dnaihao/phi-3-mini-tablebench
dnaihao/phi-3-mini-tablebench is a 4-billion parameter language model, fine-tuned from Microsoft's Phi-3-mini-4k-instruct, specifically for table-related tasks. It was trained on the tablebench_train.json corpus from dnaihao/Table-Instructs using full SFT via LLaMA-Factory. This model serves as a replication of TableBenchLLM and extends experiments from the EACL 2026 Findings paper on table LLMs, providing an additional base-model variant for research into model and data effects.
Loading preview...
Overview
dnaihao/phi-3-mini-tablebench is a 4-billion parameter language model derived from Microsoft's Phi-3-mini-4k-instruct. It is a replication of the TableBenchLLM approach, specifically instruction-tuned for tasks involving tabular data. This model was released as an artifact alongside the EACL 2026 Findings paper, "What Really Matters for Table LLMs? A Meta-Evaluation of Model and Data Effects," to expand the paper's experimental scope by adding another base-model variant trained on the same specialized corpus.
Key Capabilities
- Table-specific Instruction Following: Fine-tuned on
tablebench_train.jsonfrom thednaihao/Table-Instructsdataset, making it proficient in understanding and processing table-related instructions. - Research Extension: Provides a valuable resource for researchers studying the impact of different base models and training data on table-focused LLMs, complementing the main experiments detailed in the associated paper.
- Phi-3-mini Foundation: Leverages the compact yet capable architecture of Phi-3-mini-4k-instruct, offering a balance of performance and efficiency for table-based applications.
Good for
- Academic Research: Ideal for researchers investigating the performance of LLMs on tabular data, especially those interested in model and data effects as explored in the EACL 2026 paper.
- Benchmarking Table LLMs: Can be used as a baseline or comparison model in evaluations of other language models designed for table understanding and generation.
- Developing Table-Aware Applications: Suitable for developers building applications that require an LLM to interpret, query, or generate content based on structured tabular information.