Skip to content

Latest commit

 

History

History
80 lines (54 loc) · 1.93 KB

README.md

File metadata and controls

80 lines (54 loc) · 1.93 KB
Raw

WangChanXNER

Welcome to WangChanXNER! This repository contains a basic implementation for sequence labeling tasks. It provides functionalities for training and testing.

Installation

To install the necessary dependencies, run the following: python=3.8.16

seqeval
pythainlp
tabulate
pandas==2.0.3
torch==2.0.1
numpy==1.23.5
python=3.8.16
protobuf==3.20.0
transformers==4.29.2

Downloading the Checkpoint

Test the model with Colab.

You can download the checkpoint from our google drive: checkpoint.

Once downloaded, extract the files to ensure the appropriate directory structure.

unzip downloaded_checkpoint.zip -d storage/

Directory tree

.
├── utils
├── model
├── storage
├── trainer
├── inference.ipynb
└── inference.py

Train/Test

Training

Run the main.py script to train the model:

python main.py --device 0 -c storage/config_base.json
  • The --device flag specifies the GPU device to use during training. In this example, GPU 0 will be used.
  • The -c flag points to the configuration file config_lst20.json, which holds the hyperparameters and settings for training.

Pretrained

xlm-roberta-base
airesearch/wangchanberta-base-att-spm-uncased

Testing

To test the model, use the inference.py script:

python inference.py --resume storage/best_model/model_best.pth
  • The --resume flag points to the saved checkpoint file to load the model for testing.

Acknowledgements

This project is inspired by Tensorflow-Project-Template created by Mahmoud Gemy. We express our gratitude to Mahmoud Gemy for providing the foundation and ideas for this project.