Got an error when serializing trained model

when serializing trained model, I got an error after the conv and bn layer:

File "/home/xxh/Desktop/tvm_project/crnn_mxnet/data/ocr/", line 66, in hybrid_forward
conv = mx.nd.squeeze(conv, axis=2) # b *512 * width
File "<string>", line 39, in squeeze
AssertionError: Positional arguments must have NDArray type, but got <Symbol batchnorm2_fwd>

What’s that mean? Does it not support squeeze op or something else?

I change conv = mx.nd.squeeze(conv, axis=2) to conv = conv.squeeze(axis=2).
And it works.
but got another, T, b, h = recurrent.shape, how to change this line of code?

This link says we need to avoid printing shapes in hybrid_forward.