Why is `ortvalue_from_numpy_with_onnx_type` needed to support bfloat16

[justinchuby]

Describe the issue

When I was using ortvalue_from_numpy_with_onnx_type, it occurred to me that the input dtypes are already known by the inference session which could simply reinterpret cast any 16-bit input (like uint16) into the desired onnx type. I wonder why we need to specify the types again in ortvalue_from_numpy_with_onnx_type?

To reproduce

N/A

Urgency

No response

Platform

Linux

OS Version

N/A

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.21

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response