{"slug": "turnslide-scalable-multi-turn-data-synthesis-by-walking-a-finite-state-machine", "title": "Turnslide: Scalable Multi-Turn Data Synthesis by Walking a Finite-State Machine", "summary": "A new arXiv paper (2610.07070v1) introduces Turnslide, a fully automated multi-turn data synthesis framework that models each API as a finite-state machine to generate state-valid tool-call sequences, then translates them into complete training examples with a single LLM call. Fine-tuning small language models on Turnslide-generated trajectories reached 70.7% full accuracy versus 63.4% and 53.7% for existing works, using 3.6-6.6x fewer tokens, and significantly improved downstream accuracy over an unmutated baseline. The authors target the shortage of per-API multi-turn tool-calling data needed to fine-tune cheap, privately hosted SLMs.", "body_md": "arXiv:2610.07070v1 Announce Type: new \nAbstract: Small language models are inexpensive to serve and can run on private infrastructure, but base models are often not good enough at multi-turn tool calling, and fine-tuning them needs per-API data that rarely exists. Existing synthesis methods are too expensive for high-scale fine-tuning, as they often require mock operational environments for different domains and multiple LLM calls per generated conversation turn. We introduce a fully automated, lightweight synthesis framework that models each API as a finite-state machine, representing the system as abstract states that determine when each tool may be called, producing state-valid sequences of tools; sequences are translated into complete examples with a single LLM call. Rather than optimize diversity, we set a target distribution over the number of turns, the tool sequence and task complexity. We measure data quality by fine-tuning SLMs on generated trajectories, showing that our FSM-based generation significantly improves downstream accuracy over an unmutated baseline and, against existing works, reaches 70.7% full accuracy over 63.4% and 53.7% with 3.6-6.6$\\times$ fewer tokens.", "url": "https://wpnews.pro/news/turnslide-scalable-multi-turn-data-synthesis-by-walking-a-finite-state-machine", "canonical_source": "https://arxiv.org/abs/2610.07070", "published_at": "2026-10-07 04:00:00+00:00", "updated_at": "2026-10-07 04:19:41.707760+00:00", "lang": "en", "topics": ["large-language-models", "ai-agents", "ai-research", "machine-learning", "ai-tools"], "entities": ["Turnslide", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/turnslide-scalable-multi-turn-data-synthesis-by-walking-a-finite-state-machine", "markdown": "https://wpnews.pro/news/turnslide-scalable-multi-turn-data-synthesis-by-walking-a-finite-state-machine.md", "text": "https://wpnews.pro/news/turnslide-scalable-multi-turn-data-synthesis-by-walking-a-finite-state-machine.txt", "jsonld": "https://wpnews.pro/news/turnslide-scalable-multi-turn-data-synthesis-by-walking-a-finite-state-machine.jsonld"}}