This repository contains
TAMA-QWen3, an instruction-tuned model based on
Qwen3-8B optimized for table understanding and reasoning. It was introduced and evaluated as part of the work:
MMTU: A Massive Multi-Task Table Understanding and Reasoning Benchmark.
In [
1], we reveal that with limited instruction tuning data, we can achieve competitive performance on table tasks. This compact setup enables quick instruction tuning with advanced base models.
We present TAMA models built on Qwen 2.5 and Qwen 3. These models achieve strong results on the MMTU benchmark [
2], outperforming recent table reasoning models [
3] and competitive table LLMs like Table-GPT 2 [
4], which is tuned on 2.36M datapoints.
We adopt the official
MMTU evaluation script to compute scores. For overall performance, we use the evaluation function described
here. Category scores are the arithmetic mean across datasets in that category.
For QWen 3 model and TAMA-QWen3, we turned off the thinking mode.
1@article{mmtu,
2 title={{MMTU}: A Massive Multi-Task Table Understanding and Reasoning Benchmark},
3 author={Junjie Xing and Yeye He and Mengyu Zhou and Haoyu Dong and Shi Han and Lingjiao Chen and Dongmei Zhang and Surajit Chaudhuri and H. V. Jagadish},
4 journal={arXiv preprint arXiv:2506.05587},
5 year={2025}
6}
