i transformed MultiHeadAttention to ONXX format, but failed when loaded by CPP.
i'm not sure, but the error seems to be caused by Param key_padding_mask.
from pytorch source code, i know Byte & Bool is supported by key_padding_mask, but both type will cause error(loaded by CPP)
the errs are shown as below:
# when i use Bool in param key_padding_mask
[8] Assertion failed: inputTensor.getType() != nvinfer1::DataType::kBOOL
# when use Byte：
Unsupported ONNX data type: UINT8 (2)
ERROR: input3:188 In function importInput:
[8] Assertion failed: convertDtype(onnxDtype.elemLtypeO, &trtDtype)