keep-gpu 0.3.0

Keep GPU is a simple CLI app that keeps your GPUs running

Keep GPU

Keep GPU is a simple CLI app that keeps your GPUs running.

• 🧾 License: MIT
• 📚 Documentation: https://keepgpu.readthedocs.io

---

Contributions Welcome!

If you have ideas for new features or improvements, feel free to open an issue or submit a pull request.

This project does not yet fully support ROCm GPUs, so any contributions, suggestions, or testing help in that area are especially welcome!

---

Features

• Simple command-line interface
• Uses PyTorch and nvidia-smi to monitor and load GPUs
• Easy to extend for your own keep-alive logic

---

Installation

pip install keep-gpu

Usage

Use keep-gpu as a cli tool

keep-gpu

Specify the interval in microseconds between GPU usage checks (default is 300 seconds):

keep-gpu --interval 100

Specify GPU IDs to run on (default is all available GPUs):

keep-gpu --gpu-ids 0,1,2

Use keep-gpu api in your code

Non-blocking gpu keeping logic with CudaGPUController:

from keep_gpu.single_gpu_controller.cuda_gpu_controller import CudaGPUController
ctrl = CudaGPUController(rank=0, interval=0.5)
# occupy GPU while you do CPU-only work
# this is non-blocking
ctrl.keep()
dataset.process()
ctrl.release()        # give GPU memory back
model.train_start()   # now run real GPU training

Use CudaGPUController as a context manager:

from keep_gpu.single_gpu_controller.cuda_gpu_controller import CudaGPUController
with CudaGPUController(rank=0, interval=0.5):
    dataset.process()  # GPU occupied inside this block
model.train_start()    # GPU free after exiting block

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

Contributors