Contrastive Attribution in the Wild: An Interpretability Analysis of LLM Failures on Realistic Benchmarks

• URL: http://arxiv.org/abs/2604.17761v1
• Date: Mon, 20 Apr 2026 03:24:11 GMT
• Title: Contrastive Attribution in the Wild: An Interpretability Analysis of LLM Failures on Realistic Benchmarks
• Abstract summary: Interpretability tools are increasingly used to analyze failures of Large Language Models (LLMs)<n>We study contrastive, LRP-based attribution as a practical tool for analyzing LLM failures in realistic settings.<n>Our results show that this token-level contrastive attribution can yield informative signals in some failure cases, but is not universally applicable.
• Score: 42.92210265283373
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Interpretability tools are increasingly used to analyze failures of Large Language Models (LLMs), yet prior work largely focuses on short prompts or toy settings, leaving their behavior on commonly used benchmarks underexplored. To address this gap, we study contrastive, LRP-based attribution as a practical tool for analyzing LLM failures in realistic settings. We formulate failure analysis as \textit{contrastive attribution}, attributing the logit difference between an incorrect output token and a correct alternative to input tokens and internal model states, and introduce an efficient extension that enables construction of cross-layer attribution graphs for long-context inputs. Using this framework, we conduct a systematic empirical study across benchmarks, comparing attribution patterns across datasets, model sizes, and training checkpoints. Our results show that this token-level contrastive attribution can yield informative signals in some failure cases, but is not universally applicable, highlighting both its utility and its limitations for realistic LLM failure analysis. Our code is available at: https://aka.ms/Debug-XAI.

Related papers

• HPFA: Hypergraph-Based Paired Failure Attribution for LLM Reasoning [15.068861766259412]
  We propose a hypergraph-based paired failure attribution framework that attributes the failure root cause by comparing the hyperedges of the targeted failure reasoning path against a reference successful path.<n>By reducing the search space, our method efficiently localizes root causes and enables scalable synthesis of attribution data.<n>Experiments on mathematical reasoning and agentic coding tasks demonstrate that HPFA can dramatically increase attribution accuracy and efficiency.
  arXiv  Detail & Related papers  (2026-08-03T10:23:48Z)
• How LLMs Fail and Generalize in RTL Coding for Hardware Design? [56.361436215029045]
  We introduce a new error taxonomy grounded in problem solvability, inspired by cognitive theory.<n>Our taxonomy categorizes failures into syntactic, semantic, solvable functional, and unsolvable functional types.
  arXiv  Detail & Related papers  (2026-04-26T14:34:49Z)
• METER: Evaluating Multi-Level Contextual Causal Reasoning in Large Language Models [61.33372454250959]
  Contextual causal reasoning is a critical yet challenging capability for Large Language Models.<n>Existing benchmarks fail to ensure context consistency or cover the full causal hierarchy.<n>We pioneer METER to systematically benchmark LLMs across all three levels of the causal ladder.
  arXiv  Detail & Related papers  (2026-04-13T14:07:11Z)
• From Early Encoding to Late Suppression: Interpreting LLMs on Character Counting Tasks [49.57538588967748]
  Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks.<n>We uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma.<n>We show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph.
  arXiv  Detail & Related papers  (2026-04-01T11:40:12Z)
• LogICL: Distilling LLM Reasoning to Bridge the Semantic Gap in Cross-Domain Log Anomaly Detection [4.319103554448838]
  We propose LogICL, a framework distilling Large Language Model (LLM) reasoning into a lightweight encoder for cross-domain anomaly detection.<n>At inference, the optimized encoder retrieves reasoning-aware demonstrations using semantic similarity and delta scores.<n>Experiments on few-shot and zero-shot cross-domain benchmarks confirm LogICL achieves state-of-the-art performance across heterogeneous systems.
  arXiv  Detail & Related papers  (2025-12-10T13:13:30Z)
• Where Did It Go Wrong? Attributing Undesirable LLM Behaviors via Representation Gradient Tracing [12.835224376066769]
  Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their deployment is frequently undermined by undesirable behaviors.<n>We introduce a novel and efficient framework that diagnoses a range of undesirable LLM behaviors by analyzing representation and its gradients.<n>We systematically evaluate our method for tasks that include tracking harmful content, detecting backdoor poisoning, and identifying knowledge contamination.
  arXiv  Detail & Related papers  (2025-09-26T12:07:47Z)
• CLEAR: Error Analysis via LLM-as-a-Judge Made Easy [9.285203198113917]
  We introduce CLEAR, an interactive, open-source package for LLM-based error analysis.<n> CLEAR first generates per-instance textual feedback, then creates a set of system-level error issues, and quantifies the prevalence of each identified issue.<n>Our package also provides users with an interactive dashboard that allows for a comprehensive error analysis through aggregate visualizations.
  arXiv  Detail & Related papers  (2025-07-24T13:15:21Z)
• IDA-Bench: Evaluating LLMs on Interactive Guided Data Analysis [60.32962597618861]
  IDA-Bench is a novel benchmark evaluating large language models in multi-round interactive scenarios.<n>Agent performance is judged by comparing its final numerical output to the human-derived baseline.<n>Even state-of-the-art coding agents (like Claude-3.7-thinking) succeed on 50% of the tasks, highlighting limitations not evident in single-turn tests.
  arXiv  Detail & Related papers  (2025-05-23T09:37:52Z)
• Mitigating Forgetting in LLM Fine-Tuning via Low-Perplexity Token Learning [65.23593936798662]
  We show that fine-tuning with LLM-generated data improves target task performance and reduces non-target task degradation.<n>This is the first work to provide an empirical explanation based on token perplexity reduction to mitigate catastrophic forgetting in LLMs after fine-tuning.
  arXiv  Detail & Related papers  (2025-01-24T08:18:56Z)
• Benchmarking Causal Study to Interpret Large Language Models for Source Code [6.301373791541809]
  This paper introduces a benchmarking strategy named Galeras comprised of curated testbeds for three SE tasks. We illustrate the insights of our benchmarking strategy by conducting a case study on the performance of ChatGPT under distinct prompt engineering methods.
  arXiv  Detail & Related papers  (2023-08-23T20:32:12Z)
• Competence-Based Analysis of Language Models [21.43498764977656]
  CALM (Competence-based Analysis of Language Models) is designed to investigate LLM competence in the context of specific tasks.<n>We develop a new approach for performing causal probing interventions using gradient-based adversarial attacks.<n>We carry out a case study of CALM using these interventions to analyze and compare LLM competence across a variety of lexical inference tasks.
  arXiv  Detail & Related papers  (2023-03-01T08:53:36Z)

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.