cvic

Local, automated hyperparameter search for image classifiers - from dataset to tuned model with one command, distributed across your local GPUs.

cvic uses off-the-shelf models and packages, so you won't get SOTA performance. But it can get surprisingly close, with almost zero effort. Useful as a baseline, or for experimentation with architectures and GPUs.

Built on Ray Tune, Optuna, and timm. Requires Python ≥ 3.12.

It ships two commands:

• cvic — k-fold cross-validation hyperparameter search
• tunic — hold-out hyperparameter tuning (single train/validation split)

Install

pipx install cvic

or with uv:

uv tool install cvic

Run from source

The project is fully managed by uv with a committed uv.lock, so the exact dependency versions are reproducible across machines. You need an NVIDIA GPU with a reasonably recent driver to use CUDA; the PyTorch wheels bundle their own CUDA runtime, so no system CUDA toolkit installation is required and you do not pick a CUDA version — uv resolves the right wheel for your platform automatically.

git clone https://github.com/ljbuturovic/cvic.git
cd cvic
uv sync                       # creates .venv and installs the locked dependencies
source .venv/bin/activate     # now `cvic` and `tunic` are on your PATH

Verify the GPU is visible:

python -c "import torch; print(torch.cuda.is_available())"

Then run the commands directly (no uv run prefix needed once the venv is activated):

cvic --smoke-test
tunic --smoke-test

To run the test suite:

pytest tests/ -k "not test_smoke"

Quick start

Hold-out tuning:

tunic --data /path/to/dataset --model resnet50 --n_trials 30 --epochs 30 --output results.json

Cross-validation tuning:

cvic --data /path/to/dataset --model resnet50 --n-trials 30 --epochs 30 --folds 5

Train final model from tuning results:

tunic --final results.json --data /path/to/dataset --epochs 50 --amp

Smoke test (synthetic data, no dataset needed):

tunic --smoke-test
cvic --smoke-test

Dataset format

The dataset format is auto-detected:

• ImageFolder — standard split/class/image.ext layout
• WebDataset — sharded TAR files; detected when wds/dataset_info.json exists

tunic — hold-out hyperparameter search

tunic --data PATH --model MODEL [options]

cvic — cross-validation hyperparameter search

cvic --data PATH --model MODEL [options]

Run cvic --help / tunic --help for the full list of flags.

Search space

Override any part with a YAML file via --search-space.

License

MIT