C++ autograd tutorial #921
Conversation
|
Deploy preview for pytorch-tutorials-preview ready! Built with commit 7a1634d https://deploy-preview-921--pytorch-tutorials-preview.netlify.com |
yf225
force-pushed
the
cpp_autograd_tutorial
branch
from
75c48f5 to
ea0a7a0
Compare
yf225
force-pushed
the
cpp_autograd_tutorial
branch
from
ea0a7a0 to
cf844b7
Compare
|
|
||
| 1 1 | ||
| 1 1 | ||
| [ CPUFloatType{2,2} ] |
There was a problem hiding this comment.
The cpp Tensor print does not show requires_grad and the grad_fn like we do in python?
There was a problem hiding this comment.
No it doesn't :( We might want to fix it though to be consistent with Python tensor print. I filed an issue here: pytorch/pytorch#36004.
| } | ||
|
|
||
| static variable_list backward(AutogradContext *ctx, variable_list grad_outputs) { | ||
| auto saved = ctx->get_saved_variables(); |
There was a problem hiding this comment.
Why is this not get_saved_tensors() ?
There was a problem hiding this comment.
Yeah I think this should be get_saved_tensors(), and my guess is that this API was created before the Variable/Tensor merge was completed, and the author wanted to make a distinction between tensors and variables. Now that we don't have this distinction anymore, I think we should change it to match the Python API: pytorch/pytorch#36099.
There was a problem hiding this comment.
I renamed all Variables to Tensor now, to minimize the inconsistency with Python to only this get_saved_variables() function :)
There was a problem hiding this comment.
Yes I guess it is just a remain of the cpp Tensor/Variable merge.
advanced_source/cpp_autograd.rst
Outdated
| ---------- | ||
|
|
||
| You should now have a good overview of PyTorch's C++ autograd API. | ||
| You can find the code examples displayed in this note (yf225 TODO update link) `here |
There was a problem hiding this comment.
Will merge the examples PR and update the link here, when everything else looks good on this PR :)
albanD
left a comment
albanD
left a comment
There was a problem hiding this comment.
LGTM thanks for the updates
|
Note to self: merge pytorch/examples#745 first and update link in this PR Done |
C++ autograd tutorial
3 participants
Accompanying example code: pytorch/examples#745.