I can not use auto tvm to tuning tensorflow model


#1

hello
i can not use auto tvm to tuning tensorflow model . the error is

%46 = nn.conv2d(%44, %45, groups=64, channels=64, kernel_size=[3, 3])an internal invariant was violated while typechecking your program [07:38:22] /home/lisas/tvm/tvm/src/relay/pass/type_solver.cc:119: Check failed: resolved.defined(): Unable to unify parent types: TensorType([64, 0, 3, 3], float32) and TensorType([16, 1, 3, 3], float32)

best
lisa shi


#2

Did you end up resolving this? I am seeing the same issue when importing a model from ONNX.


#3

What is the input channel number of this conv2d?


#4

In my case, the input shape is (1,3,224,224)


#5

@shixinli, did you find a solution to this?