Use TVM to process a model with dynamic slice such like:
x = x[:, :, y1:y2, x1:x2] # shape [N, C, H, W]
y = model(x)
when use TVM to convert layout to NHWC
desired_layouts = {
"nn.conv2d": ["NHWC", "HWIO"],
}
seq = tvm.transform.Sequential([
# PyTorch frontend assumes NCHW layout
tvm.relay.transform.ConvertLayout(desired_layouts),
])
occur error
The Relay type checker is unable to show the following types match:
Tensor[(3, 3, meta[tir.FloorDiv][0], 16), float32]
Tensor[(3, 3, 1, 16), float32]
In particular:
dimension 2 conflicts: floordiv(?, 16) does not match 1.
The Relay type checker is unable to show the following types match.
In particular `Tensor[(3, 3, 1, 16), float32]` does not match `Tensor[(3, 3, meta[tir.FloorDiv][0], 16), float32]
`
The Relay type checker is unable to show the following types match:
Tensor[(3, 3, meta[tir.FloorDiv][0], 64), float32]
Tensor[(3, 3, 1, 64), float32]
What dose meta[tir.FloorDiv][0]
mean?
Is there any suggestion to solve this error?