[Web] FP8 is not supported

[lsb]

Describe the issue

When trying to create an inference session with a minimal model that casts input from fp32 into fp8 and back again, the wasm._OrtCreateSession fails with

Can't create a session. ERROR_CODE: 10, ERROR_MESSAGE: This is an invalid model. Type Error: Type 'tensor(float8e4m3fn)' of input parameter (_to_copy) of operator (Neg) in node (node_Neg_1) is invalid.

negf8.html

To reproduce

Running the attached html file (uploaded as markdown) will attempt to create an inference session with a model that uses FP8 precision for casting.

Urgency

No response

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

v1.20.1

Execution Provider

'wasm'/'cpu' (WebAssembly CPU)

[lsb]

But this works fine in the newest Python runtime, for example

[fs-eire]

I am not sure why ONNX Runtime Python binding can run this model.

According to https://github.com/onnx/onnx/blob/main/docs/Operators.md#Neg, the operator Neg does not support type Float8E4M3FN, which is reported in the error message.

[fs-eire]

I tried the same model in python binding and it says the same error:

onnxruntime.capi.onnxruntime_pybind11_state.InvalidGraph: [ONNXRuntimeError] : 10 : INVALID_GRAPH : Load model from c:/code/o2/js/web/a.onnx failed:This is an invalid model. Type Error: Type 'tensor(float8e4m3fn)' of input parameter (_to_copy) of operator (Neg) in node (node_Neg_1) is invalid.