Create initializers with the constant folding pass #2598

justinchuby · 2025-10-03T05:32:35Z

Also Create default values for the constant folding pass

Signed-off-by: Justin Chu <[email protected]>

codecov · 2025-10-03T05:35:13Z

❌ 31 Tests Failed:

Tests completed	Failed	Passed	Skipped
3983	31	3952	5403

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

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_fold_add

Stack Traces | 0.003s run time

onnxscript/optimizer/_constant_folding_test.py:38: in test_fold_add
    optimized = self._fold(model)
                ^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'four' of node: 
E   name:  OpType: Mul
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_fold_if_propagate

Stack Traces | 0.003s run time

onnxscript/optimizer/_constant_folding_test.py:156: in test_fold_if_propagate
    optimized = self._fold(model)
                ^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'zero_cast' of node: 
E   name:  OpType: Equal
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_fold_if_cond

Stack Traces | 0.004s run time

onnxscript/optimizer/_constant_folding_test.py:105: in test_fold_if_cond
    optimized = self._fold(model)
                ^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'zero_cast' of node: 
E   name:  OpType: Equal
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_fold_shape

Stack Traces | 0.004s run time

onnxscript/optimizer/_constant_folding_test.py:69: in test_fold_shape
    optimized = self._fold(model)
                ^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'two_float' of node: 
E   name:  OpType: Add
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_input_size_limit

Stack Traces | 0.004s run time

onnxscript/optimizer/_constant_folding_test.py:615: in test_input_size_limit
    optimized = self._fold(model, input_size_limit=256 * 256, output_size_limit=256 * 256)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'w_squared' of node: 
E   name:  OpType: Add
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_shape_inference

Stack Traces | 0.005s run time

onnxscript/optimizer/_constant_folding_test.py:211: in test_shape_inference
    optimized = self._fold(model, onnx_shape_inference=True)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'cond' of node: 
E   name:  OpType: If
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_fold_inside_if_branch

Stack Traces | 0.008s run time

onnxscript/optimizer/_constant_folding_test.py:130: in test_fold_inside_if_branch
    optimized = self._fold(model)
                ^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'temp' of node: 
E   name:  OpType: Mul
E    is not output of any previous nodes.
E   
E   ==> Context: Bad node spec for node. Name:  OpType: If

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_fold_shape_slice

Stack Traces | 0.008s run time

onnxscript/optimizer/_constant_folding_test.py:85: in test_fold_shape_slice
    optimized = self._fold(model)
                ^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'two_float' of node: 
E   name:  OpType: Add
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_fold_cast_like

Stack Traces | 0.009s run time

onnxscript/optimizer/_constant_folding_test.py:53: in test_fold_cast_like
    optimized = self._fold(model)
                ^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'two_float' of node: 
E   name:  OpType: Add
E    is not output of any previous nodes.

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_transpose_is_always_folded

Stack Traces | 0.012s run time

onnxscript/optimizer/_constant_folding_test.py:634: in test_transpose_is_always_folded
    optimized = self._fold(model, input_size_limit=1)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Graph output 'z' is not an output of any node in graph.

onnxscript.optimizer._function_folding_test.FunctionFoldingTest::test_fold_nested_if_function_succeeds

Stack Traces | 0.038s run time

onnxscript/optimizer/_function_folding_test.py:155: in test_fold_nested_if_function_succeeds
    self.assertNotIn("If", {n.op_type for n in optimized.graph})
E   AssertionError: 'If' unexpectedly found in {'If'}

onnxscript.rewriter.ort_fusions.gelu_test.GeluFusionTest::test_gelu_fusion

Stack Traces | 0.059s run time

.../rewriter/ort_fusions/gelu_test.py:44: in test_gelu_fusion
    original_output = test_utils.ort_run("Original", model, input)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'const_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.cos_sin_cache_test.TestCosSinCacheTransform::test_cos_sin_fusion_1_test_case_1

Stack Traces | 0.163s run time

..../test/lib/python3.11.../site-packages/parameterized/parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/cos_sin_cache_test.py:45: in test_cos_sin_fusion
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'inv_freq_3d' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.cos_sin_cache_test.TestCosSinCacheTransform::test_cos_sin_fusion_2_test_case_2

Stack Traces | 0.333s run time

..../test/lib/python3.11.../site-packages/parameterized/parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/cos_sin_cache_test.py:45: in test_cos_sin_fusion
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'inv_freq_3d' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.mha_test.TestMultiHeadAttention::test_whisper_encoder

Stack Traces | 0.41s run time

.../rewriter/ort_fusions/mha_test.py:58: in test_whisper_encoder
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'encoder_encoder_conv1_bias_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.gqa_test.GQAFusionTest::test_fusion

Stack Traces | 0.497s run time

.../rewriter/ort_fusions/gqa_test.py:355: in test_fusion
    session = ort.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'int64_1_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.skip_normalization_test.TestSkipNormalization::test_bart_encoder

Stack Traces | 2.09s run time

.../rewriter/ort_fusions/skip_normalization_test.py:71: in test_bart_encoder
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'encoder_layernorm_embedding_weight' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.attention_test.TestAttentionFusion::test_whisper_encoder

