Use ir methods to replace onnx helper

[justinchuby]

Ban onnx.helper and onnx.numpy_helper because they can be slow. Selectively enable usages of some with noqa: TID251 and updated usages of the rest.

Fix ir.tensor to generate float32 tensors when a plain python float is provided.

[codecov]

❌ 8 Tests Failed:

Tests completed	Failed	Passed	Skipped
13892	8	13884	1901

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

::tests.functions.onnxfns2_test

Stack Traces | 0s run time

onnxscript/ir/serde.py:99: in wrapper
    return func(*args, **kwargs)
onnxscript/ir/serde.py:1497: in serialize_tensor_into
    tensor_proto.raw_data = from_.tobytes()
onnxscript/ir/_core.py:453: in tobytes
    assert self.dtype.itemsize == array.itemsize, "Bug: The itemsize should match"
E   AssertionError: Bug: The itemsize should match

The above exception was the direct cause of the following exception:
tests/functions/onnxfns2_test.py:6: in <module>
    from tests.models import onnxfns2
tests/models/onnxfns2.py:87: in <module>
    @script()
onnxscript/main.py:89: in transform
    result = script_check(f_ast, opset, env, src, default_opset=default_opset)
onnxscript/main.py:33: in script_check
    return convert.translate_function_def(f)
onnxscript/converter.py:1460: in translate_function_def
    fn_ir = self._translate_function_def_common(stmt)
onnxscript/converter.py:1447: in _translate_function_def_common
    self._translate_stmt(s, index_of_stmt=i)
onnxscript/converter.py:981: in _translate_stmt
    return self._translate_if_stmt(node)
onnxscript/converter.py:1106: in _translate_if_stmt
    test = self._translate_expr(stmt.test, "cond").name
onnxscript/converter.py:559: in _translate_expr
    r = self._translate_compare_expr(node)
onnxscript/converter.py:906: in _translate_compare_expr
    right = self._translate_expr(node.comparators[0])
onnxscript/converter.py:565: in _translate_expr
    r = self._emit_const(self._eval_constant_expr(node), target, self._source_of(node))
onnxscript/converter.py:415: in _emit_const
    tensor = autocast.pyvalue_to_onnx_tensor(ovar, pyvalue)
onnxscript/_internal/autocast.py:27: in pyvalue_to_onnx_tensor
    return ir.serde.serialize_tensor(ir.tensor(pyvalue, name=tensor_name))
onnxscript/ir/serde.py:1461: in serialize_tensor
    serialize_tensor_into(tensor_proto, from_=tensor)
onnxscript/ir/serde.py:101: in wrapper
    raise SerdeError(
E   onnxscript.ir.serde.SerdeError: Error calling serialize_tensor_into with: Tensor<STRING,[]>(array('DCR', dtype='<U3'), name='const_0')

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0884_test_reduce_min_default_axes_keepdims_example

Stack Traces | 0.004s 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_reduce_min_default_axes_keepdims_example'

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_reduce_min_default_axes_keepdims_example' (e=No module named 'tests.onnx_backend_test_code.test_reduce_min_default_axes_keepdims_example') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_reduce_min_default_axes_keepdims_example.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_reduce_min_default_axes_keepdims_example.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 opset18
E   
E   @script()
E   def bck_test_reduce_min_default_axes_keepdims_example(data: FLOAT[3,2,2]) -> (FLOAT[1,1,1]):
E       reduced = opset18.ReduceMin(data, keepdims=1)
E       return reduced

tests.operator_test.TestConverter::test_const_promotion

Stack Traces | 0.004s run time

tests/operator_test.py:41: in test_const_promotion
    onnxscript.testing.assert_isomorphic_function(explicit_plus1, implicit_plus1)
onnxscript/testing/__init__.py:38: in assert_isomorphic_function
    assert _isomorphic(_to_function_proto(fn1), _to_function_proto(fn2))
E   AssertionError

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 aims to deprecate the use of onnx.helper and onnx.numpy_helper in favor of ir-based methods to improve performance. Key changes include replacing multiple onnx.helper calls with either their ir or ir.serde counterparts (and adding “# noqa: TID251” comments) and updating imports and utility functions accordingly.

Reviewed Changes

Copilot reviewed 21 out of 21 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
onnxscript/tensor.py	Updated import and added noqa comment on onnx.helper usage.
onnxscript/_internal/utils.py	Applied noqa comments to onnx.helper calls in utility functions.
onnxscript/optimizer/_constant_folding.py	Replaced onnx.numpy_helper.from_array with ir.serde.serialize_tensor.
onnxscript/main.py	Updated model export logic to use ir.Model and ir.save.
onnxscript/optimizer/_legacy/constant_folding.py, etc.	Added ruff: noqa: TID251 markers and adjusted helper calls across files.
onnxscript/_internal/autocast.py	Replaced tensor conversion via numpy_helper with ir.serde.serialize_tensor.
onnxscript/rewriter/llama_rule_sets.py	Modified allowed tensor types check to use as_int() instead of value.
onnxscript/rewriter/cast_constant_of_shape.py and others	Changed onnx.helper.make_tensor calls to use ir.tensor.
onnxscript/function_libs/torch_lib/ops/nn.py	Updated empty-tensor creation to use ir.tensor.

Comments suppressed due to low confidence (2)

onnxscript/rewriter/llama_rule_sets.py:65

• Ensure that converting the attribute value with as_int() returns the expected integer representation that matches the entries in _allowed_tensor_types. Verify that edge cases or unexpected types do not cause the check to fail.

to.as_int() in cls._allowed_tensor_types

onnxscript/_internal/autocast.py:39

• Confirm that the new serialization using ir.serde.serialize_tensor produces outputs consistent with the previous behavior of numpy_helper.from_array, ensuring compatibility with downstream tensor utilities.

return ir.serde.serialize_tensor(ir.tensor(pyvalue, name=tensor_name))