Analyze the impact of learning rate and batch size

Using the PyTorch baseline, experiment with 3 different learning rates and 3 different batch sizes. Create a grid of loss curves or a summary table. Select the optimal combination that prevents noisy gradients but converges efficiently.

Start from the PyTorch baseline and identify how training runs and loss are recorded. Run the 3×3 learning-rate and batch-size combinations, then create loss curves or a summary table and document the combination that balances low gradient noise with efficient convergence.

Written by the indexing model from the issue text.