Stack Traces | 4.14s run time

.../rewriter/ort_fusions/attention_test.py:170: in test_whisper_encoder
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'encoder_encoder_conv1_bias_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.mha_test.TestMultiHeadAttention::test_whisper_decoder

Stack Traces | 7.42s run time

.../rewriter/ort_fusions/mha_test.py:84: in test_whisper_decoder
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'sym_size_int_42' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.skip_normalization_test.TestSkipNormalization::test_whisper_decoder

Stack Traces | 13.2s run time

.../rewriter/ort_fusions/skip_normalization_test.py:56: in test_whisper_decoder
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'sym_size_int_42' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.mha_test.TestMultiHeadAttention::test_smollm

Stack Traces | 31.5s run time

.../rewriter/ort_fusions/mha_test.py:34: in test_smollm
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'int64_16_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.rms_normalization_test.TestRmsNormalization::test_smollm

Stack Traces | 66.2s run time

.../rewriter/ort_fusions/rms_normalization_test.py:19: in test_smollm
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'const_13_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.rotary_embedding_test.TestRotaryEmbedding::test_rotary_embedding_fusion_1_smollm_test_1

Stack Traces | 72.4s run time

..../test/lib/python3.11.../site-packages/parameterized/parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/rotary_embedding_test.py:33: in test_rotary_embedding_fusion
    self.assertIn("RotaryEmbedding", op_types)
E   AssertionError: 'RotaryEmbedding' not found in ['Gather', 'ConstantOfShape', 'Trilu', 'Reshape', 'Unsqueeze', 'Add', 'Equal', 'Where', 'Unsqueeze', 'Cast', 'MatMul', 'Transpose', 'Concat', 'Cos', 'Sin', 'Pow', 'ReduceMean', 'Add', 'Sqrt', 'Reciprocal', 'Mul', 'Mul', 'MatMul', 'MatMul', 'MatMul', 'Reshape', 'Transpose', 'Reshape', 'Transpose', 'Reshape', 'Transpose', 'Unsqueeze', 'Unsqueeze', 'Mul', 'Slice', 'Slice', 'Neg', 'Concat', 'Mul', 'Add', 'Mul', 'Slice', 'Slice', 'Neg', 'Concat', 'Mul', 'Add', 'Reshape', 'Transpose', 'Reshape', 'Mul', 'Mul', 'MatMul', 'Add', 'Softmax', 'MatMul', 'Transpose', 'Reshape', 'MatMul', 'Add', 'Pow', 'ReduceMean', 'Add', 'Sqrt', 'Reciprocal', 'Mul', 'Mul', 'MatMul', 'Sigmoid', 'Mul', 'MatMul', 'Mul', 'MatMul', 'Add', 'Pow', 'ReduceMean', 'Add', 'Sqrt', 'Reciprocal', 'Mul', 'Mul', 'Transpose', 'MatMul']

onnxscript.rewriter.ort_fusions.fuse_xformers_test.TestFuseXformers::test_fuse_xformers

Stack Traces | 78s run time

.../rewriter/ort_fusions/fuse_xformers_test.py:19: in test_fuse_xformers
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'const_13_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.cos_sin_cache_test.TestCosSinCacheTransform::test_cos_sin_fusion_0_smollm_test_1

Stack Traces | 79.5s run time

..../test/lib/python3.11.../site-packages/parameterized/parameterized.py:620: in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/cos_sin_cache_test.py:45: in test_cos_sin_fusion
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'const_13_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.skip_normalization_test.TestSkipNormalization::test_smollm

Stack Traces | 112s run time

.../rewriter/ort_fusions/skip_normalization_test.py:26: in test_smollm
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'const_13_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.ort_fusions.mha_test.TestMultiHeadAttention::test_phi2lm

Stack Traces | 260s run time

.../rewriter/ort_fusions/mha_test.py:106: in test_phi2lm
    self.assertEqual(
E   AssertionError: 0 != 1 : Expected exactly one MultiHeadAttention node after optimization

View the full list of 6 ❄️ flaky test(s)

onnxscript.optimizer._constant_folding_test.FoldConstantsTest::test_node_is_folded_if_specified_as_should_fold

Flake rate in main: 94.59% (Passed 17 times, Failed 297 times)

Stack Traces | 0.002s run time

onnxscript/optimizer/_constant_folding_test.py:655: in test_node_is_folded_if_specified_as_should_fold
    optimized = self._fold(
onnxscript/optimizer/_constant_folding_test.py:25: in _fold
    onnx.checker.check_model(ir.serde.serialize_model(model))
..../test/lib/python3.11.../site-packages/onnx/checker.py:179: in check_model
    C.check_model(
E   onnx.onnx_cpp2py_export.checker.ValidationError: Graph output 'z' is not an output of any node in graph.

onnxscript.rewriter.ort_fusions.gelu_test.GeluFusionTest::test_gelu_erf_fusion

Flake rate in main: 78.87% (Passed 82 times, Failed 306 times)

