Empty val batch CUDA device fix

[glenn-jocher]

Verified fix for #7525 (comment)

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved device consistency in validation code.

📊 Key Changes

• Ensured tensors are initialized on the correct device during validation.

🎯 Purpose & Impact

• 🛠️ Purpose: To avoid errors that occur when tensors are created on a different device than the model (e.g., model on GPU but tensor on CPU).
• 💻 Impact: Provides more stable and consistent code execution, especially for those using different compute devices (GPUs and CPUs).
• 🌐 Benefits users by reducing potential device mismatch bugs and enhancing the overall experience with smoother validation processes.