Summary:
🔹 register_buffer stores stats in state_dict
🔹 register_forward_hook captures activations
🔹 torch.no_grad() ensures inference safety
🔹 Works in both training & inference modes

Perfect for post-training calibration! ✅🔥 #PyTorch #DeepLearning

Example: Register a buffer to store activation stats & use a forward hook to update it: