Question about training the transformer T

[1ho0jin1]

First of all, thank you for sharing your great work :)
I tried running the codes with the default setting on MiniImageNet.
I found out that although the encoders are pre-trained models (downloaded from FEAT repository), the transformer (self attention module) is just a randomly initialized one.
So my question is:

1. Can you kindly share the pre-trained weights for the transformer? (SNaTCHerF will do.)
2. When we train the transformer, do we:
   1. fix the encoder
   2. train only the transformer using eq.12 ?

Thank you :)