I have a GNN(Graphic neural network) Model to deploy into TDA4VM, the core feature of a gnn model is that the shapes of every op's inputs and output smay change every time during inference.For example, the input shapes of a Gemm Op in a traditional CNN model are alway MxK, and KxN. However , in a gnn model, the M may change every time during inference。
Does TIDL support dynamic shape Inference?
