Set num teammates to 2

ml-agents/mlagents/trainers/torch/networks.py

         )
         self.stream_names = stream_names
         self.critic = CentralizedValueNetwork(
-            stream_names, observation_shapes, network_settings, num_agents=3
+            stream_names, observation_shapes, network_settings, num_agents=2
         )
 
     @property
     ) -> Tuple[
         AgentAction, ActionLogProbs, torch.Tensor, Dict[str, torch.Tensor], torch.Tensor
     ]:
+        print(len(critic_obs))
         if self.use_lstm:
             # Use only the back half of memories for critic and actor
             actor_mem, critic_mem = torch.split(memories, self.memory_size // 2, dim=-1)