InMindCNN

Train and evaluate a simple CNN on CIFAR-10 using PyTorch.

Quickstart

1. Clone the repository:

   git clone <repo-url>
   cd inmindCNN

2. Install the uv Python package manager (faster than pip):

   pip install uv

3. Install all dependencies defined in pyproject.toml:

   uv sync

4. Edit config.yaml for hyperparameters and paths if needed.

   • val_split controls the fraction of training data used for validation (default: 0.1).

5. Run training: Run inside uv venv:

   uv run train.py

   • Uses GPU if available
   • CIFAR-10 is auto-downloaded to data/cifar10/
   • Model weights saved to weights/checkpoint.pth by default.
   • After each epoch, validation loss and accuracy are reported.

Validation

• A portion of the training set is used for validation (see val_split in config).
• After each epoch, validation loss and accuracy are printed.
• Final test loss and accuracy are printed after training.