[IR] Display constant tensors for Value

[justinchuby]

Display constant values and simplify the value repr string when fields are empty.

>>> from onnxscript import ir
>>> v = ir.Value(name="v1", const_value=ir.tensor(1))
>>> v
Value(name='v1', const_value=Tensor<INT64,[]>(array(1), name=None))
>>> v = ir.Value(name="v1", const_value=ir.tensor([[1]]))
>>> v
Value(name='v1', const_value=Tensor<INT64,[1,1]>(array([[1]]), name=None))
>>> print(v)
%"v1"<?,?>{Tensor<INT64,[1,1]>(array([[1]]), name=None)}

Fix #2073

[Copilot]

Pull Request Overview

This PR updates the repr method for the Value class so that it displays constant tensor values and simplifies the representation when optional fields are empty. Key changes include:

• Conditionally appending type, shape, producer, index, and const_value fields to the repr string.
• Removing the default "None" string for missing producer information.
• Improving the overall consistency of field formatting in the repr output.

[codecov]

❌ 3 Tests Failed:

Tests completed	Failed	Passed	Skipped
14461	3	14458	1879

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

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0467_test_hardsigmoid_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_hardsigmoid_example'

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_hardsigmoid_example' (e=No module named 'tests.onnx_backend_test_code.test_hardsigmoid_example') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_hardsigmoid_example.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_hardsigmoid_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 opset22
E   
E   @script()
E   def bck_test_hardsigmoid_example(x: FLOAT[3]) -> (FLOAT[3]):
E       y = opset22.HardSigmoid(x, alpha=0.5, beta=0.6000000238418579)
E       return y

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0641_test_max_uint64

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_max_uint64'

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_max_uint64' (e=No module named 'tests.onnx_backend_test_code.test_max_uint64') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_max_uint64.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_max_uint64.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 UINT64
E   from onnxscript.onnx_opset import opset13
E   
E   @script()
E   def bck_test_max_uint64(data_0: UINT64[3], data_1: UINT64[3]) -> (UINT64[3]):
E       result = opset13.Max(data_0, data_1)
E       return result

onnxscript.backend.onnx_export_test.TestOnnxBackEnd::test_export2python_produces_correct_onnx_script_model_0791_test_quantizelinear_int16

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_quantizelinear_int16'

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_quantizelinear_int16' (e=No module named 'tests.onnx_backend_test_code.test_quantizelinear_int16') (file: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_quantizelinear_int16.py', absolute path: 'D:\\a\\onnxscript\\onnxscript\\tests\\onnx_backend_test_code\\test_quantizelinear_int16.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, INT16
E   from onnxscript.onnx_opset import opset21
E   
E   @script()
E   def bck_test_quantizelinear_int16(x: FLOAT[16], y_scale: FLOAT, y_zero_point: INT16) -> (INT16[16]):
E       y = opset21.QuantizeLinear(x, y_scale, y_zero_point)
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 improves the string representations of constant Value objects by including tensor details when available and simplifying the output when fields are empty. Key changes include:

• Adding conditional inclusion of type, shape, and producer information in repr.
• Introducing a shortened representation for const_value in repr and a similar change in str when the tensor size is small.

[gramalingam]

Does this show up in the displayed form of a Node? Eg., if we have a node Y = Add(X, c) where c is a constant 1, for example? Would the displayed representation inside a node for an (input) value be too dense (with type/shape etc.)? If so, can we have a more compact representation for an input-value representation inside a node? May be just the name "c" and the value "1"?

[justinchuby]

In a node, it won't show. In a graph it will show in the initializers section

graph(
    name=anonymous_graph:124652572372048,
    inputs=(

    ),
    outputs=(

    ),
    initializers=(
        %"init"<?,?>{Tensor<INT64,[5]>(array([1, 1, 1, 1, 1]), name=None)}
    ),
) {
    0 |  # node_Add_0
         %"val_0"<?,?> ⬅️ ::Add(%"init", %"init")
    return 
}

A Constant node looks like

%"anonymous:124652572372688"<?,?> ⬅️ ::Constant() {value=Tensor<INT64,[5]>(array([1, 1, 1, 1, 1]), name=None)}

Is that ok? Or maybe we want to see it directly in the node?

[justinchuby]

I updated for it to show

%"anonymous:132887616237136"<?,?> ⬅️ ::Add(%"init"{Tensor<INT64,[5]>(array([1, 1, 1, 1, 1]), name=None)}, %"init"{Tensor<INT64,[5]>(array([1, 1, 1, 1, 1]), name=None)})

[titaiwangms]

What do we have when it's larger than 10 elements?

[justinchuby]

It will show {...}

[gramalingam]

I updated for it to show

%"anonymous:132887616237136"<?,?> ⬅️ ::Add(%"init"{Tensor<INT64,[5]>(array([1, 1, 1, 1, 1]), name=None)}, %"init"{Tensor<INT64,[5]>(array([1, 1, 1, 1, 1]), name=None)})

Great! (My original request, at least, was for the ability to see it in the node.) What would be even better (IMO) is if this can be made "less busy" ... in typical usage, just the values will suffice (you want to see "0" or "1" or "0.36712") ... given that the full information is available elsewhere, if a user really wants to know, a shorter version should work better here.

I do think there is a difference between seeing %"Y" <- ::Add (%"X", %"init"{1}) and %"Y" <- ::Add (%"X", %"init"{Tensor<INT64,[1]>(array([1]), name=None)}) ... in favor of the first.

[justinchuby]

Sounds goo - I will create a follow up