AXELRAM: Quantize Once, Never Dequantize

• URL: http://arxiv.org/abs/2604.02638v1
• Date: Fri, 03 Apr 2026 02:03:38 GMT
• Title: AXELRAM: Quantize Once, Never Dequantize
• Abstract summary: AXELRAM computes attention scores directly from quantized KV cache indices without dequantization.<n>We trace the root cause to layer-wise norm and propose a gradient-free sign pattern selection (200 candidates, 8 calibration samples, one-time) that eliminates catastrophic spikes with zero additional hardware cost.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We propose AXELRAM, a smart SRAM macro architecture that computes attention scores directly from quantized KV cache indices without dequantization. The key enabler is a design-time fixed codebook: orthogonal-transform-based quantization concentrates each coordinate's distribution to N(0,1/d), so the optimal quantizer depends only on dimension d and bit-width b, not on input data. The asymmetric path design -- transform on write, table-lookup on read with no inverse transform -- reduces per-query multiplications by 102.4x (a mathematical identity). Through multi-seed evaluation (10 seeds x 3 models), we discover that sign pattern sensitivity causes catastrophic PPL spikes (Delta > 50) on certain models (Qwen2.5-3B), while others (LLaMA-3.1-8B) are fully stable. This phenomenon extends SpinQuant's observation of rotation variance in weight quantization to the KV cache domain, where the effect is qualitatively more severe. We trace the root cause to layer-wise norm heterogeneity and propose a gradient-free sign pattern selection (200 candidates, 8 calibration samples, one-time) that eliminates catastrophic spikes with zero additional hardware cost. All source code is available at https://github.com/Axelidea/AXELRAM.

