fix reduce min max backward bug #5651
Merged
Conversation
This file contains 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
hjchen2
reviewed
Jul 29, 2021
| const auto& bcast_like = | ||
| JUST(OpInterpUtil::Dispatch<Tensor>(*bcast_like_op_, {output, input}, bcast_attrs)); | ||
| const auto& bcast_eq = | ||
| JUST(OpInterpUtil::Dispatch<Tensor>(*bcast_equal_op_, {input, bcast_like})); | ||
| const auto& cast_like = JUST(OpInterpUtil::Dispatch<Tensor>(*cast_like_op_, {bcast_eq, input})); | ||
| const auto& reduce_sum_ = | ||
| JUST(OpInterpUtil::Dispatch<Tensor>(*reduce_sum_op_, {cast_like}, reduce_sum_attrs)); |
There was a problem hiding this comment.
这里用functional接口重写一下吧，
const auto& reduce_sum_ = JUST(functional::ReduceSum(cast_like, ctx->axis, ctx->keepdims));其他地方也一样改一下，CastLike目前还没有functional接口，也一起加一个。
hjchen2
reviewed
Jul 29, 2021
| @@ -19,6 +19,7 @@ limitations under the License. | |||
| #include "oneflow/core/framework/op_expr.h" | |||
| #include "oneflow/core/framework/op_expr_helper.h" | |||
There was a problem hiding this comment.
op_expr_helper.h 这个头文件可以删了
| }; | ||
|
|
||
| Maybe<void> ReduceMaxOrMinOp::Init(const OpExpr& op) { | ||
| const auto* fw_op_expr = dynamic_cast<const UserOpExpr*>(&op); | ||
| CHECK_NOTNULL_OR_RETURN(fw_op_expr); | ||
| base_attrs_ = MakeAttrMapFromUserOpConf(fw_op_expr->proto()); | ||
| const std::string& op_name = fw_op_expr->op_name(); | ||
| bcast_like_op_ = | ||
| JUST(op_expr_helper::BroadcastLikeOp(/*axis=*/{-1}, GradientOpName(op_name + "_bcast_like"))); |
There was a problem hiding this comment.
顺便把这个文件中其他地方的op_expr_helper::XXXOp也一起改成functional的吧
| const auto& bcast_eq = JUST(functional::BroadcastEqual(input, bcast_like)); | ||
| const auto& cast_like = JUST(functional::CastLike(bcast_eq, input)); | ||
| const auto& reduce_sum_ = JUST(functional::ReduceSum(cast_like, ctx->axis, ctx->keepdims)); | ||
| const auto& bcast_div_ = JUST(functional::BroadcastDiv(dy, reduce_sum_)); |
hjchen2
approved these changes
Jul 31, 2021
hjchen2
approved these changes
Jul 31, 2021
Speed stats: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
When I referred to some implementations in reduce_min backward, I found there is different behaviour between oneflow and pytorch:
So I fix it!