Closed Ezra-Yu closed 1 year ago
Hi, we do not use distillation by default, as suggested in the config file:
distillation_type: none # do not use KD by default
.
You may turn it on by using:
distillation_type: hard # should be better than soft
.
Thanks
I notice there a teacher model in your config https://github.com/JierunChen/FasterNet/blob/9a30b0e0af2afb2364c18d74469b94f527cc2de3/cfg/fasternet_l.yaml#L30