reduced steps cubewars

config/trainer_config.yaml

 
 SmallCubeSoldier:
     normalize: false
-    max_steps: 5.0e7
+    max_steps: 2.0e7
     learning_rate_schedule: constant
     batch_size: 2048
     buffer_size: 20480
 
 LargeCubeSoldier:
     normalize: false
-    max_steps: 5.0e7
+    max_steps: 2.0e7
     learning_rate_schedule: constant
     batch_size: 2048
     buffer_size: 20480