Thanks!
This is interesting, because I can't at a glance tell if .reshape(&dims[1..]) is always identical to .squeeze(0) (after applying .narrow(0, index, 1) of course) . It makes sense that it is, and from looking at the code it does seem like it is. I also modified a unit test to check quickly and the results were equal, and they were, but I don't know if I've covered all edge cases.
Also important to note that even if it isn't always identical that doesn't necessarily mean that this is wrong, it could still be an improvement.

If you could provide some kind of additional proof that would be great.
For example add a test in tensor_tests.rs that uses get on a bunch of different tensors with expected results, and ensure the results are the same on main and on this branch.