When Flat Minima Fail: Characterizing INT4 Quantization Collapse After FP32 Convergence

• URL: http://arxiv.org/abs/2604.15167v1
• Date: Thu, 16 Apr 2026 15:46:36 GMT
• Title: When Flat Minima Fail: Characterizing INT4 Quantization Collapse After FP32 Convergence
• Abstract summary: Post-training quantization assumes that a well-converged model is a quantization-ready model.<n>We show this assumption fails in a structured, measurable, and previously uncharacterized way.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Post-training quantization (PTQ) assumes that a well-converged model is a quantization-ready model. We show this assumption fails in a structured, measurable, and previously uncharacterized way. Using a calibration-free per-group INT4 probe applied to all 154 publicly available Pythia-160m training checkpoints, we identify a three-phase divergence structure: a rapid-learning phase where both FP32 perplexity and quantization robustness improve together, a meta-stable plateau lasting roughly 70,000 steps where FP32 perplexity stagnates but INT4 gap remains bounded, and an explosive divergence phase where the INT4 gap compounds from 11% to 517% while FP32 perplexity barely moves. Critically, this divergence begins not when the learning rate starts decaying, but precisely when FP32 perplexity converges a finer-grained onset predictor that implies post-convergence weight updates, rather than decay magnitude alone, are the proximate cause. We further show that INT8 quantization is entirely immune throughout all three phases, constraining the mechanism to the coarseness of the 16-level INT4 grid specifically, and rule out weight outlier accumulation as the mechanism via direct kurtosis measurement. Finally, we conduct a controlled fork experiment from the pre-divergence checkpoint comparing three learning rate schedules (cosine continuation, SGDR warm restarts, and our proposed Oscillatory Lock-In) across nine independent runs. SGDR uniformly accelerates divergence (0/9 pairwise wins against cosine), while OLI's settled cool phases reduce the INT4 gap by 2.2 percentage points on average (t = -5.46, p < 0.0001), demonstrating that schedule amplitude calibration, not oscillation alone, determines whether perturbation helps or hurts. Our code, probe implementation, and all 154-checkpoint audit results are released publicly.

Related papers

• Steering Under Compression: Dose-Response, Capability Cost, and Failure Asymmetry in Quantized LLMs [0.0]
  Inference-time steering enables behavioral control of large language models without parameter modification.<n>Post-training quantization reduces memory and compute costs for deployment.<n>We systematically study activation steering under weight-only quantization.
  arXiv  Detail & Related papers  (2026-09-06T08:41:42Z)
• Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM [0.0]
  Minima: NVFP4 W4A4 on all 496 linear layers, GDN included.<n>A four-part mechanism study explains why NVFP4's 16-element block scaling localizes the residual stream's extreme outliers.<n>A mechanistic account of why the recurrent half of a hybrid LLM is the easy half to quantize.
  arXiv  Detail & Related papers  (2026-09-03T17:04:26Z)
• On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability [57.03393882933515]
  We describe the architecture and ablations of Qwen3.8-Flash-Next, a sparse mixture-of-experts model with 125B parameters, 6B activated per token, and 51B parameters of n-gram embedding tables held off the accelerator.<n>On fourteen pre-training benchmarks the model leads the 397B-A17B predecessor on eight and trails it on the rest by at most 2.6 points, at 1/3 the activated parameters, 1/3 the training tokens, and roughly 1/9 the training FLOPs.
  arXiv  Detail & Related papers  (2026-08-31T06:35:07Z)
• Robust CHSH Self-Testing with Finite-Energy GKP States [42.71287663143178]
  We present a full-oscillator analysis of a finite-energy GKP CHSH test.<n>The observed score yields robust Bell-pair self-testing.
  arXiv  Detail & Related papers  (2026-08-11T16:33:28Z)
• Silent Failures in Quantized LLM Reasoning: A Taxonomy-Based Analysis of Hollow Convergence and Failure Mode Shifts [0.0]
  We show that post-training quantization can silently alter how large language models reason even when task accuracy is preserved.<n>We classify 30,000 chain-of-thought outputs from five instruction-tuned LLMs across three quantization precisions.<n>We find that while accuracy is robust across precisions, Hollow Convergence shows a significant size-dependent shift under NF4.
  arXiv  Detail & Related papers  (2026-07-10T21:55:05Z)
• When Do Autoregressive Sequence Models Forecast Physical Wavefields? A Controlled Study on Synthetic Seismograms [43.8784307709823]
  Long-horizon autoregressive forecasting of physical signals is limited by error accumulation.<n>We ask when such rollout stays stable using synthetic three-component seismograms as a physically structured testbed.
  arXiv  Detail & Related papers  (2026-06-09T13:46:14Z)
• The Illusion of Equivalence: Systematic FP16 Divergence in KV-Cached Autoregressive Inference [2.954681536906518]
  KV caching is a ubiquitous optimization in autoregressive transformer inference.<n>We show that FP16 KV cache inference is fundamentally non-equivalent to recomputation.
  arXiv  Detail & Related papers  (2026-04-16T15:59:40Z)
• Auxiliary Finite-Difference Residual-Gradient Regularization for PINNs [0.0]
  We study a hybrid design in which the governing PDE residual remains automatic-differentiation based, while finite differences appear only in a weak auxiliary term.<n>The FD term regularizes the residual field without replacing the PDE residual itself.<n>Across seeds 0-5 and 100k epochs, the most reliable tested configuration is a fixed shell weight of 5e-4 under the Kourkoutas-beta regime.
  arXiv  Detail & Related papers  (2026-04-15T23:09:19Z)
• Quantization Dominates Rank Reduction for KV-Cache Compression [0.0]
  quantization consistently outperforms rank reduction by 4-364 PPL depending on model and compression level.<n>We formalize this via a result showing projection damage exceeds quantization damage by 3 x 2 (2b) per direction under the softmax Fisher metric.
  arXiv  Detail & Related papers  (2026-04-13T14:06:18Z)
• Quantization Impact on the Accuracy and Communication Efficiency Trade-off in Federated Learning for Aerospace Predictive Maintenance [0.0]
  Federated learning (FL) enables privacy-preserving predictive maintenance across distributed aerospace fleets.<n>This paper investigates the impact of symmetric uniform quantization on the accuracy--efficiency trade-off of a custom-designed lightweight convolutional model.<n>Int4 achieves accuracy emphstatistically indistinguishable from FP32 on both FD001 and FD002.
  arXiv  Detail & Related papers  (2026-04-09T17:13:15Z)
• Activation Outliers in Transformer Quantization: Reproduction, Statistical Analysis, and Deployment Tradeoffs [0.0]
  Post-training quantization (PTQ) of transformers is known to suffer from severe accuracy degradation due to structured activation outliers.<n>This paper provides a reproducible empirical reproduction and systems-level extension of that phenomenon in BERT-base fine-tuned on QNLI.
  arXiv  Detail & Related papers  (2026-03-04T17:26:29Z)
• 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)
• Dissecting Outlier Dynamics in LLM NVFP4 Pretraining [46.10969678564592]
  This study conducts a longitudinal analysis of outlier dynamics across architecture during NVFP4 pretraining.<n>We find that, compared with Softmax Attention (SA), Linear Attention (LA) reduces per-tensor heavy tails but still exhibits persistent block-level spikes under block quantization.<n>We then develop CHON, an NVFP4 training recipe integrating with post-QK operation protection.
  arXiv  Detail & Related papers  (2026-02-02T12:50:27Z)
• Calibrated Decomposition of Aleatoric and Epistemic Uncertainty in Deep Features for Inference-Time Adaptation [3.018583625592182]
  Most estimators collapse all uncertainty modes into a single confidence score, preventing reliable reasoning about when to allocate more compute or adjust inference.<n>We introduce Uncertainty-Guided Inference-Time Selection, a lightweight inference time framework that disentangles aleatoric (data-driven) and model-driven uncertainty directly in deep feature space.
  arXiv  Detail & Related papers  (2025-11-15T23:47:30Z)
• Enhancing Kerr-Cat Qubit Coherence with Controlled Dissipation [64.05054054401175]
  Kerr-cat qubit (KCQ) is a bosonic quantum processor.<n>KCQs are experimentally compatible with on-chip architectures and high-fidelity operations.<n>We present direct evidence that the bit-flip time in a KCQ is limited by leakage out of the qubit manifold.
  arXiv  Detail & Related papers  (2025-11-02T17:58:36Z)
• Speeding Up MACE: Low-Precision Tricks for Equivarient Force Fields [51.95157731126864]
  Machine-learning force fields can deliver accurate molecular dynamics (MD) at high computational cost.<n>This thesis aims to make MACE cheaper and faster by identifying computational bottlenecks and evaluating low-precision execution policies.
  arXiv  Detail & Related papers  (2025-10-23T14:02:34Z)
• FlatQuant: Flatness Matters for LLM Quantization [58.28221892035609]
  We propose FlatQuant, a new post-training quantization approach that enhances the flatness of weights and activations.<n>Our approach identifies optimal affine transformations for each linear layer, calibrated in hours via a lightweight objective.<n>It achieves less than 1% accuracy drop for W4A4 quantization on the LLaMA-3-70B model, surpassing SpinQuant by 7.5%.
  arXiv  Detail & Related papers  (2024-10-12T08:10:28Z)
• Clustered Switchback Designs for Experimentation Under Spatio-temporal Interference [44.644520116360106]
  We estimate the global average treatment effect (GATE), the difference between average outcomes having exposed all units at all times to treatment or to control.<n>We propose a clustered switchback design, where units are grouped into clusters and time steps are grouped into blocks.<n>We show that for graphs that admit good clustering, a truncated Horvitz-Thompson estimator achieves a $tilde O(1/NT)$ mean squared error (MSE)<n>Our results simultaneously generalize the results from citethu2022switchback,ugander2013graph and citetleung2022rate
  arXiv  Detail & Related papers  (2023-12-25T01:00:58Z)

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.