quantize_static fail with HistogramCalibrater.

[guotuofeng]

Describe the issue

In quantize_static, the TensorsData is created by using

                if len(v) == 4:
                    self.data[k] = TensorData(lowest=v[0], highest=v[1], histogram=v[2], bins=v[3])

However, in TensorData.__init__, the following logic is check the keyword args.

class TensorData:
    _allowed = frozenset(["avg", "std", "lowest", "highest", "hist", "hist_edges"])

    def __init__(self, **kwargs):
        for k, v in kwargs.items():
            if k not in TensorData._allowed:
                raise ValueError(f"Unexpected value {k!r} not in {TensorData._allowed}.")
            setattr(self, k, v)

    @property
    def range_value(self):
        if not hasattr(self, "lowest") or not hasattr(self, "highest"):
            raise AttributeError(f"Attributes 'lowest' and/or 'highest' missing in {dir(self)}.")
        return (self.lowest, self.highest)

    @property
    def avg_std(self):
        if not hasattr(self, "avg") or not hasattr(self, "std"):
            raise AttributeError(f"Attributes 'avg' and/or 'std' missing in {dir(self)}.")
        return (self.avg, self.std)

We can see the histogram key is not in the allowed keyword, therefore the following errors are raised.

Traceback (most recent call last):
  File "/mnt/vss/_work/1/s/***/passes/onnx/quantization.py", line 408, in _run_for_config
    quantize_static(
  File "/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/onnxruntime/quantization/quantize.py", line 436, in quantize_static
    tensors_range = calibrator.compute_data()
  File "/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/onnxruntime/quantization/calibrate.py", line 504, in compute_data
    return TensorsData(cal, self.collector.compute_collection_result())
  File "/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/onnxruntime/quantization/calibrate.py", line 58, in __init__
    self.data[k] = TensorData(lowest=v[0], highest=v[1], histogram=v[2], bins=v[3])
  File "/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/onnxruntime/quantization/calibrate.py", line 30, in __init__
    raise ValueError(f"Unexpected value {k!r} not in {TensorData._allowed}.")
ValueError: Unexpected value 'histogram' not in frozenset({'hist', 'lowest', 'hist_edges', 'avg', 'std', 'highest'}).

To reproduce

See https://aiinfra.visualstudio.com/PublicPackages/_build/results?buildId=357286&view=logs&jobId=35ea891e-7d20-5395-2445-be2fa0298764&j=5f33dc1f-8f24-5b97-62d3-f49845ddf6c9&t=d902c85d-ad1e-5c9a-ae64-6b490d39ff57.

Urgency

This bug break the onnxruntime quantization. it is a blocker.

Platform

Windows

OS Version

10

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.6.0

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response