Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[onert] Introduce BiasInsertionPass #13842

Closed
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 46 additions & 0 deletions runtime/onert/core/src/compiler/train/pass/BiasInsertionPass.cc
Original file line number Diff line number Diff line change
@@ -0,0 +1,46 @@
/*
* Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

#include "BiasInsertionPass.h"

#include "ir/Graph.h"

namespace onert
{
namespace compiler
{
namespace train
{
namespace pass
{

void BiasInsertionPass::run()
{
_graph.operations().iterate([&](const ir::OperationIndex &op_index, const ir::IOperation &node) {
_current_op_index = op_index;
node.accept(*this);
});
}

void BiasInsertionPass::visit(const ir::operation::FullyConnected &)
{
// TODO Implement bias insertion for FullyConnected
}

} // namespace pass
} // namespace train
} // namespace compiler
} // namespace onert
53 changes: 53 additions & 0 deletions runtime/onert/core/src/compiler/train/pass/BiasInsertionPass.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,53 @@
/*
* Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

#ifndef __ONERT_COMPILER_TRAIN_PASS_BIAS_INSERTION_PASS_H__
#define __ONERT_COMPILER_TRAIN_PASS_BIAS_INSERTION_PASS_H__

#include "../../pass/Pass.h"
#include "ir/OperationVisitor.h"

namespace onert
{
namespace compiler
{
namespace train
{
namespace pass
{

class BiasInsertionPass final : public compiler::pass::Pass, public ir::OperationVisitor
{
public:
BiasInsertionPass(ir::Graph &graph) : compiler::pass::Pass{graph} {}

public:
std::string id() final { return "BiasInsertionPass"; }
void run() final;

public:
void visit(const ir::operation::FullyConnected &node) override;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we have to insert bias into each node?
I'd like to review this PR, But I feel hard to understand the background 🥲

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sometimes TFLite converter may generate a inference model that does not have bias input(i.e. optional input) when the bias data has all zero values. It's reasonable in inference because there is no reason to add zero values. But, in training, bias is necessary because bias needs to be trained and updated.
As a result, we need to insert bias input if a model to be trained does have bias input.

Copy link
Contributor Author

@ragmani ragmani Aug 30, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@zetwhite
Thanks for your advise offline. I made a note about that at #13808 (comment).


private:
ir::OperationIndex _current_op_index;
};

} // namespace pass
} // namespace train
} // namespace compiler
} // namespace onert

#endif // __ONERT_COMPILER_TRAIN_PASS_BIAS_INSERTION_PASS_H__