frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-21 13:44:15 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimize min, max gradient behavior description (#160312)

Browse Source 
Fixes #160273

## Test Result
<img width="897" height="593" alt="image" src="https://github.com/user-attachments/assets/6ebcdb2c-8a2c-4f0d-8195-656089e88325" />
<img width="985" height="653" alt="image" src="https://github.com/user-attachments/assets/606a7264-e223-4d2b-8c3f-f153ce43b208" />
<img width="903" height="607" alt="image" src="https://github.com/user-attachments/assets/0ae2f56f-820f-4194-b15c-a02a078c0487" />
<img width="903" height="607" alt="image" src="https://github.com/user-attachments/assets/79c38a17-45ac-4808-829f-d538178de36b" />

Pull Request resolved: https://github.com/pytorch/pytorch/pull/160312
Approved by: https://github.com/ngimel
This commit is contained in:
zeshengzong
2025-08-14 04:18:46 +00:00
committed by PyTorch MergeBot
parent 45ba7ecda8
commit fe3f5fe4ea
1 changed files with 34 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
torch/_torch_docs.py
View File

@ -6579,6 +6579,18 @@ max(input, *, out=None) -> Tensor
Returns the maximum value of all elements in the ``input`` tensor.
.. note::
The difference between ``max``/``min`` and ``amax``/``amin`` is:
- ``amax``/``amin`` supports reducing on multiple dimensions,
- ``amax``/``amin`` does not return indices.
Both ``amax``/``amin`` evenly distribute gradients between equal values
when there are multiple input elements with the same minimum or maximum value.
For ``max``/``min``:
- If reduce over all dimensions(no dim specified), gradients evenly distribute between equally ``max``/``min`` values.
- If reduce over one specified axis, only propagate to the indexed element.
Args:
{input}
@ -6717,9 +6729,13 @@ dimension(s) :attr:`dim`.
- ``amax``/``amin`` supports reducing on multiple dimensions,
- ``amax``/``amin`` does not return indices.
Both ``max``/``min`` and ``amax``/``amin`` evenly distribute gradients between equal values
Both ``amax``/``amin`` evenly distribute gradients between equal values
when there are multiple input elements with the same minimum or maximum value.
For ``max``/``min``:
- If reduce over all dimensions(no dim specified), gradients evenly distribute between equally ``max``/``min`` values.
- If reduce over one specified axis, only propagate to the indexed element.
{keepdim_details}
Args:
@ -7197,6 +7213,18 @@ min(input, *, out=None) -> Tensor
Returns the minimum value of all elements in the :attr:`input` tensor.
.. note::
The difference between ``max``/``min`` and ``amax``/``amin`` is:
- ``amax``/``amin`` supports reducing on multiple dimensions,
- ``amax``/``amin`` does not return indices.
Both ``amax``/``amin`` evenly distribute gradients between equal values
when there are multiple input elements with the same minimum or maximum value.
For ``max``/``min``:
- If reduce over all dimensions(no dim specified), gradients evenly distribute between equally ``max``/``min`` values.
- If reduce over one specified axis, only propagate to the indexed element.
Args:
{input}
@ -7325,9 +7353,13 @@ dimension(s) :attr:`dim`.
- ``amax``/``amin`` supports reducing on multiple dimensions,
- ``amax``/``amin`` does not return indices.
Both ``max``/``min`` and ``amax``/``amin`` evenly distribute gradients between equal values
Both ``amax``/``amin`` evenly distribute gradients between equal values
when there are multiple input elements with the same minimum or maximum value.
For ``max``/``min``:
- If reduce over all dimensions(no dim specified), gradients evenly distribute between equally ``max``/``min`` values.
- If reduce over one specified axis, only propagate to the indexed element.
{keepdim_details}
Args: