This is the repository for our paper Optimization Techniques for Unsupervised Complex Table Reasoning via Self-Training Framework
We experiment on four benchmarks: TAT-QA, WiKiSQL, SEM-TAB-FACTS and FEVEROUS, and we provide scripts for data generation and the self-training procedure in the corresponding directory.
Since some of the open-source repositories we use have no license, we can't upload them here directly. You may need to clone them yourself following the instructions.
We release our synthetic data on four benchmark and fine-tuned models (corresponding to the results of UCTR-ST in the paper.)
| Dataset | Synthetic Data | Fine-tuned Models |
|---|---|---|
| TAT-QA | tatqa_syn.jsonl | tatqa_checkpoint |
| WiKiSQL | wikisql_syn.jsonl | wikisql_checkpoint |
| SEM-TAB-FACTS | sem_tab_facts_syn.tsv | sem_tab_fact_checkpoint |
| FEVEROUS | feverous_syn.jsonl | feverous_checkpoint |
The code is open-sourced under MIT License
Please create an issue or email us at: Zhenyu Li lizhenyucs@gmail.com