[bh-official]

Closes #5

What was wrong: docs/api-reference/tensor.md described Tensor.sum() and Tensor.mean() as "full reduction, not per-axis", but both accept axis and keepdims and reduce per-axis.

What changed: Replaced that sentence for both methods with a parameter table (axis: int, tuple of int or None, negative axes allowed; keepdims: bool) and the resulting shapes for a (2, 3) input. Only the sum/mean sections were touched; line endings (CRLF) preserved.

Checked by running (fresh venv, pip install -e .), on Tensor(np.arange(6.).reshape(2,3), requires_grad=True):

• .sum() -> shape (), 15.0; .mean() -> 2.5
• .sum(axis=0) -> shape (3,); .sum(axis=1, keepdims=True) -> shape (2, 1)
• .mean(axis=-1) -> shape (2,); .sum(axis=(0,1), keepdims=True) -> shape (1, 1)
• backward gives gradients of the input's shape for all of these (1 for sum, 1/n for mean)
• pytest: 22 passed

[Terminay]

Congrats you're the first documentation contributor to leanpass! The changes are well addressed and very elegantly implemented.

Your name will appear in the readme as a contributor soon, thanks for joining leanpass.