[torchlib] Fix scatter reduce on error cases

[justinchuby]

Fix three errors

value = ir.tensor([np.iinfo(dtype.numpy()).min], dtype=dtype)
                       ^^^^^^^^^^^^^^^^^^^^^^^
  File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/site-packages/numpy/_core/getlimits.py", line 706, in __init__
    raise ValueError("Invalid integer data type %r." % (self.kind,))
ValueError: Invalid integer data type 'b'.

Traceback (most recent call last):
  File "/Users/runner/work/torch-onnx-op-matrix/torch-onnx-op-matrix/op_matrix/onnx_dynamo_op_survey.py", line 54, in check_single_op
    onnx.checker.check_model(onnx_model, full_check=True)  # type: ignore
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/site-packages/onnx/checker.py", line 180, in check_model
    C.check_model(
onnx.onnx_cpp2py_export.checker.ValidationError: Mismatched attribute type in 'node_ConstantOfShape_1 : value'. Expected: 'TENSOR', actual: 'INT'

==> Context: Bad node spec for node. Name: node_ConstantOfShape_1 OpType: ConstantOfShape

Fix a case for bfloat16 when min should be max.

[codecov]

❌ 3 Tests Failed:

Tests completed	Failed	Passed	Skipped
14468	3	14465	1880

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

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_1000_test_scatternd_multiply

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_scatternd_multiply'

The above exception was the direct cause of the following exception:
.nox\test_ort_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_scatternd_multiply' (e=No module named 'tests.onnx_backend_test_code.test_scatternd_multiply') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_scatternd_multiply.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_scatternd_multiply.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, INT64
E   from onnxscript.onnx_opset import opset18
E   
E   @script()
E   def bck_test_scatternd_multiply(data: FLOAT[4,4,4], indices: INT64[2,1], updates: FLOAT[2,4,4]) -> (FLOAT[4,4,4]):
E       y = opset18.ScatterND(data, indices, updates, reduction='mul')
E       return y

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0059_test_argmin_keepdims_random_select_last_index

Stack Traces | 0.006s run time

onnxscript\backend\onnx_export_test.py:137: in extract_functions
    mod = importlib.import_module(import_name)
C:\hostedtoolcache\windows\Python\3.12.10\x64\Lib\importlib\__init__.py:90: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
E   ModuleNotFoundError: No module named 'tests.onnx_backend_test_code.test_argmin_keepdims_random_select_last_index'

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_argmin_keepdims_random_select_last_index' (e=No module named 'tests.onnx_backend_test_code.test_argmin_keepdims_random_select_last_index') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_argmin_keepdims_random_select_last_index.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_argmin_keepdims_random_select_last_index.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, INT64
E   from onnxscript.onnx_opset import opset13
E   
E   @script()
E   def bck_test_argmin_keepdims_random_select_last_index(data: FLOAT[2,3,4]) -> (INT64[2,1,4]):
E       result = opset13.ArgMin(data, axis=1, keepdims=1, select_last_index=1)
E       return result

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_1151_test_softmax_negative_axis_expanded_ver18

Stack Traces | 0.006s 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_softmax_negative_axis_expanded_ver18'

The above exception was the direct cause of the following exception:
.nox\test_ort_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_softmax_negative_axis_expanded_ver18' (e=No module named 'tests.onnx_backend_test_code.test_softmax_negative_axis_expanded_ver18') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_softmax_negative_axis_expanded_ver18.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_softmax_negative_axis_expanded_ver18.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_softmax_negative_axis_expanded_ver18(x: FLOAT[3,4,5]) -> (FLOAT[3,4,5]):
E       Softmax_test_softmax_negative_axis_expanded_function_axes = opset18.Constant(value=make_tensor("value", 7, dims=[1], vals=[-1]))
E       Softmax_test_softmax_negative_axis_expanded_function_X_ReduceMax = opset18.ReduceMax(x, Softmax_test_softmax_negative_axis_expanded_function_axes, keepdims=1)
E       Softmax_test_softmax_negative_axis_expanded_function_X_Sub = opset18.Sub(x, Softmax_test_softmax_negative_axis_expanded_function_X_ReduceMax)
E       Softmax_test_softmax_negative_axis_expanded_function_X_Exp = opset18.Exp(Softmax_test_softmax_negative_axis_expanded_function_X_Sub)
E       Softmax_test_softmax_negative_axis_expanded_function_X_ReduceSum = opset18.ReduceSum(Softmax_test_softmax_negative_axis_expanded_function_X_Exp, Softmax_test_softmax_negative_axis_expanded_function_axes, keepdims=1)
E       y = opset18.Div(Softmax_test_softmax_negative_axis_expanded_function_X_Exp, Softmax_test_softmax_negative_axis_expanded_function_X_ReduceSum)
E       return y

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 fixes error cases in the scatter reduce operation by correcting the tensor initialization values for BFLOAT16, BOOL, and non-reduction modes.

• Fixes the BFLOAT16 branch in the max reduction case by using torch.finfo(torch.bfloat16).max.
• Adds proper handling for BOOL type in both min (False) and max (True) branches and ensures a consistent tensor type is returned for the "none" reduction case.

Comments suppressed due to low confidence (3)

onnxscript/function_libs/torch_lib/ops/core.py:7630

• The addition of boolean handling in the 'min' branch for scatter reduce appears to resolve the error with BOOL types; please confirm that using 'False' is semantically correct for a minimum reduction with boolean values.

elif dtype == ir.DataType.BOOL:
                value = ir.tensor([False], dtype=dtype)

onnxscript/function_libs/torch_lib/ops/core.py:7643

• Changing the BFLOAT16 case to use the maximum value for a max reduction correctly addresses the error, but please double-check that this change aligns with the intended behavior compared to the PyTorch semantics.

value = ir.tensor([torch.finfo(torch.bfloat16).max], dtype=dtype)

onnxscript/function_libs/torch_lib/ops/core.py:7656

• Replacing the literal 0 with a tensor wrapping ensures consistency in type handling; please verify that this change maintains the expected behavior in downstream operations.

value = ir.tensor([0], dtype=dtype)