`convert_to_numpy=True` not working with `output_value="token_embeddings"`; returns `torch.Tensor` type

[bhavnicksm]

Hey ST Team! 👋

I was trying to get the "token_embeddings" from a bge-base-en-v1.5 model, and converting it to numpy.ndarray but it was not functioning as expected.

Have a look at the screenshot below 👇

As a workaround, I was using .cpu().numpy() which was working.

Env:

sentence-transformers              3.3.1

Thanks! 😊

[bhavnicksm]

The issue probably lies at this line