analiza aiqbench
This study provides a critical deployability analysis of on-device Named-Entity Recognition (NER), comparing classical encoders against local generative LLMs across accuracy, latency, and output validity. The findings reveal that while large instruct models achieve competitive accuracy, specialized bidirectional encoders offer superior efficiency with zero malformed outputs, whereas small generative models frequently produce invalid results unless scaled up. Furthermore, the research highlights significant overconfidence issues in encoder-based confidence scores that can be mitigated through temperature scaling, offering practical guidance for building reliable local AI systems.
On-Device Named-Entity Recognition: A Deployability Study of Accuracy, Cost, Reliability, and Confidence
arXiv:2610.00007v1 Announce Type: cross Abstract: Named-entity recognition (NER) is increasingly wanted on-device (no API, low latency, data kept local). The practitioner's question is not the leaderboard but which model is deployable, how to evaluate it without human annotation, and whether its confidence can be trusted. We answer these jointly. We place nine systems across three paradigms and 13 M to 8 B parameters: a classical tagger (spaCy), bidirectional-encoder specialists (GLiNER, 166 to 460 M), and generative LLMs run locally (Qwen3-0.6B/1.7B/4B-Instruct, DeepSeek-R1-1.5B/8B), on three datasets of differing character, and report accuracy plus two axes the literature omits: latency and output validity. Because our corpus (RSS-News) had no gold, we built silver gold from a cross-family LLM judge panel, then measured its fidelity against benchmark gold and a full human re-validation of the corpus (strict F1 0.95, an upper bound since the human gold was silver-seeded); gold provenance flips the paradigm ranking, moving from LLM-authored silver to human gold raises every encoder and lowers every generative model. On accuracy alone a 4 B instruct LLM is competitive (it leads on clean newswire), so the encoder's case is deployability: it matches or slightly trails at one-ninth to one-twenty-fourth the size, at millisecond-to-second latency, with zero malformed output, while the smallest generative models emit up to 27% invalid output on long inputs, a failure fixed by scale, not output budget. We then characterize GLiNER's per-span confidence: it ranks correctness well (AUROC 0.76 to 0.86) but is overconfident (ECE 0.24 to 0.47, halved by temperature scaling); thresholding gives a small honest out-of-sample F1 gain; an all-local small-to-large cascade gives a modest, corpus-dependent gain over cost-matched random routing; and confidence tracks correctness but not novelty. Every number recomputes offline from per-span records.
Abstract:Named-entity recognition (NER) is increasingly wanted on-device (no API, low latency, data kept local). The practitioner's question is not the leaderboard but which model is deployable, how to evaluate it without human annotation, and whether its confidence can be trusted. We answer these jointly. We place nine systems across three paradigms and 13 M to 8 B parameters: a classical tagger (spaCy), bidirectional-encoder specialists (GLiNER, 166 to 460 M), and generative LLMs run locally (Qwen3-0.6B/1.7B/4B-Instruct, DeepSeek-R1-1.5B/8B), on three datasets of differing character, and report accuracy plus two axes the literature omits: latency and output validity. Because our corpus (RSS-News) had no gold, we built silver gold from a cross-family LLM judge panel, then measured its fidelity against benchmark gold and a full human re-validation of the corpus (strict F1 0.95, an upper bound since the human gold was silver-seeded); gold provenance flips the paradigm ranking, moving from LLM-authored silver to human gold raises every encoder and lowers every generative model. On accuracy alone a 4 B instruct LLM is competitive (it leads on clean newswire), so the encoder's case is deployability: it matches or slightly trails at one-ninth to one-twenty-fourth the size, at millisecond-to-second latency, with zero malformed output, while the smallest generative models emit up to 27% invalid output on long inputs, a failure fixed by scale, not output budget. We then characterize GLiNER's per-span confidence: it ranks correctness well (AUROC 0.76 to 0.86) but is overconfident (ECE 0.24 to 0.47, halved by temperature scaling); thresholding gives a small honest out-of-sample F1 gain; an all-local small-to-large cascade gives a modest, corpus-dependent gain over cost-matched random routing; and confidence tracks correctness but not novelty. Every number recomputes offline from per-span records.
7 pages, 5 figures, 12 tables. Code and per-span records reproduce all reported numbers offline
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2610.00007 [cs.CL]
(or arXiv:2610.00007v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.00007 Focus to learn more arXiv-issued DOI via DataCite
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.
Text integral sub licența deschisă a publicației · 392 cuvinte
Subiecte conexe
Digital States 2026: AI Moves From Experiment to Infrastructure
govtech.com · 24 sept. 2026
News
Quantum Computing Report · 25 iul. 2026
Presentation: Running AI at the Edge: Running Real Workloads Directly in the Browser
InfoQ · 31 aug. 2026
Researchers Evaluate Quantum and Classical Models Achieving 90 Per Cent Digit Recognition Accuracy
Quantum Zeitgeist · 30 mai 2026