increase batch/buff/erbeta

config/trainer_config.yaml

     normalize: true
     max_steps: 5.0e7
     learning_rate_schedule: constant
-    batch_size: 1024
-    buffer_size: 10240
+    batch_size: 2048
+    buffer_size: 20480
+    beta: 1.0e-2
     self_play:
         window: 10
         play_against_latest_model_ratio: 0.5