|
About the autograd category
|
|
0
|
3598
|
May 13, 2017
|
|
Autograd through two networks
|
|
0
|
14
|
May 3, 2024
|
|
Gradiant cusom variables
|
|
3
|
28
|
May 3, 2024
|
|
Why detach probs in symmetric_kl function (deberta shift)
|
|
1
|
25
|
May 3, 2024
|
|
Parameters stuck at zero for custom layer
|
|
8
|
52
|
May 2, 2024
|
|
Custom gradients only for inputs that require grad
|
|
3
|
41
|
May 1, 2024
|
|
Is it safe to modify output's grad and return as input's grad?
|
|
8
|
78
|
May 1, 2024
|
|
Skip gradient propagation if gradients less than a threshold
|
|
5
|
107
|
May 1, 2024
|
|
What does the function wrapper @once_differentiable do?
|
|
5
|
3505
|
May 1, 2024
|
|
In what situation does the decorator once_differentiable help?
|
|
3
|
55
|
May 1, 2024
|
|
2x slowdown when directly operating on parameter tensor in forward function
|
|
2
|
40
|
April 30, 2024
|
|
Compute gradient with respect to the input for high-dimensional output
|
|
1
|
32
|
April 30, 2024
|
|
Backprop through indices based on some operations
|
|
3
|
53
|
April 29, 2024
|
|
My network's weights get updated despite using torch.no_grad()
|
|
4
|
52
|
April 29, 2024
|
|
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [200, 2]], which is output 0 of AsStridedBackward0, is at version 1401; expected version 1400 instead
|
|
5
|
168
|
April 28, 2024
|
|
How to concatenate feature vectors from different extractors inplace
|
|
0
|
35
|
April 27, 2024
|
|
How to make my faster GPU execute two batches while my other GPU does a batch?
|
|
0
|
42
|
April 26, 2024
|
|
`zero_grad` before `step` causes gradient explosion?
|
|
3
|
68
|
April 26, 2024
|
|
Meta learning weight update
|
|
0
|
33
|
April 26, 2024
|
|
Propagate gradient to parameters not directly involved in computation
|
|
9
|
178
|
April 25, 2024
|
|
Is it safe to set flag for single autograd.Function.backward call
|
|
3
|
61
|
April 24, 2024
|
|
Diverging gradients obtained through: 1) tensor.register_hook in a class vs 2) module.register_full_backward_hook outside of a class
|
|
0
|
222
|
April 24, 2024
|
|
Fixing the Training Function for an LSTM model built from scratch
|
|
4
|
60
|
April 24, 2024
|
|
What is the detailed operations of `torchfunc.jvp`
|
|
2
|
54
|
April 24, 2024
|
|
Getting gradients with respect to input features for a batched input
|
|
3
|
50
|
April 23, 2024
|
|
CUDA Out of Memory Error with jvp() in Large Model Training - Need Insights or Fixes
|
|
0
|
44
|
April 23, 2024
|
|
Mixing pytorch autograd with custom vjp for optimization
|
|
2
|
28
|
April 23, 2024
|
|
Loss.backward() call not updating loss
|
|
2
|
40
|
April 23, 2024
|
|
Out.backward() and weight access
|
|
1
|
39
|
April 23, 2024
|
|
Backpropagating w.r.t detached tensor
|
|
2
|
37
|
April 23, 2024
|