CORTEXA
← Browse
arxivcs.CLcs.LG2026-07-10

Silent Failures in Quantized LLM Reasoning: A Taxonomy-Based Analysis of Hollow Convergence and Failure Mode Shifts

Renuka Oladri, Mohan Vamsi Varadaraju Priya, Jerry Wu

We show that post-training quantization can silently alter how large language models reason even when task accuracy is preserved. Using a six-category failure taxonomy validated by two independent human annotators (Cohen's $κ$ = 0.906), we classify 30,000 chain-of-thought outputs from five instruction-tuned LLMs (3B--14B parameters) across three quantization precisions (FP32, FP16, NF4) and four reasoning benchmarks. We find that while accuracy is robust across precisions (maximum 3.1 pp drop), Hollow Convergence (correct answers reached through incomplete or unverifiable reasoning) shows a significant size-dependent shift under NF4, dropping sharply for the two smallest models tested but remaining invariant for models at 12B parameters and above. This effect is also benchmark-specific: GSM8K is categorically immune while LogiQA and ARC-Challenge show the largest shifts. Furthermore, under NF4, Shortcut Collapse rises from 44% to 78% of wrong-answer failures in LLaMA 3.2-3B while Confidence Snowballing collapses from 15.8% to near zero, a qualitative shift invisible to accuracy metrics. Finally, we show Hollow Convergence cannot be reliably detected from surface-level text features (best F1 = 0.53), establishing it as a deployment-relevant failure mode that standard evaluation pipelines cannot catch.

View free PDFSource page

Related papers

arxivcs.DBcs.AIcs.CLcs.LG2026-07-24

DBA-Bench: A Production-Fidelity Benchmark for LLM-Based Database Operations Agents

Junming Chen, Junyang Jiang, Xu Chen, Zibo Liang, Kai Zheng

LLM-based database agents show promise, but differing task scopes, testbeds, and metrics hinder comparison. We identify four gaps between evaluation and production operations: live-environment fidelity (multi-turn read-write interaction with a running database); observation-space…

View free PDFSource page
arxivcs.CLcs.LG2026-07-06

SalAngaBhava: A Sinhala Market Dataset for Aspect-based Sentiment Analysis

Lakshani Galwatta, Nisansa de Silva, Sarangi Aththanayake, Adithya Galwatta

Sentiment analysis has been a primary domain under Natural Language Processing (NLP) from its inception as it plays a vital role in both real-world and research applications. In high-resource languages, this has been extended a step further, and instead of predicting sentiment at…

View free PDFSource page
arxivcs.CLcs.AIcs.LG2026-07-03

KARMA: Knowledge graph-based Automated Reasoning Materialization and Alignment

Jinkyeong Choi, Chaebin Jeong, Donghyeon Park

Template-based contrastive synthesis is scalable, but its candidates often differ only in a few entity-slots while sequence-level optimization spreads supervision over mostly shared templates. We formalize this as the Resolution Mismatch Problem and propose KARMA, which enumerate…

View free PDFSource page
arxivcs.CLcs.AIcs.CYcs.LG2026-06-26

Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty Prediction

Chenguang Wang, Ming Li, Xinyue Zeng, Zhuochun Li, Hong Jiao, Tianyi Zhou, et al.

Predicting human item difficulty is central to educational assessment, where reliable estimates support fairness and effective test construction. Existing methods often depend on costly human calibration or item-level textual representations, providing limited evidence about the…

View free PDFSource page