Cross-Layer Error Compensation and Finite-Sample Feature-Statistics Matching for Extreme Low-Bit Quantization of Large Language Models

• URL: http://arxiv.org/abs/2607.14630v1
• Date: Thu, 16 Jul 2026 06:47:52 GMT
• Title: Cross-Layer Error Compensation and Finite-Sample Feature-Statistics Matching for Extreme Low-Bit Quantization of Large Language Models
• Abstract summary: We formulate quantization as a joint optimization over the discrete codes and scales of all layers.<n>We show that error compensation alone reaches a perplexity ratio of 9.56 +/- 0.15 over the FP16 teacher.<n>On Qwen2.5-1.5B with 1.125-bit group-binary weights, error compensation alone reaches a perplexity ratio of 9.56 +/- 0.15 over the FP16 teacher.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Layer-wise post-training quantization of large language models minimizes each layer's reconstruction error in isolation, allowing quantization errors to accumulate across depth and causing severe degradation in extreme low-bit regimes. We formulate quantization as a joint optimization over the discrete codes and scales of all layers, driven by two mechanisms: (i) cross-layer error compensation, which maintains the network-level accumulated error through the recursion e_{l+1} = A_l e_l + q_l, with a propagation operator A_l derived from the layer's input differential and a local quantization residual q_l evaluated at teacher features; and (ii) finite-sample feature-statistics matching, which aligns means, projected covariances, and centered empirical kernels between the full-precision and quantized networks under relative normalization. We prove that instantiating the propagation operator as a finite difference of the quantized network makes the recursion exact for arbitrary nonlinear layers, enabling an efficient forward-difference implementation. Binary weights are optimized via a mirror-descent parameterization u = tanh(beta*z) with annealed inverse temperature and group-wise log-scales. On Qwen2.5-1.5B with 1.125-bit group-binary weights, error compensation alone reaches a perplexity ratio of 9.56 +/- 0.15 over the FP16 teacher, outperforming logit distillation (14.09 +/- 0.53; 32 percent relative, more than 8 sigma over 3 seeds) and layer-local reconstruction by two orders of magnitude. The same objective transfers unchanged to 4-bit quantization (1.060 vs. 1.088 for layer-local). Out-of-domain evaluations (C4, CNN/DailyMail) show the advantage of error compensation grows off-domain, while statistics matching keeps feature-statistics discrepancy low off-domain (0.42-0.88 vs. 1.41-2.99 without it), revealing a complementary division of labor between the two mechanisms.

