arXiv:2605.17774v2 Announce Type: replace 
Abstract: Large language models are increasingly used as planning components in agentic systems, but current tool-use pipelines often require full tool schemas to be included in every prompt, creating substantial token overhead and limiting the practicality of smaller models. This paper investigates whether tool-use knowledge can be internalized into small language models through parameter-efficient fine-tuning, enabling structured planning without explicit tool descriptions at inference time. Using AssetOpsBench as the primary benchmark, we fine-tune Gemma 4 E4B and Qwen3-4B with 8-bit QLoRA on approximately 1,700 tool-use examples spanning tool knowledge, question-to-plan mappings, and execution-style traces. We evaluate the resulting models under description-free inference, where the prompt omits the tool catalog entirely. The fine-tuned models outperform an informed unfine-tuned baseline that receives full tool descriptions, reducing input length by 82.6\% while improving structural and LLM-judge planning scores. In the best Gemma run, the model achieves an AT-F1 of 0.65 and an overall judge score of 3.88, compared with 0.47 and 2.88 for the informed baseline. Qwen3-4B achieves a strong overall judge score of 3.78 while using 62\% less memory and running 2.5$\times$ faster than Gemma, though it also exhibits greater catastrophic forgetting on general multiple-choice benchmarks. Additional ablations show that LoRA rank controls a quality--retention trade-off, with $r=32$ maximizing planning quality and smaller ranks preserving more general knowledge. These results suggest that, for fixed tool catalogs, QLoRA fine-tuning can shift tool knowledge from prompt context into model weights, substantially reducing inference overhead while maintaining or improving tool-planning quality.

أظهرت دراسة حديثة أن نماذج اللغة الصغيرة يمكن أن تستوعب معرفة الأدوات من خلال تحسين QLoRA، مما يسمح لها بإجراء تخطيط منظم دون الحاجة إلى أوصاف أدوات صريحة أثناء الاستدلال. استخدمت الدراسة معيار AssetOpsBench وقامت بتحسين نماذج مثل Gemma 4 E4B و Qwen3-4B على حوالي 1700 مثال لاستخدام الأدوات.

Un estudio reciente ha demostrado que los pequeños modelos de lenguaje pueden internalizar el conocimiento sobre herramientas a través del ajuste fino QLoRA, lo que les permite realizar una planificación estructurada sin necesidad de descripciones explícitas de herramientas durante la inferencia. La investigación utilizó el benchmark AssetOpsBench y ajustó modelos como Gemma 4 E4B y Qwen3-4B en aproximadamente 1,700 ejemplos de uso de herramientas.

Une étude récente a démontré que les petits modèles de langage peuvent internaliser les connaissances sur les outils grâce à un affinement QLoRA, leur permettant d'effectuer une planification structurée sans avoir besoin de descriptions explicites des outils lors de l'inférence. La recherche a utilisé le benchmark AssetOpsBench et a affiné des modèles tels que Gemma 4 E4B et Qwen3-4B sur environ 1 700 exemples d'utilisation d'outils.

A recent study has demonstrated that small language models can internalize tool knowledge through QLoRA fine-tuning, allowing them to perform structured planning without needing explicit tool descriptions during inference. The research utilized the AssetOpsBench benchmark and fine-tuned models like Gemma 4 E4B and Qwen3-4B on approximately 1,700 examples of tool usage.