You are on page 1of 1

weight modulation

Task embedding procedure


procedure

Task z FiLM Generator


ID 1 1 1 2 2 3 3 3

weight modulation
nn.linear(dim_in, dim_out) task embedding

Transformer sub-
network

BERT emneddings
Putting it
together
Attention FiLM Generator
Different layers of Transformer is
modulated by FilM generator
Layer Normalization

You might also like