arXiv:2605.13076v2 Announce Type: replace 
Abstract: The LLM-based generation of machine-readable outputs such as JSON has attracted significant attention for integration with external systems. However, existing approaches cannot strictly enforce the maximum number of tokens to be generated, leading to infinite generation or truncated outputs that cause a system malfunction. To address this limitation, we propose TruncProof, a novel grammar-constrained generation method that enables LLMs to produce grammatically valid JSONs while adhering to a predefined token limit. By leveraging the properties of LL(1) parsers, TruncProof efficiently approximates the minimum number of tokens required to complete a grammatically valid output at each decoding step. Experiments on the Text-to-JSON instruction tasks demonstrate that TruncProof successfully generates syntactically correct outputs even under strict token constraints. Furthermore, we show that TruncProof can be effectively combined with advanced decoding strategies, resulting in outputs that are not only grammatically valid but also semantically accurate.

تم تقديم TruncProof كطريقة جديدة لتوليد مخرجات JSON من نماذج اللغة الكبيرة (LLMs) مع الالتزام بقيود صارمة على طول الرموز، مما يعالج مشكلات التوليد اللانهائي والمخرجات المقطوعة التي يمكن أن تعطل وظيفة النظام.

TruncProof se ha presentado como un método novedoso para generar salidas JSON a partir de modelos de lenguaje de gran tamaño (LLM) mientras se cumplen estrictas limitaciones de longitud de tokens, abordando problemas de generación infinita y salidas truncadas que pueden interrumpir la funcionalidad del sistema.

TruncProof a été présenté comme une méthode novatrice pour générer des sorties JSON à partir de modèles de langage de grande taille (LLM) tout en respectant des contraintes strictes de longueur de jeton, abordant les problèmes de génération infinie et de sorties tronquées qui peuvent perturber le fonctionnement du système.

TruncProof has been introduced as a novel method for generating JSON outputs from large language models (LLMs) while adhering to strict token-length constraints, addressing issues of infinite generation and truncated outputs that can disrupt system functionality.