LLM Merging methods evaluated on GSM8K mathematical reasoning

This table presents LLM merging methods evaluated on the GSM8K benchmark, a widely used dataset for multi-step arithmetic reasoning. It compares merge space, training dependencies, resource requirements, and reported key findings.

本表格呈现了在GSM8K基准测试集上开展评估的大语言模型（Large Language Model, LLM）融合方法。GSM8K是一款广泛用于多步算术推理任务的基准数据集，本表格对比了各方法的融合空间、训练依赖、资源需求与已报道的核心研究结论。