Change dimensions of recurrent in to BxN for ONNX

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

         # cause problem to barracuda import.
         self.policy = policy
         batch_dim = [1]
-        seq_len_dim = [1]
         dummy_vec_obs = [torch.zeros(batch_dim + [self.policy.vec_obs_size])]
         # create input shape of NCHW
         # (It's NHWC in self.policy.behavior_spec.observation_shapes)
             if len(shape) == 3
         ]
         dummy_masks = torch.ones(batch_dim + [sum(self.policy.actor_critic.act_size)])
-        dummy_memories = torch.zeros(
-            batch_dim + seq_len_dim + [self.policy.export_memory_size]
-        )
+        # Assume sequence length is 1
+        dummy_memories = torch.zeros(batch_dim + [self.policy.export_memory_size])
 
         self.dummy_input = (dummy_vec_obs, dummy_vis_obs, dummy_masks, dummy_memories)
 

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

         """
         Note: This forward() method is required for exporting to ONNX. Don't modify the inputs and outputs.
         """
+        # Barracuda prefers memories to be in format (batch, enc_size) when seq_len is 1
+        if memories is not None:
+            memories = memories.unsqueeze(1)
         dists, _ = self.get_dists(vec_inputs, vis_inputs, masks, memories, 1)
         if self.act_type == ActionType.CONTINUOUS:
             action_list = self.sample_action(dists)