How do I extract tensors in a tensor, into a 2D-numpy array?

I'm trying to extract tensors in a larger tensor, into a 2D-numpy array. (The tensor of tensors holds node embeddings after passing through a graph neural network). I'm using PyTorch (Geometric) for my project. I need the individual embeddings to go further.

This is my tensor:

tensor([[-0.7863,  0.8097],
        [-1.0679,  1.1331],
        [-1.8162,  1.9160],
        [ 2.0584, -2.2741],
        [-1.8818,  1.9333],
        [ 0.7870, -0.8974],
        [ 6.1731, -6.8074],
        [ 7.3219, -8.0852],
        [-0.9933,  0.9439],
        [ 4.6217, -5.1856],
        [-1.3747,  1.4614],
        [ 4.6429, -5.0861],
        [ 3.1141, -3.4420],
        [ 2.6417, -2.9173],
        [-2.9696,  3.0740],
        [ 4.0654, -4.5340],
        [ 1.7143, -1.9558],
        [-1.7497,  1.8496],
        [-1.9055,  1.9934],
        [ 3.9273, -4.3356],
        [ 4.0350, -4.4137],
        [ 1.2770, -1.4225],
        [-1.7447,  1.8458],
        [ 1.3937, -1.5936],
        [ 3.2471, -3.5991],
        [ 2.2516, -2.6034],
        [ 1.3096, -1.4573],
        [-1.7823,  1.8775],
        [ 0.9923, -1.2175],
        [-1.1818,  1.2430],
        [ 1.0997, -1.2466],
        [ 0.4841, -0.5800],
        [ 4.1609, -4.5518],
        [ 3.6211, -3.9535],
        [-1.6287,  1.7216],
        [ 2.1960, -2.5067],
        [ 1.9977, -2.2448],
        [-0.9295,  0.9438],
        [ 2.2512, -2.5578],
        [-2.5360,  2.6436],
        [-1.8890,  1.9787],
        [ 2.4500, -2.7050],
        [ 3.5502, -3.9974],
        [ 7.8740, -8.7413],
        [ 1.9768, -2.2287],
        [-0.9723,  1.0192],
        [ 5.3840, -5.9153],
        [-1.2483,  1.2866],
        [-1.4501,  1.5467],
        [-1.0471,  1.0899],
        [ 2.3409, -2.5763],
        [ 3.1816, -3.5639],
        [-1.8847,  1.9865],
        [-2.2041,  2.2781],
        [-2.7572,  2.8656],
        [-2.3390,  2.4441],
        [ 3.0862, -3.3945],
        [ 1.0977, -1.2327],
        [-1.7125,  1.7395],
        [ 2.8744, -3.2442],
        [ 1.8027, -2.0044],
        [-0.7821,  0.7521]], grad_fn=<AddmmBackward0>)

This is the code I wrote to get the embeddings as numpy arrays:

final = []
for element in final_embeddings:
    element.detach().numpy()
    final.append(element)

print(final)

This still gives me a list of tensors, not a 2D-numpy array. Using just element.numpy() gives me an error:

RuntimeError                              Traceback (most recent call last)
Cell In[139], line 3
      1 final = []
      2 for element in final_embeddings:
----> 3     element.numpy()
      4     final.append(element)
      6 print(final)

RuntimeError: Can't call numpy() on Tensor that requires grad. Use tensor.detach().numpy() instead.

Can someone tell me what might be going wrong?

Solution

There is no need to iterate through the entries. You should be able to just do:

final = final_embeddings.detach().numpy()