Related papers

• Scalable Lindblad Noise Learning via Stochastic Tensor-Network Simulation [48.887627688666974]
  Learning dissipation rates in large-scale open quantum systems is a major obstacle for near-term quantum technologies.<n>Here, we propose a scalable noise-learning framework for Lindblad dissipation rates that combines a simulation method, the Jump Method (TJM)<n>We demonstrate the approach on two noise models in the Ising model, in which independent dissipation rates are learned for each site up to $N_mathrmsite=16$.<n>This combination of scalable numerics and rigorous theoretical guarantees positions TJM-based noise learning as a practical foundation for characterizing dissipation in large quantum devices.
  arXiv  Detail & Related papers  (2026-08-25T15:07:11Z)
• Multi-Stage Mamba-Based Architecture for Fast and Scalable Superconducting Qubit Readout [2.7180447022087932]
  In superconducting quantum processors, readout operations are both error-prone and high-latency.<n>We present a multi-stage qubit state discriminator based on the Mamba model.<n>Our model achieves a geometric mean readout fidelity of 0.906, outperforming the best-reported state-of-the-art method.
  arXiv  Detail & Related papers  (2026-07-13T15:12:07Z)
• SPARC-Net: A Spectral, Causality-Aware, and Hard-Constrained Physics-Informed Architecture for Stiff and Shock-Dominated Partial Differential Equations [0.0654051537828115]
  Physics-Informed Neural Networks (PINNs) provide a meshless approach for solving partial differential equations (PDEs)<n>PINNs suffer severe degradation in stiff and shock-dominated problems, where small PDE residuals can correspond to globally inaccurate solutions.<n>We show these failures are multi-causal, arising from the concurrent interplay of (i) spectral bias against sharp features, (ii) imbalanced multi-term optimization and loss-weight collapse, (iii) violation of temporal causality, and (iv) under-resolved collocation.<n>We present-Net, a unified architecture and training
  arXiv  Detail & Related papers  (2026-07-13T09:23:17Z)
• Direct/adaptive-mixture phase-gradient learning for neural-network quantum states with complex phase structure [3.1413021865737765]
  Estimator design emerges as a first-class lever for complex neural quantum states.<n>We show how a small network trained this way reaches $0.89%$ median error, where tuned standard baselines plateau at $1.8%$.<n>The advantage carries over to chiral XXX chains: the direct estimator again converges to a markedly lower error than the standard one, across $$ and size.
  arXiv  Detail & Related papers  (2026-06-11T21:05:28Z)
• Closed-Form Spectral Regularization for Multi-Task Model Merging [96.82449201305234]
  Model merging combines several independently fine-tuned experts into a single multi-task model without any training data.<n>State-of-the-art merging methods formulate merging as a layer-wise interference problem.<n>We propose SWUDI, a closed-form method that combines a soft exponential filter, which matches the gradient-flow trajectory of iterative descent.
  arXiv  Detail & Related papers  (2026-06-05T14:00:47Z)
• RUQuant: Towards Refining Uniform Quantization for Large Language Models [17.258420059228808]
  Post-training quantization (PTQ) has emerged as a practical solution by compressing models without requiring retraining.<n>Existing methods often suffer from substantial accuracy degradation due to the non-uniform nature of activation distributions.<n>In this work, we revisit the activation quantization problem from a theoretical perspective grounded in the Lloyd-Max optimality conditions.
  arXiv  Detail & Related papers  (2026-04-05T08:04:39Z)
• RaBiT: Residual-Aware Binarization Training for Accurate and Efficient LLMs [5.782015253162346]
  Residual binarization enables matmul-free inference by stacking binary layers.<n>We propose RaBiT, a novel quantization framework that resolves coadaptation by algorithmically enforcing a residual hierarchy.<n>RaBiT achieves state-of-the-art performance, rivals even hardware-intensive Vector Quantization (VQ) methods, and delivers a $4.49times$ inference speed-up over full-precision models.
  arXiv  Detail & Related papers  (2026-02-05T06:41:11Z)
• 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)
• Structure-Informed Estimation for Pilot-Limited MIMO Channels via Tensor Decomposition [51.56484100374058]
  This paper formulates pilot-limited channel estimation as low-rank tensor completion from sparse observations.<n>Experiments on synthetic channels demonstrate 10-20,dB normalized mean-square error (NMSE) improvement over least-squares (LS)<n> evaluations on DeepMIMO ray-tracing channels show 24-44% additional NMSE reduction over pure tensor-based methods.
  arXiv  Detail & Related papers  (2026-02-03T23:38:05Z)
• Variational Entropic Optimal Transport [67.76725267984578]
  We propose Variational Entropic Optimal Transport (VarEOT) for domain translation problems.<n>VarEOT is based on an exact variational reformulation of the log-partition $log mathbbE[exp(cdot)$ as a tractable generalization over an auxiliary positive normalizer.<n> Experiments on synthetic data and unpaired image-to-image translation demonstrate competitive or improved translation quality.
  arXiv  Detail & Related papers  (2026-02-02T15:48:44Z)
• ParetoQ: Improving Scaling Laws in Extremely Low-bit LLM Quantization [73.60493264901359]
  We present a unified framework for rigorous comparisons across 1-bit, 1.58-bit, 2-bit, 3-bit, and 4-bit quantization settings.<n>We show that ternary, 2-bit, and 3-bit quantization maintains comparable performance in the size-accuracy trade-off.<n>Considering hardware constraints, 2-bit quantization offers promising potential for memory reduction and speedup.
  arXiv  Detail & Related papers  (2025-02-04T18:59:26Z)
• Deriving Differential Target Propagation from Iterating Approximate Inverses [91.3755431537592]
  We show that a particular form of target propagation, relying on learned inverses of each layer, which is differential, gives rise to an update rule which corresponds to an approximate Gauss-Newton gradient-based optimization. We consider several iterative calculations based on local auto-encoders at each layer in order to achieve more precise inversions for more accurate target propagation.
  arXiv  Detail & Related papers  (2020-07-29T22:34:45Z)

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.