4x model

experiments/custom-ppg.yaml

@@ -45,9 +45,9 @@ procgen-ppo:
         no_done_at_end: False
         
         # Custom switches
-        skips: 2
+        skips: 0
         n_pi: 16
-        num_retunes: 14
+        num_retunes: 16
         retune_epochs: 6
         standardize_rewards: True
         aux_mbsize: 4
@@ -88,8 +88,10 @@ procgen-ppo:
         model:
             custom_model: impala_torch_ppg
             custom_model_config:
-                depths: [32, 64, 64]
-                nlatents: 512
+#                 depths: [32, 64, 64]
+#                 nlatents: 512
+                depths: [64, 128, 128]
+                nlatents: 1024
                 init_normed: True
                 use_layernorm: False
                 diff_framestack: True