Socratic guiding questions in synthetic arithmetic data: matched LoRA runs (revision v2)

Abstract

Supporting data, adapters, predictions and code for the article *Low-Cost LoRA Fine-Tuning of Small Language Models for Multi-Step Arithmetic Reasoning* by Jake O'Grady, Asena Isik Gürhan, Chee Fong Ting and Effirul Ramlan (University of Galway). We generated 20,000 GSM8K-derived arithmetic problems with step-by-step solutions and rendered each solution twice: with a guiding question before every reasoning step (Socratic) and without those questions (non-Socratic). Qwen3-0.6B, Qwen3-1.7B and Llama-3.2-1B-Instruct were fine-tuned with LoRA on both versions, and Qwen3-0.6B also on nested subsets of 5,000 and 10,000 examples. All models were evaluated on GSM8K, MultiArith, SVAMP and GSM-Hard under a protocol fixed before any run.See README.md for the file list, licences and how to verify every number.