Restrict the inputs of ops.pad()

[Grvzard]

If the dimensions of pad_width and input are not the same, other backends will throw a corresponding exception, but PyTorch can still work, which may result in an unrelated exception being raised.

Users may assume that the pad_width = ((0, 0), (1, 2), (0, 0)) can be shortened to pad_width = ((0, 0), (1, 2)).
However, due to torch's use of reverse order, the latter will be processed to ((1, 2),) (reversed and with higher dimensions removed).

keras/keras/src/backend/torch/numpy.py

Line 1089 in 8e7843c

pad_width = list(pad_width)[::-1] # torch uses reverse order

keras/keras/src/backend/torch/numpy.py

Lines 1096 to 1097 in 8e7843c

	if pad_width_sum == 0: # early break when no padding in higher order
	break

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.

Project coverage is 78.88%. Comparing base (8e7843c) to head (b652ccc).

Files with missing lines	Patch %	Lines
keras/src/ops/numpy.py	0.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #20348      +/-   ##
==========================================
- Coverage   78.88%   78.88%   -0.01%     
==========================================
  Files         512      512              
  Lines       49250    49252       +2     
  Branches     7954     7955       +1     
==========================================
  Hits        38853    38853              
- Misses       8539     8540       +1     
- Partials     1858     1859       +1     

Flag	Coverage Δ
keras	78.75% <0.00%> (-0.01%)	⬇️
keras-jax	62.40% <0.00%> (-0.01%)	⬇️
keras-numpy	57.42% <0.00%> (-0.01%)	⬇️
keras-tensorflow	63.64% <0.00%> (-0.01%)	⬇️
keras-torch	62.37% <0.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.