Copertina

Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Alimaskina, Ekaterina, Rudas, Darya, Shveykin, Denis, Molodtsov, Gleb, Vasiliev, Pavel, Beznosikov, Aleksandr
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Artificial Intelligence Machine Learning
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

HARP: Hadamard-Preconditioned Adaptive Rotation Processor for Extreme LLM Quantization
di: Zagitov, Artur, et al.
Pubblicazione: (2026)

Hierarchical Mixture-of-Experts with Two-Stage Optimization
di: Molodtsov, Gleb, et al.
Pubblicazione: (2026)

Optimal Data Splitting in Distributed Optimization for Machine Learning
di: Medyakov, Daniil, et al.
Pubblicazione: (2024)

Effective Method with Compression for Distributed and Federated Cocoercive Variational Inequalities
di: Medyakov, Daniil, et al.
Pubblicazione: (2024)

Shuffling Heuristic in Variational Inequalities: Establishing New Convergence Guarantees
di: Medyakov, Daniil, et al.
Pubblicazione: (2025)

Variance Reduction Methods Do Not Need to Compute Full Gradients: Improved Efficiency through Shuffling
di: Medyakov, Daniil, et al.
Pubblicazione: (2025)

LCD: Advancing Extreme Low-Bit Clustering for Large Language Models via Knowledge Distillation
di: Liu, Fangxin, et al.
Pubblicazione: (2025)

Quant-dLLM: Post-Training Extreme Low-Bit Quantization for Diffusion Large Language Models
di: Zhang, Tianao, et al.
Pubblicazione: (2025)

Thinking like a CHEMIST: Combined Heterogeneous Embedding Model Integrating Structure and Tokens
di: Rekut, Nikolai, et al.
Pubblicazione: (2025)

Quantization Meets Reasoning: Exploring and Mitigating Degradation of Low-Bit LLMs in Mathematical Reasoning
di: Li, Zhen, et al.
Pubblicazione: (2025)

Diagonal-Tiled Mixed-Precision Attention for Efficient Low-Bit MXFP Inference
di: Ding, Yifu, et al.
Pubblicazione: (2026)

HESTIA: A Hessian-Guided Differentiable Quantization-Aware Training Framework for Extremely Low-Bit LLMs
di: Wang, Guoan, et al.
Pubblicazione: (2026)

PTQ1.61: Push the Real Limit of Extremely Low-Bit Post-Training Quantization Methods for Large Language Models
di: Zhao, Jiaqi, et al.
Pubblicazione: (2025)

Metropolis-Scale Road Network Datasets for Fine-Grained Urban Traffic Modeling
di: Velikonivtsev, Fedor, et al.
Pubblicazione: (2025)

LUT-DLA: Lookup Table as Efficient Extreme Low-Bit Deep Learning Accelerator
di: Li, Guoyu, et al.
Pubblicazione: (2025)

Communication-Efficient Federated Learning with Adaptive Number of Participants
di: Skorik, Sergey, et al.
Pubblicazione: (2025)

I-LLM: Efficient Integer-Only Inference for Fully-Quantized Low-Bit Large Language Models
di: Hu, Xing, et al.
Pubblicazione: (2024)

Degradation Modeling and Prognostic Analysis Under Unknown Failure Modes
di: Fu, Ying, et al.
Pubblicazione: (2024)

Sign-SGD via Parameter-Free Optimization
di: Medyakov, Daniil, et al.
Pubblicazione: (2025)

Revealing Interpretable Failure Modes of VLMs
di: Chaudhary, Isha, et al.
Pubblicazione: (2026)

Unleashing Low-Bit Inference on Ascend NPUs: A Comprehensive Evaluation of HiFloat Formats
di: Zhao, Pengxiang, et al.
Pubblicazione: (2026)

LittleBit: Ultra Low-Bit Quantization via Latent Factorization
di: Lee, Banseok, et al.
Pubblicazione: (2025)

Dynamic Low-rank Approximation of Full-Matrix Preconditioner for Training Generalized Linear Models
di: Matveeva, Tatyana, et al.
Pubblicazione: (2025)

Memory Injections: Correcting Multi-Hop Reasoning Failures during Inference in Transformer-Based Language Models
di: Sakarvadia, Mansi, et al.
Pubblicazione: (2023)

Large Language Model Reasoning Failures
di: Song, Peiyang, et al.
Pubblicazione: (2026)

More Than Bits: Multi-Envelope Double Binary Factorization for Extreme Quantization
di: Ichikawa, Yuma, et al.
Pubblicazione: (2025)

Influence-Inspired Spectral Rotations for Extreme Low-Bit LLM Quantization
di: Pavlov, Gorgi
Pubblicazione: (2026)

What Makes Low-Bit Quantization-Aware Training Work for Reasoning LLMs? A Systematic Study
di: Lv, Keyu, et al.
Pubblicazione: (2026)

SHARP: Accelerating Language Model Inference by SHaring Adjacent layers with Recovery Parameters
di: Wang, Yiping, et al.
Pubblicazione: (2025)

Learning Grouped Lattice Vector Quantizers for Low-Bit LLM Compression
di: Zhang, Xi, et al.
Pubblicazione: (2025)

InfoQuant: Shaping Activation Distributions for Low-Bit LLM Quantization
di: Li, Ke, et al.
Pubblicazione: (2026)

Why Do Some Inputs Break Low-Bit LLM Quantization?
di: Chang, Ting-Yun, et al.
Pubblicazione: (2025)

Pushing the Limits of Low-Bit Optimizers: A Focus on EMA Dynamics
di: Xu, Cong, et al.
Pubblicazione: (2025)

SplitQuant: Layer Splitting for Low-Bit Neural Network Quantization
di: Song, Jaewoo, et al.
Pubblicazione: (2025)

Solving Dual Sourcing Problems with Supply Mode Dependent Failure Rates
di: Akkerman, Fabian, et al.
Pubblicazione: (2024)

MemFail: Stress-Testing Failure Modes of LLM Memory Systems
di: Garg, Ishir, et al.
Pubblicazione: (2026)

Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking
di: Feuer, Benjamin, et al.
Pubblicazione: (2024)

Target Concept Tuning Improves Extreme Weather Forecasting
di: Ren, Shijie, et al.
Pubblicazione: (2026)

Bayesian Inverse Problems Meet Flow Matching: Efficient and Flexible Inference via Transformers
di: Sherki, Daniil, et al.
Pubblicazione: (2025)

ECQ$^{\text{x}}$: Explainability-Driven Quantization for Low-Bit and Sparse DNNs
di: Becking, Daniel, et al.
Pubblicazione: (2021)