4e2d88b75f
## Description
This pull request aims to enhance the efficiency of the inference
session creation by eliminating unnecessary `Node::ToProto` invocations.
The current codebase presents opportunities for optimization,
particularly in the removal of superfluous `Node::ToProto` calls, along
with their subsequent `~NodeProto` invocations.
## Motivation and Context
The optimization focus of this pull request is on addressing low-hanging
fruit in the inference session creation process. By strategically
removing undesired `Node::ToProto` calls, we aim to streamline the
codebase and enhance the overall performance. The flame graphs
illustrate the notable improvements achieved by reducing the percentage
of `Node::ToProto` calls, thereby optimizing the execution flow.
### Code Snippet
```cpp
TEST(InferenceSessionTests, Bench) {
// Initialize logging manager
auto logging_manager = std::make_unique<logging::LoggingManager>(
std::unique_ptr<ISink>(new CLogSink()), logging::Severity::kVERBOSE, false,
LoggingManager::InstanceType::Temporal);
// Create environment
std::unique_ptr<Environment> env;
auto st = Environment::Create(std::move(logging_manager), env);
ASSERT_TRUE(st.IsOK());
// Configure session options
SessionOptions so;
so.execution_mode = ExecutionMode::ORT_SEQUENTIAL;
so.graph_optimization_level = TransformerLevel::Level2;
so.intra_op_param.thread_pool_size = 1;
// Initialize and load the InferenceSession
InferenceSessionTestGlobalThreadPools session1{so, *env};
ASSERT_STATUS_OK(session1.Load("big.onnx"));
ASSERT_STATUS_OK(session1.Initialize());
}
```
### `big.onnx` model creation
```python
import onnx
import numpy as np
from spox import argument, build, Tensor, Var
from spox.opset.ai.onnx import v17 as op
from spox.opset.ai.onnx.ml.v3 import label_encoder
a = argument(Tensor(np.int64, ('N',)))
c = a
for x in range(1000):
c = op.mul(c, op.const(np.ones(10000, dtype=np.int64)))
for x in range(3000):
all_strings = list("random_string" + str(i) for i in range(100))
all_ints = list(range(len(all_strings)))
c = label_encoder(
c,
keys_int64s=all_ints,
values_strings=all_strings
)
c = label_encoder(c, keys_strings=all_strings, values_int64s=all_ints)
model: onnx.ModelProto = build(inputs={'a': a}, outputs={'c': c})
onnx.save(model, "big.onnx")
```
Testing in `Release` with `perf` yields:
Before: 3.3% spent in `Node::ToProto`
After: 1.6% spent in `Node::ToProto`
---------
Co-authored-by: Atanas Dimitrov <atanasdimitrov@Atanass-MacBook-Pro.local>
|
||
|---|---|---|
| .config | ||
| .devcontainer | ||
| .gdn | ||
| .github | ||
| .pipelines | ||
| .vscode | ||
| cgmanifests | ||
| cmake | ||
| csharp | ||
| dockerfiles | ||
| docs | ||
| include/onnxruntime/core | ||
| java | ||
| js | ||
| objectivec | ||
| onnxruntime | ||
| orttraining | ||
| rust | ||
| samples | ||
| tools | ||
| winml | ||
| .clang-format | ||
| .clang-tidy | ||
| .dockerignore | ||
| .gitattributes | ||
| .gitignore | ||
| .gitmodules | ||
| .lintrunner.toml | ||
| build.bat | ||
| build.sh | ||
| build_arm64x.bat | ||
| CITATION.cff | ||
| CODEOWNERS | ||
| CONTRIBUTING.md | ||
| lgtm.yml | ||
| LICENSE | ||
| NuGet.config | ||
| ort.wprp | ||
| ORT_icon_for_light_bg.png | ||
| packages.config | ||
| pyproject.toml | ||
| README.md | ||
| requirements-dev.txt | ||
| requirements-doc.txt | ||
| requirements-lintrunner.txt | ||
| requirements-training.txt | ||
| requirements.txt.in | ||
| SECURITY.md | ||
| setup.py | ||
| ThirdPartyNotices.txt | ||
| VERSION_NUMBER | ||
ONNX Runtime is a cross-platform inference and training machine-learning accelerator.
ONNX Runtime inference can enable faster customer experiences and lower costs, supporting models from deep learning frameworks such as PyTorch and TensorFlow/Keras as well as classical machine learning libraries such as scikit-learn, LightGBM, XGBoost, etc. ONNX Runtime is compatible with different hardware, drivers, and operating systems, and provides optimal performance by leveraging hardware accelerators where applicable alongside graph optimizations and transforms. Learn more →
ONNX Runtime training can accelerate the model training time on multi-node NVIDIA GPUs for transformer models with a one-line addition for existing PyTorch training scripts. Learn more →
Get Started & Resources
-
General Information: onnxruntime.ai
-
Usage documentation and tutorials: onnxruntime.ai/docs
-
YouTube video tutorials: youtube.com/@ONNXRuntime
-
Companion sample repositories:
- ONNX Runtime Inferencing: microsoft/onnxruntime-inference-examples
- ONNX Runtime Training: microsoft/onnxruntime-training-examples
Builtin Pipeline Status
| System | Inference | Training |
|---|---|---|
| Windows | ||
| Linux | ||
| Mac | ||
| Android | ||
| iOS | ||
| Web | ||
| Other |
Third-party Pipeline Status
| System | Inference | Training |
|---|---|---|
| Linux |  |
Data/Telemetry
Windows distributions of this project may collect usage data and send it to Microsoft to help improve our products and services. See the privacy statement for more details.
Contributions and Feedback
We welcome contributions! Please see the contribution guidelines.
For feature requests or bug reports, please file a GitHub Issue.
For general discussion or questions, please use GitHub Discussions.
Code of Conduct
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.
License
This project is licensed under the MIT License.