amd-torch-device-gfx90a

Create a machine learning model training utility using the 'amd-torch-device-gfx90a' package. This utility will enable users to train neural networks on AMD GPUs with GFX90A architecture, leveraging the specific optimizations provided by the package. The utility should include the following features:

1. **Model Selection**: Allow users to choose from predefined neural network architectures (e.g., CNNs for image classification, RNNs for sequence data).
2. **Dataset Integration**: Support loading datasets from common formats such as CSV, JSON, or directly from popular frameworks like PyTorch's torchvision.
3. **Training Configuration**: Provide options to configure training parameters such as batch size, learning rate, number of epochs, and loss functions.
4. **Performance Monitoring**: Implement real-time monitoring of training progress, including accuracy and loss metrics, and visualize these metrics using Matplotlib or similar libraries.
5. **Optimization Insights**: Use the 'amd-torch-device-gfx90a' package to optimize the model for AMD GPUs with GFX90A architecture, and display insights into how these optimizations improve performance compared to non-optimized runs.
6. **Save & Load Models**: Enable saving trained models to disk and loading them for inference or further training.
7. **Documentation & Help**: Include comprehensive documentation and a help section within the utility to guide users through setup and usage.

The 'amd-torch-device-gfx90a' package is crucial for this project as it provides optimized support for running PyTorch models on AMD GPUs with GFX90A architecture. It ensures that the neural networks are efficiently executed on these hardware platforms, taking advantage of their unique capabilities. Your task is to integrate this package seamlessly into your utility, ensuring that users can easily leverage its benefits without needing deep knowledge of GPU optimization.