I’m new to TVM.
When I converted TF model to Relay, It crashed.
The related code snippet.
input_shape = (1, 3, 32, 32)
shape_dict = {'input_1': input_shape}
irmod, params = relay.frontend.from_tensorflow(graph_def, layout='NCHW', shape=shape_dict)
The crash message is so confusing:
After a long time of debugging, I found the root cause leading to this crash is that Relay does not support 'NCHW’ format for TensorFlow now.
However, The crash message is related with tensor shape mismatch. It very confused.
The big gap between the error location and the real error location makes it difficult for users to debug. Maybe we need to provide more friendly crash message.
By the way, do we need to improve this crash message?