Keep GPU

Keep GPU keeps shared GPUs from being reclaimed while you prep data, debug, or coordinate multi-stage pipelines. It allocates just enough VRAM and issues lightweight CUDA work so schedulers observe an “active” device—without running a full training job.

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

Why it exists

On many clusters, idle GPUs are reaped or silently shared after a short grace period. The cost of losing your reservation (or discovering another job has taken your card) can dwarf the cost of a tiny keep-alive loop. KeepGPU is a minimal, auditable guardrail:

• Predictable – Single-purpose controller with explicit resource knobs (VRAM size, interval, utilization backoff).
• Polite – Uses NVML to read utilization and backs off when the GPU is busy.
• Portable – Typer/Rich CLI for humans; Python API for orchestrators and notebooks.
• Observable – Structured logging and optional file logs for auditing what kept the GPU alive.
• Power-aware – Uses intervalled elementwise ops instead of heavy matmul floods to present “busy” utilization while keeping power and thermals lower (see CudaGPUController._run_mat_batch for the loop).
• NVML-backed – GPU telemetry comes from nvidia-ml-py (the pynvml module), with optional rocm-smi support when you install the rocm extra.

Quick start (CLI)

pip install keep-gpu

# Hold GPU 0 with 1 GiB VRAM and throttle if utilization exceeds 25%
keep-gpu --gpu-ids 0 --vram 1GiB --busy-threshold 25 --interval 60

Platform installs at a glance

• CUDA (example: cu121)

  pip install --index-url https://download.pytorch.org/whl/cu121 torch
  pip install keep-gpu

• ROCm (example: rocm6.1)

  pip install --index-url https://download.pytorch.org/whl/rocm6.1 torch
  pip install keep-gpu[rocm]

• CPU-only

  pip install torch
  pip install keep-gpu

Flags that matter:

• --vram (1GiB, 750MB, or bytes): how much memory to pin.
• --interval (seconds): sleep between keep-alive bursts.
• --busy-threshold: skip work when NVML reports higher utilization.
• --gpu-ids: target a subset; otherwise all visible GPUs are guarded.

Embed in Python

from keep_gpu.single_gpu_controller.cuda_gpu_controller import CudaGPUController

with CudaGPUController(rank=0, interval=0.5, vram_to_keep="1GiB", busy_threshold=20):
    preprocess_dataset()   # GPU is marked busy while you run CPU-heavy code

train_model()              # GPU freed after exiting the context

Need multiple devices?

from keep_gpu.global_gpu_controller.global_gpu_controller import GlobalGPUController

with GlobalGPUController(gpu_ids=[0, 1], vram_to_keep="750MB", interval=90, busy_threshold=30):
    run_pipeline_stage()

What you get

• Battle-tested keep-alive loop built on PyTorch.
• NVML-based utilization monitoring (by way of nvidia-ml-py) to avoid hogging busy GPUs; optional ROCm SMI support by way of pip install keep-gpu[rocm].
• CLI + API parity: same controllers power both code paths.
• Continuous docs + CI: mkdocs + mkdocstrings build in CI to keep guidance up to date.

For developers

• Install dev extras: pip install -e ".[dev]" (add .[rocm] if you need ROCm SMI).
• Fast CUDA checks: pytest tests/cuda_controller tests/global_controller tests/utilities/test_platform_manager.py tests/test_cli_thresholds.py
• ROCm-only tests carry @pytest.mark.rocm; run with pytest --run-rocm tests/rocm_controller.
• Markers: rocm (needs ROCm stack) and large_memory (opt-in locally).

MCP endpoint (experimental)

• Start a simple JSON-RPC server on stdin/stdout (default):

  keep-gpu-mcp-server

• Or expose it over HTTP (JSON-RPC 2.0 by way of POST):

  keep-gpu-mcp-server --mode http --host 0.0.0.0 --port 8765

• Example request (one per line):

  {"id": 1, "method": "start_keep", "params": {"gpu_ids": [0], "vram": "512MB", "interval": 60, "busy_threshold": 20}}

• Methods: start_keep, stop_keep (optional job_id, default stops all), status (optional job_id), list_gpus (basic info).
• Minimal client config (stdio MCP):

  servers:
    keepgpu:
      command: ["keep-gpu-mcp-server"]
      adapter: stdio

• Minimal client config (HTTP MCP):

  servers:
    keepgpu:
      url: http://127.0.0.1:8765/
      adapter: http

• Remote/SSH tunnel example (HTTP):

  keep-gpu-mcp-server --mode http --host 0.0.0.0 --port 8765

  Client config (replace hostname/tunnel as needed):

  servers:
    keepgpu:
      url: http://gpu-box.example.com:8765/
      adapter: http

  For untrusted networks, put the server behind your own auth/reverse-proxy or tunnel by way of SSH (for example, ssh -L 8765:localhost:8765 gpu-box).

Contributing

Contributions are welcome—especially around ROCm support, platform fallbacks, and scheduler-specific recipes. Open an issue or PR if you hit edge cases on your cluster. See docs/contributing.md for dev setup, test commands, and PR tips.

Credits

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

Contributors

📖 Citation

If you find KeepGPU useful in your research or work, please cite it as:

@software{Wangmerlyn_KeepGPU_2025,
  author       = {Wang, Siyuan and Shi, Yaorui and Liu, Yida and Yin, Yuqi},
  title        = {KeepGPU: a simple CLI app that keeps your GPUs running},
  year         = {2025},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17129114},
  url          = {https://github.com/Wangmerlyn/KeepGPU},
  note         = {GitHub repository},
  keywords     = {ai, hpc, gpu, cluster, cuda, torch, debug}
}