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.
$ docker pull ufoym/deepo
# You should edit some options for your environments
$ bash docker/generate_container.sh --image_name ufoym/deepo --container_name $CONTAINER_NAME --external_port 8888
# Run this code on the root directory
$ pip install -r requirement.txt
$ sh run.sh
Task | Dataset | Model | Test Accuracy |
---|---|---|---|
Sentiment Classification | IMDB | BERT-base | 93.52% |