In the forward pass of my network, I encode two modalities, concatenate them, and use a decoder to produce a final output. Here is a sample code:
out_a = self.encoder_a(input_a)
out_v = self.encoder_v(input_v)
out = torch.cat((out_a, out_v), dim=-1)
pred = self.decoder(out)
I wanted to know if it was possible to easily parallelise the encoding calls, as they can me computed separately.
The computations shown here are done on a GPU.
