aparler/PyNorch
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix max min sum access elements autograd backward running on device

Browse source
This commit is contained in:
lucasdelimanogueira 2024-05-23 12:11:59 -03:00
parent 27feb8a366
commit 9b49a07d2a
1 changed files with 5 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

10
norch/autograd/functions.py
View file

@ -136,7 +136,7 @@ class SumBackward:
input_shape = self.input[0].shape.copy()
if self.axis == -1:
# If axis is None, sum reduces the tensor to a scalar.
grad_output = float(gradient.tensor.contents.data[0]) * self.input[0].ones_like()
grad_output = float(gradient[[0] * len(gradient.shape)]) * self.input[0].ones_like()
else:
if self.keepdim:
@ -214,9 +214,9 @@ class MaxBackward:
max_value = self.input[0].max()
mask = self.input[0].equal(max_value)
grad_output = float(gradient.tensor.contents.data[0]) * self.input[0].ones_like()
grad_output = float(gradient[[0] * len(gradient.shape)]) * self.input[0].ones_like()
grad_output = (grad_output * mask) / mask.sum().tensor.contents.data[0]
grad_output = (grad_output * mask) / mask.sum()[0]
else:
@ -250,9 +250,9 @@ class MinBackward:
min_value = self.input[0].min()
mask = self.input[0].equal(min_value)
grad_output = float(gradient.tensor.contents.data[0]) * self.input[0].ones_like()
grad_output = float(gradient[[0] * len(gradient.shape)]) * self.input[0].ones_like()
grad_output = (grad_output * mask) / mask.sum().tensor.contents.data[0]
grad_output = (grad_output * mask) / mask.sum()[0]
else:
if self.keepdim: