search

LeelaChessZero / lczero-training

For code etc relating to the network training process.
147 stars 120 forks source link

Allow to choose which blocks have SE units #188

Open Tilps opened 2 years ago

Tilps commented 2 years ago

Inspired by testing which suggests a 15 block net only needs 3 SE units to get the same loss values.

Tilps commented 2 years ago

yaml options. No se_blocks section means same as before, all blocks have SE. se_blocks: [] means no SE at all.

se_blocks:
  - 1 based block numbers
  - listed here
  - to control exactly which