity analysis of NVFP4 and MXFP4

Quantization addresses the high resource demand for large language models (LLMs) by alleviating memory pressure and bandwidth congestion and providing significantly scaled compute power with a tolerable impact on accuracy. Four-bit floating point (FP4), the lowest-precision format that preserves essential numerical properties such as exponent and sign, has begun to be adopted in cutting-edge architectures, including Blackwell and AMD CDNA, to support LLM quantization and reduce deployment costs. Although aggressive quantization can yield efficiency gains, the quantization sensitivity of within-transformer layers and whether these sensitivities generalize across existing FP4 formats and model scales remain underexplored. To elucidate quantization sensitivity, this study conducts a systematic analysis of two FP4 formats, MXFP4 and NVFP4, across three Qwen2.5 model scales (0.5B, 7B, and 14B), using controlled component-wise and block-wise isolation methodologies. We observe that MLP up- and down-projection layers consistently dominate in terms of sensitivity, while gate and attention projections are moderately and substantially less sensitive to FP4 quantization, respectively. We further find that sensitivity does not universally localize to the final blocks, but early blocks can be highly sensitive, particularly under MXFP4. Our results provide a diagnostic characterization of the inference behavior of FP4 across components, depths, and FP4 formats.
Mar 5, 2026

Sparse-BitNet: 1.58-bit LLMs are Naturally Friendly to Semi-Structured Sparsity

Semi-structured N:M sparsity and low-bit quantization (e.g., 1.58-bit BitNet) are two promising approaches for improving the efficiency of large language models (LLMs), yet they have largely been studied in isolation. In this work, we investigate their interaction and show that 1.58-bit BitNet is naturally more compatible with N:M sparsity than full-precision models. To study this effect, we propose Sparse-BitNet, a unified framework that jointly applies 1.58-bit quantization and dynamic N:M sparsification while ensuring stable training for the first time. Across multiple model scales and training regimes (sparse pretraining and dense-to-sparse schedules), 1.58-bit BitNet consistently exhibits smaller performance degradation than full-precision baselines at the same sparsity levels and can tolerate higher structured sparsity before accuracy collapse. Moreover, using our custom sparse tensor core, Sparse-BitNet achieves substantial speedups in both training and inference, reaching up to 1.30X. These results highlight that combining extremely low-bit quantization with semi-structured N:M sparsity is a promising direction for efficient LLMs. Code available at https://github.com/AAzdi/Sparse-BitNet
Mar 5, 2026

WaterSIC: Information-Theoretically (Near) Optimal Linear Layer Quantization

This paper considers the problem of converting a given dense linear layer to low precision. The tradeoff between compressed length and output discrepancy is analyzed information theoretically (IT). It is shown that a popular GPTQ algorithm may have an arbitrarily large gap to the IT limit. To alleviate this problem, a novel algorithm, termed ``WaterSIC'', is proposed and is shown to be within a rate gap of 0.255 bits to the IT limit, uniformly over all possible covariance matrices of input activations. The key innovation of WaterSIC's is to allocate different quantization rates to different columns (in-features) of the weight matrix, mimicking the classical IT solution known as "waterfilling". Applying WaterSIC to the Llama and Qwen family of LLMs establishes new state-of-the-art performance for all quantization rates from 1 to 4 bits. Our code is available at https://github.com/egorlifar/watersic.
Mar 4, 2026

Dissecting Quantization Error: A Concentration-Alignment Perspective

Quantization can drastically increase the efficiency of large language and vision models, but typically incurs an accuracy drop. Recently, function-preserving transforms (e.g. rotations, Hadamard transform, channel-wise scaling) have been successfully applied to reduce post-training quantization error, yet a principled explanation remains elusive. We analyze linear-layer quantization via the signal-to-quantization-noise ratio (SQNR), showing that for uniform integer quantization at a fixed bit width, SQNR decomposes into (i) the concentration of weights and activations (capturing spread and outliers), and (ii) the alignment of their dominant variation directions. This reveals an actionable insight: beyond concentration - the focus of most prior transforms (e.g. rotations or Hadamard) - improving alignment between weight and activation can further reduce quantization error. Motivated by this, we introduce block Concentration-Alignment Transforms (CAT), a lightweight linear transformation that uses a covariance estimate from a small calibration set to jointly improve concentration and alignment, approximately maximizing SQNR. Experiments across several LLMs show that CAT consistently matches or outperforms prior transform-based quantization methods at 4-bit precision, confirming the insights gained in our framework.
Mar 4, 2026

Bielik-Q2-Sharp: A Comparative Study of Extreme 2-bit Quantization Methods for a Polish 11B Language Model

