An Autonomous Parallelization of Transformer Model Inference on Heterogeneous Edge Devices | Synapse