Llama-Polya: Instruction Tuning for Large Language Model based on Polya's Problem-solving

Researchers have introduced Llama-Polya, an instruction-tuned LLM based on Llama-3.1-8B that integrates Polya's four-step problem-solving framework to improve mathematical reasoning. Fine-tuned on synthetic data from GSM8K, the model was designed to act as a structured educational tutor. Results show it produces more balanced reasoning and fewer premature answers, with expert evaluators noting improved pedagogical coherence and metacognitive prompting compared to general-purpose or domain-only tuned models.