Newest papers

Predictable performance is suboptimal, and what it means for model routing

Guanhua Zhang, Moritz Hardt · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31428 · cited by 0 works (OpenAlex) · open-access PDF

Hindi

बड़े fin rohibing भाषा मॉडल (LLMs) क्षमताओं और लागतों में भिन्न होते हैं, मार्गदर्शन के तरीकों को प्रेरित करते हैं जो प्रत्येक पूछताछ के लिए अनुकूलित रूप से एक मॉडल चुनते हैं। ये तरीके प्रदर्शन पूर्वानुमान पर निर्भर करते हैं: एक ही पूछताछ के माध्यम से, यह अनुमान लगाते हैं कि क्या एक मॉडल इसे सही ढंग से जवाब देगा। अधिकांश मार्गदर्शन तरीकों को चुनने वाले मॉडल के प्रदर्शन के अलग-अलग पूर्वानुमानों को मार्गदर्शन करने के लिए प्रशिक्षित किया जाता है। हमारे काम से पता चलता है कि प्रदर्शन पूर्वानुमान केवल एक सरल बुनियादी लाइन को हराता है जब उम्मीदवार मॉडल कार्य या प्रदर्शन पूर्वानुमान कार्य पर अयोग्य है। विशेष रूप से, हम मॉडल के आत्मविश्वास को लेते हैं,

Machine translation (Meta M2M100); the English original follows.

Abstract

Large language models (LLMs) differ in capabilities and costs, motivating routing methods that adaptively choose a model for each query. These methods rely on performance prediction: given a query, estimate whether a model will answer it correctly. Most routing methods train separate predictors of candidate models' performance to guide model selection. Our work shows that performance prediction only beats a simple baseline when the candidate model is suboptimal on the task or the performance predictor adds significant capacity relative to the candidate model. Specifically, we take a model's self-confidence, the probability it assigns to its chosen answer, as a baseline and study how much a performance predictor can improve on it. First, we show that any performance predictor that outperforms a model's self-confidence can be used to improve the model's loss: the predictor's advantage transfers exactly in log loss through recalibration. Second, accounting for the predictor's additional parameters, we bound its advantage by the model's suboptimality plus the gain from those parameters. Third, we extend our theoretical results to two-model cascades, which route queries to a more capable model when the first model is predicted to fail. Experiments with 16 language models on four benchmarks show that predictors gain a larger advantage over self-confidence on more suboptimal models. This advantage shrinks as models improve through finetuning. Cascading exhibits the same two patterns: predictor-based routing offers larger gains over confidence-based routing for more suboptimal models, and these gains diminish after finetuning.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Task (project management), Baseline (sea), Routing (electronic design automation), Performance prediction, Performance improvement

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange