Input and output datatype should be INT16. FLOAT32 data is not supported currently. So, you need to quantize or dequantize the data accordingly. The shift factors can be accessed from the XMODEL using XIR API. The following is a Python example:
graph = xir.Graph.deserialize(“rnn.xmodel”)
in_pos = graph.get_root_subgraph().get_attr('input_float2fix')
out_pos = graph.get_root_subgraph().get_attr('output_fix2float')