I’m looking to perform fusions on a model’s torch.fx graph, but don’t know exactly what makes nodes in the forward pass trace “fusable”. I see that some do it for conv->BatchNorm and linear->BatchNorm, and even with weights and biases of layers. I was thinking to do so with Linear->LayerNorm and then try with other combinations of nodes, but am not sure about which are the conditions for 2 nodes to be fused.