62c7ab518214286a4721dd7410978effd6d05471,rllib/execution/train_ops.py,TrainTFMultiGPU,__call__,#TrainTFMultiGPU#Any#,159

Before Change

                policy._debug_vars()
                tuples = policy._get_loss_inputs_dict(
                    batch, shuffle=self.shuffle_sequences)
                data_keys = [ph for _, ph in policy._loss_inputs]
                if policy._state_inputs:
                    state_keys = policy._state_inputs + [policy._seq_lens]
                else:
                    state_keys = []

After Change

                policy._debug_vars()
                tuples = policy._get_loss_inputs_dict(
                    batch, shuffle=self.shuffle_sequences)
                data_keys = list(policy._loss_input_dict_no_rnn.values())
                if policy._state_inputs:
                    state_keys = policy._state_inputs + [policy._seq_lens]
                else:
                    state_keys = []

Instances