Add add/remove initializers passes

[justinchuby]

Implement RemoveInitializersFromInputsPass and AddInitializersToInputsPass in onnxscript/ir/passes/common/constant_manipulation.py.

• RemoveInitializersFromInputsPass

  • Add RemoveInitializersFromInputsPass class to find and remove graph inputs with const_value.
  • Implement call method to remove inputs with const_value from graph.inputs.
  • Register RemoveInitializersFromInputsPass in the __all__ list.

• AddInitializersToInputsPass

  • Add AddInitializersToInputsPass class to find and add initializers to the graph inputs.
  • Implement call method to add all initializers to the graph inputs if not already present.
  • Register AddInitializersToInputsPass in the __all__ list.

• Tests

  • Add test cases for RemoveInitializersFromInputsPass in onnxscript/ir/passes/common/constant_manipulation_test.py to verify removal of inputs with const_value.
  • Add test cases for AddInitializersToInputsPass in onnxscript/ir/passes/common/constant_manipulation_test.py to verify addition of initializers to the graph inputs.

---

For more details, open the Copilot Workspace session.

[codecov]

❌ 7 Tests Failed:

Tests completed	Failed	Passed	Skipped
13873	7	13866	2338

View the top 3 failed test(s) by shortest run time

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0385_test_gathernd_example_int32

Stack Traces | 0.003s run time

onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_gathernd_example_int32'

The above exception was the direct cause of the following exception:
.nox\test\lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_gathernd_example_int32' (e=No module named 'tests.onnx_backend_test_code.test_gathernd_example_int32') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_gathernd_example_int32.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_gathernd_example_int32.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import INT32, INT64
E   from onnxscript.onnx_opset import opset13
E   
E   @script()
E   def bck_test_gathernd_example_int32(data: INT32[2,2], indices: INT64[2,2]) -> (INT32[2]):
E       output = opset13.GatherND(data, indices)
E       return output

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0507_test_gemm_default_matrix_bias

Stack Traces | 0.003s run time

onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
C:\hostedtoolcache\windows\Python\3.11.9\x64\Lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_gemm_default_matrix_bias'

The above exception was the direct cause of the following exception:
.nox\test_torch_nightly\Lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_gemm_default_matrix_bias' (e=No module named 'tests.onnx_backend_test_code.test_gemm_default_matrix_bias') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_gemm_default_matrix_bias.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_gemm_default_matrix_bias.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import FLOAT
E   from onnxscript.onnx_opset import opset13
E   
E   @script()
E   def bck_test_gemm_default_matrix_bias(a: FLOAT[3,6], b: FLOAT[6,4], c: FLOAT[3,4]) -> (FLOAT[3,4]):
E       y = opset13.Gemm(a, b, c)
E       return y

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0649_test_min_float32

Stack Traces | 0.003s run time

onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\importlib\__init__.py:126: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_min_float32'

The above exception was the direct cause of the following exception:
.nox\test\lib\site-packages\parameterized\parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
onnxscript\backend\onnx_export_test.py:271: in test_export2python_produces_correct_onnx_script_model
    functions = extract_functions(backend_test.name, code, self.test_folder)
onnxscript\backend\onnx_export_test.py:139: in extract_functions
    raise AssertionError(
E   AssertionError: Unable to import 'tests.onnx_backend_test_code.test_min_float32' (e=No module named 'tests.onnx_backend_test_code.test_min_float32') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_min_float32.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_min_float32.py', current folder: D:\a\onnxscript\onnxscript
E   ---- CONTENT --
E   import numpy
E   from onnx import TensorProto
E   from onnx.helper import make_tensor
E   from onnxscript import script, external_tensor
E   from onnxscript.values import Opset
E   from onnxscript.onnx_types import FLOAT
E   from onnxscript.onnx_opset import opset13
E   
E   @script()
E   def bck_test_min_float32(data_0: FLOAT[3], data_1: FLOAT[3]) -> (FLOAT[3]):
E       result = opset13.Min(data_0, data_1)
E       return result

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[Copilot]

Pull Request Overview

This PR implements two passes for constant manipulation in ONNX graphs by removing and adding initializers from/to graph inputs, and updates test cases accordingly.

• Introduces RemoveInitializersFromInputsPass to filter initializer values from graph inputs.
• Implements AddInitializersToInputsPass to add missing initializers to graph inputs.
• Updates and expands tests in constant_manipulation_test.py.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
onnxscript/ir/passes/common/constant_manipulation.py	Adds two new pass classes and registers them in all.
onnxscript/ir/passes/common/constant_manipulation_test.py	Adds tests to verify the behavior of both RemoveInitializersFromInputsPass and AddInitializersToInputsPass.

[titaiwangms]

Put it into optimizer?

[justinchuby]

Put it into optimizer?

For now pytorch exporter doesn't need it because we don't produce models with initializers being inputs by default. When users turn the option on they are requesting initializers to the inputs explicitly so we don't want to undo it for users. Maybe put into ort optimizations if needed?