TabM Model Instructions

Model Overview

1.1 Method Description

This implementation utilizes the approach from the research paper:
"TabM: Advancing Tabular Deep Learning With Parameter-Efficient Ensembling" (ICLR 2025)

• Paper Link: https://arxiv.org/abs/2410.24210
• Original Repository: https://github.com/yandex-research/tabm

1.2 Motivation

Compared to traditional machine learning methods used in NeoRanking, we aim to explore the performance of Tabular Deep Learning approaches on this type of data for classification tasks.

Training Details

2.1 Training Data

The detailed training dataset is located at: data/tabm_train.tsv

2.2 Training Parameters

• Model parameters were optimized using the hyper framework for parameter tuning
• Training script: src/tabm_train.py
• Testing script: src/tabm_test.py

2.3 Testing Data

• Test dataset: data/tabm_test.tsv
• Evaluation metrics have been updated in spaces/leaderboard

Model Usage

3.1 Installation

pip3 install tabm

3.2 Training

bash scripts/tabm_train.sh

3.3 Testing

bash scripts/tabm_test.sh

Special Notes

Our use of TabM fully complies with the Apache-2.0 license. If you need to reference or reuse this model, please adhere to the original author's citation requirements and properly attribute the source.