Hey, I have a model with multiple inputs and outputs, and all the op are annotated as Supported Annotations, but after partition, it’s still split to multiple parts. Below is part of the model, both strided_slice and concatenate are annotated as supported annotations.
Below are corresponding results of the model, kendryte_0 ends with strided_slice and kendryte_13 starts with concatenate.
def @main(%i0: Tensor[(1, 3, 224, 224), float32], %i1: Tensor[(1, 3, 56, 56), float32], %i2: Tensor[(1, 4, 28, 28), float32], %i3: Tensor[(1, 4, 28, 28), float32], %i4: Tensor[(1, 8, 14, 14), float32], %i5: Tensor[(1, 8, 14, 14), float32], %i6: Tensor[(1, 8, 14, 14), float32], %i7: Tensor[(1, 12, 14, 14), float32], %i8: Tensor[(1, 12, 14, 14), float32], %i9: Tensor[(1, 20, 7, 7), float32], %i10: Tensor[(1, 20, 7, 7), float32]) → (Tensor[(1, 27), float32], Tensor[(1, 3, 56, 56), float32], Tensor[(1, 4, 28, 28), float32], Tensor[(1, 4, 28, 28), float32], Tensor[(1, 8, 14, 14), float32], Tensor[(1, 8, 14, 14), float32], Tensor[(1, 8, 14, 14), float32], Tensor[(1, 12, 14, 14), float32], Tensor[(1, 12, 14, 14), float32], Tensor[(1, 20, 7, 7), float32], Tensor[(1, 20, 7, 7), float32]) {
%106 = @kendryte_0(%i0);
%107 = %106.0;
%108 = %106.1;
%109 = (%i1, %108);
%110 = @kendryte_13(%107, %109);
Does this have something to do with the input_op(i1) of the concatenate? Thanks very much.