Transformer-Machine-Translation

👀 Demos Video

TODO

🔧 Dependencies and Installation

• Python >= 3.11
• PyTorch >= 2.4.0

Installation

1. Clone repo

git clone https://github.com/qin1122/Transformer-Machine-Translation.git
cd Transformer-Machine-Translation

2. Install dependent packages (use conda)

conda create --name transformer
conda activate transformer
conda install python
pip install -r requirements.txt

🗂️ Datasets

I use a dataset consisting of 21,621 pairs of English and Chinese short sentences. Original dataset can be downloaded here.

I applied a better tokenizing method for Chinese sentences, which led to a higher BLEU score. The new dataset and the preprocessed original dataset can be downloaded here

⚙️ Train

Create a new yaml config file or use my config files (e.g. './configs/train_1.yaml'), put the yaml file in './configs'.

Then run:

python main.py --config 'configs/config.yaml'

You can monitor the training process in real time on Weights & Biases (wandb).

To use wandb, you have to login in terminal first, use the command below to login:

wandb login

⚡️ Quick Test

Create a new yaml config file or use my config files (e.g. './configs/test_origin.yaml'), put the yaml file in './configs'.

Then run:

python test.py --config 'configs/config.yaml'

🏰 Model Zoo

We conducted a total of 19 experiments, and the best model parameters can be downloaded here.