Stack Traces | 0.038s run time

.../rewriter/ort_fusions/gelu_test.py:77: in test_gelu_erf_fusion
    original_output = test_utils.ort_run("Original", model, input)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.../rewriter/ort_fusions/_test_utils.py:18: in ort_run
    session = onnxruntime.InferenceSession(
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input '_sqrt_two_cast' is not a graph input, initializer, or output of a previous node.

onnxscript.rewriter.rules.common._redundant_scatter_nd_test.RedundantScatterNdTest::test_redundant_scatter_nd_dynamic_indices

Flake rate in main: 94.59% (Passed 17 times, Failed 297 times)

Stack Traces | 0.023s run time

.../rules/common/_redundant_scatter_nd_test.py:52: in test_redundant_scatter_nd_dynamic_indices
    self.assertEqual(count, 1)
E   AssertionError: 0 != 1

onnxscript.rewriter.rules.fusion._gqa_test.GQAFusionTest::test_basic_gqa_fusion

Flake rate in main: 95.16% (Passed 3 times, Failed 59 times)

Stack Traces | 0.1s run time

.../rules/fusion/_gqa_test.py:81: in test_basic_gqa_fusion
    self.assertGreater(count, 0, "GQA fusion should have occurred")
E   AssertionError: 0 not greater than 0 : GQA fusion should have occurred

onnxscript.tools.transformers_models.phi_test.TestExportPhi::test_phi_export_cpu

Flake rate in main: 95.74% (Passed 16 times, Failed 360 times)

Stack Traces | 39.2s run time

onnxscript/_internal/version_utils.py:94: in call_f
    return fct(self)
           ^^^^^^^^^
.../tools/transformers_models/phi_test.py:34: in test_phi_export_cpu
    sess = onnxruntime.InferenceSession(
..../test_torch_nightly/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test_torch_nightly/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'val_3' is not a graph input, initializer, or output of a previous node.

onnxscript.tools.transformers_models.phi_test.TestExportPhi::test_phi_export_cpu_export_api

Flake rate in main: 95.74% (Passed 16 times, Failed 360 times)

Stack Traces | 16.6s run time

onnxscript/_internal/version_utils.py:94: in call_f
    return fct(self)
           ^^^^^^^^^
.../tools/transformers_models/phi_test.py:54: in test_phi_export_cpu_export_api
    sess = onnxruntime.InferenceSession(
..../test_torch_nightly/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:465: in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
..../test_torch_nightly/lib/python3.11.../onnxruntime/capi/onnxruntime_inference_collection.py:528: in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_bytes, False, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Invalid model. Node input 'val_3' is not a graph input, initializer, or output of a previous node.

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

xadupre · 2025-10-28T09:51:51Z

onnxscript/optimizer/_constant_folding.py

-        output_size_limit: int,
+        shape_inference: bool = True,
+        input_size_limit: int = DEFAULT_CONSTANT_FOLD_INPUT_SIZE_LIMIT,
+        output_size_limit: int = DEFAULT_CONSTANT_FOLD_OUTPUT_SIZE_LIMIT,


I wonder why you need two limits. Maybe we could set up the second one to None and replace None and the first value if it is still None.

justinchuby · 2025-10-28T13:58:51Z

Replaced by #2650

Partially From #2598 This provides a better optimized graph after constant folding in terms of the number of nodes, which is better for debugging.

justinchuby added 3 commits October 2, 2025 22:01

Create default values for the constant folding pass

bfb26eb

Signed-off-by: Justin Chu <[email protected]>

Support multi outputs and store replacement as initializers

b7f51ad

Signed-off-by: Justin Chu <[email protected]>

WIP create initializers

c62f45f

Signed-off-by: Justin Chu <[email protected]>

github-project-automation bot added this to ONNX Script Review Board Oct 3, 2025

github-project-automation bot moved this to Todo in ONNX Script Review Board Oct 3, 2025

justinchuby added this to the 0.5.4 milestone Oct 7, 2025

titaiwangms self-requested a review October 10, 2025 15:58

titaiwangms self-assigned this Oct 10, 2025

justinchuby removed this from the 0.5.4 milestone Oct 16, 2025

titaiwangms mentioned this pull request Oct 20, 2025

Create initializers not constant nodes in constant folding pass #2650

Merged

xadupre reviewed Oct 28, 2025

View reviewed changes

titaiwangms closed this Oct 28, 2025

github-project-automation bot moved this from Todo to Done in ONNX Script Review Board Oct 28, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Create initializers with the constant folding pass #2598

Create initializers with the constant folding pass #2598

Uh oh!

justinchuby commented Oct 3, 2025

Uh oh!

codecov bot commented Oct 3, 2025 •

edited

Loading

Uh oh!

xadupre Oct 28, 2025

Uh oh!

justinchuby commented Oct 28, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Create initializers with the constant folding pass #2598

Create initializers with the constant folding pass #2598

Uh oh!

Conversation

justinchuby commented Oct 3, 2025

Uh oh!

codecov bot commented Oct 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

❌ 31 Tests Failed:

Uh oh!

xadupre Oct 28, 2025

Choose a reason for hiding this comment

Uh oh!

justinchuby commented Oct 28, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

codecov bot commented Oct 3, 2025 •

edited

Loading