Minimal PyTorch-Lightning NLP Template with WandB ⚡️

Hope to customize the pipeline in detail? How about this template with vanila PyTorch?

PyTorch-Lightning template for easy use! (actually, for my own😉)

• This template is for fast prototyping of research ideas.
• This template follows Black code formatting.
• This template contains a simple BERT classification code.
• This template contains a simple WandB application.
• Use Sweep for hyperparameter tuning.
• Note: You should enter WandB account information on the first run.

1. Set Environments

1.1. Pull Docker Image

$ docker pull ufoym/deepo

1.2. Make Container

# You should edit some options for your environments
$ bash docker/generate_container.sh --image_name ufoym/deepo --container_name $CONTAINER_NAME --external_port 8888 

1.3. Install Required Dependencies

# Run this code on the root directory
$ pip install -r requirement.txt

2. Run Code

$ sh run.sh

3. Sample Experiment Results

3.1. Model Performance

Task	Dataset	Model	Test Accuracy
Sentiment Classification	IMDB	BERT-base	93.52%

3.2. WandB Logging Results

4. LICENSE

MIT License