Describe the issue

I have simple einsum equation 'ijk->kj' with input shape [2,3,4] and output shape as [4,2]. However, onnx runtime can't get it computed with error ('onnxruntime::common::Status onnxruntime::EinsumOp::DeviceHelpers::CpuDeviceHelpers::DataCopy(const onnxruntime::Tensor&, onnxruntime::Tensor&, void*) output.SizeInBytes() == input.SizeInBytes() was false. Einsum op: The candidate output does not match the actual output's shape).

Looking through code, I would think:
EinsumTypedComputeProcessor::FinalizeOutput(const Tensor& candidate_output,
const gsl::span& ordered_subscript_indices_in_candidate)
would implicitly require candidate_output has the same rank as ordered_subscript_indices_in_candidate. But in this case, rank of candidate_output would still be 3 with shape {2,1,4} while ordered_subscript_indices_in_candidate would be 2 with shape{0, 2} , this will generate candidate_output_shape_without_reduced_dims as [2,1], which is incorrect and should be [2,4].

To reproduce

(py11) sjt@sjt:~/piano1$ cat test.py
from onnx import helper,checker
from onnx import TensorProto
import numpy as np
import onnxruntime as rt
import onnx

def generate_onnx(node_name, op_set):
Eqn = "ijk->ki"
test_node = helper.make_node(
node_name,
inputs=["X"],
outputs=["Y"],
equation = Eqn
)

inputs = [
    helper.make_tensor_value_info("X", TensorProto.DOUBLE, [2, 3, 4]),
]

outputs = [helper.make_tensor_value_info("Y", TensorProto.DOUBLE, [4, 2])]

graph = helper.make_graph(
    [test_node],
    "test",
    inputs,
    outputs,
)

model = helper.make_model(graph)
model.opset_import[0].version = op_set
checker.check_model(model)
onnx.save(model, node_name + ".onnx")
print(f"model is generated: {node_name}.onnx")

generate_onnx('Einsum', op_set=12)

input = np.random.randn(2,3,4)
sess = rt.InferenceSession("./Einsum.onnx")
output = sess.run(None, {'X': input})

(py11) sjt@sjt:~/piano1$ python test.py
model is generated: Einsum.onnx
2023-11-30 16:28:27.019210785 [E:onnxruntime:, sequential_executor.cc:514 ExecuteKernel] Non-zero status code returned while running Einsum node. Name:'' Status Message: /home/conda/feedstock_root/build_artifacts/onnxruntime_1697223480578/work/onnxruntime/core/providers/cpu/math/einsum_utils/einsum_auxiliary_ops.cc:18 onnxruntime::common::Status onnxruntime::EinsumOp::DeviceHelpers::CpuDeviceHelpers::DataCopy(const onnxruntime::Tensor&, onnxruntime::Tensor&, void*) output.SizeInBytes() == input.SizeInBytes() was false. Einsum op: The candidate output does not match the actual output's shape

Traceback (most recent call last):
File "/home/sjt/piano1/test.py", line 41, in
output = sess.run(None, {'X': input})
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/sjt/anaconda3/envs/py11/lib/python3.11/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 220, in run
return self._sess.run(output_names, input_feed, run_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running Einsum node. Name:'' Status Message: /home/conda/feedstock_root/build_artifacts/onnxruntime_1697223480578/work/onnxruntime/core/providers/cpu/math/einsum_utils/einsum_auxiliary_ops.cc:18 onnxruntime::common::Status onnxruntime::EinsumOp::DeviceHelpers::CpuDeviceHelpers::DataCopy(const onnxruntime::Tensor&, onnxruntime::Tensor&, void*) output.SizeInBytes() == input.SizeInBytes() was false. Einsum op: The candidate output does not match the actual output's shape

Urgency

No response

Platform

Linux

OS Version

Ubuntu 20.04.6 LTS

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.16.1

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response