Issues when converting model to float16

[toothache]

Some issues observed when converting model to float16:

• When converting model with keep_io_types set to True, onnx.checker will complain:

  onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted,
  

  The fix would be fairly simple, inserting the Cast operator for input tensors to the beginning of node list.
• Some operators (e.g., Resize, CumSum) support float16, however it's listed in DEFAULT_OP_BLOCK_LIST:
  • Should we update this list or auto generated from onnx opschema?
  • For operator like Resize, some of the optional inputs doesn't support float16. We need exclude fp16 conversion to them.

[xiaowuhu]

@toothache does this issue still existed?

[1615070057]

Hi, I also encountered the same problem, The fix would be fairly simple, inserting the Cast operator for input tensors to the beginning of node list. How should I do it, thanks!

[1615070057]

The onnx.checker verification is successful, but the onnxruntime loading model reports an error: onnxruntime.capi.onnxruntime_pybind11_state.InvalidGraph: [ONNXRuntimeError] : 10 : INVALID_GRAPH : Load model from model_fp16.onnx failed:This is an invalid model. Type Error: Type 'tensor(float16 )' of input parameter (const_empty_float__132) of operator (Resize) in node (Resize__139) is invalid.