cs.LGNov 13, 2026

AdaMenu: Query-Adaptive Menus for Decision Models over Large Label Spaces

Authors: Jiangze Yan, Yi Shen, Jieyun Huang, Wenjing Zhang, Zhaoxiang Liu, Ning Wang, Kai Wang, Shiguo Lian

Organizations: Unicom Data Intelligence, China Unicom · Data Science & Artificial Intelligence Research Institute, China Unicom

Abstract

Decision models such as Jev return probabilities over caller-defined options and are increasingly used for classification. Large label spaces challenge both accuracy and scalability: larger menus can impair decisions, increase input costs, and exceed option limits. Hierarchical classification reduces menu size but introduces serial calls and error propagation. We propose AdaMenu, which builds a compact candidate menu from retrieved labeled examples and makes a single decision-model request. Its candidate labels adapt to the query, and its default setting also adjusts menu size using aggregated retrieval scores. We evaluate AdaMenu with open and closed decision models on four label sets containing up to 1,016 labels. It outperforms TypeSafe's beam-search baseline on all 16 model and label-set pairs where both methods run, improving accuracy by 16.6-34.3 percentage points. Relative to the full menu on the three smaller label sets, accuracy differences range from -0.5 to +7.2 percentage points, while input token usage falls by factors of 3.2-22.9. AdaMenu also runs when option limits exclude the full menu, including settings where the hierarchical baseline is unavailable.

Figures & tables

Explore similar work

Oct 6, 2026

JevCascade: Readout Stability and Inference-Time Compute Allocation for Prefill-Only Decision Models

We present JevCascade, an inference-time compute allocation framework for prefill-only decision models, built on readout stability and zero-label restricted readout prediction. Prefill-only decision models score every candidate in a menu in a single forward pass and never decode, which makes one call one to two orders of magnitude cheaper than a same-scale generative language model. We show that this read-out structure carries a testable property. When an intervention changes only the menu and leaves the input text fixed, the post-intervention accuracy is already determined by the cached first-pass distribution. The restriction predictor restricts the pass-1 probabilities to the menu, renormalizes and reads off the argmax, with no labels and no second pass. Across six model families, ten core datasets (sixteen in total) and two task types, menu-only interventions are predicted to within 4.2 points, exactly for one family. The property is ordinal, not probabilistic, since a probability-level variant of the same estimator errs by 21.0 points, and class-specific, since same-scale generative models err by 1.6 to 15.8 points and degrade as they grow. It makes inference-time compute decidable before deployment. Uniform extra passes buy calibration but almost no accuracy; at matched cost a confidence cascade beats every scheme that re-asks the same model, and curating the menu beats enlarging the model, a 0.8B model on a curated 5-candidate menu reaching 95.4% on CLINC150 against 80.0% for a 4B model on the full 150-label menu.
Oct 12, 2026

Releasing the Inner Jev: A Simple Recipe for Open Decision Models

LLM-based decision-making systems, also called System One models, have recently gained popularity, most notably through Jev (Almeida, 2026). Given a typed input state, a question, and a set of options, they return a probability for each option in a single forward pass, without text generation. We show that existing open instruction-tuned language models can achieve comparable performance without any additional training, by reading out answers from letters (A, B, C, etc.) in the model's output logits at the last token position. Furthermore, we show how a simple supervised fine-tuning recipe can improve their performance further. We evaluate four open instruction-tuned language models: Gemma 4 31B, Gemma 4 12B, Qwen3.6-27B, and Qwen3.5-4B. We fine-tune each using LoRA and a decision head initialized from a slice of the language model head. Furthermore, we introduce OpenDecisions, a novel 2,352-question validation dataset designed to reflect the task composition of JevBench (JevBench main- tainers, 2026) and which computes a Capability score based on accuracy and calibration metrics. Without fine-tuning, Gemma 4 31B scores 79.6 on OpenDecisions, compared to Jev's 80.1. After undergoing our fine-tuning recipe, it scores 84.4 on OpenDecisions and 81.5 on JevBench v1.7, outperforming Jev by 4.3 and 4.4 Capability points respectively and setting a new state-of-the-art on OpenDecisions. On a suite of public decision datasets, our models achieve accuracy comparable to Jev, with lower calibration error. Our results show that open language models already possess System One capabilities, and with simple fine-tuning they outperform Jev on a wide range of decision tasks. We release our models, serving code, and OpenDecisions.
Oct 12, 2026

FunDM: Cheap & Label-Free Decision Models from Frozen Language Models

Decision models such as Jev score candidate answers or actions. A small decision head trained label-free can recover about 99% of a frozen language model's multiple-choice accuracy while using only two-thirds of its transformer layers. It learns the full model's answer probabilities with KL divergence, without correct-answer labels for training. One backbone pass reads the question and all options together; the head returns probabilities for every option. We call this model FunDM (Fun Decision Model). Using 24 of Qwen3-8B's 36 layers and skipping the LM head, its +1T variant reaches 75.0% mean accuracy over eight benchmarks, versus 75.7% for the full model. On MMLU-Pro, both heads achieve 1.56 to 1.57 times batched vLLM throughput with about one-third lower peak GPU allocator memory. Both heads score higher than released Contrastive Language Models (CLM), fine-tuned CLM, and CLM-shaped heads trained from scratch on the eight multiple-choice benchmarks and coding-agent selection. Separately trained agent heads learn judge probabilities without pass/fail labels and select passing runs for 74.4 to 74.6 of 100 held-out tasks on average, versus 74 for the full-model judge and 52.0 for a CLM-shaped action-trained head. Code is available at: https://github.com/Ujjawal-K-Panchal/FunDM