Template-Driven LLM-Paraphrased Framework for Tabular Math Word Problem Generation

Authors: Xiaoqiang Kang, Zimu Wang, Xiaobo Jin, Wei Wang, Kaizhu Huang, Qiufeng Wang

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through the proposed framework, we construct a high-quality dataset Tab MWPTe LL by adhering to the question types in the Tab MWP dataset, and we conduct extensive experiments on a variety of LLMs to demonstrate the effectiveness of Tab MWP-Te LL in improving TMWP-solving performance.
Researcher Affiliation Academia 1School of Advanced Technology, Xi an Jiaotong-Liverpool University 2University of Liverpool 3Duke Kunshan University EMAIL, EMAIL
Pseudocode No The paper describes methods and processes through figures (Figure 2, Figure 3, Figure 4) and textual descriptions, but it does not contain explicit pseudocode or algorithm blocks.
Open Source Code Yes Code https://github.com/Jason8Kang/TELL
Open Datasets Yes We conduct evaluations on Tab MWP (Lu et al. 2023b), a recent large-scale dataset containing 38, 431 grade-level MWPs with tabular context, whose statistics are presented in Table 1.
Dataset Splits Yes We conduct evaluations on Tab MWP (Lu et al. 2023b), a recent large-scale dataset containing 38, 431 grade-level MWPs with tabular context, whose statistics are presented in Table 1. Table 1: Statistics of the Tab MWP dataset. Train Valid Test Total #Question 23, 059 7, 686 7, 686 38, 431
Hardware Specification Yes All experiments are conducted on 8 NVIDIA Ge Force RTX 3090 graphics cards.
Software Dependencies No The paper mentions using XTuner for QLoRA, and specific LLMs (Yi, Mistral, Qwen 2, Llama 3) but does not provide specific version numbers for these or other key software libraries like Python or PyTorch, which would be necessary for reproducibility.
Experiment Setup Yes During the fine-tuning process, we set the number of epochs as 2, the batch size per device as 12, the gradient accumulation steps as 4, and the learning rate as 2e 4.