titleLLM-as-Jev: LLMs Are Already Jev-Style Decision Models -- When and How to Fine-Tune Them
authorsYinheng Li, Justin Wagle
date2026-10-04
venuearXiv (preprint, not peer reviewed)
idarXiv:2610.02076
affiliationMicrosoft
upvotes14 (Hugging Face Daily Papers, 2026-11-01)
linksabs pdf hf
abstractJev-style decision models return categorical probability distributions over predefined options without generating free-form text, enabling software systems to act on their outputs directly. In this work, we investigate the extent to which general-purpose LLMs already possess this capability out of the box, and when fine-tuning is actually necessary. We present LLM-as-Jev, an architecture-preserving framework that extracts calibrated decisions directly from next-token probabilities over bracketed numeric identifiers. LLM-as-Jev provides both a training-free inference recipe and a fine-tuning objective that optimizes candidate selection via a tree-factorized listwise loss while anchoring auxiliary predictions to the base model using KL divergence penalties. Evaluating on Qwen3.5-4B and Qwen3-0.6B, we find that modern LLMs are inherently effective decision models: without training, the 4B model matches community Jev-style models built on the same backbone, outperforms letter-logit readouts, supports arbitrary option counts, and natively handles multimodal decisions over images. Fine-tuning provides targeted rather than universal benefits -- substantially improving weaker models and specific tasks (such as many-option intent routing), but offering diminishing returns for strong backbones. Crucially, our KL anchors prevent behavioral degradation in conversational text generation, with LoRA delivering the strongest performance on capable models.

cite

@misc{li2026llm,
  title={LLM-as-Jev: LLMs Are Already Jev-Style Decision Models -- When and How to Fine-Tune Them},
  author={Yinheng Li and Justin Wagle},
  year={2026},
  eprint={2610.02076},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2610.02076}
}