The output of a RNN layer is a 3-dim tensor, but the input of the Conv1D layer is a 4-dim tensor. This cause an error at inference if I put a Conv1D layer after RNN layer. Honestly, I use the Conv1D layer with kernel size 1 instead of a Dense layer as the output layer, so per channel quantization can be applied. Are there any solutions to fix this?
The output of a RNN layer is a 3-dim tensor, but the input of the Conv1D layer is a 4-dim tensor. This cause an error at inference if I put a Conv1D layer after RNN layer. Honestly, I use the Conv1D layer with kernel size 1 instead of a Dense layer as the output layer, so per channel quantization can be applied. Are there any solutions to fix this?