[BUG] - unpack error in CUDA extension tutorial

Add Link

https://pytorch.org/tutorials/advanced/cpp_extension.html

Describe the bug

Code to reproduce the issue.

import time

import torch

batch_size = 16
input_features = 32
state_size = 128

# Check if CUDA (GPU) is available
if torch.cuda.is_available():
    # Set the device to CUDA
    device = torch.device("cuda")
    print("CUDA is available. Using GPU.")
else:
    # If CUDA is not available, fall back to CPU
    device = torch.device("cpu")
    print("CUDA is not available. Using CPU.")

X = torch.randn(batch_size, input_features, device=device)
h = torch.randn(batch_size, state_size, device=device)
C = torch.randn(batch_size, state_size, device=device)

rnn = LLTM(input_features, state_size).to(device)

forward = 0
backward = 0
for _ in range(2):
    start = time.time()
    new_h, new_C = rnn(X, (h, C))
    forward += time.time() - start

    start = time.time()
    (new_h.sum() + new_C.sum()).backward()
    backward += time.time() - start

print('Forward: {:.3f} s | Backward {:.3f} s'.format(forward, backward))

Error info:

Traceback (most recent call last):
  File "toy_model.py", line 74, in <module>
    (new_h.sum() + new_C.sum()).backward()
  File "/home/username/miniconda3/envs/tmp/lib/python3.8/site-packages/torch/_tensor.py", line 492, in backward
    torch.autograd.backward(
  File "/home/username/miniconda3/envs/tmp/lib/python3.8/site-packages/torch/autograd/__init__.py", line 251, in backward
    Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
  File "/home/username/miniconda3/envs/tmp/lib/python3.8/site-packages/torch/autograd/function.py", line 288, in apply
    return user_fn(self, *args)
  File "toy_model.py", line 25, in backward
    d_old_h, d_input, d_weights, d_bias, d_old_cell = outputs
ValueError: too many values to unpack (expected 5)

Describe your environment

• Platform: Ubuntu
• CUDA: yes
• Pytorch Version: 2.1.0+cu118

Open the linked CUDA extension tutorial and run the provided reproducer in the stated PyTorch 2.1.0 environment. Start at the custom backward method where outputs are unpacked, then compare its expected values with the tutorial's forward result. Done means the example completes backward without the unpacking error and the tutorial remains runnable.