isuco / SERM-master

Experimental code of SERM-master

This code repo contains the preprocess, training, inference and evaluation code for SERM-master on TACRED relation extraction dataset.

Requirements

• Python 3.6.3
• PyTorch 1.2.0
• Transformers 2.2.1
• CUDA 9.2

Perparation

First, download and unzip TACRED dataset from LDC.

Training

Please run

python train.py

Evaluation

Evaluate the trained model on TACRED dataset:

python evaluation.py

You will get:

Model	Precision	Recall	micro-F1
SERM(ALBERT-large)	0.828	0.811	0.819