The Quantization Trap: Breaking Linear Scaling Laws in Multi-Hop Reasoning

• URL: http://arxiv.org/abs/2602.13595v1
• Date: Sat, 14 Feb 2026 04:25:27 GMT
• Title: The Quantization Trap: Breaking Linear Scaling Laws in Multi-Hop Reasoning
• Abstract summary: We demonstrate that this scaling law breaks in the context of multi-hop reasoning.<n>We reveal a 'quantization trap' where reducing precision from 16-bit to 8/4-bit paradoxically increases more net energy consumption while degrading reasoning accuracy.
• Score: 10.91558162376728
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Neural scaling laws provide a predictable recipe for AI advancement: reducing numerical precision should linearly improve computational efficiency and energy profile (E proportional to bits). In this paper, we demonstrate that this scaling law breaks in the context of multi-hop reasoning. We reveal a 'quantization trap' where reducing precision from 16-bit to 8/4-bit paradoxically increases more net energy consumption while degrading reasoning accuracy. We provide a rigorous theoretical decomposition that attributes this failure to hardware casting overhead, the hidden latency cost of dequantization kernels, which becomes a dominant bottleneck in sequential reasoning chains, as well as to a sequential energy amortization failure. As a result, scaling law breaking is unavoidable in practice. Our findings suggest that the industry's "smaller-is-better" heuristic is mathematically counterproductive for complex reasoning tasks.

Related papers

• The Unreasonable Effectiveness of Scaling Laws in AI [0.0]
  scaling laws describe how training loss decreases with compute in a power-law form.<n>Despite diminishing returns they predict, progress in practice has often continued through rapidly improving efficiency.<n>This paper argues that both features arise from the same source: scaling laws are unusually effective because they abstract away from many realization details.
  arXiv  Detail & Related papers  (2026-03-30T14:42:53Z)
• Towards Robust Scaling Laws for Optimizers [89.21160945066737]
  Empirical scaling laws are widely used to predict loss as model size and training data grow.<n>We show that Chinchilla-style scaling laws emerge naturally as a result of loss decomposition into irreducible, approximation, and optimization errors.
  arXiv  Detail & Related papers  (2026-02-07T21:40:33Z)
• Effective Frontiers: A Unification of Neural Scaling Laws [19.808117554175013]
  We propose a unified framework that abstracts general learning tasks as the progressive coverage of patterns from a long-tail (Zipfian) distribution.<n>We derive the precise scaling laws for $N$, $D$, and $C$, attributing them to capacity, coverage, and optimization bottlenecks.
  arXiv  Detail & Related papers  (2026-02-01T10:44:46Z)
• ReGuLaR: Variational Latent Reasoning Guided by Rendered Chain-of-Thought [49.203970812338916]
  Explicit reasoning chains introduce substantial computational redundancy.<n>Recent latent reasoning methods attempt to mitigate this by compressing reasoning processes into latent space.<n>We propose Rendered CoT-Guided variational Latent Reasoning (ReGuLaR)
  arXiv  Detail & Related papers  (2026-01-30T17:08:06Z)
• ENTRA: Entropy-Based Redundancy Avoidance in Large Language Model Reasoning [30.786062954495403]
  Large Reasoning Models (LRMs) often suffer from overthinking, generating unnecessarily long reasoning chains even for simple tasks.<n>We propose ENTRA, an entropy-based training framework that suppresses redundant reasoning while preserving performance.
  arXiv  Detail & Related papers  (2026-01-12T01:26:30Z)
• Arbitrage: Efficient Reasoning via Advantage-Aware Speculation [71.45710345765528]
  Speculative Decoding accelerates inference by employing a fast but inaccurate draft model to autoregressively propose tokens.<n>But due to unnecessary rejections caused by token mismatches in semantically equivalent steps, traditional token-level Speculative Decoding struggles in reasoning tasks.<n>We propose Arbitrage, a novel step-level speculative generation framework that routes generation dynamically based on the relative advantage between draft and target models.
  arXiv  Detail & Related papers  (2025-12-04T17:50:53Z)
• Compute-Optimal LLMs Provably Generalize Better With Scale [102.29926217670926]
  We develop generalization bounds on the pretraining objective of large language models (LLMs) in the compute-optimal regime.<n>We introduce a novel, fully empirical Freedman-type martingale concentration that tightens existing bounds by accounting for the variance of the loss function.<n>We produce a scaling law for the generalization gap, with bounds that become predictably stronger with scale.
  arXiv  Detail & Related papers  (2025-04-21T16:26:56Z)
• Energy Scale Degradation in Sparse Quantum Solvers: A Barrier to Quantum Utility [0.8340329709052821]
  Quantum computing offers a promising route for tackling hard optimization problems by encoding them as Ising models.<n>Minor-embedding, mapping logical qubits onto chains of physical qubits, requires stronger intra-chain to maintain consistency.<n>This elevated coupling strength forces a rescaling of the Hamiltonian due to hardware-imposed limits on the allowable ranges of coupling strengths.<n>We show that as the connectivity degree increases, the effective temperature rises as a function, resulting in a success probability that decays exponentially.
  arXiv  Detail & Related papers  (2025-03-11T11:14:05Z)
• O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning [98.3430004984531]
  We propose Length-Harmonizing Fine-Tuning (O1-Pruner) to minimize reasoning overhead while maintaining accuracy.<n>Our code is coming soon at https://github.com/StarDewXXX/O1-Pruner.
  arXiv  Detail & Related papers  (2025-01-22T01:35:11Z)
• Non-convex Bayesian Learning via Stochastic Gradient Markov Chain Monte Carlo [4.656426393230839]
  The rise of artificial intelligence (AI) hinges on efficient of modern deep neural networks (DNNs) for non-trips and uncertainty. In this thesis we propose a tool to handle the problem of Monte Carlo exploitation. We also propose two dynamic importance sampling algorithms for the underlying ordinary equation (ODE) system.
  arXiv  Detail & Related papers  (2023-05-30T18:25:11Z)
• Scaling Laws Beyond Backpropagation [64.0476282000118]
  We study the ability of Direct Feedback Alignment to train causal decoder-only Transformers efficiently. We find that DFA fails to offer more efficient scaling than backpropagation.
  arXiv  Detail & Related papers  (2022-10-26T10:09:14Z)
• Truncated phase-based quantum arithmetic: error propagation and resource reduction [0.0]
  We present a modification of the Draper quantum Fourier adder which eliminates small-angle rotations to highly coarse levels. We show that the inherited loss of fidelity is directly given by the rate of carry and borrow bits in the subroutine. Surprisingly, we find that each of the $7times 107$ quantum Fourier transforms may be truncated down to $pi/64$, with additive rotations left only slightly finer.
  arXiv  Detail & Related papers  (2021-10-01T05:19:03Z)

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.