wppkg is a package I developed for my daily work.

🔥About Me

• CV: https://wppcv.readthedocs.io

---

Installation

Environment Setup

Step 1: Set up a python environment

We recommend creating a virtual Python environment with Anaconda:

• Required version: python >= 3.10

conda create -n wppkg python=3.10
conda activate wppkg

Step 2: Install pytorch

Install PyTorch based on your system configuration. Refer to PyTorch installation instructions

For the exact command, for example:

• You may choose any version to install, but make sure the PyTorch version is not too old.
• We recommend torch ≥ 2.6.

# Installation Example: torch v2.8.0
# CUDA 12.6
pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu126
# CUDA 12.8
pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu128
# CUDA 12.9
pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu129

Step 3: Install deepspeed (optional)

Install DeepSpeed based on your system configuration. Refer to DeepSpeed installation instructions

For the exact command, for example:

pip install deepspeed

Step 4: Install wppkg and dependencies

To install wppkg, run:

pip install wppkg

Or install from github:

git clone https://github.com/Peg-Wu/wppkg
cd wppkg
pip install [-e] .

# w/o dependencies
pip install [-e] . --no-deps

Update wppkg

If you want to update all dependencies of wppkg except torch, you can run the following command:

pip install -U $(pip show wppkg | sed -n 's/^Requires: //p' | tr ',' ' ' | xargs -n1 | grep -vi '^torch$')

---

Trainer Tips

• Early stopping does not currently support resuming training. If training is forcibly resumed, the early stopping callback will be reinitialized.
• If you enable early stopping, ensure that eval_every_n_epochs and checkpointing_steps are aligned, as the Trainer does not automatically save the best model.
• The final model is always saved at the end of training, even if early stopping is triggered.

---

Star History