English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

This paper addresses fine-tuning Large Language Models (LLMs) for functioncalling tasks when real user interaction data is unavailable. In digitalcontent creation tools, where users express their needs through naturallanguage queries that must be mapped to API calls, the lack of real-worldtask-specific data and privacy constraints for training on it necessitatesynthetic data generation. Existing approaches to synthetic data generationfall short in diversity and complexity, failing to replicate real-world datadistributions and leading to suboptimal performance after LLM fine-tuning. Wepresent a novel router-based architecture that leverages domain resources likecontent metadata and structured knowledge graphs, along with text-to-text andvision-to-text language models to generate high-quality synthetic trainingdata. Our architecture's flexible routing mechanism enables synthetic datageneration that matches observed real-world distributions, addressing afundamental limitation of traditional approaches. Evaluation on a comprehensiveset of real user queries demonstrates significant improvements in both functionclassification accuracy and API parameter selection. Models fine-tuned with oursynthetic data consistently outperform traditional approaches, establishing newbenchmarks for function calling tasks.

RouteNator: A Router-Based Multi-Modal Architecture for Generating Synthetic Training Data for Function Calling LLMs