Implement BatchNorm double backwards#2207
Merged
soumith merged 4 commits intopytorch:masterfrom Jul 27, 2017
Merged
Conversation
…ctly from C++. This will be converted to C++ code once ATen is integrated with autograd.
Contributor
Author
|
The math isn't that clean here; I'm sure it can be simplified and made more efficient (as one example, I didn't use in-place functions). |
albanD
reviewed
Jul 26, 2017
| all_inputs.push_back(input_var); | ||
| if (weight.get()) { | ||
| all_inputs.push_back(weight_var); | ||
| all_inputs.push_back(bias_var); |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
|
||
| def sum_exclude_dim1(to_sum, keepdim=True): | ||
| to_sum = to_sum.sum(dim=0, keepdim=True) | ||
| dim = 2 |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| # define some terms we will reuse | ||
| M = reduce(mul, input.size()[0:1] + input.size()[2:]) | ||
| mu = sum_exclude_dim1(input).div_(M) | ||
| input_sub_mu = input - mu |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| gG = None | ||
| if affine and ggI is not None: | ||
| # gG is just the first backwards with the gamma term removed (then shaped properly) | ||
| gG = ggI * first_back_grad_input(gO, 1) |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
Contributor
Author
|
updated for @albanD's suggestions. |
Collaborator
|
woohoo! thanks Greg. |
ruotianluo
added a commit
to ruotianluo/pytorch-1
that referenced
this pull request
Jul 31, 2017
* commit '8262920b72374b1d9643f35057663ab02ab20330': (272 commits) Add ATen overload to AutoGPU. (pytorch#2234) Add comments for default value (pytorch#2242) Remove dead THPP code that has been replaced with ATen objects. (pytorch#2235) fix a bug where an uninitialized at::Tensor was passed to createPyObject (pytorch#2239) Replace thpp::Tensor with ATen Tensor in autograd csrc (pytorch#2170) Added aarch64 support (pytorch#2226) Increase tol. for float tensor qr big test. Improve Variable.retain_grad add `retain_grad` method, to variable, so gradient gets stored during backpop, on non-user variables Implement BatchNorm double backwards (pytorch#2207) [bugfix] in bce_with_logits logsumexp calculation (pytorch#2221) fix for ATen API Change Opt into Trusty builds. (pytorch#2214) allow retain to be specified for unsafeTensorFromTH Deduplicate THPUtils_checkLong/THPUtils_unpackLong (pytorch#2218) fix osx build errors related to long/int64_t Note [Undefined-dim versus 0-dim] Remove __func__ hack in auto nn. Enable Conv groups gradgradchecks. (pytorch#2216) fix a bug where some scalars were getting truncated to integers incorrectly. ...
ruotianluo
added a commit
to ruotianluo/pytorch-1
that referenced
this pull request
Aug 1, 2017
* commit '8262920b72374b1d9643f35057663ab02ab20330': (272 commits) Add ATen overload to AutoGPU. (pytorch#2234) Add comments for default value (pytorch#2242) Remove dead THPP code that has been replaced with ATen objects. (pytorch#2235) fix a bug where an uninitialized at::Tensor was passed to createPyObject (pytorch#2239) Replace thpp::Tensor with ATen Tensor in autograd csrc (pytorch#2170) Added aarch64 support (pytorch#2226) Increase tol. for float tensor qr big test. Improve Variable.retain_grad add `retain_grad` method, to variable, so gradient gets stored during backpop, on non-user variables Implement BatchNorm double backwards (pytorch#2207) [bugfix] in bce_with_logits logsumexp calculation (pytorch#2221) fix for ATen API Change Opt into Trusty builds. (pytorch#2214) allow retain to be specified for unsafeTensorFromTH Deduplicate THPUtils_checkLong/THPUtils_unpackLong (pytorch#2218) fix osx build errors related to long/int64_t Note [Undefined-dim versus 0-dim] Remove __func__ hack in auto nn. Enable Conv groups gradgradchecks. (pytorch#2216) fix a bug where some scalars were getting truncated to integers incorrectly. ...
zou3519
pushed a commit
to zou3519/pytorch
that referenced
this pull request
Mar 30, 2018
* update comments in segment_reduction_op * Update segment_reduction_op.cc
jagadish-amd
pushed a commit
to jagadish-amd/pytorch
that referenced
this pull request
Jul 9, 2025
…efault on ROCm7.0+ (pytorch#2207) Cherry-pick of ROCm#2180 Co-authored-by: Dmitry Nikolaev <139769634+dnikolaev-amd@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The double backwards function is currently implemented as a python function called direction from C++, but the plan is to convert it to C++ code once ATen is integrated with autograd.