Naturally author ONNX functions and models using a subset of Python
Project description
ONNX Script
ONNX Script enables developers to naturally author ONNX functions and models using a subset of Python. ONNX Script is:
- Expressive: enables the authoring of all ONNX functions.
- Simple and concise: function code is natural and simple.
- Debuggable: allows for eager-mode evaluation that provides for a more delightful ONNX model debugging experience.
Note however that ONNX Script does not intend to support the entirety of the Python language.
Design Overview
ONNX Script provides a few major capabilities for authoring and debugging ONNX models and functions:
-
A converter which translates a Python ONNX Script function into an ONNX graph, accomplished by traversing the Python Abstract Syntax Tree to build an ONNX graph equivalent of the function.
-
A converter that operates inversely, translating ONNX models and functions into ONNX Script. This capability can be used to fully round-trip ONNX Script ↔ ONNX graph.
-
A runtime shim that allows such functions to be evaluated (in an "eager mode"). This functionality currently relies on ONNX Runtime for executing every ONNX Operator, and there is a Python-only reference runtime for ONNX underway that will also be supported.
Note that the runtime is intended to help understand and debug function definitions. Performance is not a goal here.
Installing ONNX Script
pip install --upgrade onnxscript
Install for Development
pip install onnx onnxruntime pytest
git clone https://github.com/microsoft/onnxscript
cd onnxscript
pip install -e .
Run Unit Tests
pytest onnxscript
Example
import onnx
# We use ONNX opset 15 to define the function below.
from onnxscript import FLOAT
from onnxscript import opset15 as op
from onnxscript import script
# We use the script decorator to indicate that
# this is meant to be translated to ONNX.
@script()
def onnx_hardmax(X, axis: int):
"""Hardmax is similar to ArgMax, with the result being encoded OneHot style."""
# The type annotation on X indicates that it is a float tensor of
# unknown rank. The type annotation on axis indicates that it will
# be treated as an int attribute in ONNX.
#
# Invoke ONNX opset 15 op ArgMax.
# Use unnamed arguments for ONNX input parameters, and named
# arguments for ONNX attribute parameters.
argmax = op.ArgMax(X, axis=axis, keepdims=False)
xshape = op.Shape(X, start=axis)
# use the Constant operator to create constant tensors
zero = op.Constant(value_ints=[0])
depth = op.GatherElements(xshape, zero)
empty_shape = op.Constant(value_ints=[0])
depth = op.Reshape(depth, empty_shape)
values = op.Constant(value_ints=[0, 1])
cast_values = op.CastLike(values, X)
return op.OneHot(argmax, depth, cast_values, axis=axis)
# We use the script decorator to indicate that
# this is meant to be translated to ONNX.
@script()
def sample_model(X: FLOAT[64, 128], Wt: FLOAT[128, 10], Bias: FLOAT[10]) -> FLOAT[64, 10]:
matmul = op.MatMul(X, Wt) + Bias
return onnx_hardmax(matmul, axis=1)
# onnx_model is an in-memory ModelProto
onnx_model = sample_model.to_model_proto()
# Save the ONNX model at a given path
onnx.save(onnx_model, "sample_model.onnx")
# Check the model
try:
onnx.checker.check_model(onnx_model)
except onnx.checker.ValidationError as e:
print(f"The model is invalid: {e}")
else:
print("The model is valid!")
The decorator parses the code of the function, converting it into an
intermediate representation. If it fails, it produces an error message
indicating the line where the error was detected. If it succeeds, the
intermediate representation can be converted into an ONNX graph
structure of type FunctionProto:
Hardmax.to_function_proto()returns aFunctionProto
Eager Mode Evaluation
Eager mode is mostly used to debug and validate that intermediate results are as expected. The function defined above can be called as below, executing in an eager-evaluation mode:
import numpy as np
v = np.array([[0, 1], [2, 3]], dtype=np.float32)
result = Hardmax(v)
More examples can be found in the docs/examples directory.
Development Guidelines
Every change impacting the converter or the eager evaluation must be
unit tested with class OnnxScriptTestCase to ensure both systems do
return the same results with the same inputs.
Coding Style
We use ruff, black, isort, and mypy etc. to check code formatting and use lintrunner to run all linters.
You can install the dependencies and initialize with
pip install lintrunner lintrunner-adapters
lintrunner init
This will install lintrunner on your system and download all the necessary dependencies to run linters locally.
If you want to see what lintrunner init will install, run lintrunner init --dry-run.
To lint local changes:
lintrunner
To format files:
lintrunner f
To lint all files:
lintrunner --all-files
Use --output oneline to produce a compact list of lint errors, useful when
there are many errors to fix.
See all available options with lintrunner -h.
To read more about lintrunner, see wiki.
To update an existing linting rule or create a new one, modify .lintrunner.toml or create a
new adapter following examples in https://github.com/justinchuby/lintrunner-adapters.
Contributing
We're always looking for your help to improve the product (bug fixes, new features, documentation, etc). Currently ONNX Script is under early and heavy development, so we encourage proposing any major changes by filing an issue to discuss your idea with the team first.
Report a Security Issue
Please do not report security vulnerabilities through public GitHub issues.
Please refer to our guidance on filing Security Issues.
Licensing Guidelines
This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.
When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repositories using our CLA.
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.
Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos is subject to those third-party's policies.
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file onnxscript-0.1.0.dev20230924.tar.gz.
File metadata
- Download URL: onnxscript-0.1.0.dev20230924.tar.gz
- Upload date:
- Size: 443.8 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: RestSharp/106.13.0.0
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 936611d1064431a50e46afb05a0af8d0f924148e061868300fc447551d4b6776 |
|
| MD5 | ad75a88c3f8c474a893de8052017f9a6 |
|
| BLAKE2b-256 | 234111cff640e728d884f17be9b9b79e6c1f45aceb9afb68fa5a61fd3e223e18 |
File details
Details for the file onnxscript-0.1.0.dev20230924-py3-none-any.whl.
File metadata
- Download URL: onnxscript-0.1.0.dev20230924-py3-none-any.whl
- Upload date:
- Size: 534.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: RestSharp/106.13.0.0
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 599ad92fb134ba38b569b07e192c17cf7ead1f253360e858b15e8bce143a9f61 |
|
| MD5 | 698c717485e37361ac2212a603672795 |
|
| BLAKE2b-256 | f22f5baaffbba8c3e2915a091a46ef8dd61fca1a354097aa2e22f1e6b49cf24c |
