Table-R1

Table-R1数据集是耶鲁大学自然语言处理实验室首次探索推理时扩展在表格推理任务上的应用。该数据集通过DeepSeek-R1生成推理轨迹，用于微调LLMs，以实现Table-R1-SFT模型。数据集的创建过程包括从DeepSeek-R1生成推理轨迹，并使用LLM-based annotators进行验证。Table-R1数据集应用于推理时扩展，旨在解决表格推理任务中的推理能力问题，包括数据处理、推理链生成等。The Table-R1 dataset is the first exploration of applying inference-time expansion to table reasoning tasks by the Natural Language Processing Laboratory of Yale University. This dataset generates reasoning traces via DeepSeek-R1, and is used for fine-tuning large language models (LLMs) to build the Table-R1-SFT model. The dataset creation process includes generating reasoning traces from DeepSeek-R1 and validating these traces using LLM-based annotators. The Table-R1 dataset is applied to inference-time expansion, aiming to address reasoning capability issues in table reasoning tasks, including data processing and reasoning chain generation.