NumericJev: Jev-like LLM Numerical Decoding with Multiway Decision Trees

2026-09-23

AI Summary

This study addresses the inherent difficulty of large language models (LLMs) in directly generating high-precision numerical values. To overcome this limitation, we propose a training-free decoding algorithm that requires neither parameter updates nor access to hidden states. The method reformulates numerical generation as a structured decision-making problem by employing a multi-way decision tree to recursively refine numerical ranges, thereby enabling LLMs to produce precise numerical outputs through a discrete selection interface. This framework significantly outperforms direct candidate-list selection strategies. On arithmetic benchmarks, it achieves a 2.93% performance improvement, reduces the normalized mean absolute error to 1.84%, and yields a 0% readout error on historical indices, effectively overcoming the numerical reasoning bottleneck of LLMs.

Abstract

Large language models can interpret natural lan- guage, yet robust decisions remain challenging. Jev-like models expose structured choices, but these interfaces do not directly provide numeri- cal values at a requested precision. We propose NUMERICJEV, a training-free numerical decod- ing algorithm that enables numerical output from any LLM with a Jev-like structured-choice in- terface. Surprisingly, on our arithmetic bench- mark, it outperforms direct selection from a can- didate list containing the correct answer by 2.93 percentage points (Figure 1). Our motivation comes from the observation that numerical range selection is itself a decision problem that Jev- like LLMs can address. NUMERICJEV recur- sively refines a range through a multiway deci- sion tree while retaining the original question in context, without parameter updates or hidden- state access. On a 100-value grid, a ten-way tree requires only two decision rounds. Range- normalized MAE is 1.84% versus 5.18% for di- rect choice. A separate three-date historical- index study yields 4.58% mean relative recall er- ror and 0% readout error when the value is sup- plied. Code is available at https://github. com/Bring-AI/jev-numeric.

Problem

• Large Language Models
• Numerical Decoding
• Structured-Choice Interface
• Precision Output
• Decision Problem

Innovation

• Training-free numerical decoding
• Multiway decision trees
• Jev-like LLMs
• Recursive range refinement
• Structured-choice interface