Skip to content
AI News HubLIVE
Original source2 min read

Does the Ranking Hold? A Practitioner-Facing Audit of Hallucination Detectors on Small, Quantized, On-Device LLMs

Summary

A reproducible single-GPU audit finds that hallucination detector rankings invert across small 4-bit quantized on-device models (Spearman rho = -0.80), and a cheap lexical baseline is statistically indistinguishable from sampling-based methods. The study also shows the standard TruthfulQA-Generation labeler yields near-chance AUROC (0.46–0.59), a labeler failure rather than a detector failure.

SourcearXiv AIAuthor: Shubhra Mittal, Gaurav Naresh Mittal
Does the Ranking Hold? A Practitioner-Facing Audit of Hallucination Detectors on Small, Quantized, On-Device LLMs
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 31 Aug 2026]

Title:Does the Ranking Hold? A Practitioner-Facing Audit of Hallucination Detectors on Small, Quantized, On-Device LLMs

View a PDF of the paper titled Does the Ranking Hold? A Practitioner-Facing Audit of Hallucination Detectors on Small, Quantized, On-Device LLMs, by Shubhra Mittal and Gaurav Naresh Mittal

View PDF

Abstract:Hallucination detection for large language models is dominated by methods evaluated on 7B-70B research models with abundant sampling budgets and exact-match or ROUGE correctness labels. The small models practitioners actually ship on-device, Phi-3-mini, Llama-3.2-3B-Instruct, Qwen2.5-3B-Instruct, and Gemma-2-2B-Instruct, typically at 4-bit precision, sit outside this evidence base. We present a reproducible one-GPU audit of four detectors spanning three paradigms: length-normalized token-margin calibration, a lexical n-gram baseline, and two sampling-based consistency methods (semantic entropy and SelfCheckGPT-NLI) on two small open instruction-tuned models (Llama-3.2-3B-Instruct and Qwen2.5-3B-Instruct) loaded at 4-bit NF4 quantization on a single Colab T4 GPU, evaluated on closed-book TriviaQA. Cross-model ranking transfer is poor: Spearman rho is -0.80 and Kendall tau-b is -0.667, meaning the detector ranking actively inverts between the two models, and the cheap lexical baseline is statistically indistinguishable from the sampling-based consistency detectors on both models under bootstrap 95 percent confidence intervals. As a second, methodological contribution, we show that the standard token-overlap-to-multiple-choice correctness labeler for TruthfulQA-Generation produces near-chance AUROC (0.46-0.59) for all four detectors on both models, a labeler failure, not a detector failure, since the same detectors reach 0.69-0.89 AUROC on TriviaQA under the identical pipeline. We release the harness, prompts, and per-item scores and frame the detector choice explicitly as a fairness-of-error-rates allocation problem for practitioners deploying on resource-constrained, on-device hardware.

Subjects:

Artificial Intelligence (cs.AI)

Cite as: arXiv:2611.17339 [cs.AI]

(or arXiv:2611.17339v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2611.17339

arXiv-issued DOI via DataCite

Submission history

From: Shubhra Mittal [view email] [v1] Mon, 31 Aug 2026 19:11:30 UTC (217 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Does the Ranking Hold? A Practitioner-Facing Audit of Hallucination Detectors on Small, Quantized, On-Device LLMs, by Shubhra Mittal and Gaurav Naresh Mittal

View PDF

view license

Additional Features

Audio Summary

Current browse context:

cs.AI

new | recent | 2026-11

Change to browse by:

cs

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Detector rankings actively invert between Llama-3.2-3B-Instruct and Qwen2.5-3B-Instruct, so rankings do not transfer across models.
  • A cheap lexical n-gram baseline is statistically indistinguishable from semantic entropy and SelfCheckGPT-NLI on TriviaQA.
  • The standard token-overlap labeler for TruthfulQA-Generation produces near-chance AUROC (0.46–0.59) for all detectors, while the same pipeline reaches 0.69–0.89 on TriviaQA.
  • The authors release the harness, prompts, and per-item scores and frame detector choice as a fairness-of-error-rates allocation problem for resource-constrained hardware.

Highlights and analysis are generated automatically and may contain errors. Check the original source.