This pull request includes several important updates and fixes to ensure compatibility with the latest version of PyTorch and to resolve issues related to tensor operations while training AlexNet on the CIFAR dataset.

The key changes are as follows:

1. Removedasync=True argument fromcuda() calls: This argument is deprecated in PyTorch 1.0.0 and has been removed to maintain compatibility.

   • Commit:5c70d06

2. Refactored code to usetorch.no_grad() instead ofvolatile=True: Updated to the new context manager approach to handle operations without tracking gradients for better efficiency with the latest PyTorch version.

   • Commit:4d393ff

3. Fixed tensor reshaping in the accuracy function: Replacedview withreshape to correct issues when reshaping tensors, improving reliability.

   • Commit:d378d87

4. Addressed all 0-dim tensor indexing errors in cifar.py: Accessing 0-dim tensors caused index errors, ensuring smooth operation during training and metrics update.

   • Commits:c570bd6,2935578

Additional Information:

• Files Changed:
  • cifar.py: Updated to remove deprecated or obsolete arguments and fix tensor operations
  • eval.py: Adjusted tensor reshaping logic for accuracy calculation