Related papers

• The ZZ feature map induces a signless Laplacian metric: a closed-form classical surrogate for quantum kernel regression [0.0]
  We prove that in the small-bandwidth regime the induced kernel is, to leading order, an anisotropic Gaussian kernel with metric M = I + pi2 Q.<n>The corresponding classical kernel requires no fitted parameters and no quantum simulation.
  arXiv  Detail & Related papers  (2026-08-29T19:59:00Z)
• What Does a Discrete Diffusion Model Learn? [71.03603607324338]
  discrete diffusion model learns: a denoiser, a score ratio, or a bridge plug-in predictor.<n>We start with a rigorous derivation of the continuous-time Markov chain (CTMC) ELBO for any noising process.<n>Every identity is verified numerically, without approximation, on an exactly solvable model.
  arXiv  Detail & Related papers  (2026-07-06T17:56:11Z)
• Hierarchical Transformer Preconditioning for Interactive Physics Simulation [28.137076331332413]
  Hierarchical Transformer Preconditioner, a neural preconditioner anchored to a weak-admissibility H-matrix partition.<n>Network models the inverse through low-rank far-field factors.<n>Uses highway connections to propagate context across depth sparse.
  arXiv  Detail & Related papers  (2026-05-13T11:02:27Z)
• ITQ3_S: High-Fidelity 3-bit LLM Inference via Interleaved Ternary Quantization with Rotation-Domain Smoothing [0.0]
  We present ITQ3_S (Interleaved Ternary Quantization -- Specialized), a novel 3-bit weight quantization format for LLMs integrating TurboQuant (TQ)<n>Conventional 3-bit methods suffer precision loss from heavy-tailed weight distributions and inter-channel outliers.<n>ITQ3_S pre-rotates the weight space via FWHT before quantization, spreading energy across the vector and inducing a near-Gaussian distribution.
  arXiv  Detail & Related papers  (2026-03-30T00:03:22Z)
• Layer-wise QUBO-Based Training of CNN Classifiers for Quantum Annealing [0.0]
  We propose an iterative framework based on Quadratic Un Binary Optimization (QUBO) for training the head of convolutional neural networks (CNNs)<n>A per-output decomposition splits the $C$-class problem into $C$ independent QUBOs, each with $(d+1)K$ binary variables, where $d$ is the feature dimension and $K$ is the bit precision.<n>We evaluate the method on six image-classification benchmarks (sklearn digits, MNIST, Fashion-MNIST, CIFAR-10, EMNIST, KMNIST)
  arXiv  Detail & Related papers  (2026-03-03T13:10:36Z)
• BPDQ: Bit-Plane Decomposition Quantization on a Variable Grid for Large Language Models [56.504879072674015]
  We propose Bit-Plane Decomposition Quantization (BPDQ), which constructs a variable quantization grid via bit-planes and scalar coefficients.<n>BPDQ enables serving Qwen2.5-72B on a single GTX 3090 with 83.85% GSM8K accuracy (vs. 90.83% at 16-bit)
  arXiv  Detail & Related papers  (2026-02-04T02:54:37Z)
• High-Rate Quantized Matrix Multiplication: Theory and Practice [29.75700570685703]
  This work investigates the problem of quantized matrix multiplication (MatMul)<n>We consider two settings: 1) Generic MatMul, where both matrices must be quantized (weight+ quantization); and 2) weight-only quantization, where the second matrix is only known through covariance matrix $_X$ of its columns.
  arXiv  Detail & Related papers  (2026-01-23T21:32:44Z)
• ButterflyQuant: Ultra-low-bit LLM Quantization through Learnable Orthogonal Butterfly Transforms [21.010238822100135]
  Large language models require massive memory footprints, severely limiting deployment on consumer hardware.<n> Quantization reduces memory through lower numerical precision, but extreme 2-bit quantization suffers from catastrophic performance loss due to outliers in activations.<n>We propose ButterflyQuant, which replaces Hadamard rotations with learnable butterfly transforms parameterized by continuous Givens rotation angles.
  arXiv  Detail & Related papers  (2025-09-11T17:59:51Z)
• SpinQuant: LLM quantization with learned rotations [49.07335692298487]
  Post-training quantization (PTQ) techniques applied to weights, activations, and the KV cache greatly reduce memory usage, latency, and power consumption of Large Language Models (LLMs)<n>We identify a collection of applicable rotation parameterizations that lead to identical outputs in full-precision Transformer architectures while enhancing quantization accuracy.<n>We propose SpinQuant, a novel approach that incorporates learned rotation matrices for optimal quantized network accuracy.
  arXiv  Detail & Related papers  (2024-05-26T02:15:49Z)
• SqueezeLLM: Dense-and-Sparse Quantization [80.32162537942138]
  Main bottleneck for generative inference with LLMs is memory bandwidth, rather than compute, for single batch inference. We introduce SqueezeLLM, a post-training quantization framework that enables lossless compression to ultra-low precisions of up to 3-bit. Our framework incorporates two novel ideas: (i) sensitivity-based non-uniform quantization, which searches for the optimal bit precision assignment based on second-order information; and (ii) the Dense-and-Sparse decomposition that stores outliers and sensitive weight values in an efficient sparse format.
  arXiv  Detail & Related papers  (2023-06-13T08:57:54Z)
• Transformers meet Stochastic Block Models: Attention with Data-Adaptive Sparsity and Cost [53.746169882193456]
  Recent works have proposed various sparse attention modules to overcome the quadratic cost of self-attention. We propose a model that resolves both problems by endowing each attention head with a mixed-membership Block Model. Our model outperforms previous efficient variants as well as the original Transformer with full attention.
  arXiv  Detail & Related papers  (2022-10-27T15:30:52Z)
• Scaling Structured Inference with Randomization [64.18063627155128]
  We propose a family of dynamic programming (RDP) randomized for scaling structured models to tens of thousands of latent states. Our method is widely applicable to classical DP-based inference. It is also compatible with automatic differentiation so can be integrated with neural networks seamlessly.
  arXiv  Detail & Related papers  (2021-12-07T11:26:41Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.