Enregistré dans:
Détails bibliographiques
Auteur principal: Özbay, Oğuzhan
Format: Recurso digital
Langue:
Publié: Zenodo 2026
Accès en ligne:https://doi.org/10.5281/zenodo.18626464
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866901746206375936
author Özbay, Oğuzhan
author_facet Özbay, Oğuzhan
contents <p><span lang="EN-GB">We introduce <strong>Entropy Bandwidth Y-Axis (EBY)</strong>, a scalar diagnostic metric designed to quantify the behavioural quality of autoregressive language models during training. While conventional metrics such as cross-entropy loss and perplexity measure predictive accuracy, they provide no direct signal about the shape of the model’s output distribution—specifically, the balance between diversity and confidence.</span></p> <p><span lang="EN-GB">EBY captures this balance by combining normalized Shannon entropy with top-</span><span lang="EN-GB"></span><span lang="EN-GB"><span> </span>probability mass (sharpness), yielding a bounded signal in </span><span lang="EN-GB"></span><span lang="EN-GB">. Crucially, both degenerate regimes—deterministic collapse and incoherent randomness—map to low EBY values, while only balanced, human-like predictive behavior yields high EBY.</span></p> <p><span lang="EN-GB">We formalize EBY, analyse its asymptotic behaviour, introduce positional variants (EBY-Head and EBY-Tail) for temporal degradation detection, and describe a trend-based monitoring system suitable for online training. Empirical results from training a 381M-parameter Turkish language model demonstrate that EBY detects distributional collapse several hundred optimization steps before perplexity degradation becomes visible, enabling early corrective intervention. EBY thus provides a complementary behavioural axis for model diagnostics, orthogonal to loss-based metrics.</span></p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18626464
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Entropy Bandwidth Y-Axis (EBY)
Özbay, Oğuzhan
<p><span lang="EN-GB">We introduce <strong>Entropy Bandwidth Y-Axis (EBY)</strong>, a scalar diagnostic metric designed to quantify the behavioural quality of autoregressive language models during training. While conventional metrics such as cross-entropy loss and perplexity measure predictive accuracy, they provide no direct signal about the shape of the model’s output distribution—specifically, the balance between diversity and confidence.</span></p> <p><span lang="EN-GB">EBY captures this balance by combining normalized Shannon entropy with top-</span><span lang="EN-GB"></span><span lang="EN-GB"><span> </span>probability mass (sharpness), yielding a bounded signal in </span><span lang="EN-GB"></span><span lang="EN-GB">. Crucially, both degenerate regimes—deterministic collapse and incoherent randomness—map to low EBY values, while only balanced, human-like predictive behavior yields high EBY.</span></p> <p><span lang="EN-GB">We formalize EBY, analyse its asymptotic behaviour, introduce positional variants (EBY-Head and EBY-Tail) for temporal degradation detection, and describe a trend-based monitoring system suitable for online training. Empirical results from training a 381M-parameter Turkish language model demonstrate that EBY detects distributional collapse several hundred optimization steps before perplexity degradation becomes visible, enabling early corrective intervention. EBY thus provides a complementary behavioural axis for model diagnostics, orthogonal to loss-based metrics.</span></p>
title Entropy Bandwidth Y-Axis (EBY)
url https://doi.org/10.5281/zenodo.18626464