In Sparse Conv2d Implementation for 3x3 kernels, I read the sample usages and try to use it on my code, but some errors occurs.
I try to use the function
relay.transform._ffi_api.Conv2dToSparse2 in the sample usages to convert conv2d to sparse_conv2d, but the error message says that:
AttributeError: module 'tvm.relay.transform._ffi_api' has no attribute 'Conv2dToSparse2'
Then I trace to _ffi_api.py and found that there are nothing inside.
and I found that the function
Conv2dToSparse2 lies in transform.py, so I replace it as
relay.transform.Conv2dToSparse2, but another error occurs:
TypeError: Conv2dToSparse2() takes 4 positional arguments but 5 were given
I have no idea why the function here just take 4 arguments but the code in sample usages give it 5 arguments.
I wonder how to use
Conv2dToSparse2 correctly to do the conversion.