We present Bielik-Q2-Sharp, the first systematic academic evaluation of extreme 2-bit quantization applied to a Polish large language model. Using Bielik-11B-v2.3-Instruct (11B parameters, Mistral architecture) as our base model, we compare six state-of-the-art post-training quantization methods -- QuIP#, SpinQuant+GPTQ, ButterflyQuant, QTIP, VPTQ, and AQLM -- all calibrated on a Polish-language corpus (CulturaX-PL) with shared Hessian matrices. Our best variant (QuIP# E8P12) achieves 71.92% across 22 Polish benchmarks versus 72.07% for the IQ2_XXS baseline -- within statistical noise, at a modest size premium (3.26 GB vs. ~2.6 GB). On eq_bench, our method scores 47.14 versus 43.53 (+3.6pp), suggesting superior preservation of higher-order reasoning. QTIP achieves the best per-bit efficiency (79.4% MC acc_norm at ~2.4 bpw, 3.27 GB), matching VPTQ's quality at 35% smaller size. We additionally document a MC-generation dissociation phenomenon where rotation-based methods preserve log-likelihood quality but fail catastrophically at autoregressive generation. The entire project was conducted by a single independent researcher on cloud GPUs (vast.ai) within a $285 budget. All models, Hessians, and evaluation logs are publicly available.
Mar 2, 2026

FreeAct: Freeing Activations for LLM Quantization

Quantization is pivotal for mitigating the significant memory and computational overhead of Large Language Models (LLMs). While emerging transformation-based methods have successfully enhanced quantization by projecting feature spaces onto smoother manifolds using orthogonal matrices, they typically enforce a rigid one-to-one transformation constraint. This static approach fails to account for the dynamic patterns inherent in input activations, particularly within diffusion LLMs (dLLMs) and Multimodal LLMs (MLLMs), where varying token types exhibit distinct distributions. To advance this, we propose FreeAct, a novel quantization framework that relaxes the static one-to-one constraint to accommodate dynamic activation disparities. Theoretically, we leverage the rank-deficient nature of activations to derive a solution space that extends beyond simple inverse matrices, enabling the decoupling of activation transformations from weights. Methodologically, FreeAct identifies token-specific dynamics (i.e., vision v.s. text, or masked tokens) and allocates distinct transformation matrices to the activation side, while maintaining a unified, static transformation for the weights. Extensive experiments across dLLMs and MLLMs demonstrate that FreeAct significantly outperforms baselines, up to 5.3% performance improvement, with in-depth analyses. Our code will be publicly released.
Feb 26, 2026

pQuant: Towards Effective Low-Bit Language Models via Decoupled Linear Quantization-Aware Training

Quantization-Aware Training from scratch has emerged as a promising approach for building efficient large language models (LLMs) with extremely low-bit weights (sub 2-bit), which can offer substantial advantages for edge deployment. However, existing methods still fail to achieve satisfactory accuracy and scalability. In this work, we identify a parameter democratization effect as a key bottleneck: the sensitivity of all parameters becomes homogenized, severely limiting expressivity. To address this, we propose pQuant, a method that decouples parameters by splitting linear layers into two specialized branches: a dominant 1-bit branch for efficient computation and a compact high-precision branch dedicated to preserving the most sensitive parameters. Through tailored feature scaling, we explicitly guide the model to allocate sensitive parameters to the high-precision branch. Furthermore, we extend this branch into multiple, sparsely-activated experts, enabling efficient capacity scaling. Extensive experiments indicate our pQuant achieves state-of-the-art performance in extremely low-bit quantization.
Feb 21, 2026

MoBiQuant: Mixture-of-Bits Quantization for Token-Adaptive Any-Precision LLM

Dynamic runtime latency and memory constraints necessitate flexible large language model (LLM) deployment, where an LLM can be inferred with various quantization precisions based on available computational resources. Recent work on such any-precision quantization either relies on hardware-inefficient vector quantization or induces additional scaling factors when switching between bit-widths. Meanwhile, existing post-training quantization (PTQ) methods calibrated for a fixed low precision show poor generalizability under runtime precision change. In this work, we attribute the source of poor generalization across bit-widths to a precision-dependent outlier migration phenomenon where the distribution of PTQ-sensitive tokens changes across precisions. Motivated by this observation, we propose MoBiQuant, a novel any-precision Mixture-of-Bits quantization framework that adjusts weight precision for flexible LLM inference based on token sensitivity. Specifically, we propose a many-in-one recursive residual quantization that can iteratively reconstruct higher-precision weights at runtime and mitigates outlier migration with a token-aware router to dynamically select the optimal inference precision of each token.Extensive experiments show that MoBiQuant matches or surpasses frontier single-precision PTQ while exhibiting strong elasticity, achieving significant memory savings and throughput gains of up to 1.34×1.34\times over state-of-the-art any-precision methods. Our code is available at https://github.com/SimWangArizona/MoBiQuant.