Shuqi-li / PretrainTST

Pretrained PatchTST

Requirements

# install pytorch according to the cuda version
pip install transformers accelerate wandb gpustat pandas matplotlib scikit-learn ipywidgets
conda install ipykernel

TODO

Finish the ##TODO tag in pretrain.py(mainly about how to prepare data and how to calculate loss).

Get Started

wandb login
accelerate config
## multi GPU training
accelerate launch --num_processes 4 --gpu_ids 0,1,2,3 pretrain.py