No check for types matching is expected for qnn matmul inputs due to there is no constraint for QLinearMatMul operation in ONNX documentation
Actual behavior
Quantized model from Hugging face failed during compilation due to matching of input tensor types.
Environment
Linux 20.04 LTE
Steps to reproduce
Usual step for compilation and launching of the onnx-model by VirtualMachine by python front-end
Triage
frontend:onnx
relay:qnn
Notes
There is similar issue and discuss for qnn conv2d. Unfortunately there is no any solution or minimal discussion. It looks like the problem is more general: tvm relays matches input types for qnn operations, but it does not assume by ONNX op description.
Expected behavior
No check for types matching is expected for qnn matmul inputs due to there is no constraint for QLinearMatMul operation in ONNX documentation
Actual behavior
Quantized model from Hugging face failed during compilation due to matching of input tensor types.
Environment
Linux 20.04 LTE
Steps to reproduce
Usual step for compilation and launching of the onnx-model by VirtualMachine by python front-end
Triage
Notes
There is similar issue and discuss for qnn conv2d. Unfortunately there is no any solution or minimal discussion. It looks like the problem is more general: tvm relays matches input types for qnn operations, but it does not assume by ONNX op description.
cc @KJlaccHoeUM9l @ehsanmok