Hi folks: I am new to TVM. I have a question. Currently, I am trying to compile a tensorflow model which has dynamic batch size. Like:
input_names = ["input_ids_1:0", "input_mask_1:0", "segment_ids_1:0"]
output_names = ["loss/Softmax:0"]
input_shapes = [[-1,256],[-1,256],[-1,256]]
I am trying to pass my graph_def into relay frontend, like:
mod, params = relay.frontend.from_tensorflow(graph_def, shape = None, layout=None, outputs=output_names)
However, I found that I got the following error:
return ufunc.reduce(obj, axis, dtype, out, **passkwargs)
TypeError: unsupported operand type(s) for *: 'Any' and 'Any'
I am not quite sure whether current tvm has supported dynamic shape? Thanks!!!