Multi-GPU Training Does Not Work Properly

Multi-GPU support appears to be available, but training does not work correctly when multiple GPUs are selected. Current issue When using more than one GPU, the training process does not properly utilize the available GPUs or does not behave as expected. Suggested improvements Properly initialize all selected GPUs. Ensure the training workload is distributed across the GPUs. Verify that model parameters and batches are correctly synchronized. Provide accurate GPU utilization during multi-GPU training. Handle multi-GPU configuration automatically where possible. Display a clear error when multi-GPU training cannot be initialized. The goal is to make multi-GPU training reliable and ensure all selected GPUs are actually used during training.