How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failures

Code and data for the EMNLP 2025 paper submission "How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failures."

Overview

We characterize LLM reasoning failures through token-level uncertainty signals extracted from chain-of-thought traces, finding that failures emerge through two empirically distinguishable modes: committed failure, where a model locks onto an incorrect reasoning path early, and persistent uncertainty, where uncertainty accumulates throughout the trace. We identify the commitment point, the position in a reasoning trace beyond which additional tokens hurt rather than help failure detection, and demonstrate direct implications for self-consistency.

Repository Structure

├── *.py                                    # Inference pipelines for all model-dataset configurations
├── analyze_updated_dataset_agnostic.py     # Computes PR-AUC curves, bootstrap CIs (10,000 samples),
│                                           # and failure mode classification
└── results/
    ├── *.csv / *.xlsx                      # Inference outputs: token-level uncertainty features
    │                                       # and failure labels per configuration
    └── analysis/
        ├── failure_modes/                  # Per-configuration failure mode classifications
        │   └── *.txt                       # Committed/persistent classification with bootstrap
        │                                   # CIs and ΔPR-AUC statistics
        ├── self_consistency/               # Self-consistency triage and complementarity results
        └── plots/                          # PR-AUC curve plots for all configurations

Reproducing Results

Step 1: Run inference

python <dataset>_async_vllm.py

Output .csv / .xlsx files are saved to results/.

Step 2: Run failure mode analysis

python analyze_updated_dataset_agnostic.py --input results/<output_file>

This computes token-level uncertainty features over early windows, fits a stratified logistic regression classifier with 5-fold cross-validation, computes PR-AUC at each window size and runs 10,000-iteration paired bootstrap to produce confidence intervals on ΔPR-AUC. Classification outputs are saved to results/analysis/failure_modes/.

Requirements

pip install -r requirements.txt

Open-weight models are served via vLLM. GPT-4o and Gemini2.5-Pro experiments use the OpenAI API with top_logprobs=20. All experiments use temperature=0.6.