_

config/sac/Walker.yaml

       learning_rate: 0.0003
       learning_rate_schedule: constant
       batch_size: 1024
-      buffer_size: 200000 #2000000
+      buffer_size: 2000000
       buffer_init_steps: 0
       tau: 0.005
       steps_per_update: 30.0
     network_settings:
       normalize: true
-      hidden_units: 256
-      num_layers: 3
+      hidden_units: 512
+      num_layers: 4
       vis_encode_type: simple
       goal_conditioning_type: none
     reward_signals:
     keep_checkpoints: 5
-    max_steps: 15000000
+    max_steps: 150000000
     time_horizon: 1000
     summary_freq: 30000

ml-agents/mlagents/trainers/sac/optimizer_torch.py

             with torch.no_grad():
                 cont_log_probs = log_probs.continuous_tensor
                 target_current_diff = torch.sum(
-                    cont_log_probs, dim=1) + self.target_entropy.continuous
+                    cont_log_probs, dim=1) + 10 * self.target_entropy.continuous
                 # print(self.target_entropy.continuous, cont_log_probs, torch.sum(
                 #     cont_log_probs, dim=1) + self.target_entropy.continuous)
             # We update all the _cont_ent_coef as one block